1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/frame.inline.hpp"
37 #include "runtime/safepointMechanism.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "runtime/synchronizer.hpp"
41 #include "utilities/macros.hpp"
42
43 #define __ _masm->
44
45 // Global Register Names
46 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
47 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
48
49 // Platform-dependent initialization
50 void TemplateTable::pd_initialize() {
51 // No x86 specific initialization
52 }
53
54 // Address Computation: local variables
55 static inline Address iaddress(int n) {
56 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
57 }
58
59 static inline Address laddress(int n) {
60 return iaddress(n + 1);
61 }
62
63 #ifndef _LP64
64 static inline Address haddress(int n) {
65 return iaddress(n + 0);
66 }
67 #endif
68
69 static inline Address faddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address daddress(int n) {
74 return laddress(n);
75 }
76
77 static inline Address aaddress(int n) {
78 return iaddress(n);
79 }
80
81 static inline Address iaddress(Register r) {
82 return Address(rlocals, r, Address::times_ptr);
83 }
84
85 static inline Address laddress(Register r) {
86 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
87 }
88
89 #ifndef _LP64
90 static inline Address haddress(Register r) {
91 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
92 }
93 #endif
94
95 static inline Address faddress(Register r) {
96 return iaddress(r);
97 }
98
99 static inline Address daddress(Register r) {
100 return laddress(r);
101 }
102
103 static inline Address aaddress(Register r) {
104 return iaddress(r);
105 }
106
107
108 // expression stack
109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
110 // data beyond the rsp which is potentially unsafe in an MT environment;
111 // an interrupt may overwrite that data.)
112 static inline Address at_rsp () {
113 return Address(rsp, 0);
114 }
115
116 // At top of Java expression stack which may be different than esp(). It
117 // isn't for category 1 objects.
118 static inline Address at_tos () {
119 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
120 }
121
122 static inline Address at_tos_p1() {
123 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
124 }
125
126 static inline Address at_tos_p2() {
127 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
128 }
129
130 // Condition conversion
131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
132 switch (cc) {
133 case TemplateTable::equal : return Assembler::notEqual;
134 case TemplateTable::not_equal : return Assembler::equal;
135 case TemplateTable::less : return Assembler::greaterEqual;
136 case TemplateTable::less_equal : return Assembler::greater;
137 case TemplateTable::greater : return Assembler::lessEqual;
138 case TemplateTable::greater_equal: return Assembler::less;
139 }
140 ShouldNotReachHere();
141 return Assembler::zero;
142 }
143
144
145
146 // Miscelaneous helper routines
147 // Store an oop (or NULL) at the address described by obj.
148 // If val == noreg this means store a NULL
149
150
151 static void do_oop_store(InterpreterMacroAssembler* _masm,
152 Address dst,
153 Register val,
154 DecoratorSet decorators = 0) {
155 assert(val == noreg || val == rax, "parameter is just for looks");
156 __ store_heap_oop(dst, val, rdx, rbx, decorators);
157 }
158
159 static void do_oop_load(InterpreterMacroAssembler* _masm,
160 Address src,
161 Register dst,
162 DecoratorSet decorators = 0) {
163 __ load_heap_oop(dst, src, rdx, rbx, decorators);
164 }
165
166 Address TemplateTable::at_bcp(int offset) {
167 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
168 return Address(rbcp, offset);
169 }
170
171
172 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
173 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
174 int byte_no) {
175 if (!RewriteBytecodes) return;
176 Label L_patch_done;
177
178 switch (bc) {
179 case Bytecodes::_fast_aputfield:
180 case Bytecodes::_fast_bputfield:
181 case Bytecodes::_fast_zputfield:
182 case Bytecodes::_fast_cputfield:
183 case Bytecodes::_fast_dputfield:
184 case Bytecodes::_fast_fputfield:
185 case Bytecodes::_fast_iputfield:
186 case Bytecodes::_fast_lputfield:
187 case Bytecodes::_fast_sputfield:
188 {
189 // We skip bytecode quickening for putfield instructions when
190 // the put_code written to the constant pool cache is zero.
191 // This is required so that every execution of this instruction
192 // calls out to InterpreterRuntime::resolve_get_put to do
193 // additional, required work.
194 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
195 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
196 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
197 __ movl(bc_reg, bc);
198 __ cmpl(temp_reg, (int) 0);
199 __ jcc(Assembler::zero, L_patch_done); // don't patch
200 }
201 break;
202 default:
203 assert(byte_no == -1, "sanity");
204 // the pair bytecodes have already done the load.
205 if (load_bc_into_bc_reg) {
206 __ movl(bc_reg, bc);
207 }
208 }
209
210 if (JvmtiExport::can_post_breakpoint()) {
211 Label L_fast_patch;
212 // if a breakpoint is present we can't rewrite the stream directly
213 __ movzbl(temp_reg, at_bcp(0));
214 __ cmpl(temp_reg, Bytecodes::_breakpoint);
215 __ jcc(Assembler::notEqual, L_fast_patch);
216 __ get_method(temp_reg);
217 // Let breakpoint table handling rewrite to quicker bytecode
218 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
219 #ifndef ASSERT
220 __ jmpb(L_patch_done);
221 #else
222 __ jmp(L_patch_done);
223 #endif
224 __ bind(L_fast_patch);
225 }
226
227 #ifdef ASSERT
228 Label L_okay;
229 __ load_unsigned_byte(temp_reg, at_bcp(0));
230 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
231 __ jcc(Assembler::equal, L_okay);
232 __ cmpl(temp_reg, bc_reg);
233 __ jcc(Assembler::equal, L_okay);
234 __ stop("patching the wrong bytecode");
235 __ bind(L_okay);
236 #endif
237
238 // patch bytecode
239 __ movb(at_bcp(0), bc_reg);
240 __ bind(L_patch_done);
241 }
242 // Individual instructions
243
244
245 void TemplateTable::nop() {
246 transition(vtos, vtos);
247 // nothing to do
248 }
249
250 void TemplateTable::shouldnotreachhere() {
251 transition(vtos, vtos);
252 __ stop("shouldnotreachhere bytecode");
253 }
254
255 void TemplateTable::aconst_null() {
256 transition(vtos, atos);
257 __ xorl(rax, rax);
258 }
259
260 void TemplateTable::iconst(int value) {
261 transition(vtos, itos);
262 if (value == 0) {
263 __ xorl(rax, rax);
264 } else {
265 __ movl(rax, value);
266 }
267 }
268
269 void TemplateTable::lconst(int value) {
270 transition(vtos, ltos);
271 if (value == 0) {
272 __ xorl(rax, rax);
273 } else {
274 __ movl(rax, value);
275 }
276 #ifndef _LP64
277 assert(value >= 0, "check this code");
278 __ xorptr(rdx, rdx);
279 #endif
280 }
281
282
283
284 void TemplateTable::fconst(int value) {
285 transition(vtos, ftos);
286 if (UseSSE >= 1) {
287 static float one = 1.0f, two = 2.0f;
288 switch (value) {
289 case 0:
290 __ xorps(xmm0, xmm0);
291 break;
292 case 1:
293 __ movflt(xmm0, ExternalAddress((address) &one));
294 break;
295 case 2:
296 __ movflt(xmm0, ExternalAddress((address) &two));
297 break;
298 default:
299 ShouldNotReachHere();
300 break;
301 }
302 } else {
303 #ifdef _LP64
304 ShouldNotReachHere();
305 #else
306 if (value == 0) { __ fldz();
307 } else if (value == 1) { __ fld1();
308 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
309 } else { ShouldNotReachHere();
310 }
311 #endif // _LP64
312 }
313 }
314
315 void TemplateTable::dconst(int value) {
316 transition(vtos, dtos);
317 if (UseSSE >= 2) {
318 static double one = 1.0;
319 switch (value) {
320 case 0:
321 __ xorpd(xmm0, xmm0);
322 break;
323 case 1:
324 __ movdbl(xmm0, ExternalAddress((address) &one));
325 break;
326 default:
327 ShouldNotReachHere();
328 break;
329 }
330 } else {
331 #ifdef _LP64
332 ShouldNotReachHere();
333 #else
334 if (value == 0) { __ fldz();
335 } else if (value == 1) { __ fld1();
336 } else { ShouldNotReachHere();
337 }
338 #endif
339 }
340 }
341
342 void TemplateTable::bipush() {
343 transition(vtos, itos);
344 __ load_signed_byte(rax, at_bcp(1));
345 }
346
347 void TemplateTable::sipush() {
348 transition(vtos, itos);
349 __ load_unsigned_short(rax, at_bcp(1));
350 __ bswapl(rax);
351 __ sarl(rax, 16);
352 }
353
354 void TemplateTable::ldc(bool wide) {
355 transition(vtos, vtos);
356 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
357 Label call_ldc, notFloat, notClass, notInt, Done;
358
359 if (wide) {
360 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
361 } else {
362 __ load_unsigned_byte(rbx, at_bcp(1));
363 }
364
365 __ get_cpool_and_tags(rcx, rax);
366 const int base_offset = ConstantPool::header_size() * wordSize;
367 const int tags_offset = Array<u1>::base_offset_in_bytes();
368
369 // get type
370 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
371
372 // unresolved class - get the resolved class
373 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
374 __ jccb(Assembler::equal, call_ldc);
375
376 // unresolved class in error state - call into runtime to throw the error
377 // from the first resolution attempt
378 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
379 __ jccb(Assembler::equal, call_ldc);
380
381 // resolved class - need to call vm to get java mirror of the class
382 __ cmpl(rdx, JVM_CONSTANT_Class);
383 __ jcc(Assembler::notEqual, notClass);
384
385 __ bind(call_ldc);
386
387 __ movl(rarg, wide);
388 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
389
390 __ push(atos);
391 __ jmp(Done);
392
393 __ bind(notClass);
394 __ cmpl(rdx, JVM_CONSTANT_Float);
395 __ jccb(Assembler::notEqual, notFloat);
396
397 // ftos
398 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
399 __ push(ftos);
400 __ jmp(Done);
401
402 __ bind(notFloat);
403 __ cmpl(rdx, JVM_CONSTANT_Integer);
404 __ jccb(Assembler::notEqual, notInt);
405
406 // itos
407 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
408 __ push(itos);
409 __ jmp(Done);
410
411 // assume the tag is for condy; if not, the VM runtime will tell us
412 __ bind(notInt);
413 condy_helper(Done);
414
415 __ bind(Done);
416 }
417
418 // Fast path for caching oop constants.
419 void TemplateTable::fast_aldc(bool wide) {
420 transition(vtos, atos);
421
422 Register result = rax;
423 Register tmp = rdx;
424 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
425 int index_size = wide ? sizeof(u2) : sizeof(u1);
426
427 Label resolved;
428
429 // We are resolved if the resolved reference cache entry contains a
430 // non-null object (String, MethodType, etc.)
431 assert_different_registers(result, tmp);
432 __ get_cache_index_at_bcp(tmp, 1, index_size);
433 __ load_resolved_reference_at_index(result, tmp);
434 __ testptr(result, result);
435 __ jcc(Assembler::notZero, resolved);
436
437 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
438
439 // first time invocation - must resolve first
440 __ movl(rarg, (int)bytecode());
441 __ call_VM(result, entry, rarg);
442 __ bind(resolved);
443
444 { // Check for the null sentinel.
445 // If we just called the VM, that already did the mapping for us,
446 // but it's harmless to retry.
447 Label notNull;
448 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
449 __ movptr(tmp, null_sentinel);
450 __ cmpptr(tmp, result);
451 __ jccb(Assembler::notEqual, notNull);
452 __ xorptr(result, result); // NULL object reference
453 __ bind(notNull);
454 }
455
456 if (VerifyOops) {
457 __ verify_oop(result);
458 }
459 }
460
461 void TemplateTable::ldc2_w() {
462 transition(vtos, vtos);
463 Label notDouble, notLong, Done;
464 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
465
466 __ get_cpool_and_tags(rcx, rax);
467 const int base_offset = ConstantPool::header_size() * wordSize;
468 const int tags_offset = Array<u1>::base_offset_in_bytes();
469
470 // get type
471 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
472 __ cmpl(rdx, JVM_CONSTANT_Double);
473 __ jccb(Assembler::notEqual, notDouble);
474
475 // dtos
476 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
477 __ push(dtos);
478
479 __ jmp(Done);
480 __ bind(notDouble);
481 __ cmpl(rdx, JVM_CONSTANT_Long);
482 __ jccb(Assembler::notEqual, notLong);
483
484 // ltos
485 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
486 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
487 __ push(ltos);
488 __ jmp(Done);
489
490 __ bind(notLong);
491 condy_helper(Done);
492
493 __ bind(Done);
494 }
495
496 void TemplateTable::condy_helper(Label& Done) {
497 const Register obj = rax;
498 const Register off = rbx;
499 const Register flags = rcx;
500 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
501 __ movl(rarg, (int)bytecode());
502 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
503 #ifndef _LP64
504 // borrow rdi from locals
505 __ get_thread(rdi);
506 __ get_vm_result_2(flags, rdi);
507 __ restore_locals();
508 #else
509 __ get_vm_result_2(flags, r15_thread);
510 #endif
511 // VMr = obj = base address to find primitive value to push
512 // VMr2 = flags = (tos, off) using format of CPCE::_flags
513 __ movl(off, flags);
514 __ andl(off, ConstantPoolCacheEntry::field_index_mask);
515 const Address field(obj, off, Address::times_1, 0*wordSize);
516
517 // What sort of thing are we loading?
518 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
519 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
520
521 switch (bytecode()) {
522 case Bytecodes::_ldc:
523 case Bytecodes::_ldc_w:
524 {
525 // tos in (itos, ftos, stos, btos, ctos, ztos)
526 Label notInt, notFloat, notShort, notByte, notChar, notBool;
527 __ cmpl(flags, itos);
528 __ jcc(Assembler::notEqual, notInt);
529 // itos
530 __ movl(rax, field);
531 __ push(itos);
532 __ jmp(Done);
533
534 __ bind(notInt);
535 __ cmpl(flags, ftos);
536 __ jcc(Assembler::notEqual, notFloat);
537 // ftos
538 __ load_float(field);
539 __ push(ftos);
540 __ jmp(Done);
541
542 __ bind(notFloat);
543 __ cmpl(flags, stos);
544 __ jcc(Assembler::notEqual, notShort);
545 // stos
546 __ load_signed_short(rax, field);
547 __ push(stos);
548 __ jmp(Done);
549
550 __ bind(notShort);
551 __ cmpl(flags, btos);
552 __ jcc(Assembler::notEqual, notByte);
553 // btos
554 __ load_signed_byte(rax, field);
555 __ push(btos);
556 __ jmp(Done);
557
558 __ bind(notByte);
559 __ cmpl(flags, ctos);
560 __ jcc(Assembler::notEqual, notChar);
561 // ctos
562 __ load_unsigned_short(rax, field);
563 __ push(ctos);
564 __ jmp(Done);
565
566 __ bind(notChar);
567 __ cmpl(flags, ztos);
568 __ jcc(Assembler::notEqual, notBool);
569 // ztos
570 __ load_signed_byte(rax, field);
571 __ push(ztos);
572 __ jmp(Done);
573
574 __ bind(notBool);
575 break;
576 }
577
578 case Bytecodes::_ldc2_w:
579 {
580 Label notLong, notDouble;
581 __ cmpl(flags, ltos);
582 __ jcc(Assembler::notEqual, notLong);
583 // ltos
584 __ movptr(rax, field);
585 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
586 __ push(ltos);
587 __ jmp(Done);
588
589 __ bind(notLong);
590 __ cmpl(flags, dtos);
591 __ jcc(Assembler::notEqual, notDouble);
592 // dtos
593 __ load_double(field);
594 __ push(dtos);
595 __ jmp(Done);
596
597 __ bind(notDouble);
598 break;
599 }
600
601 default:
602 ShouldNotReachHere();
603 }
604
605 __ stop("bad ldc/condy");
606 }
607
608 void TemplateTable::locals_index(Register reg, int offset) {
609 __ load_unsigned_byte(reg, at_bcp(offset));
610 __ negptr(reg);
611 }
612
613 void TemplateTable::iload() {
614 iload_internal();
615 }
616
617 void TemplateTable::nofast_iload() {
618 iload_internal(may_not_rewrite);
619 }
620
621 void TemplateTable::iload_internal(RewriteControl rc) {
622 transition(vtos, itos);
623 if (RewriteFrequentPairs && rc == may_rewrite) {
624 Label rewrite, done;
625 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
626 LP64_ONLY(assert(rbx != bc, "register damaged"));
627
628 // get next byte
629 __ load_unsigned_byte(rbx,
630 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
631 // if _iload, wait to rewrite to iload2. We only want to rewrite the
632 // last two iloads in a pair. Comparing against fast_iload means that
633 // the next bytecode is neither an iload or a caload, and therefore
634 // an iload pair.
635 __ cmpl(rbx, Bytecodes::_iload);
636 __ jcc(Assembler::equal, done);
637
638 __ cmpl(rbx, Bytecodes::_fast_iload);
639 __ movl(bc, Bytecodes::_fast_iload2);
640
641 __ jccb(Assembler::equal, rewrite);
642
643 // if _caload, rewrite to fast_icaload
644 __ cmpl(rbx, Bytecodes::_caload);
645 __ movl(bc, Bytecodes::_fast_icaload);
646 __ jccb(Assembler::equal, rewrite);
647
648 // rewrite so iload doesn't check again.
649 __ movl(bc, Bytecodes::_fast_iload);
650
651 // rewrite
652 // bc: fast bytecode
653 __ bind(rewrite);
654 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
655 __ bind(done);
656 }
657
658 // Get the local value into tos
659 locals_index(rbx);
660 __ movl(rax, iaddress(rbx));
661 }
662
663 void TemplateTable::fast_iload2() {
664 transition(vtos, itos);
665 locals_index(rbx);
666 __ movl(rax, iaddress(rbx));
667 __ push(itos);
668 locals_index(rbx, 3);
669 __ movl(rax, iaddress(rbx));
670 }
671
672 void TemplateTable::fast_iload() {
673 transition(vtos, itos);
674 locals_index(rbx);
675 __ movl(rax, iaddress(rbx));
676 }
677
678 void TemplateTable::lload() {
679 transition(vtos, ltos);
680 locals_index(rbx);
681 __ movptr(rax, laddress(rbx));
682 NOT_LP64(__ movl(rdx, haddress(rbx)));
683 }
684
685 void TemplateTable::fload() {
686 transition(vtos, ftos);
687 locals_index(rbx);
688 __ load_float(faddress(rbx));
689 }
690
691 void TemplateTable::dload() {
692 transition(vtos, dtos);
693 locals_index(rbx);
694 __ load_double(daddress(rbx));
695 }
696
697 void TemplateTable::aload() {
698 transition(vtos, atos);
699 locals_index(rbx);
700 __ movptr(rax, aaddress(rbx));
701 }
702
703 void TemplateTable::locals_index_wide(Register reg) {
704 __ load_unsigned_short(reg, at_bcp(2));
705 __ bswapl(reg);
706 __ shrl(reg, 16);
707 __ negptr(reg);
708 }
709
710 void TemplateTable::wide_iload() {
711 transition(vtos, itos);
712 locals_index_wide(rbx);
713 __ movl(rax, iaddress(rbx));
714 }
715
716 void TemplateTable::wide_lload() {
717 transition(vtos, ltos);
718 locals_index_wide(rbx);
719 __ movptr(rax, laddress(rbx));
720 NOT_LP64(__ movl(rdx, haddress(rbx)));
721 }
722
723 void TemplateTable::wide_fload() {
724 transition(vtos, ftos);
725 locals_index_wide(rbx);
726 __ load_float(faddress(rbx));
727 }
728
729 void TemplateTable::wide_dload() {
730 transition(vtos, dtos);
731 locals_index_wide(rbx);
732 __ load_double(daddress(rbx));
733 }
734
735 void TemplateTable::wide_aload() {
736 transition(vtos, atos);
737 locals_index_wide(rbx);
738 __ movptr(rax, aaddress(rbx));
739 }
740
741 void TemplateTable::index_check(Register array, Register index) {
742 // Pop ptr into array
743 __ pop_ptr(array);
744 index_check_without_pop(array, index);
745 }
746
747 void TemplateTable::index_check_without_pop(Register array, Register index) {
748 // destroys rbx
749 // check array
750 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
751 // sign extend index for use by indexed load
752 __ movl2ptr(index, index);
753 // check index
754 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
755 if (index != rbx) {
756 // ??? convention: move aberrant index into rbx for exception message
757 assert(rbx != array, "different registers");
758 __ movl(rbx, index);
759 }
760 Label skip;
761 __ jccb(Assembler::below, skip);
762 // Pass array to create more detailed exceptions.
763 #ifdef IA32
764 __ mov(rax, array);
765 #else // AMD64
766 __ mov(c_rarg1, array);
767 #endif
768 __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
769 __ bind(skip);
770 }
771
772
773 void TemplateTable::iaload() {
774 transition(itos, itos);
775 // rax: index
776 // rdx: array
777 index_check(rdx, rax); // kills rbx
778 __ movl(rax, Address(rdx, rax,
779 Address::times_4,
780 arrayOopDesc::base_offset_in_bytes(T_INT)));
781 }
782
783 void TemplateTable::laload() {
784 transition(itos, ltos);
785 // rax: index
786 // rdx: array
787 index_check(rdx, rax); // kills rbx
788 NOT_LP64(__ mov(rbx, rax));
789 // rbx,: index
790 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
791 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
792 }
793
794
795
796 void TemplateTable::faload() {
797 transition(itos, ftos);
798 // rax: index
799 // rdx: array
800 index_check(rdx, rax); // kills rbx
801 __ load_float(Address(rdx, rax,
802 Address::times_4,
803 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
804 }
805
806 void TemplateTable::daload() {
807 transition(itos, dtos);
808 // rax: index
809 // rdx: array
810 index_check(rdx, rax); // kills rbx
811 __ load_double(Address(rdx, rax,
812 Address::times_8,
813 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
814 }
815
816 void TemplateTable::aaload() {
817 transition(itos, atos);
818 // rax: index
819 // rdx: array
820 index_check(rdx, rax); // kills rbx
821 do_oop_load(_masm,
822 Address(rdx, rax,
823 UseCompressedOops ? Address::times_4 : Address::times_ptr,
824 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
825 rax,
826 IN_HEAP_ARRAY);
827 }
828
829 void TemplateTable::baload() {
830 transition(itos, itos);
831 // rax: index
832 // rdx: array
833 index_check(rdx, rax); // kills rbx
834 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
835 }
836
837 void TemplateTable::caload() {
838 transition(itos, itos);
839 // rax: index
840 // rdx: array
841 index_check(rdx, rax); // kills rbx
842 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
843 }
844
845 // iload followed by caload frequent pair
846 void TemplateTable::fast_icaload() {
847 transition(vtos, itos);
848 // load index out of locals
849 locals_index(rbx);
850 __ movl(rax, iaddress(rbx));
851
852 // rax: index
853 // rdx: array
854 index_check(rdx, rax); // kills rbx
855 __ load_unsigned_short(rax,
856 Address(rdx, rax,
857 Address::times_2,
858 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
859 }
860
861
862 void TemplateTable::saload() {
863 transition(itos, itos);
864 // rax: index
865 // rdx: array
866 index_check(rdx, rax); // kills rbx
867 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
868 }
869
870 void TemplateTable::iload(int n) {
871 transition(vtos, itos);
872 __ movl(rax, iaddress(n));
873 }
874
875 void TemplateTable::lload(int n) {
876 transition(vtos, ltos);
877 __ movptr(rax, laddress(n));
878 NOT_LP64(__ movptr(rdx, haddress(n)));
879 }
880
881 void TemplateTable::fload(int n) {
882 transition(vtos, ftos);
883 __ load_float(faddress(n));
884 }
885
886 void TemplateTable::dload(int n) {
887 transition(vtos, dtos);
888 __ load_double(daddress(n));
889 }
890
891 void TemplateTable::aload(int n) {
892 transition(vtos, atos);
893 __ movptr(rax, aaddress(n));
894 }
895
896 void TemplateTable::aload_0() {
897 aload_0_internal();
898 }
899
900 void TemplateTable::nofast_aload_0() {
901 aload_0_internal(may_not_rewrite);
902 }
903
904 void TemplateTable::aload_0_internal(RewriteControl rc) {
905 transition(vtos, atos);
906 // According to bytecode histograms, the pairs:
907 //
908 // _aload_0, _fast_igetfield
909 // _aload_0, _fast_agetfield
910 // _aload_0, _fast_fgetfield
911 //
912 // occur frequently. If RewriteFrequentPairs is set, the (slow)
913 // _aload_0 bytecode checks if the next bytecode is either
914 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
915 // rewrites the current bytecode into a pair bytecode; otherwise it
916 // rewrites the current bytecode into _fast_aload_0 that doesn't do
917 // the pair check anymore.
918 //
919 // Note: If the next bytecode is _getfield, the rewrite must be
920 // delayed, otherwise we may miss an opportunity for a pair.
921 //
922 // Also rewrite frequent pairs
923 // aload_0, aload_1
924 // aload_0, iload_1
925 // These bytecodes with a small amount of code are most profitable
926 // to rewrite
927 if (RewriteFrequentPairs && rc == may_rewrite) {
928 Label rewrite, done;
929
930 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
931 LP64_ONLY(assert(rbx != bc, "register damaged"));
932
933 // get next byte
934 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
935
936 // if _getfield then wait with rewrite
937 __ cmpl(rbx, Bytecodes::_getfield);
938 __ jcc(Assembler::equal, done);
939
940 // if _igetfield then rewrite to _fast_iaccess_0
941 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
942 __ cmpl(rbx, Bytecodes::_fast_igetfield);
943 __ movl(bc, Bytecodes::_fast_iaccess_0);
944 __ jccb(Assembler::equal, rewrite);
945
946 // if _agetfield then rewrite to _fast_aaccess_0
947 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
948 __ cmpl(rbx, Bytecodes::_fast_agetfield);
949 __ movl(bc, Bytecodes::_fast_aaccess_0);
950 __ jccb(Assembler::equal, rewrite);
951
952 // if _fgetfield then rewrite to _fast_faccess_0
953 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
954 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
955 __ movl(bc, Bytecodes::_fast_faccess_0);
956 __ jccb(Assembler::equal, rewrite);
957
958 // else rewrite to _fast_aload0
959 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
960 __ movl(bc, Bytecodes::_fast_aload_0);
961
962 // rewrite
963 // bc: fast bytecode
964 __ bind(rewrite);
965 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
966
967 __ bind(done);
968 }
969
970 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
971 aload(0);
972 }
973
974 void TemplateTable::istore() {
975 transition(itos, vtos);
976 locals_index(rbx);
977 __ movl(iaddress(rbx), rax);
978 }
979
980
981 void TemplateTable::lstore() {
982 transition(ltos, vtos);
983 locals_index(rbx);
984 __ movptr(laddress(rbx), rax);
985 NOT_LP64(__ movptr(haddress(rbx), rdx));
986 }
987
988 void TemplateTable::fstore() {
989 transition(ftos, vtos);
990 locals_index(rbx);
991 __ store_float(faddress(rbx));
992 }
993
994 void TemplateTable::dstore() {
995 transition(dtos, vtos);
996 locals_index(rbx);
997 __ store_double(daddress(rbx));
998 }
999
1000 void TemplateTable::astore() {
1001 transition(vtos, vtos);
1002 __ pop_ptr(rax);
1003 locals_index(rbx);
1004 __ movptr(aaddress(rbx), rax);
1005 }
1006
1007 void TemplateTable::wide_istore() {
1008 transition(vtos, vtos);
1009 __ pop_i();
1010 locals_index_wide(rbx);
1011 __ movl(iaddress(rbx), rax);
1012 }
1013
1014 void TemplateTable::wide_lstore() {
1015 transition(vtos, vtos);
1016 NOT_LP64(__ pop_l(rax, rdx));
1017 LP64_ONLY(__ pop_l());
1018 locals_index_wide(rbx);
1019 __ movptr(laddress(rbx), rax);
1020 NOT_LP64(__ movl(haddress(rbx), rdx));
1021 }
1022
1023 void TemplateTable::wide_fstore() {
1024 #ifdef _LP64
1025 transition(vtos, vtos);
1026 __ pop_f(xmm0);
1027 locals_index_wide(rbx);
1028 __ movflt(faddress(rbx), xmm0);
1029 #else
1030 wide_istore();
1031 #endif
1032 }
1033
1034 void TemplateTable::wide_dstore() {
1035 #ifdef _LP64
1036 transition(vtos, vtos);
1037 __ pop_d(xmm0);
1038 locals_index_wide(rbx);
1039 __ movdbl(daddress(rbx), xmm0);
1040 #else
1041 wide_lstore();
1042 #endif
1043 }
1044
1045 void TemplateTable::wide_astore() {
1046 transition(vtos, vtos);
1047 __ pop_ptr(rax);
1048 locals_index_wide(rbx);
1049 __ movptr(aaddress(rbx), rax);
1050 }
1051
1052 void TemplateTable::iastore() {
1053 transition(itos, vtos);
1054 __ pop_i(rbx);
1055 // rax: value
1056 // rbx: index
1057 // rdx: array
1058 index_check(rdx, rbx); // prefer index in rbx
1059 __ movl(Address(rdx, rbx,
1060 Address::times_4,
1061 arrayOopDesc::base_offset_in_bytes(T_INT)),
1062 rax);
1063 }
1064
1065 void TemplateTable::lastore() {
1066 transition(ltos, vtos);
1067 __ pop_i(rbx);
1068 // rax,: low(value)
1069 // rcx: array
1070 // rdx: high(value)
1071 index_check(rcx, rbx); // prefer index in rbx,
1072 // rbx,: index
1073 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1074 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1075 }
1076
1077
1078 void TemplateTable::fastore() {
1079 transition(ftos, vtos);
1080 __ pop_i(rbx);
1081 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1082 // rbx: index
1083 // rdx: array
1084 index_check(rdx, rbx); // prefer index in rbx
1085 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1086 }
1087
1088 void TemplateTable::dastore() {
1089 transition(dtos, vtos);
1090 __ pop_i(rbx);
1091 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1092 // rbx: index
1093 // rdx: array
1094 index_check(rdx, rbx); // prefer index in rbx
1095 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1096 }
1097
1098 void TemplateTable::aastore() {
1099 Label is_null, ok_is_subtype, done;
1100 transition(vtos, vtos);
1101 // stack: ..., array, index, value
1102 __ movptr(rax, at_tos()); // value
1103 __ movl(rcx, at_tos_p1()); // index
1104 __ movptr(rdx, at_tos_p2()); // array
1105
1106 Address element_address(rdx, rcx,
1107 UseCompressedOops? Address::times_4 : Address::times_ptr,
1108 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1109
1110 index_check_without_pop(rdx, rcx); // kills rbx
1111 __ testptr(rax, rax);
1112 __ jcc(Assembler::zero, is_null);
1113
1114 // Move subklass into rbx
1115 __ load_klass(rbx, rax);
1116 // Move superklass into rax
1117 __ load_klass(rax, rdx);
1118 __ movptr(rax, Address(rax,
1119 ObjArrayKlass::element_klass_offset()));
1120 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1121 __ lea(rdx, element_address);
1122
1123 // Generate subtype check. Blows rcx, rdi
1124 // Superklass in rax. Subklass in rbx.
1125 __ gen_subtype_check(rbx, ok_is_subtype);
1126
1127 // Come here on failure
1128 // object is at TOS
1129 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1130
1131 // Come here on success
1132 __ bind(ok_is_subtype);
1133
1134 // Get the value we will store
1135 __ movptr(rax, at_tos());
1136 // Now store using the appropriate barrier
1137 do_oop_store(_masm, Address(rdx, 0), rax, IN_HEAP_ARRAY);
1138 __ jmp(done);
1139
1140 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1141 __ bind(is_null);
1142 __ profile_null_seen(rbx);
1143
1144 // Store a NULL
1145 do_oop_store(_masm, element_address, noreg, IN_HEAP_ARRAY);
1146
1147 // Pop stack arguments
1148 __ bind(done);
1149 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1150 }
1151
1152 void TemplateTable::bastore() {
1153 transition(itos, vtos);
1154 __ pop_i(rbx);
1155 // rax: value
1156 // rbx: index
1157 // rdx: array
1158 index_check(rdx, rbx); // prefer index in rbx
1159 // Need to check whether array is boolean or byte
1160 // since both types share the bastore bytecode.
1161 __ load_klass(rcx, rdx);
1162 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1163 int diffbit = Klass::layout_helper_boolean_diffbit();
1164 __ testl(rcx, diffbit);
1165 Label L_skip;
1166 __ jccb(Assembler::zero, L_skip);
1167 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1168 __ bind(L_skip);
1169 __ movb(Address(rdx, rbx,
1170 Address::times_1,
1171 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1172 rax);
1173 }
1174
1175 void TemplateTable::castore() {
1176 transition(itos, vtos);
1177 __ pop_i(rbx);
1178 // rax: value
1179 // rbx: index
1180 // rdx: array
1181 index_check(rdx, rbx); // prefer index in rbx
1182 __ movw(Address(rdx, rbx,
1183 Address::times_2,
1184 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1185 rax);
1186 }
1187
1188
1189 void TemplateTable::sastore() {
1190 castore();
1191 }
1192
1193 void TemplateTable::istore(int n) {
1194 transition(itos, vtos);
1195 __ movl(iaddress(n), rax);
1196 }
1197
1198 void TemplateTable::lstore(int n) {
1199 transition(ltos, vtos);
1200 __ movptr(laddress(n), rax);
1201 NOT_LP64(__ movptr(haddress(n), rdx));
1202 }
1203
1204 void TemplateTable::fstore(int n) {
1205 transition(ftos, vtos);
1206 __ store_float(faddress(n));
1207 }
1208
1209 void TemplateTable::dstore(int n) {
1210 transition(dtos, vtos);
1211 __ store_double(daddress(n));
1212 }
1213
1214
1215 void TemplateTable::astore(int n) {
1216 transition(vtos, vtos);
1217 __ pop_ptr(rax);
1218 __ movptr(aaddress(n), rax);
1219 }
1220
1221 void TemplateTable::pop() {
1222 transition(vtos, vtos);
1223 __ addptr(rsp, Interpreter::stackElementSize);
1224 }
1225
1226 void TemplateTable::pop2() {
1227 transition(vtos, vtos);
1228 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1229 }
1230
1231
1232 void TemplateTable::dup() {
1233 transition(vtos, vtos);
1234 __ load_ptr(0, rax);
1235 __ push_ptr(rax);
1236 // stack: ..., a, a
1237 }
1238
1239 void TemplateTable::dup_x1() {
1240 transition(vtos, vtos);
1241 // stack: ..., a, b
1242 __ load_ptr( 0, rax); // load b
1243 __ load_ptr( 1, rcx); // load a
1244 __ store_ptr(1, rax); // store b
1245 __ store_ptr(0, rcx); // store a
1246 __ push_ptr(rax); // push b
1247 // stack: ..., b, a, b
1248 }
1249
1250 void TemplateTable::dup_x2() {
1251 transition(vtos, vtos);
1252 // stack: ..., a, b, c
1253 __ load_ptr( 0, rax); // load c
1254 __ load_ptr( 2, rcx); // load a
1255 __ store_ptr(2, rax); // store c in a
1256 __ push_ptr(rax); // push c
1257 // stack: ..., c, b, c, c
1258 __ load_ptr( 2, rax); // load b
1259 __ store_ptr(2, rcx); // store a in b
1260 // stack: ..., c, a, c, c
1261 __ store_ptr(1, rax); // store b in c
1262 // stack: ..., c, a, b, c
1263 }
1264
1265 void TemplateTable::dup2() {
1266 transition(vtos, vtos);
1267 // stack: ..., a, b
1268 __ load_ptr(1, rax); // load a
1269 __ push_ptr(rax); // push a
1270 __ load_ptr(1, rax); // load b
1271 __ push_ptr(rax); // push b
1272 // stack: ..., a, b, a, b
1273 }
1274
1275
1276 void TemplateTable::dup2_x1() {
1277 transition(vtos, vtos);
1278 // stack: ..., a, b, c
1279 __ load_ptr( 0, rcx); // load c
1280 __ load_ptr( 1, rax); // load b
1281 __ push_ptr(rax); // push b
1282 __ push_ptr(rcx); // push c
1283 // stack: ..., a, b, c, b, c
1284 __ store_ptr(3, rcx); // store c in b
1285 // stack: ..., a, c, c, b, c
1286 __ load_ptr( 4, rcx); // load a
1287 __ store_ptr(2, rcx); // store a in 2nd c
1288 // stack: ..., a, c, a, b, c
1289 __ store_ptr(4, rax); // store b in a
1290 // stack: ..., b, c, a, b, c
1291 }
1292
1293 void TemplateTable::dup2_x2() {
1294 transition(vtos, vtos);
1295 // stack: ..., a, b, c, d
1296 __ load_ptr( 0, rcx); // load d
1297 __ load_ptr( 1, rax); // load c
1298 __ push_ptr(rax); // push c
1299 __ push_ptr(rcx); // push d
1300 // stack: ..., a, b, c, d, c, d
1301 __ load_ptr( 4, rax); // load b
1302 __ store_ptr(2, rax); // store b in d
1303 __ store_ptr(4, rcx); // store d in b
1304 // stack: ..., a, d, c, b, c, d
1305 __ load_ptr( 5, rcx); // load a
1306 __ load_ptr( 3, rax); // load c
1307 __ store_ptr(3, rcx); // store a in c
1308 __ store_ptr(5, rax); // store c in a
1309 // stack: ..., c, d, a, b, c, d
1310 }
1311
1312 void TemplateTable::swap() {
1313 transition(vtos, vtos);
1314 // stack: ..., a, b
1315 __ load_ptr( 1, rcx); // load a
1316 __ load_ptr( 0, rax); // load b
1317 __ store_ptr(0, rcx); // store a in b
1318 __ store_ptr(1, rax); // store b in a
1319 // stack: ..., b, a
1320 }
1321
1322 void TemplateTable::iop2(Operation op) {
1323 transition(itos, itos);
1324 switch (op) {
1325 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1326 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1327 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1328 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1329 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1330 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1331 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1332 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1333 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1334 default : ShouldNotReachHere();
1335 }
1336 }
1337
1338 void TemplateTable::lop2(Operation op) {
1339 transition(ltos, ltos);
1340 #ifdef _LP64
1341 switch (op) {
1342 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1343 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1344 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1345 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1346 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1347 default : ShouldNotReachHere();
1348 }
1349 #else
1350 __ pop_l(rbx, rcx);
1351 switch (op) {
1352 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1353 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1354 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1355 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1356 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1357 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1358 default : ShouldNotReachHere();
1359 }
1360 #endif
1361 }
1362
1363 void TemplateTable::idiv() {
1364 transition(itos, itos);
1365 __ movl(rcx, rax);
1366 __ pop_i(rax);
1367 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1368 // they are not equal, one could do a normal division (no correction
1369 // needed), which may speed up this implementation for the common case.
1370 // (see also JVM spec., p.243 & p.271)
1371 __ corrected_idivl(rcx);
1372 }
1373
1374 void TemplateTable::irem() {
1375 transition(itos, itos);
1376 __ movl(rcx, rax);
1377 __ pop_i(rax);
1378 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1379 // they are not equal, one could do a normal division (no correction
1380 // needed), which may speed up this implementation for the common case.
1381 // (see also JVM spec., p.243 & p.271)
1382 __ corrected_idivl(rcx);
1383 __ movl(rax, rdx);
1384 }
1385
1386 void TemplateTable::lmul() {
1387 transition(ltos, ltos);
1388 #ifdef _LP64
1389 __ pop_l(rdx);
1390 __ imulq(rax, rdx);
1391 #else
1392 __ pop_l(rbx, rcx);
1393 __ push(rcx); __ push(rbx);
1394 __ push(rdx); __ push(rax);
1395 __ lmul(2 * wordSize, 0);
1396 __ addptr(rsp, 4 * wordSize); // take off temporaries
1397 #endif
1398 }
1399
1400 void TemplateTable::ldiv() {
1401 transition(ltos, ltos);
1402 #ifdef _LP64
1403 __ mov(rcx, rax);
1404 __ pop_l(rax);
1405 // generate explicit div0 check
1406 __ testq(rcx, rcx);
1407 __ jump_cc(Assembler::zero,
1408 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1409 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1410 // they are not equal, one could do a normal division (no correction
1411 // needed), which may speed up this implementation for the common case.
1412 // (see also JVM spec., p.243 & p.271)
1413 __ corrected_idivq(rcx); // kills rbx
1414 #else
1415 __ pop_l(rbx, rcx);
1416 __ push(rcx); __ push(rbx);
1417 __ push(rdx); __ push(rax);
1418 // check if y = 0
1419 __ orl(rax, rdx);
1420 __ jump_cc(Assembler::zero,
1421 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1422 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1423 __ addptr(rsp, 4 * wordSize); // take off temporaries
1424 #endif
1425 }
1426
1427 void TemplateTable::lrem() {
1428 transition(ltos, ltos);
1429 #ifdef _LP64
1430 __ mov(rcx, rax);
1431 __ pop_l(rax);
1432 __ testq(rcx, rcx);
1433 __ jump_cc(Assembler::zero,
1434 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1435 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1436 // they are not equal, one could do a normal division (no correction
1437 // needed), which may speed up this implementation for the common case.
1438 // (see also JVM spec., p.243 & p.271)
1439 __ corrected_idivq(rcx); // kills rbx
1440 __ mov(rax, rdx);
1441 #else
1442 __ pop_l(rbx, rcx);
1443 __ push(rcx); __ push(rbx);
1444 __ push(rdx); __ push(rax);
1445 // check if y = 0
1446 __ orl(rax, rdx);
1447 __ jump_cc(Assembler::zero,
1448 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1450 __ addptr(rsp, 4 * wordSize);
1451 #endif
1452 }
1453
1454 void TemplateTable::lshl() {
1455 transition(itos, ltos);
1456 __ movl(rcx, rax); // get shift count
1457 #ifdef _LP64
1458 __ pop_l(rax); // get shift value
1459 __ shlq(rax);
1460 #else
1461 __ pop_l(rax, rdx); // get shift value
1462 __ lshl(rdx, rax);
1463 #endif
1464 }
1465
1466 void TemplateTable::lshr() {
1467 #ifdef _LP64
1468 transition(itos, ltos);
1469 __ movl(rcx, rax); // get shift count
1470 __ pop_l(rax); // get shift value
1471 __ sarq(rax);
1472 #else
1473 transition(itos, ltos);
1474 __ mov(rcx, rax); // get shift count
1475 __ pop_l(rax, rdx); // get shift value
1476 __ lshr(rdx, rax, true);
1477 #endif
1478 }
1479
1480 void TemplateTable::lushr() {
1481 transition(itos, ltos);
1482 #ifdef _LP64
1483 __ movl(rcx, rax); // get shift count
1484 __ pop_l(rax); // get shift value
1485 __ shrq(rax);
1486 #else
1487 __ mov(rcx, rax); // get shift count
1488 __ pop_l(rax, rdx); // get shift value
1489 __ lshr(rdx, rax);
1490 #endif
1491 }
1492
1493 void TemplateTable::fop2(Operation op) {
1494 transition(ftos, ftos);
1495
1496 if (UseSSE >= 1) {
1497 switch (op) {
1498 case add:
1499 __ addss(xmm0, at_rsp());
1500 __ addptr(rsp, Interpreter::stackElementSize);
1501 break;
1502 case sub:
1503 __ movflt(xmm1, xmm0);
1504 __ pop_f(xmm0);
1505 __ subss(xmm0, xmm1);
1506 break;
1507 case mul:
1508 __ mulss(xmm0, at_rsp());
1509 __ addptr(rsp, Interpreter::stackElementSize);
1510 break;
1511 case div:
1512 __ movflt(xmm1, xmm0);
1513 __ pop_f(xmm0);
1514 __ divss(xmm0, xmm1);
1515 break;
1516 case rem:
1517 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1518 // modulo operation. The frem method calls the function
1519 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1520 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1521 // (signalling or quiet) is returned.
1522 //
1523 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1524 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1525 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1526 // The fprem instruction used on x86_32 is functionally equivalent to
1527 // SharedRuntime::frem in that it returns a NaN.
1528 #ifdef _LP64
1529 __ movflt(xmm1, xmm0);
1530 __ pop_f(xmm0);
1531 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1532 #else
1533 __ push_f(xmm0);
1534 __ pop_f();
1535 __ fld_s(at_rsp());
1536 __ fremr(rax);
1537 __ f2ieee();
1538 __ pop(rax); // pop second operand off the stack
1539 __ push_f();
1540 __ pop_f(xmm0);
1541 #endif
1542 break;
1543 default:
1544 ShouldNotReachHere();
1545 break;
1546 }
1547 } else {
1548 #ifdef _LP64
1549 ShouldNotReachHere();
1550 #else
1551 switch (op) {
1552 case add: __ fadd_s (at_rsp()); break;
1553 case sub: __ fsubr_s(at_rsp()); break;
1554 case mul: __ fmul_s (at_rsp()); break;
1555 case div: __ fdivr_s(at_rsp()); break;
1556 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1557 default : ShouldNotReachHere();
1558 }
1559 __ f2ieee();
1560 __ pop(rax); // pop second operand off the stack
1561 #endif // _LP64
1562 }
1563 }
1564
1565 void TemplateTable::dop2(Operation op) {
1566 transition(dtos, dtos);
1567 if (UseSSE >= 2) {
1568 switch (op) {
1569 case add:
1570 __ addsd(xmm0, at_rsp());
1571 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1572 break;
1573 case sub:
1574 __ movdbl(xmm1, xmm0);
1575 __ pop_d(xmm0);
1576 __ subsd(xmm0, xmm1);
1577 break;
1578 case mul:
1579 __ mulsd(xmm0, at_rsp());
1580 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1581 break;
1582 case div:
1583 __ movdbl(xmm1, xmm0);
1584 __ pop_d(xmm0);
1585 __ divsd(xmm0, xmm1);
1586 break;
1587 case rem:
1588 // Similar to fop2(), the modulo operation is performed using the
1589 // SharedRuntime::drem method (on x86_64 platforms) or using the
1590 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1591 #ifdef _LP64
1592 __ movdbl(xmm1, xmm0);
1593 __ pop_d(xmm0);
1594 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1595 #else
1596 __ push_d(xmm0);
1597 __ pop_d();
1598 __ fld_d(at_rsp());
1599 __ fremr(rax);
1600 __ d2ieee();
1601 __ pop(rax);
1602 __ pop(rdx);
1603 __ push_d();
1604 __ pop_d(xmm0);
1605 #endif
1606 break;
1607 default:
1608 ShouldNotReachHere();
1609 break;
1610 }
1611 } else {
1612 #ifdef _LP64
1613 ShouldNotReachHere();
1614 #else
1615 switch (op) {
1616 case add: __ fadd_d (at_rsp()); break;
1617 case sub: __ fsubr_d(at_rsp()); break;
1618 case mul: {
1619 Label L_strict;
1620 Label L_join;
1621 const Address access_flags (rcx, Method::access_flags_offset());
1622 __ get_method(rcx);
1623 __ movl(rcx, access_flags);
1624 __ testl(rcx, JVM_ACC_STRICT);
1625 __ jccb(Assembler::notZero, L_strict);
1626 __ fmul_d (at_rsp());
1627 __ jmpb(L_join);
1628 __ bind(L_strict);
1629 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1630 __ fmulp();
1631 __ fmul_d (at_rsp());
1632 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1633 __ fmulp();
1634 __ bind(L_join);
1635 break;
1636 }
1637 case div: {
1638 Label L_strict;
1639 Label L_join;
1640 const Address access_flags (rcx, Method::access_flags_offset());
1641 __ get_method(rcx);
1642 __ movl(rcx, access_flags);
1643 __ testl(rcx, JVM_ACC_STRICT);
1644 __ jccb(Assembler::notZero, L_strict);
1645 __ fdivr_d(at_rsp());
1646 __ jmp(L_join);
1647 __ bind(L_strict);
1648 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1649 __ fmul_d (at_rsp());
1650 __ fdivrp();
1651 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1652 __ fmulp();
1653 __ bind(L_join);
1654 break;
1655 }
1656 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1657 default : ShouldNotReachHere();
1658 }
1659 __ d2ieee();
1660 // Pop double precision number from rsp.
1661 __ pop(rax);
1662 __ pop(rdx);
1663 #endif
1664 }
1665 }
1666
1667 void TemplateTable::ineg() {
1668 transition(itos, itos);
1669 __ negl(rax);
1670 }
1671
1672 void TemplateTable::lneg() {
1673 transition(ltos, ltos);
1674 LP64_ONLY(__ negq(rax));
1675 NOT_LP64(__ lneg(rdx, rax));
1676 }
1677
1678 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1679 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1680 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1681 // of 128-bits operands for SSE instructions.
1682 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1683 // Store the value to a 128-bits operand.
1684 operand[0] = lo;
1685 operand[1] = hi;
1686 return operand;
1687 }
1688
1689 // Buffer for 128-bits masks used by SSE instructions.
1690 static jlong float_signflip_pool[2*2];
1691 static jlong double_signflip_pool[2*2];
1692
1693 void TemplateTable::fneg() {
1694 transition(ftos, ftos);
1695 if (UseSSE >= 1) {
1696 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1697 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1698 } else {
1699 LP64_ONLY(ShouldNotReachHere());
1700 NOT_LP64(__ fchs());
1701 }
1702 }
1703
1704 void TemplateTable::dneg() {
1705 transition(dtos, dtos);
1706 if (UseSSE >= 2) {
1707 static jlong *double_signflip =
1708 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1709 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1710 } else {
1711 #ifdef _LP64
1712 ShouldNotReachHere();
1713 #else
1714 __ fchs();
1715 #endif
1716 }
1717 }
1718
1719 void TemplateTable::iinc() {
1720 transition(vtos, vtos);
1721 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1722 locals_index(rbx);
1723 __ addl(iaddress(rbx), rdx);
1724 }
1725
1726 void TemplateTable::wide_iinc() {
1727 transition(vtos, vtos);
1728 __ movl(rdx, at_bcp(4)); // get constant
1729 locals_index_wide(rbx);
1730 __ bswapl(rdx); // swap bytes & sign-extend constant
1731 __ sarl(rdx, 16);
1732 __ addl(iaddress(rbx), rdx);
1733 // Note: should probably use only one movl to get both
1734 // the index and the constant -> fix this
1735 }
1736
1737 void TemplateTable::convert() {
1738 #ifdef _LP64
1739 // Checking
1740 #ifdef ASSERT
1741 {
1742 TosState tos_in = ilgl;
1743 TosState tos_out = ilgl;
1744 switch (bytecode()) {
1745 case Bytecodes::_i2l: // fall through
1746 case Bytecodes::_i2f: // fall through
1747 case Bytecodes::_i2d: // fall through
1748 case Bytecodes::_i2b: // fall through
1749 case Bytecodes::_i2c: // fall through
1750 case Bytecodes::_i2s: tos_in = itos; break;
1751 case Bytecodes::_l2i: // fall through
1752 case Bytecodes::_l2f: // fall through
1753 case Bytecodes::_l2d: tos_in = ltos; break;
1754 case Bytecodes::_f2i: // fall through
1755 case Bytecodes::_f2l: // fall through
1756 case Bytecodes::_f2d: tos_in = ftos; break;
1757 case Bytecodes::_d2i: // fall through
1758 case Bytecodes::_d2l: // fall through
1759 case Bytecodes::_d2f: tos_in = dtos; break;
1760 default : ShouldNotReachHere();
1761 }
1762 switch (bytecode()) {
1763 case Bytecodes::_l2i: // fall through
1764 case Bytecodes::_f2i: // fall through
1765 case Bytecodes::_d2i: // fall through
1766 case Bytecodes::_i2b: // fall through
1767 case Bytecodes::_i2c: // fall through
1768 case Bytecodes::_i2s: tos_out = itos; break;
1769 case Bytecodes::_i2l: // fall through
1770 case Bytecodes::_f2l: // fall through
1771 case Bytecodes::_d2l: tos_out = ltos; break;
1772 case Bytecodes::_i2f: // fall through
1773 case Bytecodes::_l2f: // fall through
1774 case Bytecodes::_d2f: tos_out = ftos; break;
1775 case Bytecodes::_i2d: // fall through
1776 case Bytecodes::_l2d: // fall through
1777 case Bytecodes::_f2d: tos_out = dtos; break;
1778 default : ShouldNotReachHere();
1779 }
1780 transition(tos_in, tos_out);
1781 }
1782 #endif // ASSERT
1783
1784 static const int64_t is_nan = 0x8000000000000000L;
1785
1786 // Conversion
1787 switch (bytecode()) {
1788 case Bytecodes::_i2l:
1789 __ movslq(rax, rax);
1790 break;
1791 case Bytecodes::_i2f:
1792 __ cvtsi2ssl(xmm0, rax);
1793 break;
1794 case Bytecodes::_i2d:
1795 __ cvtsi2sdl(xmm0, rax);
1796 break;
1797 case Bytecodes::_i2b:
1798 __ movsbl(rax, rax);
1799 break;
1800 case Bytecodes::_i2c:
1801 __ movzwl(rax, rax);
1802 break;
1803 case Bytecodes::_i2s:
1804 __ movswl(rax, rax);
1805 break;
1806 case Bytecodes::_l2i:
1807 __ movl(rax, rax);
1808 break;
1809 case Bytecodes::_l2f:
1810 __ cvtsi2ssq(xmm0, rax);
1811 break;
1812 case Bytecodes::_l2d:
1813 __ cvtsi2sdq(xmm0, rax);
1814 break;
1815 case Bytecodes::_f2i:
1816 {
1817 Label L;
1818 __ cvttss2sil(rax, xmm0);
1819 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1820 __ jcc(Assembler::notEqual, L);
1821 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1822 __ bind(L);
1823 }
1824 break;
1825 case Bytecodes::_f2l:
1826 {
1827 Label L;
1828 __ cvttss2siq(rax, xmm0);
1829 // NaN or overflow/underflow?
1830 __ cmp64(rax, ExternalAddress((address) &is_nan));
1831 __ jcc(Assembler::notEqual, L);
1832 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1833 __ bind(L);
1834 }
1835 break;
1836 case Bytecodes::_f2d:
1837 __ cvtss2sd(xmm0, xmm0);
1838 break;
1839 case Bytecodes::_d2i:
1840 {
1841 Label L;
1842 __ cvttsd2sil(rax, xmm0);
1843 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1844 __ jcc(Assembler::notEqual, L);
1845 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1846 __ bind(L);
1847 }
1848 break;
1849 case Bytecodes::_d2l:
1850 {
1851 Label L;
1852 __ cvttsd2siq(rax, xmm0);
1853 // NaN or overflow/underflow?
1854 __ cmp64(rax, ExternalAddress((address) &is_nan));
1855 __ jcc(Assembler::notEqual, L);
1856 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1857 __ bind(L);
1858 }
1859 break;
1860 case Bytecodes::_d2f:
1861 __ cvtsd2ss(xmm0, xmm0);
1862 break;
1863 default:
1864 ShouldNotReachHere();
1865 }
1866 #else
1867 // Checking
1868 #ifdef ASSERT
1869 { TosState tos_in = ilgl;
1870 TosState tos_out = ilgl;
1871 switch (bytecode()) {
1872 case Bytecodes::_i2l: // fall through
1873 case Bytecodes::_i2f: // fall through
1874 case Bytecodes::_i2d: // fall through
1875 case Bytecodes::_i2b: // fall through
1876 case Bytecodes::_i2c: // fall through
1877 case Bytecodes::_i2s: tos_in = itos; break;
1878 case Bytecodes::_l2i: // fall through
1879 case Bytecodes::_l2f: // fall through
1880 case Bytecodes::_l2d: tos_in = ltos; break;
1881 case Bytecodes::_f2i: // fall through
1882 case Bytecodes::_f2l: // fall through
1883 case Bytecodes::_f2d: tos_in = ftos; break;
1884 case Bytecodes::_d2i: // fall through
1885 case Bytecodes::_d2l: // fall through
1886 case Bytecodes::_d2f: tos_in = dtos; break;
1887 default : ShouldNotReachHere();
1888 }
1889 switch (bytecode()) {
1890 case Bytecodes::_l2i: // fall through
1891 case Bytecodes::_f2i: // fall through
1892 case Bytecodes::_d2i: // fall through
1893 case Bytecodes::_i2b: // fall through
1894 case Bytecodes::_i2c: // fall through
1895 case Bytecodes::_i2s: tos_out = itos; break;
1896 case Bytecodes::_i2l: // fall through
1897 case Bytecodes::_f2l: // fall through
1898 case Bytecodes::_d2l: tos_out = ltos; break;
1899 case Bytecodes::_i2f: // fall through
1900 case Bytecodes::_l2f: // fall through
1901 case Bytecodes::_d2f: tos_out = ftos; break;
1902 case Bytecodes::_i2d: // fall through
1903 case Bytecodes::_l2d: // fall through
1904 case Bytecodes::_f2d: tos_out = dtos; break;
1905 default : ShouldNotReachHere();
1906 }
1907 transition(tos_in, tos_out);
1908 }
1909 #endif // ASSERT
1910
1911 // Conversion
1912 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1913 switch (bytecode()) {
1914 case Bytecodes::_i2l:
1915 __ extend_sign(rdx, rax);
1916 break;
1917 case Bytecodes::_i2f:
1918 if (UseSSE >= 1) {
1919 __ cvtsi2ssl(xmm0, rax);
1920 } else {
1921 __ push(rax); // store int on tos
1922 __ fild_s(at_rsp()); // load int to ST0
1923 __ f2ieee(); // truncate to float size
1924 __ pop(rcx); // adjust rsp
1925 }
1926 break;
1927 case Bytecodes::_i2d:
1928 if (UseSSE >= 2) {
1929 __ cvtsi2sdl(xmm0, rax);
1930 } else {
1931 __ push(rax); // add one slot for d2ieee()
1932 __ push(rax); // store int on tos
1933 __ fild_s(at_rsp()); // load int to ST0
1934 __ d2ieee(); // truncate to double size
1935 __ pop(rcx); // adjust rsp
1936 __ pop(rcx);
1937 }
1938 break;
1939 case Bytecodes::_i2b:
1940 __ shll(rax, 24); // truncate upper 24 bits
1941 __ sarl(rax, 24); // and sign-extend byte
1942 LP64_ONLY(__ movsbl(rax, rax));
1943 break;
1944 case Bytecodes::_i2c:
1945 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1946 LP64_ONLY(__ movzwl(rax, rax));
1947 break;
1948 case Bytecodes::_i2s:
1949 __ shll(rax, 16); // truncate upper 16 bits
1950 __ sarl(rax, 16); // and sign-extend short
1951 LP64_ONLY(__ movswl(rax, rax));
1952 break;
1953 case Bytecodes::_l2i:
1954 /* nothing to do */
1955 break;
1956 case Bytecodes::_l2f:
1957 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1958 // 64-bit long values to floats. On 32-bit platforms it is not possible
1959 // to use that instruction with 64-bit operands, therefore the FPU is
1960 // used to perform the conversion.
1961 __ push(rdx); // store long on tos
1962 __ push(rax);
1963 __ fild_d(at_rsp()); // load long to ST0
1964 __ f2ieee(); // truncate to float size
1965 __ pop(rcx); // adjust rsp
1966 __ pop(rcx);
1967 if (UseSSE >= 1) {
1968 __ push_f();
1969 __ pop_f(xmm0);
1970 }
1971 break;
1972 case Bytecodes::_l2d:
1973 // On 32-bit platforms the FPU is used for conversion because on
1974 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1975 // instruction with 64-bit operands.
1976 __ push(rdx); // store long on tos
1977 __ push(rax);
1978 __ fild_d(at_rsp()); // load long to ST0
1979 __ d2ieee(); // truncate to double size
1980 __ pop(rcx); // adjust rsp
1981 __ pop(rcx);
1982 if (UseSSE >= 2) {
1983 __ push_d();
1984 __ pop_d(xmm0);
1985 }
1986 break;
1987 case Bytecodes::_f2i:
1988 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1989 // as it returns 0 for any NaN.
1990 if (UseSSE >= 1) {
1991 __ push_f(xmm0);
1992 } else {
1993 __ push(rcx); // reserve space for argument
1994 __ fstp_s(at_rsp()); // pass float argument on stack
1995 }
1996 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1997 break;
1998 case Bytecodes::_f2l:
1999 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2000 // as it returns 0 for any NaN.
2001 if (UseSSE >= 1) {
2002 __ push_f(xmm0);
2003 } else {
2004 __ push(rcx); // reserve space for argument
2005 __ fstp_s(at_rsp()); // pass float argument on stack
2006 }
2007 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2008 break;
2009 case Bytecodes::_f2d:
2010 if (UseSSE < 1) {
2011 /* nothing to do */
2012 } else if (UseSSE == 1) {
2013 __ push_f(xmm0);
2014 __ pop_f();
2015 } else { // UseSSE >= 2
2016 __ cvtss2sd(xmm0, xmm0);
2017 }
2018 break;
2019 case Bytecodes::_d2i:
2020 if (UseSSE >= 2) {
2021 __ push_d(xmm0);
2022 } else {
2023 __ push(rcx); // reserve space for argument
2024 __ push(rcx);
2025 __ fstp_d(at_rsp()); // pass double argument on stack
2026 }
2027 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2028 break;
2029 case Bytecodes::_d2l:
2030 if (UseSSE >= 2) {
2031 __ push_d(xmm0);
2032 } else {
2033 __ push(rcx); // reserve space for argument
2034 __ push(rcx);
2035 __ fstp_d(at_rsp()); // pass double argument on stack
2036 }
2037 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2038 break;
2039 case Bytecodes::_d2f:
2040 if (UseSSE <= 1) {
2041 __ push(rcx); // reserve space for f2ieee()
2042 __ f2ieee(); // truncate to float size
2043 __ pop(rcx); // adjust rsp
2044 if (UseSSE == 1) {
2045 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2046 // the conversion is performed using the FPU in this case.
2047 __ push_f();
2048 __ pop_f(xmm0);
2049 }
2050 } else { // UseSSE >= 2
2051 __ cvtsd2ss(xmm0, xmm0);
2052 }
2053 break;
2054 default :
2055 ShouldNotReachHere();
2056 }
2057 #endif
2058 }
2059
2060 void TemplateTable::lcmp() {
2061 transition(ltos, itos);
2062 #ifdef _LP64
2063 Label done;
2064 __ pop_l(rdx);
2065 __ cmpq(rdx, rax);
2066 __ movl(rax, -1);
2067 __ jccb(Assembler::less, done);
2068 __ setb(Assembler::notEqual, rax);
2069 __ movzbl(rax, rax);
2070 __ bind(done);
2071 #else
2072
2073 // y = rdx:rax
2074 __ pop_l(rbx, rcx); // get x = rcx:rbx
2075 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2076 __ mov(rax, rcx);
2077 #endif
2078 }
2079
2080 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2081 if ((is_float && UseSSE >= 1) ||
2082 (!is_float && UseSSE >= 2)) {
2083 Label done;
2084 if (is_float) {
2085 // XXX get rid of pop here, use ... reg, mem32
2086 __ pop_f(xmm1);
2087 __ ucomiss(xmm1, xmm0);
2088 } else {
2089 // XXX get rid of pop here, use ... reg, mem64
2090 __ pop_d(xmm1);
2091 __ ucomisd(xmm1, xmm0);
2092 }
2093 if (unordered_result < 0) {
2094 __ movl(rax, -1);
2095 __ jccb(Assembler::parity, done);
2096 __ jccb(Assembler::below, done);
2097 __ setb(Assembler::notEqual, rdx);
2098 __ movzbl(rax, rdx);
2099 } else {
2100 __ movl(rax, 1);
2101 __ jccb(Assembler::parity, done);
2102 __ jccb(Assembler::above, done);
2103 __ movl(rax, 0);
2104 __ jccb(Assembler::equal, done);
2105 __ decrementl(rax);
2106 }
2107 __ bind(done);
2108 } else {
2109 #ifdef _LP64
2110 ShouldNotReachHere();
2111 #else
2112 if (is_float) {
2113 __ fld_s(at_rsp());
2114 } else {
2115 __ fld_d(at_rsp());
2116 __ pop(rdx);
2117 }
2118 __ pop(rcx);
2119 __ fcmp2int(rax, unordered_result < 0);
2120 #endif // _LP64
2121 }
2122 }
2123
2124 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2125 __ get_method(rcx); // rcx holds method
2126 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2127 // holds bumped taken count
2128
2129 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2130 InvocationCounter::counter_offset();
2131 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2132 InvocationCounter::counter_offset();
2133
2134 // Load up edx with the branch displacement
2135 if (is_wide) {
2136 __ movl(rdx, at_bcp(1));
2137 } else {
2138 __ load_signed_short(rdx, at_bcp(1));
2139 }
2140 __ bswapl(rdx);
2141
2142 if (!is_wide) {
2143 __ sarl(rdx, 16);
2144 }
2145 LP64_ONLY(__ movl2ptr(rdx, rdx));
2146
2147 // Handle all the JSR stuff here, then exit.
2148 // It's much shorter and cleaner than intermingling with the non-JSR
2149 // normal-branch stuff occurring below.
2150 if (is_jsr) {
2151 // Pre-load the next target bytecode into rbx
2152 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2153
2154 // compute return address as bci in rax
2155 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2156 in_bytes(ConstMethod::codes_offset())));
2157 __ subptr(rax, Address(rcx, Method::const_offset()));
2158 // Adjust the bcp in r13 by the displacement in rdx
2159 __ addptr(rbcp, rdx);
2160 // jsr returns atos that is not an oop
2161 __ push_i(rax);
2162 __ dispatch_only(vtos, true);
2163 return;
2164 }
2165
2166 // Normal (non-jsr) branch handling
2167
2168 // Adjust the bcp in r13 by the displacement in rdx
2169 __ addptr(rbcp, rdx);
2170
2171 assert(UseLoopCounter || !UseOnStackReplacement,
2172 "on-stack-replacement requires loop counters");
2173 Label backedge_counter_overflow;
2174 Label profile_method;
2175 Label dispatch;
2176 if (UseLoopCounter) {
2177 // increment backedge counter for backward branches
2178 // rax: MDO
2179 // rbx: MDO bumped taken-count
2180 // rcx: method
2181 // rdx: target offset
2182 // r13: target bcp
2183 // r14: locals pointer
2184 __ testl(rdx, rdx); // check if forward or backward branch
2185 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2186
2187 // check if MethodCounters exists
2188 Label has_counters;
2189 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2190 __ testptr(rax, rax);
2191 __ jcc(Assembler::notZero, has_counters);
2192 __ push(rdx);
2193 __ push(rcx);
2194 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2195 rcx);
2196 __ pop(rcx);
2197 __ pop(rdx);
2198 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2199 __ testptr(rax, rax);
2200 __ jcc(Assembler::zero, dispatch);
2201 __ bind(has_counters);
2202
2203 if (TieredCompilation) {
2204 Label no_mdo;
2205 int increment = InvocationCounter::count_increment;
2206 if (ProfileInterpreter) {
2207 // Are we profiling?
2208 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2209 __ testptr(rbx, rbx);
2210 __ jccb(Assembler::zero, no_mdo);
2211 // Increment the MDO backedge counter
2212 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2213 in_bytes(InvocationCounter::counter_offset()));
2214 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2215 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2216 rax, false, Assembler::zero, &backedge_counter_overflow);
2217 __ jmp(dispatch);
2218 }
2219 __ bind(no_mdo);
2220 // Increment backedge counter in MethodCounters*
2221 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2222 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2223 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2224 rax, false, Assembler::zero, &backedge_counter_overflow);
2225 } else { // not TieredCompilation
2226 // increment counter
2227 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2228 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2229 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2230 __ movl(Address(rcx, be_offset), rax); // store counter
2231
2232 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2233
2234 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2235 __ addl(rax, Address(rcx, be_offset)); // add both counters
2236
2237 if (ProfileInterpreter) {
2238 // Test to see if we should create a method data oop
2239 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2240 __ jcc(Assembler::less, dispatch);
2241
2242 // if no method data exists, go to profile method
2243 __ test_method_data_pointer(rax, profile_method);
2244
2245 if (UseOnStackReplacement) {
2246 // check for overflow against rbx which is the MDO taken count
2247 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2248 __ jcc(Assembler::below, dispatch);
2249
2250 // When ProfileInterpreter is on, the backedge_count comes
2251 // from the MethodData*, which value does not get reset on
2252 // the call to frequency_counter_overflow(). To avoid
2253 // excessive calls to the overflow routine while the method is
2254 // being compiled, add a second test to make sure the overflow
2255 // function is called only once every overflow_frequency.
2256 const int overflow_frequency = 1024;
2257 __ andl(rbx, overflow_frequency - 1);
2258 __ jcc(Assembler::zero, backedge_counter_overflow);
2259
2260 }
2261 } else {
2262 if (UseOnStackReplacement) {
2263 // check for overflow against rax, which is the sum of the
2264 // counters
2265 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2266 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2267
2268 }
2269 }
2270 }
2271 __ bind(dispatch);
2272 }
2273
2274 // Pre-load the next target bytecode into rbx
2275 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2276
2277 // continue with the bytecode @ target
2278 // rax: return bci for jsr's, unused otherwise
2279 // rbx: target bytecode
2280 // r13: target bcp
2281 __ dispatch_only(vtos, true);
2282
2283 if (UseLoopCounter) {
2284 if (ProfileInterpreter) {
2285 // Out-of-line code to allocate method data oop.
2286 __ bind(profile_method);
2287 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2288 __ set_method_data_pointer_for_bcp();
2289 __ jmp(dispatch);
2290 }
2291
2292 if (UseOnStackReplacement) {
2293 // invocation counter overflow
2294 __ bind(backedge_counter_overflow);
2295 __ negptr(rdx);
2296 __ addptr(rdx, rbcp); // branch bcp
2297 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2298 __ call_VM(noreg,
2299 CAST_FROM_FN_PTR(address,
2300 InterpreterRuntime::frequency_counter_overflow),
2301 rdx);
2302
2303 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2304 // rdx: scratch
2305 // r14: locals pointer
2306 // r13: bcp
2307 __ testptr(rax, rax); // test result
2308 __ jcc(Assembler::zero, dispatch); // no osr if null
2309 // nmethod may have been invalidated (VM may block upon call_VM return)
2310 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2311 __ jcc(Assembler::notEqual, dispatch);
2312
2313 // We have the address of an on stack replacement routine in rax.
2314 // In preparation of invoking it, first we must migrate the locals
2315 // and monitors from off the interpreter frame on the stack.
2316 // Ensure to save the osr nmethod over the migration call,
2317 // it will be preserved in rbx.
2318 __ mov(rbx, rax);
2319
2320 NOT_LP64(__ get_thread(rcx));
2321
2322 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2323
2324 // rax is OSR buffer, move it to expected parameter location
2325 LP64_ONLY(__ mov(j_rarg0, rax));
2326 NOT_LP64(__ mov(rcx, rax));
2327 // We use j_rarg definitions here so that registers don't conflict as parameter
2328 // registers change across platforms as we are in the midst of a calling
2329 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2330
2331 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2332 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2333
2334 // pop the interpreter frame
2335 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2336 __ leave(); // remove frame anchor
2337 __ pop(retaddr); // get return address
2338 __ mov(rsp, sender_sp); // set sp to sender sp
2339 // Ensure compiled code always sees stack at proper alignment
2340 __ andptr(rsp, -(StackAlignmentInBytes));
2341
2342 // unlike x86 we need no specialized return from compiled code
2343 // to the interpreter or the call stub.
2344
2345 // push the return address
2346 __ push(retaddr);
2347
2348 // and begin the OSR nmethod
2349 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2350 }
2351 }
2352 }
2353
2354 void TemplateTable::if_0cmp(Condition cc) {
2355 transition(itos, vtos);
2356 // assume branch is more often taken than not (loops use backward branches)
2357 Label not_taken;
2358 __ testl(rax, rax);
2359 __ jcc(j_not(cc), not_taken);
2360 branch(false, false);
2361 __ bind(not_taken);
2362 __ profile_not_taken_branch(rax);
2363 }
2364
2365 void TemplateTable::if_icmp(Condition cc) {
2366 transition(itos, vtos);
2367 // assume branch is more often taken than not (loops use backward branches)
2368 Label not_taken;
2369 __ pop_i(rdx);
2370 __ cmpl(rdx, rax);
2371 __ jcc(j_not(cc), not_taken);
2372 branch(false, false);
2373 __ bind(not_taken);
2374 __ profile_not_taken_branch(rax);
2375 }
2376
2377 void TemplateTable::if_nullcmp(Condition cc) {
2378 transition(atos, vtos);
2379 // assume branch is more often taken than not (loops use backward branches)
2380 Label not_taken;
2381 __ testptr(rax, rax);
2382 __ jcc(j_not(cc), not_taken);
2383 branch(false, false);
2384 __ bind(not_taken);
2385 __ profile_not_taken_branch(rax);
2386 }
2387
2388 void TemplateTable::if_acmp(Condition cc) {
2389 transition(atos, vtos);
2390 // assume branch is more often taken than not (loops use backward branches)
2391 Label not_taken;
2392 __ pop_ptr(rdx);
2393 __ cmpoop(rdx, rax);
2394 __ jcc(j_not(cc), not_taken);
2395 branch(false, false);
2396 __ bind(not_taken);
2397 __ profile_not_taken_branch(rax);
2398 }
2399
2400 void TemplateTable::ret() {
2401 transition(vtos, vtos);
2402 locals_index(rbx);
2403 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2404 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2405 __ profile_ret(rbx, rcx);
2406 __ get_method(rax);
2407 __ movptr(rbcp, Address(rax, Method::const_offset()));
2408 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2409 ConstMethod::codes_offset()));
2410 __ dispatch_next(vtos, 0, true);
2411 }
2412
2413 void TemplateTable::wide_ret() {
2414 transition(vtos, vtos);
2415 locals_index_wide(rbx);
2416 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2417 __ profile_ret(rbx, rcx);
2418 __ get_method(rax);
2419 __ movptr(rbcp, Address(rax, Method::const_offset()));
2420 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2421 __ dispatch_next(vtos, 0, true);
2422 }
2423
2424 void TemplateTable::tableswitch() {
2425 Label default_case, continue_execution;
2426 transition(itos, vtos);
2427
2428 // align r13/rsi
2429 __ lea(rbx, at_bcp(BytesPerInt));
2430 __ andptr(rbx, -BytesPerInt);
2431 // load lo & hi
2432 __ movl(rcx, Address(rbx, BytesPerInt));
2433 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2434 __ bswapl(rcx);
2435 __ bswapl(rdx);
2436 // check against lo & hi
2437 __ cmpl(rax, rcx);
2438 __ jcc(Assembler::less, default_case);
2439 __ cmpl(rax, rdx);
2440 __ jcc(Assembler::greater, default_case);
2441 // lookup dispatch offset
2442 __ subl(rax, rcx);
2443 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2444 __ profile_switch_case(rax, rbx, rcx);
2445 // continue execution
2446 __ bind(continue_execution);
2447 __ bswapl(rdx);
2448 LP64_ONLY(__ movl2ptr(rdx, rdx));
2449 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2450 __ addptr(rbcp, rdx);
2451 __ dispatch_only(vtos, true);
2452 // handle default
2453 __ bind(default_case);
2454 __ profile_switch_default(rax);
2455 __ movl(rdx, Address(rbx, 0));
2456 __ jmp(continue_execution);
2457 }
2458
2459 void TemplateTable::lookupswitch() {
2460 transition(itos, itos);
2461 __ stop("lookupswitch bytecode should have been rewritten");
2462 }
2463
2464 void TemplateTable::fast_linearswitch() {
2465 transition(itos, vtos);
2466 Label loop_entry, loop, found, continue_execution;
2467 // bswap rax so we can avoid bswapping the table entries
2468 __ bswapl(rax);
2469 // align r13
2470 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2471 // this instruction (change offsets
2472 // below)
2473 __ andptr(rbx, -BytesPerInt);
2474 // set counter
2475 __ movl(rcx, Address(rbx, BytesPerInt));
2476 __ bswapl(rcx);
2477 __ jmpb(loop_entry);
2478 // table search
2479 __ bind(loop);
2480 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2481 __ jcc(Assembler::equal, found);
2482 __ bind(loop_entry);
2483 __ decrementl(rcx);
2484 __ jcc(Assembler::greaterEqual, loop);
2485 // default case
2486 __ profile_switch_default(rax);
2487 __ movl(rdx, Address(rbx, 0));
2488 __ jmp(continue_execution);
2489 // entry found -> get offset
2490 __ bind(found);
2491 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2492 __ profile_switch_case(rcx, rax, rbx);
2493 // continue execution
2494 __ bind(continue_execution);
2495 __ bswapl(rdx);
2496 __ movl2ptr(rdx, rdx);
2497 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2498 __ addptr(rbcp, rdx);
2499 __ dispatch_only(vtos, true);
2500 }
2501
2502 void TemplateTable::fast_binaryswitch() {
2503 transition(itos, vtos);
2504 // Implementation using the following core algorithm:
2505 //
2506 // int binary_search(int key, LookupswitchPair* array, int n) {
2507 // // Binary search according to "Methodik des Programmierens" by
2508 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2509 // int i = 0;
2510 // int j = n;
2511 // while (i+1 < j) {
2512 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2513 // // with Q: for all i: 0 <= i < n: key < a[i]
2514 // // where a stands for the array and assuming that the (inexisting)
2515 // // element a[n] is infinitely big.
2516 // int h = (i + j) >> 1;
2517 // // i < h < j
2518 // if (key < array[h].fast_match()) {
2519 // j = h;
2520 // } else {
2521 // i = h;
2522 // }
2523 // }
2524 // // R: a[i] <= key < a[i+1] or Q
2525 // // (i.e., if key is within array, i is the correct index)
2526 // return i;
2527 // }
2528
2529 // Register allocation
2530 const Register key = rax; // already set (tosca)
2531 const Register array = rbx;
2532 const Register i = rcx;
2533 const Register j = rdx;
2534 const Register h = rdi;
2535 const Register temp = rsi;
2536
2537 // Find array start
2538 NOT_LP64(__ save_bcp());
2539
2540 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2541 // get rid of this
2542 // instruction (change
2543 // offsets below)
2544 __ andptr(array, -BytesPerInt);
2545
2546 // Initialize i & j
2547 __ xorl(i, i); // i = 0;
2548 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2549
2550 // Convert j into native byteordering
2551 __ bswapl(j);
2552
2553 // And start
2554 Label entry;
2555 __ jmp(entry);
2556
2557 // binary search loop
2558 {
2559 Label loop;
2560 __ bind(loop);
2561 // int h = (i + j) >> 1;
2562 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2563 __ sarl(h, 1); // h = (i + j) >> 1;
2564 // if (key < array[h].fast_match()) {
2565 // j = h;
2566 // } else {
2567 // i = h;
2568 // }
2569 // Convert array[h].match to native byte-ordering before compare
2570 __ movl(temp, Address(array, h, Address::times_8));
2571 __ bswapl(temp);
2572 __ cmpl(key, temp);
2573 // j = h if (key < array[h].fast_match())
2574 __ cmov32(Assembler::less, j, h);
2575 // i = h if (key >= array[h].fast_match())
2576 __ cmov32(Assembler::greaterEqual, i, h);
2577 // while (i+1 < j)
2578 __ bind(entry);
2579 __ leal(h, Address(i, 1)); // i+1
2580 __ cmpl(h, j); // i+1 < j
2581 __ jcc(Assembler::less, loop);
2582 }
2583
2584 // end of binary search, result index is i (must check again!)
2585 Label default_case;
2586 // Convert array[i].match to native byte-ordering before compare
2587 __ movl(temp, Address(array, i, Address::times_8));
2588 __ bswapl(temp);
2589 __ cmpl(key, temp);
2590 __ jcc(Assembler::notEqual, default_case);
2591
2592 // entry found -> j = offset
2593 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2594 __ profile_switch_case(i, key, array);
2595 __ bswapl(j);
2596 LP64_ONLY(__ movslq(j, j));
2597
2598 NOT_LP64(__ restore_bcp());
2599 NOT_LP64(__ restore_locals()); // restore rdi
2600
2601 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2602 __ addptr(rbcp, j);
2603 __ dispatch_only(vtos, true);
2604
2605 // default case -> j = default offset
2606 __ bind(default_case);
2607 __ profile_switch_default(i);
2608 __ movl(j, Address(array, -2 * BytesPerInt));
2609 __ bswapl(j);
2610 LP64_ONLY(__ movslq(j, j));
2611
2612 NOT_LP64(__ restore_bcp());
2613 NOT_LP64(__ restore_locals());
2614
2615 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2616 __ addptr(rbcp, j);
2617 __ dispatch_only(vtos, true);
2618 }
2619
2620 void TemplateTable::_return(TosState state) {
2621 transition(state, state);
2622
2623 assert(_desc->calls_vm(),
2624 "inconsistent calls_vm information"); // call in remove_activation
2625
2626 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2627 assert(state == vtos, "only valid state");
2628 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2629 __ movptr(robj, aaddress(0));
2630 __ load_klass(rdi, robj);
2631 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2632 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2633 Label skip_register_finalizer;
2634 __ jcc(Assembler::zero, skip_register_finalizer);
2635
2636 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2637
2638 __ bind(skip_register_finalizer);
2639 }
2640
2641 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2642 Label no_safepoint;
2643 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2644 #ifdef _LP64
2645 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2646 #else
2647 const Register thread = rdi;
2648 __ get_thread(thread);
2649 __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2650 #endif
2651 __ jcc(Assembler::zero, no_safepoint);
2652 __ push(state);
2653 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2654 InterpreterRuntime::at_safepoint));
2655 __ pop(state);
2656 __ bind(no_safepoint);
2657 }
2658
2659 // Narrow result if state is itos but result type is smaller.
2660 // Need to narrow in the return bytecode rather than in generate_return_entry
2661 // since compiled code callers expect the result to already be narrowed.
2662 if (state == itos) {
2663 __ narrow(rax);
2664 }
2665 __ remove_activation(state, rbcp);
2666
2667 __ jmp(rbcp);
2668 }
2669
2670 // ----------------------------------------------------------------------------
2671 // Volatile variables demand their effects be made known to all CPU's
2672 // in order. Store buffers on most chips allow reads & writes to
2673 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2674 // without some kind of memory barrier (i.e., it's not sufficient that
2675 // the interpreter does not reorder volatile references, the hardware
2676 // also must not reorder them).
2677 //
2678 // According to the new Java Memory Model (JMM):
2679 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2680 // writes act as aquire & release, so:
2681 // (2) A read cannot let unrelated NON-volatile memory refs that
2682 // happen after the read float up to before the read. It's OK for
2683 // non-volatile memory refs that happen before the volatile read to
2684 // float down below it.
2685 // (3) Similar a volatile write cannot let unrelated NON-volatile
2686 // memory refs that happen BEFORE the write float down to after the
2687 // write. It's OK for non-volatile memory refs that happen after the
2688 // volatile write to float up before it.
2689 //
2690 // We only put in barriers around volatile refs (they are expensive),
2691 // not _between_ memory refs (that would require us to track the
2692 // flavor of the previous memory refs). Requirements (2) and (3)
2693 // require some barriers before volatile stores and after volatile
2694 // loads. These nearly cover requirement (1) but miss the
2695 // volatile-store-volatile-load case. This final case is placed after
2696 // volatile-stores although it could just as well go before
2697 // volatile-loads.
2698
2699 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2700 // Helper function to insert a is-volatile test and memory barrier
2701 if(!os::is_MP()) return; // Not needed on single CPU
2702 __ membar(order_constraint);
2703 }
2704
2705 void TemplateTable::resolve_cache_and_index(int byte_no,
2706 Register Rcache,
2707 Register index,
2708 size_t index_size) {
2709 const Register temp = rbx;
2710 assert_different_registers(Rcache, index, temp);
2711
2712 Label resolved;
2713
2714 Bytecodes::Code code = bytecode();
2715 switch (code) {
2716 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2717 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2718 default: break;
2719 }
2720
2721 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2722 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2723 __ cmpl(temp, code); // have we resolved this bytecode?
2724 __ jcc(Assembler::equal, resolved);
2725
2726 // resolve first time through
2727 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2728 __ movl(temp, code);
2729 __ call_VM(noreg, entry, temp);
2730 // Update registers with resolved info
2731 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2732 __ bind(resolved);
2733 }
2734
2735 // The cache and index registers must be set before call
2736 void TemplateTable::load_field_cp_cache_entry(Register obj,
2737 Register cache,
2738 Register index,
2739 Register off,
2740 Register flags,
2741 bool is_static = false) {
2742 assert_different_registers(cache, index, flags, off);
2743
2744 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2745 // Field offset
2746 __ movptr(off, Address(cache, index, Address::times_ptr,
2747 in_bytes(cp_base_offset +
2748 ConstantPoolCacheEntry::f2_offset())));
2749 // Flags
2750 __ movl(flags, Address(cache, index, Address::times_ptr,
2751 in_bytes(cp_base_offset +
2752 ConstantPoolCacheEntry::flags_offset())));
2753
2754 // klass overwrite register
2755 if (is_static) {
2756 __ movptr(obj, Address(cache, index, Address::times_ptr,
2757 in_bytes(cp_base_offset +
2758 ConstantPoolCacheEntry::f1_offset())));
2759 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2760 __ movptr(obj, Address(obj, mirror_offset));
2761 __ resolve_oop_handle(obj);
2762 }
2763 }
2764
2765 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2766 Register method,
2767 Register itable_index,
2768 Register flags,
2769 bool is_invokevirtual,
2770 bool is_invokevfinal, /*unused*/
2771 bool is_invokedynamic) {
2772 // setup registers
2773 const Register cache = rcx;
2774 const Register index = rdx;
2775 assert_different_registers(method, flags);
2776 assert_different_registers(method, cache, index);
2777 assert_different_registers(itable_index, flags);
2778 assert_different_registers(itable_index, cache, index);
2779 // determine constant pool cache field offsets
2780 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2781 const int method_offset = in_bytes(
2782 ConstantPoolCache::base_offset() +
2783 ((byte_no == f2_byte)
2784 ? ConstantPoolCacheEntry::f2_offset()
2785 : ConstantPoolCacheEntry::f1_offset()));
2786 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2787 ConstantPoolCacheEntry::flags_offset());
2788 // access constant pool cache fields
2789 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2790 ConstantPoolCacheEntry::f2_offset());
2791
2792 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2793 resolve_cache_and_index(byte_no, cache, index, index_size);
2794 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2795
2796 if (itable_index != noreg) {
2797 // pick up itable or appendix index from f2 also:
2798 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2799 }
2800 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2801 }
2802
2803 // The registers cache and index expected to be set before call.
2804 // Correct values of the cache and index registers are preserved.
2805 void TemplateTable::jvmti_post_field_access(Register cache,
2806 Register index,
2807 bool is_static,
2808 bool has_tos) {
2809 if (JvmtiExport::can_post_field_access()) {
2810 // Check to see if a field access watch has been set before we take
2811 // the time to call into the VM.
2812 Label L1;
2813 assert_different_registers(cache, index, rax);
2814 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2815 __ testl(rax,rax);
2816 __ jcc(Assembler::zero, L1);
2817
2818 // cache entry pointer
2819 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2820 __ shll(index, LogBytesPerWord);
2821 __ addptr(cache, index);
2822 if (is_static) {
2823 __ xorptr(rax, rax); // NULL object reference
2824 } else {
2825 __ pop(atos); // Get the object
2826 __ verify_oop(rax);
2827 __ push(atos); // Restore stack state
2828 }
2829 // rax,: object pointer or NULL
2830 // cache: cache entry pointer
2831 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2832 rax, cache);
2833 __ get_cache_and_index_at_bcp(cache, index, 1);
2834 __ bind(L1);
2835 }
2836 }
2837
2838 void TemplateTable::pop_and_check_object(Register r) {
2839 __ pop_ptr(r);
2840 __ null_check(r); // for field access must check obj.
2841 __ verify_oop(r);
2842 }
2843
2844 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2845 transition(vtos, vtos);
2846
2847 const Register cache = rcx;
2848 const Register index = rdx;
2849 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2850 const Register off = rbx;
2851 const Register flags = rax;
2852 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2853
2854 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2855 jvmti_post_field_access(cache, index, is_static, false);
2856 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2857
2858 if (!is_static) pop_and_check_object(obj);
2859
2860 const Address field(obj, off, Address::times_1, 0*wordSize);
2861 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2862
2863 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2864
2865 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2866 // Make sure we don't need to mask edx after the above shift
2867 assert(btos == 0, "change code, btos != 0");
2868
2869 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2870
2871 __ jcc(Assembler::notZero, notByte);
2872 // btos
2873 __ load_signed_byte(rax, field);
2874 __ push(btos);
2875 // Rewrite bytecode to be faster
2876 if (!is_static && rc == may_rewrite) {
2877 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2878 }
2879 __ jmp(Done);
2880
2881 __ bind(notByte);
2882 __ cmpl(flags, ztos);
2883 __ jcc(Assembler::notEqual, notBool);
2884
2885 // ztos (same code as btos)
2886 __ load_signed_byte(rax, field);
2887 __ push(ztos);
2888 // Rewrite bytecode to be faster
2889 if (!is_static && rc == may_rewrite) {
2890 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2891 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2892 }
2893 __ jmp(Done);
2894
2895 __ bind(notBool);
2896 __ cmpl(flags, atos);
2897 __ jcc(Assembler::notEqual, notObj);
2898 // atos
2899 do_oop_load(_masm, field, rax);
2900 __ push(atos);
2901 if (!is_static && rc == may_rewrite) {
2902 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2903 }
2904 __ jmp(Done);
2905
2906 __ bind(notObj);
2907 __ cmpl(flags, itos);
2908 __ jcc(Assembler::notEqual, notInt);
2909 // itos
2910 __ movl(rax, field);
2911 __ push(itos);
2912 // Rewrite bytecode to be faster
2913 if (!is_static && rc == may_rewrite) {
2914 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2915 }
2916 __ jmp(Done);
2917
2918 __ bind(notInt);
2919 __ cmpl(flags, ctos);
2920 __ jcc(Assembler::notEqual, notChar);
2921 // ctos
2922 __ load_unsigned_short(rax, field);
2923 __ push(ctos);
2924 // Rewrite bytecode to be faster
2925 if (!is_static && rc == may_rewrite) {
2926 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2927 }
2928 __ jmp(Done);
2929
2930 __ bind(notChar);
2931 __ cmpl(flags, stos);
2932 __ jcc(Assembler::notEqual, notShort);
2933 // stos
2934 __ load_signed_short(rax, field);
2935 __ push(stos);
2936 // Rewrite bytecode to be faster
2937 if (!is_static && rc == may_rewrite) {
2938 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2939 }
2940 __ jmp(Done);
2941
2942 __ bind(notShort);
2943 __ cmpl(flags, ltos);
2944 __ jcc(Assembler::notEqual, notLong);
2945 // ltos
2946
2947 #ifndef _LP64
2948 // Generate code as if volatile. There just aren't enough registers to
2949 // save that information and this code is faster than the test.
2950 __ fild_d(field); // Must load atomically
2951 __ subptr(rsp,2*wordSize); // Make space for store
2952 __ fistp_d(Address(rsp,0));
2953 __ pop(rax);
2954 __ pop(rdx);
2955 #else
2956 __ movq(rax, field);
2957 #endif
2958
2959 __ push(ltos);
2960 // Rewrite bytecode to be faster
2961 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2962 __ jmp(Done);
2963
2964 __ bind(notLong);
2965 __ cmpl(flags, ftos);
2966 __ jcc(Assembler::notEqual, notFloat);
2967 // ftos
2968
2969 __ load_float(field);
2970 __ push(ftos);
2971 // Rewrite bytecode to be faster
2972 if (!is_static && rc == may_rewrite) {
2973 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2974 }
2975 __ jmp(Done);
2976
2977 __ bind(notFloat);
2978 #ifdef ASSERT
2979 __ cmpl(flags, dtos);
2980 __ jcc(Assembler::notEqual, notDouble);
2981 #endif
2982 // dtos
2983 __ load_double(field);
2984 __ push(dtos);
2985 // Rewrite bytecode to be faster
2986 if (!is_static && rc == may_rewrite) {
2987 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2988 }
2989 #ifdef ASSERT
2990 __ jmp(Done);
2991
2992
2993 __ bind(notDouble);
2994 __ stop("Bad state");
2995 #endif
2996
2997 __ bind(Done);
2998 // [jk] not needed currently
2999 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3000 // Assembler::LoadStore));
3001 }
3002
3003 void TemplateTable::getfield(int byte_no) {
3004 getfield_or_static(byte_no, false);
3005 }
3006
3007 void TemplateTable::nofast_getfield(int byte_no) {
3008 getfield_or_static(byte_no, false, may_not_rewrite);
3009 }
3010
3011 void TemplateTable::getstatic(int byte_no) {
3012 getfield_or_static(byte_no, true);
3013 }
3014
3015
3016 // The registers cache and index expected to be set before call.
3017 // The function may destroy various registers, just not the cache and index registers.
3018 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3019
3020 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3021 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3022 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3023 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3024
3025 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3026
3027 if (JvmtiExport::can_post_field_modification()) {
3028 // Check to see if a field modification watch has been set before
3029 // we take the time to call into the VM.
3030 Label L1;
3031 assert_different_registers(cache, index, rax);
3032 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3033 __ testl(rax, rax);
3034 __ jcc(Assembler::zero, L1);
3035
3036 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3037
3038
3039 if (is_static) {
3040 // Life is simple. Null out the object pointer.
3041 __ xorl(RBX, RBX);
3042
3043 } else {
3044 // Life is harder. The stack holds the value on top, followed by
3045 // the object. We don't know the size of the value, though; it
3046 // could be one or two words depending on its type. As a result,
3047 // we must find the type to determine where the object is.
3048 #ifndef _LP64
3049 Label two_word, valsize_known;
3050 #endif
3051 __ movl(RCX, Address(robj, RDX,
3052 Address::times_ptr,
3053 in_bytes(cp_base_offset +
3054 ConstantPoolCacheEntry::flags_offset())));
3055 NOT_LP64(__ mov(rbx, rsp));
3056 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3057
3058 // Make sure we don't need to mask rcx after the above shift
3059 ConstantPoolCacheEntry::verify_tos_state_shift();
3060 #ifdef _LP64
3061 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3062 __ cmpl(c_rarg3, ltos);
3063 __ cmovptr(Assembler::equal,
3064 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3065 __ cmpl(c_rarg3, dtos);
3066 __ cmovptr(Assembler::equal,
3067 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3068 #else
3069 __ cmpl(rcx, ltos);
3070 __ jccb(Assembler::equal, two_word);
3071 __ cmpl(rcx, dtos);
3072 __ jccb(Assembler::equal, two_word);
3073 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3074 __ jmpb(valsize_known);
3075
3076 __ bind(two_word);
3077 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3078
3079 __ bind(valsize_known);
3080 // setup object pointer
3081 __ movptr(rbx, Address(rbx, 0));
3082 #endif
3083 }
3084 // cache entry pointer
3085 __ addptr(robj, in_bytes(cp_base_offset));
3086 __ shll(RDX, LogBytesPerWord);
3087 __ addptr(robj, RDX);
3088 // object (tos)
3089 __ mov(RCX, rsp);
3090 // c_rarg1: object pointer set up above (NULL if static)
3091 // c_rarg2: cache entry pointer
3092 // c_rarg3: jvalue object on the stack
3093 __ call_VM(noreg,
3094 CAST_FROM_FN_PTR(address,
3095 InterpreterRuntime::post_field_modification),
3096 RBX, robj, RCX);
3097 __ get_cache_and_index_at_bcp(cache, index, 1);
3098 __ bind(L1);
3099 }
3100 }
3101
3102 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3103 transition(vtos, vtos);
3104
3105 const Register cache = rcx;
3106 const Register index = rdx;
3107 const Register obj = rcx;
3108 const Register off = rbx;
3109 const Register flags = rax;
3110 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3111
3112 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3113 jvmti_post_field_mod(cache, index, is_static);
3114 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3115
3116 // [jk] not needed currently
3117 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3118 // Assembler::StoreStore));
3119
3120 Label notVolatile, Done;
3121 __ movl(rdx, flags);
3122 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3123 __ andl(rdx, 0x1);
3124
3125 // field addresses
3126 const Address field(obj, off, Address::times_1, 0*wordSize);
3127 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3128
3129 Label notByte, notBool, notInt, notShort, notChar,
3130 notLong, notFloat, notObj, notDouble;
3131
3132 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3133
3134 assert(btos == 0, "change code, btos != 0");
3135 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3136 __ jcc(Assembler::notZero, notByte);
3137
3138 // btos
3139 {
3140 __ pop(btos);
3141 if (!is_static) pop_and_check_object(obj);
3142 __ movb(field, rax);
3143 if (!is_static && rc == may_rewrite) {
3144 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3145 }
3146 __ jmp(Done);
3147 }
3148
3149 __ bind(notByte);
3150 __ cmpl(flags, ztos);
3151 __ jcc(Assembler::notEqual, notBool);
3152
3153 // ztos
3154 {
3155 __ pop(ztos);
3156 if (!is_static) pop_and_check_object(obj);
3157 __ andl(rax, 0x1);
3158 __ movb(field, rax);
3159 if (!is_static && rc == may_rewrite) {
3160 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3161 }
3162 __ jmp(Done);
3163 }
3164
3165 __ bind(notBool);
3166 __ cmpl(flags, atos);
3167 __ jcc(Assembler::notEqual, notObj);
3168
3169 // atos
3170 {
3171 __ pop(atos);
3172 if (!is_static) pop_and_check_object(obj);
3173 // Store into the field
3174 do_oop_store(_masm, field, rax);
3175 if (!is_static && rc == may_rewrite) {
3176 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3177 }
3178 __ jmp(Done);
3179 }
3180
3181 __ bind(notObj);
3182 __ cmpl(flags, itos);
3183 __ jcc(Assembler::notEqual, notInt);
3184
3185 // itos
3186 {
3187 __ pop(itos);
3188 if (!is_static) pop_and_check_object(obj);
3189 __ movl(field, rax);
3190 if (!is_static && rc == may_rewrite) {
3191 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3192 }
3193 __ jmp(Done);
3194 }
3195
3196 __ bind(notInt);
3197 __ cmpl(flags, ctos);
3198 __ jcc(Assembler::notEqual, notChar);
3199
3200 // ctos
3201 {
3202 __ pop(ctos);
3203 if (!is_static) pop_and_check_object(obj);
3204 __ movw(field, rax);
3205 if (!is_static && rc == may_rewrite) {
3206 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3207 }
3208 __ jmp(Done);
3209 }
3210
3211 __ bind(notChar);
3212 __ cmpl(flags, stos);
3213 __ jcc(Assembler::notEqual, notShort);
3214
3215 // stos
3216 {
3217 __ pop(stos);
3218 if (!is_static) pop_and_check_object(obj);
3219 __ movw(field, rax);
3220 if (!is_static && rc == may_rewrite) {
3221 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3222 }
3223 __ jmp(Done);
3224 }
3225
3226 __ bind(notShort);
3227 __ cmpl(flags, ltos);
3228 __ jcc(Assembler::notEqual, notLong);
3229
3230 // ltos
3231 #ifdef _LP64
3232 {
3233 __ pop(ltos);
3234 if (!is_static) pop_and_check_object(obj);
3235 __ movq(field, rax);
3236 if (!is_static && rc == may_rewrite) {
3237 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3238 }
3239 __ jmp(Done);
3240 }
3241 #else
3242 {
3243 Label notVolatileLong;
3244 __ testl(rdx, rdx);
3245 __ jcc(Assembler::zero, notVolatileLong);
3246
3247 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3248 if (!is_static) pop_and_check_object(obj);
3249
3250 // Replace with real volatile test
3251 __ push(rdx);
3252 __ push(rax); // Must update atomically with FIST
3253 __ fild_d(Address(rsp,0)); // So load into FPU register
3254 __ fistp_d(field); // and put into memory atomically
3255 __ addptr(rsp, 2*wordSize);
3256 // volatile_barrier();
3257 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3258 Assembler::StoreStore));
3259 // Don't rewrite volatile version
3260 __ jmp(notVolatile);
3261
3262 __ bind(notVolatileLong);
3263
3264 __ pop(ltos); // overwrites rdx
3265 if (!is_static) pop_and_check_object(obj);
3266 __ movptr(hi, rdx);
3267 __ movptr(field, rax);
3268 // Don't rewrite to _fast_lputfield for potential volatile case.
3269 __ jmp(notVolatile);
3270 }
3271 #endif // _LP64
3272
3273 __ bind(notLong);
3274 __ cmpl(flags, ftos);
3275 __ jcc(Assembler::notEqual, notFloat);
3276
3277 // ftos
3278 {
3279 __ pop(ftos);
3280 if (!is_static) pop_and_check_object(obj);
3281 __ store_float(field);
3282 if (!is_static && rc == may_rewrite) {
3283 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3284 }
3285 __ jmp(Done);
3286 }
3287
3288 __ bind(notFloat);
3289 #ifdef ASSERT
3290 __ cmpl(flags, dtos);
3291 __ jcc(Assembler::notEqual, notDouble);
3292 #endif
3293
3294 // dtos
3295 {
3296 __ pop(dtos);
3297 if (!is_static) pop_and_check_object(obj);
3298 __ store_double(field);
3299 if (!is_static && rc == may_rewrite) {
3300 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3301 }
3302 }
3303
3304 #ifdef ASSERT
3305 __ jmp(Done);
3306
3307 __ bind(notDouble);
3308 __ stop("Bad state");
3309 #endif
3310
3311 __ bind(Done);
3312
3313 // Check for volatile store
3314 __ testl(rdx, rdx);
3315 __ jcc(Assembler::zero, notVolatile);
3316 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3317 Assembler::StoreStore));
3318 __ bind(notVolatile);
3319 }
3320
3321 void TemplateTable::putfield(int byte_no) {
3322 putfield_or_static(byte_no, false);
3323 }
3324
3325 void TemplateTable::nofast_putfield(int byte_no) {
3326 putfield_or_static(byte_no, false, may_not_rewrite);
3327 }
3328
3329 void TemplateTable::putstatic(int byte_no) {
3330 putfield_or_static(byte_no, true);
3331 }
3332
3333 void TemplateTable::jvmti_post_fast_field_mod() {
3334
3335 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3336
3337 if (JvmtiExport::can_post_field_modification()) {
3338 // Check to see if a field modification watch has been set before
3339 // we take the time to call into the VM.
3340 Label L2;
3341 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3342 __ testl(scratch, scratch);
3343 __ jcc(Assembler::zero, L2);
3344 __ pop_ptr(rbx); // copy the object pointer from tos
3345 __ verify_oop(rbx);
3346 __ push_ptr(rbx); // put the object pointer back on tos
3347 // Save tos values before call_VM() clobbers them. Since we have
3348 // to do it for every data type, we use the saved values as the
3349 // jvalue object.
3350 switch (bytecode()) { // load values into the jvalue object
3351 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3352 case Bytecodes::_fast_bputfield: // fall through
3353 case Bytecodes::_fast_zputfield: // fall through
3354 case Bytecodes::_fast_sputfield: // fall through
3355 case Bytecodes::_fast_cputfield: // fall through
3356 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3357 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3358 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3359 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3360
3361 default:
3362 ShouldNotReachHere();
3363 }
3364 __ mov(scratch, rsp); // points to jvalue on the stack
3365 // access constant pool cache entry
3366 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3367 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3368 __ verify_oop(rbx);
3369 // rbx: object pointer copied above
3370 // c_rarg2: cache entry pointer
3371 // c_rarg3: jvalue object on the stack
3372 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3373 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3374
3375 switch (bytecode()) { // restore tos values
3376 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3377 case Bytecodes::_fast_bputfield: // fall through
3378 case Bytecodes::_fast_zputfield: // fall through
3379 case Bytecodes::_fast_sputfield: // fall through
3380 case Bytecodes::_fast_cputfield: // fall through
3381 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3382 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3383 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3384 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3385 default: break;
3386 }
3387 __ bind(L2);
3388 }
3389 }
3390
3391 void TemplateTable::fast_storefield(TosState state) {
3392 transition(state, vtos);
3393
3394 ByteSize base = ConstantPoolCache::base_offset();
3395
3396 jvmti_post_fast_field_mod();
3397
3398 // access constant pool cache
3399 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3400
3401 // test for volatile with rdx but rdx is tos register for lputfield.
3402 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3403 in_bytes(base +
3404 ConstantPoolCacheEntry::flags_offset())));
3405
3406 // replace index with field offset from cache entry
3407 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3408 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3409
3410 // [jk] not needed currently
3411 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3412 // Assembler::StoreStore));
3413
3414 Label notVolatile;
3415 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3416 __ andl(rdx, 0x1);
3417
3418 // Get object from stack
3419 pop_and_check_object(rcx);
3420
3421 // field address
3422 const Address field(rcx, rbx, Address::times_1);
3423
3424 // access field
3425 switch (bytecode()) {
3426 case Bytecodes::_fast_aputfield:
3427 do_oop_store(_masm, field, rax);
3428 break;
3429 case Bytecodes::_fast_lputfield:
3430 #ifdef _LP64
3431 __ movq(field, rax);
3432 #else
3433 __ stop("should not be rewritten");
3434 #endif
3435 break;
3436 case Bytecodes::_fast_iputfield:
3437 __ movl(field, rax);
3438 break;
3439 case Bytecodes::_fast_zputfield:
3440 __ andl(rax, 0x1); // boolean is true if LSB is 1
3441 // fall through to bputfield
3442 case Bytecodes::_fast_bputfield:
3443 __ movb(field, rax);
3444 break;
3445 case Bytecodes::_fast_sputfield:
3446 // fall through
3447 case Bytecodes::_fast_cputfield:
3448 __ movw(field, rax);
3449 break;
3450 case Bytecodes::_fast_fputfield:
3451 __ store_float(field);
3452 break;
3453 case Bytecodes::_fast_dputfield:
3454 __ store_double(field);
3455 break;
3456 default:
3457 ShouldNotReachHere();
3458 }
3459
3460 // Check for volatile store
3461 __ testl(rdx, rdx);
3462 __ jcc(Assembler::zero, notVolatile);
3463 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3464 Assembler::StoreStore));
3465 __ bind(notVolatile);
3466 }
3467
3468 void TemplateTable::fast_accessfield(TosState state) {
3469 transition(atos, state);
3470
3471 // Do the JVMTI work here to avoid disturbing the register state below
3472 if (JvmtiExport::can_post_field_access()) {
3473 // Check to see if a field access watch has been set before we
3474 // take the time to call into the VM.
3475 Label L1;
3476 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3477 __ testl(rcx, rcx);
3478 __ jcc(Assembler::zero, L1);
3479 // access constant pool cache entry
3480 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3481 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3482 __ verify_oop(rax);
3483 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3484 LP64_ONLY(__ mov(c_rarg1, rax));
3485 // c_rarg1: object pointer copied above
3486 // c_rarg2: cache entry pointer
3487 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3488 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3489 __ pop_ptr(rax); // restore object pointer
3490 __ bind(L1);
3491 }
3492
3493 // access constant pool cache
3494 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3495 // replace index with field offset from cache entry
3496 // [jk] not needed currently
3497 // if (os::is_MP()) {
3498 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3499 // in_bytes(ConstantPoolCache::base_offset() +
3500 // ConstantPoolCacheEntry::flags_offset())));
3501 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3502 // __ andl(rdx, 0x1);
3503 // }
3504 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3505 in_bytes(ConstantPoolCache::base_offset() +
3506 ConstantPoolCacheEntry::f2_offset())));
3507
3508 // rax: object
3509 __ verify_oop(rax);
3510 __ null_check(rax);
3511 Address field(rax, rbx, Address::times_1);
3512
3513 // access field
3514 switch (bytecode()) {
3515 case Bytecodes::_fast_agetfield:
3516 do_oop_load(_masm, field, rax);
3517 __ verify_oop(rax);
3518 break;
3519 case Bytecodes::_fast_lgetfield:
3520 #ifdef _LP64
3521 __ movq(rax, field);
3522 #else
3523 __ stop("should not be rewritten");
3524 #endif
3525 break;
3526 case Bytecodes::_fast_igetfield:
3527 __ movl(rax, field);
3528 break;
3529 case Bytecodes::_fast_bgetfield:
3530 __ movsbl(rax, field);
3531 break;
3532 case Bytecodes::_fast_sgetfield:
3533 __ load_signed_short(rax, field);
3534 break;
3535 case Bytecodes::_fast_cgetfield:
3536 __ load_unsigned_short(rax, field);
3537 break;
3538 case Bytecodes::_fast_fgetfield:
3539 __ load_float(field);
3540 break;
3541 case Bytecodes::_fast_dgetfield:
3542 __ load_double(field);
3543 break;
3544 default:
3545 ShouldNotReachHere();
3546 }
3547 // [jk] not needed currently
3548 // if (os::is_MP()) {
3549 // Label notVolatile;
3550 // __ testl(rdx, rdx);
3551 // __ jcc(Assembler::zero, notVolatile);
3552 // __ membar(Assembler::LoadLoad);
3553 // __ bind(notVolatile);
3554 //};
3555 }
3556
3557 void TemplateTable::fast_xaccess(TosState state) {
3558 transition(vtos, state);
3559
3560 // get receiver
3561 __ movptr(rax, aaddress(0));
3562 // access constant pool cache
3563 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3564 __ movptr(rbx,
3565 Address(rcx, rdx, Address::times_ptr,
3566 in_bytes(ConstantPoolCache::base_offset() +
3567 ConstantPoolCacheEntry::f2_offset())));
3568 // make sure exception is reported in correct bcp range (getfield is
3569 // next instruction)
3570 __ increment(rbcp);
3571 __ null_check(rax);
3572 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3573 switch (state) {
3574 case itos:
3575 __ movl(rax, field);
3576 break;
3577 case atos:
3578 do_oop_load(_masm, field, rax);
3579 __ verify_oop(rax);
3580 break;
3581 case ftos:
3582 __ load_float(field);
3583 break;
3584 default:
3585 ShouldNotReachHere();
3586 }
3587
3588 // [jk] not needed currently
3589 // if (os::is_MP()) {
3590 // Label notVolatile;
3591 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3592 // in_bytes(ConstantPoolCache::base_offset() +
3593 // ConstantPoolCacheEntry::flags_offset())));
3594 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3595 // __ testl(rdx, 0x1);
3596 // __ jcc(Assembler::zero, notVolatile);
3597 // __ membar(Assembler::LoadLoad);
3598 // __ bind(notVolatile);
3599 // }
3600
3601 __ decrement(rbcp);
3602 }
3603
3604 //-----------------------------------------------------------------------------
3605 // Calls
3606
3607 void TemplateTable::count_calls(Register method, Register temp) {
3608 // implemented elsewhere
3609 ShouldNotReachHere();
3610 }
3611
3612 void TemplateTable::prepare_invoke(int byte_no,
3613 Register method, // linked method (or i-klass)
3614 Register index, // itable index, MethodType, etc.
3615 Register recv, // if caller wants to see it
3616 Register flags // if caller wants to test it
3617 ) {
3618 // determine flags
3619 const Bytecodes::Code code = bytecode();
3620 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3621 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3622 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3623 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3624 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3625 const bool load_receiver = (recv != noreg);
3626 const bool save_flags = (flags != noreg);
3627 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3628 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3629 assert(flags == noreg || flags == rdx, "");
3630 assert(recv == noreg || recv == rcx, "");
3631
3632 // setup registers & access constant pool cache
3633 if (recv == noreg) recv = rcx;
3634 if (flags == noreg) flags = rdx;
3635 assert_different_registers(method, index, recv, flags);
3636
3637 // save 'interpreter return address'
3638 __ save_bcp();
3639
3640 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3641
3642 // maybe push appendix to arguments (just before return address)
3643 if (is_invokedynamic || is_invokehandle) {
3644 Label L_no_push;
3645 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3646 __ jcc(Assembler::zero, L_no_push);
3647 // Push the appendix as a trailing parameter.
3648 // This must be done before we get the receiver,
3649 // since the parameter_size includes it.
3650 __ push(rbx);
3651 __ mov(rbx, index);
3652 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3653 __ load_resolved_reference_at_index(index, rbx);
3654 __ pop(rbx);
3655 __ push(index); // push appendix (MethodType, CallSite, etc.)
3656 __ bind(L_no_push);
3657 }
3658
3659 // load receiver if needed (after appendix is pushed so parameter size is correct)
3660 // Note: no return address pushed yet
3661 if (load_receiver) {
3662 __ movl(recv, flags);
3663 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3664 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3665 const int receiver_is_at_end = -1; // back off one slot to get receiver
3666 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3667 __ movptr(recv, recv_addr);
3668 __ verify_oop(recv);
3669 }
3670
3671 if (save_flags) {
3672 __ movl(rbcp, flags);
3673 }
3674
3675 // compute return type
3676 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3677 // Make sure we don't need to mask flags after the above shift
3678 ConstantPoolCacheEntry::verify_tos_state_shift();
3679 // load return address
3680 {
3681 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3682 ExternalAddress table(table_addr);
3683 LP64_ONLY(__ lea(rscratch1, table));
3684 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3685 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3686 }
3687
3688 // push return address
3689 __ push(flags);
3690
3691 // Restore flags value from the constant pool cache, and restore rsi
3692 // for later null checks. r13 is the bytecode pointer
3693 if (save_flags) {
3694 __ movl(flags, rbcp);
3695 __ restore_bcp();
3696 }
3697 }
3698
3699 void TemplateTable::invokevirtual_helper(Register index,
3700 Register recv,
3701 Register flags) {
3702 // Uses temporary registers rax, rdx
3703 assert_different_registers(index, recv, rax, rdx);
3704 assert(index == rbx, "");
3705 assert(recv == rcx, "");
3706
3707 // Test for an invoke of a final method
3708 Label notFinal;
3709 __ movl(rax, flags);
3710 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3711 __ jcc(Assembler::zero, notFinal);
3712
3713 const Register method = index; // method must be rbx
3714 assert(method == rbx,
3715 "Method* must be rbx for interpreter calling convention");
3716
3717 // do the call - the index is actually the method to call
3718 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3719
3720 // It's final, need a null check here!
3721 __ null_check(recv);
3722
3723 // profile this call
3724 __ profile_final_call(rax);
3725 __ profile_arguments_type(rax, method, rbcp, true);
3726
3727 __ jump_from_interpreted(method, rax);
3728
3729 __ bind(notFinal);
3730
3731 // get receiver klass
3732 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3733 __ load_klass(rax, recv);
3734
3735 // profile this call
3736 __ profile_virtual_call(rax, rlocals, rdx);
3737 // get target Method* & entry point
3738 __ lookup_virtual_method(rax, index, method);
3739 __ profile_called_method(method, rdx, rbcp);
3740
3741 __ profile_arguments_type(rdx, method, rbcp, true);
3742 __ jump_from_interpreted(method, rdx);
3743 }
3744
3745 void TemplateTable::invokevirtual(int byte_no) {
3746 transition(vtos, vtos);
3747 assert(byte_no == f2_byte, "use this argument");
3748 prepare_invoke(byte_no,
3749 rbx, // method or vtable index
3750 noreg, // unused itable index
3751 rcx, rdx); // recv, flags
3752
3753 // rbx: index
3754 // rcx: receiver
3755 // rdx: flags
3756
3757 invokevirtual_helper(rbx, rcx, rdx);
3758 }
3759
3760 void TemplateTable::invokespecial(int byte_no) {
3761 transition(vtos, vtos);
3762 assert(byte_no == f1_byte, "use this argument");
3763 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3764 rcx); // get receiver also for null check
3765 __ verify_oop(rcx);
3766 __ null_check(rcx);
3767 // do the call
3768 __ profile_call(rax);
3769 __ profile_arguments_type(rax, rbx, rbcp, false);
3770 __ jump_from_interpreted(rbx, rax);
3771 }
3772
3773 void TemplateTable::invokestatic(int byte_no) {
3774 transition(vtos, vtos);
3775 assert(byte_no == f1_byte, "use this argument");
3776 prepare_invoke(byte_no, rbx); // get f1 Method*
3777 // do the call
3778 __ profile_call(rax);
3779 __ profile_arguments_type(rax, rbx, rbcp, false);
3780 __ jump_from_interpreted(rbx, rax);
3781 }
3782
3783
3784 void TemplateTable::fast_invokevfinal(int byte_no) {
3785 transition(vtos, vtos);
3786 assert(byte_no == f2_byte, "use this argument");
3787 __ stop("fast_invokevfinal not used on x86");
3788 }
3789
3790
3791 void TemplateTable::invokeinterface(int byte_no) {
3792 transition(vtos, vtos);
3793 assert(byte_no == f1_byte, "use this argument");
3794 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
3795 rcx, rdx); // recv, flags
3796
3797 // rax: reference klass (from f1)
3798 // rbx: method (from f2)
3799 // rcx: receiver
3800 // rdx: flags
3801
3802 // Special case of invokeinterface called for virtual method of
3803 // java.lang.Object. See cpCacheOop.cpp for details.
3804 // This code isn't produced by javac, but could be produced by
3805 // another compliant java compiler.
3806 Label notMethod;
3807 __ movl(rlocals, rdx);
3808 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3809
3810 __ jcc(Assembler::zero, notMethod);
3811
3812 invokevirtual_helper(rbx, rcx, rdx);
3813 __ bind(notMethod);
3814
3815 // Get receiver klass into rdx - also a null check
3816 __ restore_locals(); // restore r14
3817 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3818 __ load_klass(rdx, rcx);
3819
3820 Label no_such_interface, no_such_method;
3821
3822 // Preserve method for throw_AbstractMethodErrorVerbose.
3823 __ mov(rcx, rbx);
3824 // Receiver subtype check against REFC.
3825 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3826 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3827 rdx, rax, noreg,
3828 // outputs: scan temp. reg, scan temp. reg
3829 rbcp, rlocals,
3830 no_such_interface,
3831 /*return_method=*/false);
3832
3833 // profile this call
3834 __ restore_bcp(); // rbcp was destroyed by receiver type check
3835 __ profile_virtual_call(rdx, rbcp, rlocals);
3836
3837 // Get declaring interface class from method, and itable index
3838 __ movptr(rax, Address(rbx, Method::const_offset()));
3839 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3840 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3841 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3842 __ subl(rbx, Method::itable_index_max);
3843 __ negl(rbx);
3844
3845 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3846 __ mov(rlocals, rdx);
3847 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3848 rlocals, rax, rbx,
3849 // outputs: method, scan temp. reg
3850 rbx, rbcp,
3851 no_such_interface);
3852
3853 // rbx: Method* to call
3854 // rcx: receiver
3855 // Check for abstract method error
3856 // Note: This should be done more efficiently via a throw_abstract_method_error
3857 // interpreter entry point and a conditional jump to it in case of a null
3858 // method.
3859 __ testptr(rbx, rbx);
3860 __ jcc(Assembler::zero, no_such_method);
3861
3862 __ profile_called_method(rbx, rbcp, rdx);
3863 __ profile_arguments_type(rdx, rbx, rbcp, true);
3864
3865 // do the call
3866 // rcx: receiver
3867 // rbx,: Method*
3868 __ jump_from_interpreted(rbx, rdx);
3869 __ should_not_reach_here();
3870
3871 // exception handling code follows...
3872 // note: must restore interpreter registers to canonical
3873 // state for exception handling to work correctly!
3874
3875 __ bind(no_such_method);
3876 // throw exception
3877 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3878 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3879 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3880 // Pass arguments for generating a verbose error message.
3881 #ifdef _LP64
3882 Register recvKlass = c_rarg1;
3883 Register method = c_rarg2;
3884 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3885 if (method != rcx) { __ movq(method, rcx); }
3886 #else
3887 Register recvKlass = rdx;
3888 Register method = rcx;
3889 #endif
3890 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3891 recvKlass, method);
3892 // The call_VM checks for exception, so we should never return here.
3893 __ should_not_reach_here();
3894
3895 __ bind(no_such_interface);
3896 // throw exception
3897 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3898 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3899 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3900 // Pass arguments for generating a verbose error message.
3901 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3902 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3903 recvKlass, rax);
3904 // the call_VM checks for exception, so we should never return here.
3905 __ should_not_reach_here();
3906 }
3907
3908 void TemplateTable::invokehandle(int byte_no) {
3909 transition(vtos, vtos);
3910 assert(byte_no == f1_byte, "use this argument");
3911 const Register rbx_method = rbx;
3912 const Register rax_mtype = rax;
3913 const Register rcx_recv = rcx;
3914 const Register rdx_flags = rdx;
3915
3916 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3917 __ verify_method_ptr(rbx_method);
3918 __ verify_oop(rcx_recv);
3919 __ null_check(rcx_recv);
3920
3921 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3922 // rbx: MH.invokeExact_MT method (from f2)
3923
3924 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3925
3926 // FIXME: profile the LambdaForm also
3927 __ profile_final_call(rax);
3928 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3929
3930 __ jump_from_interpreted(rbx_method, rdx);
3931 }
3932
3933 void TemplateTable::invokedynamic(int byte_no) {
3934 transition(vtos, vtos);
3935 assert(byte_no == f1_byte, "use this argument");
3936
3937 const Register rbx_method = rbx;
3938 const Register rax_callsite = rax;
3939
3940 prepare_invoke(byte_no, rbx_method, rax_callsite);
3941
3942 // rax: CallSite object (from cpool->resolved_references[f1])
3943 // rbx: MH.linkToCallSite method (from f2)
3944
3945 // Note: rax_callsite is already pushed by prepare_invoke
3946
3947 // %%% should make a type profile for any invokedynamic that takes a ref argument
3948 // profile this call
3949 __ profile_call(rbcp);
3950 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3951
3952 __ verify_oop(rax_callsite);
3953
3954 __ jump_from_interpreted(rbx_method, rdx);
3955 }
3956
3957 //-----------------------------------------------------------------------------
3958 // Allocation
3959
3960 void TemplateTable::_new() {
3961 transition(vtos, atos);
3962 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3963 Label slow_case;
3964 Label slow_case_no_pop;
3965 Label done;
3966 Label initialize_header;
3967 Label initialize_object; // including clearing the fields
3968
3969 __ get_cpool_and_tags(rcx, rax);
3970
3971 // Make sure the class we're about to instantiate has been resolved.
3972 // This is done before loading InstanceKlass to be consistent with the order
3973 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3974 const int tags_offset = Array<u1>::base_offset_in_bytes();
3975 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3976 __ jcc(Assembler::notEqual, slow_case_no_pop);
3977
3978 // get InstanceKlass
3979 __ load_resolved_klass_at_index(rcx, rdx, rcx);
3980 __ push(rcx); // save the contexts of klass for initializing the header
3981
3982 // make sure klass is initialized & doesn't have finalizer
3983 // make sure klass is fully initialized
3984 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3985 __ jcc(Assembler::notEqual, slow_case);
3986
3987 // get instance_size in InstanceKlass (scaled to a count of bytes)
3988 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3989 // test to see if it has a finalizer or is malformed in some way
3990 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3991 __ jcc(Assembler::notZero, slow_case);
3992
3993 // Allocate the instance:
3994 // If TLAB is enabled:
3995 // Try to allocate in the TLAB.
3996 // If fails, go to the slow path.
3997 // Else If inline contiguous allocations are enabled:
3998 // Try to allocate in eden.
3999 // If fails due to heap end, go to slow path.
4000 //
4001 // If TLAB is enabled OR inline contiguous is enabled:
4002 // Initialize the allocation.
4003 // Exit.
4004 //
4005 // Go to slow path.
4006
4007 const bool allow_shared_alloc =
4008 Universe::heap()->supports_inline_contig_alloc();
4009
4010 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4011 #ifndef _LP64
4012 if (UseTLAB || allow_shared_alloc) {
4013 __ get_thread(thread);
4014 }
4015 #endif // _LP64
4016
4017 if (UseTLAB) {
4018 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4019 __ lea(rbx, Address(rax, rdx, Address::times_1));
4020 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4021 __ jcc(Assembler::above, slow_case);
4022 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4023 if (ZeroTLAB) {
4024 // the fields have been already cleared
4025 __ jmp(initialize_header);
4026 } else {
4027 // initialize both the header and fields
4028 __ jmp(initialize_object);
4029 }
4030 } else {
4031 // Allocation in the shared Eden, if allowed.
4032 //
4033 // rdx: instance size in bytes
4034 if (allow_shared_alloc) {
4035 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4036 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4037
4038 Label retry;
4039 __ bind(retry);
4040 __ movptr(rax, heap_top);
4041 __ lea(rbx, Address(rax, rdx, Address::times_1));
4042 __ cmpptr(rbx, heap_end);
4043 __ jcc(Assembler::above, slow_case);
4044
4045 // Compare rax, with the top addr, and if still equal, store the new
4046 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4047 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4048 //
4049 // rax,: object begin
4050 // rbx,: object end
4051 // rdx: instance size in bytes
4052 __ locked_cmpxchgptr(rbx, heap_top);
4053
4054 // if someone beat us on the allocation, try again, otherwise continue
4055 __ jcc(Assembler::notEqual, retry);
4056
4057 __ incr_allocated_bytes(thread, rdx, 0);
4058 }
4059 }
4060
4061 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4062 // there is an initialize need. Otherwise, skip and go to the slow path.
4063 if (UseTLAB || allow_shared_alloc) {
4064 // The object is initialized before the header. If the object size is
4065 // zero, go directly to the header initialization.
4066 __ bind(initialize_object);
4067 __ decrement(rdx, sizeof(oopDesc));
4068 __ jcc(Assembler::zero, initialize_header);
4069
4070 // Initialize topmost object field, divide rdx by 8, check if odd and
4071 // test if zero.
4072 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4073 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4074
4075 // rdx must have been multiple of 8
4076 #ifdef ASSERT
4077 // make sure rdx was multiple of 8
4078 Label L;
4079 // Ignore partial flag stall after shrl() since it is debug VM
4080 __ jccb(Assembler::carryClear, L);
4081 __ stop("object size is not multiple of 2 - adjust this code");
4082 __ bind(L);
4083 // rdx must be > 0, no extra check needed here
4084 #endif
4085
4086 // initialize remaining object fields: rdx was a multiple of 8
4087 { Label loop;
4088 __ bind(loop);
4089 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4090 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4091 __ decrement(rdx);
4092 __ jcc(Assembler::notZero, loop);
4093 }
4094
4095 // initialize object header only.
4096 __ bind(initialize_header);
4097 if (UseBiasedLocking) {
4098 __ pop(rcx); // get saved klass back in the register.
4099 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4100 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4101 } else {
4102 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4103 (intptr_t)markOopDesc::prototype()); // header
4104 __ pop(rcx); // get saved klass back in the register.
4105 }
4106 #ifdef _LP64
4107 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4108 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4109 #endif
4110 __ store_klass(rax, rcx); // klass
4111
4112 {
4113 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4114 // Trigger dtrace event for fastpath
4115 __ push(atos);
4116 __ call_VM_leaf(
4117 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4118 __ pop(atos);
4119 }
4120
4121 __ jmp(done);
4122 }
4123
4124 // slow case
4125 __ bind(slow_case);
4126 __ pop(rcx); // restore stack pointer to what it was when we came in.
4127 __ bind(slow_case_no_pop);
4128
4129 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4130 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4131
4132 __ get_constant_pool(rarg1);
4133 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4134 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4135 __ verify_oop(rax);
4136
4137 // continue
4138 __ bind(done);
4139 }
4140
4141 void TemplateTable::newarray() {
4142 transition(itos, atos);
4143 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4144 __ load_unsigned_byte(rarg1, at_bcp(1));
4145 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4146 rarg1, rax);
4147 }
4148
4149 void TemplateTable::anewarray() {
4150 transition(itos, atos);
4151
4152 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4153 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4154
4155 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4156 __ get_constant_pool(rarg1);
4157 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4158 rarg1, rarg2, rax);
4159 }
4160
4161 void TemplateTable::arraylength() {
4162 transition(atos, itos);
4163 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4164 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4165 }
4166
4167 void TemplateTable::checkcast() {
4168 transition(atos, atos);
4169 Label done, is_null, ok_is_subtype, quicked, resolved;
4170 __ testptr(rax, rax); // object is in rax
4171 __ jcc(Assembler::zero, is_null);
4172
4173 // Get cpool & tags index
4174 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4175 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4176 // See if bytecode has already been quicked
4177 __ cmpb(Address(rdx, rbx,
4178 Address::times_1,
4179 Array<u1>::base_offset_in_bytes()),
4180 JVM_CONSTANT_Class);
4181 __ jcc(Assembler::equal, quicked);
4182 __ push(atos); // save receiver for result, and for GC
4183 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4184
4185 // vm_result_2 has metadata result
4186 #ifndef _LP64
4187 // borrow rdi from locals
4188 __ get_thread(rdi);
4189 __ get_vm_result_2(rax, rdi);
4190 __ restore_locals();
4191 #else
4192 __ get_vm_result_2(rax, r15_thread);
4193 #endif
4194
4195 __ pop_ptr(rdx); // restore receiver
4196 __ jmpb(resolved);
4197
4198 // Get superklass in rax and subklass in rbx
4199 __ bind(quicked);
4200 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4201 __ load_resolved_klass_at_index(rcx, rbx, rax);
4202
4203 __ bind(resolved);
4204 __ load_klass(rbx, rdx);
4205
4206 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4207 // Superklass in rax. Subklass in rbx.
4208 __ gen_subtype_check(rbx, ok_is_subtype);
4209
4210 // Come here on failure
4211 __ push_ptr(rdx);
4212 // object is at TOS
4213 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4214
4215 // Come here on success
4216 __ bind(ok_is_subtype);
4217 __ mov(rax, rdx); // Restore object in rdx
4218
4219 // Collect counts on whether this check-cast sees NULLs a lot or not.
4220 if (ProfileInterpreter) {
4221 __ jmp(done);
4222 __ bind(is_null);
4223 __ profile_null_seen(rcx);
4224 } else {
4225 __ bind(is_null); // same as 'done'
4226 }
4227 __ bind(done);
4228 }
4229
4230 void TemplateTable::instanceof() {
4231 transition(atos, itos);
4232 Label done, is_null, ok_is_subtype, quicked, resolved;
4233 __ testptr(rax, rax);
4234 __ jcc(Assembler::zero, is_null);
4235
4236 // Get cpool & tags index
4237 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4238 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4239 // See if bytecode has already been quicked
4240 __ cmpb(Address(rdx, rbx,
4241 Address::times_1,
4242 Array<u1>::base_offset_in_bytes()),
4243 JVM_CONSTANT_Class);
4244 __ jcc(Assembler::equal, quicked);
4245
4246 __ push(atos); // save receiver for result, and for GC
4247 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4248 // vm_result_2 has metadata result
4249
4250 #ifndef _LP64
4251 // borrow rdi from locals
4252 __ get_thread(rdi);
4253 __ get_vm_result_2(rax, rdi);
4254 __ restore_locals();
4255 #else
4256 __ get_vm_result_2(rax, r15_thread);
4257 #endif
4258
4259 __ pop_ptr(rdx); // restore receiver
4260 __ verify_oop(rdx);
4261 __ load_klass(rdx, rdx);
4262 __ jmpb(resolved);
4263
4264 // Get superklass in rax and subklass in rdx
4265 __ bind(quicked);
4266 __ load_klass(rdx, rax);
4267 __ load_resolved_klass_at_index(rcx, rbx, rax);
4268
4269 __ bind(resolved);
4270
4271 // Generate subtype check. Blows rcx, rdi
4272 // Superklass in rax. Subklass in rdx.
4273 __ gen_subtype_check(rdx, ok_is_subtype);
4274
4275 // Come here on failure
4276 __ xorl(rax, rax);
4277 __ jmpb(done);
4278 // Come here on success
4279 __ bind(ok_is_subtype);
4280 __ movl(rax, 1);
4281
4282 // Collect counts on whether this test sees NULLs a lot or not.
4283 if (ProfileInterpreter) {
4284 __ jmp(done);
4285 __ bind(is_null);
4286 __ profile_null_seen(rcx);
4287 } else {
4288 __ bind(is_null); // same as 'done'
4289 }
4290 __ bind(done);
4291 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4292 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4293 }
4294
4295
4296 //----------------------------------------------------------------------------------------------------
4297 // Breakpoints
4298 void TemplateTable::_breakpoint() {
4299 // Note: We get here even if we are single stepping..
4300 // jbug insists on setting breakpoints at every bytecode
4301 // even if we are in single step mode.
4302
4303 transition(vtos, vtos);
4304
4305 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4306
4307 // get the unpatched byte code
4308 __ get_method(rarg);
4309 __ call_VM(noreg,
4310 CAST_FROM_FN_PTR(address,
4311 InterpreterRuntime::get_original_bytecode_at),
4312 rarg, rbcp);
4313 __ mov(rbx, rax); // why?
4314
4315 // post the breakpoint event
4316 __ get_method(rarg);
4317 __ call_VM(noreg,
4318 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4319 rarg, rbcp);
4320
4321 // complete the execution of original bytecode
4322 __ dispatch_only_normal(vtos);
4323 }
4324
4325 //-----------------------------------------------------------------------------
4326 // Exceptions
4327
4328 void TemplateTable::athrow() {
4329 transition(atos, vtos);
4330 __ null_check(rax);
4331 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4332 }
4333
4334 //-----------------------------------------------------------------------------
4335 // Synchronization
4336 //
4337 // Note: monitorenter & exit are symmetric routines; which is reflected
4338 // in the assembly code structure as well
4339 //
4340 // Stack layout:
4341 //
4342 // [expressions ] <--- rsp = expression stack top
4343 // ..
4344 // [expressions ]
4345 // [monitor entry] <--- monitor block top = expression stack bot
4346 // ..
4347 // [monitor entry]
4348 // [frame data ] <--- monitor block bot
4349 // ...
4350 // [saved rbp ] <--- rbp
4351 void TemplateTable::monitorenter() {
4352 transition(atos, vtos);
4353
4354 // check for NULL object
4355 __ null_check(rax);
4356
4357 const Address monitor_block_top(
4358 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4359 const Address monitor_block_bot(
4360 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4361 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4362
4363 Label allocated;
4364
4365 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4366 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4367 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4368
4369 // initialize entry pointer
4370 __ xorl(rmon, rmon); // points to free slot or NULL
4371
4372 // find a free slot in the monitor block (result in rmon)
4373 {
4374 Label entry, loop, exit;
4375 __ movptr(rtop, monitor_block_top); // points to current entry,
4376 // starting with top-most entry
4377 __ lea(rbot, monitor_block_bot); // points to word before bottom
4378 // of monitor block
4379 __ jmpb(entry);
4380
4381 __ bind(loop);
4382 // check if current entry is used
4383 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4384 // if not used then remember entry in rmon
4385 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4386 // check if current entry is for same object
4387 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4388 // if same object then stop searching
4389 __ jccb(Assembler::equal, exit);
4390 // otherwise advance to next entry
4391 __ addptr(rtop, entry_size);
4392 __ bind(entry);
4393 // check if bottom reached
4394 __ cmpptr(rtop, rbot);
4395 // if not at bottom then check this entry
4396 __ jcc(Assembler::notEqual, loop);
4397 __ bind(exit);
4398 }
4399
4400 __ testptr(rmon, rmon); // check if a slot has been found
4401 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4402
4403 // allocate one if there's no free slot
4404 {
4405 Label entry, loop;
4406 // 1. compute new pointers // rsp: old expression stack top
4407 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4408 __ subptr(rsp, entry_size); // move expression stack top
4409 __ subptr(rmon, entry_size); // move expression stack bottom
4410 __ mov(rtop, rsp); // set start value for copy loop
4411 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4412 __ jmp(entry);
4413 // 2. move expression stack contents
4414 __ bind(loop);
4415 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4416 // word from old location
4417 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4418 __ addptr(rtop, wordSize); // advance to next word
4419 __ bind(entry);
4420 __ cmpptr(rtop, rmon); // check if bottom reached
4421 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4422 // copy next word
4423 }
4424
4425 // call run-time routine
4426 // rmon: points to monitor entry
4427 __ bind(allocated);
4428
4429 // Increment bcp to point to the next bytecode, so exception
4430 // handling for async. exceptions work correctly.
4431 // The object has already been poped from the stack, so the
4432 // expression stack looks correct.
4433 __ increment(rbcp);
4434
4435 // store object
4436 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4437 __ lock_object(rmon);
4438
4439 // check to make sure this monitor doesn't cause stack overflow after locking
4440 __ save_bcp(); // in case of exception
4441 __ generate_stack_overflow_check(0);
4442
4443 // The bcp has already been incremented. Just need to dispatch to
4444 // next instruction.
4445 __ dispatch_next(vtos);
4446 }
4447
4448 void TemplateTable::monitorexit() {
4449 transition(atos, vtos);
4450
4451 // check for NULL object
4452 __ null_check(rax);
4453
4454 const Address monitor_block_top(
4455 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4456 const Address monitor_block_bot(
4457 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4458 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4459
4460 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4461 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4462
4463 Label found;
4464
4465 // find matching slot
4466 {
4467 Label entry, loop;
4468 __ movptr(rtop, monitor_block_top); // points to current entry,
4469 // starting with top-most entry
4470 __ lea(rbot, monitor_block_bot); // points to word before bottom
4471 // of monitor block
4472 __ jmpb(entry);
4473
4474 __ bind(loop);
4475 // check if current entry is for same object
4476 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4477 // if same object then stop searching
4478 __ jcc(Assembler::equal, found);
4479 // otherwise advance to next entry
4480 __ addptr(rtop, entry_size);
4481 __ bind(entry);
4482 // check if bottom reached
4483 __ cmpptr(rtop, rbot);
4484 // if not at bottom then check this entry
4485 __ jcc(Assembler::notEqual, loop);
4486 }
4487
4488 // error handling. Unlocking was not block-structured
4489 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4490 InterpreterRuntime::throw_illegal_monitor_state_exception));
4491 __ should_not_reach_here();
4492
4493 // call run-time routine
4494 __ bind(found);
4495 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4496 __ unlock_object(rtop);
4497 __ pop_ptr(rax); // discard object
4498 }
4499
4500 // Wide instructions
4501 void TemplateTable::wide() {
4502 transition(vtos, vtos);
4503 __ load_unsigned_byte(rbx, at_bcp(1));
4504 ExternalAddress wtable((address)Interpreter::_wentry_point);
4505 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4506 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4507 }
4508
4509 // Multi arrays
4510 void TemplateTable::multianewarray() {
4511 transition(vtos, atos);
4512
4513 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4514 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4515 // last dim is on top of stack; we want address of first one:
4516 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4517 // the latter wordSize to point to the beginning of the array.
4518 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4519 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4520 __ load_unsigned_byte(rbx, at_bcp(3));
4521 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4522 }