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