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 __ cmpoop(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 __ jump_cc(Assembler::aboveEqual,
761 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
762 }
763
764
765 void TemplateTable::iaload() {
766 transition(itos, itos);
767 // rax: index
768 // rdx: array
769 index_check(rdx, rax); // kills rbx
770 __ resolve_for_read(OOP_NOT_NULL, rdx);
771 __ movl(rax, Address(rdx, rax,
772 Address::times_4,
773 arrayOopDesc::base_offset_in_bytes(T_INT)));
774 }
775
776 void TemplateTable::laload() {
777 transition(itos, ltos);
778 // rax: index
779 // rdx: array
780 index_check(rdx, rax); // kills rbx
781 NOT_LP64(__ mov(rbx, rax));
782 // rbx,: index
783 __ resolve_for_read(OOP_NOT_NULL, rdx);
784 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
785 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
786 }
787
788
789
790 void TemplateTable::faload() {
791 transition(itos, ftos);
792 // rax: index
793 // rdx: array
794 index_check(rdx, rax); // kills rbx
795 __ resolve_for_read(OOP_NOT_NULL, rdx);
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 __ resolve_for_read(OOP_NOT_NULL, rdx);
807 __ load_double(Address(rdx, rax,
808 Address::times_8,
809 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
810 }
811
812 void TemplateTable::aaload() {
813 transition(itos, atos);
814 // rax: index
815 // rdx: array
816 index_check(rdx, rax); // kills rbx
817 __ resolve_for_read(OOP_NOT_NULL, rdx);
818 do_oop_load(_masm,
819 Address(rdx, rax,
820 UseCompressedOops ? Address::times_4 : Address::times_ptr,
821 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
822 rax,
823 IN_HEAP_ARRAY);
824 }
825
826 void TemplateTable::baload() {
827 transition(itos, itos);
828 // rax: index
829 // rdx: array
830 index_check(rdx, rax); // kills rbx
831 __ resolve_for_read(OOP_NOT_NULL, rdx);
832 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
833 }
834
835 void TemplateTable::caload() {
836 transition(itos, itos);
837 // rax: index
838 // rdx: array
839 index_check(rdx, rax); // kills rbx
840 __ resolve_for_read(OOP_NOT_NULL, rdx);
841 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
842 }
843
844 // iload followed by caload frequent pair
845 void TemplateTable::fast_icaload() {
846 transition(vtos, itos);
847 // load index out of locals
848 locals_index(rbx);
849 __ movl(rax, iaddress(rbx));
850
851 // rax: index
852 // rdx: array
853 index_check(rdx, rax); // kills rbx
854 __ resolve_for_read(OOP_NOT_NULL, rdx);
855 __ load_unsigned_short(rax,
856 Address(rdx, rax,
857 Address::times_2,
858 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
859 }
860
861
862 void TemplateTable::saload() {
863 transition(itos, itos);
864 // rax: index
865 // rdx: array
866 index_check(rdx, rax); // kills rbx
867 __ resolve_for_read(OOP_NOT_NULL, rdx);
868 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
869 }
870
871 void TemplateTable::iload(int n) {
872 transition(vtos, itos);
873 __ movl(rax, iaddress(n));
874 }
875
876 void TemplateTable::lload(int n) {
877 transition(vtos, ltos);
878 __ movptr(rax, laddress(n));
879 NOT_LP64(__ movptr(rdx, haddress(n)));
880 }
881
882 void TemplateTable::fload(int n) {
883 transition(vtos, ftos);
884 __ load_float(faddress(n));
885 }
886
887 void TemplateTable::dload(int n) {
888 transition(vtos, dtos);
889 __ load_double(daddress(n));
890 }
891
892 void TemplateTable::aload(int n) {
893 transition(vtos, atos);
894 __ movptr(rax, aaddress(n));
895 }
896
897 void TemplateTable::aload_0() {
898 aload_0_internal();
899 }
900
901 void TemplateTable::nofast_aload_0() {
902 aload_0_internal(may_not_rewrite);
903 }
904
905 void TemplateTable::aload_0_internal(RewriteControl rc) {
906 transition(vtos, atos);
907 // According to bytecode histograms, the pairs:
908 //
909 // _aload_0, _fast_igetfield
910 // _aload_0, _fast_agetfield
911 // _aload_0, _fast_fgetfield
912 //
913 // occur frequently. If RewriteFrequentPairs is set, the (slow)
914 // _aload_0 bytecode checks if the next bytecode is either
915 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
916 // rewrites the current bytecode into a pair bytecode; otherwise it
917 // rewrites the current bytecode into _fast_aload_0 that doesn't do
918 // the pair check anymore.
919 //
920 // Note: If the next bytecode is _getfield, the rewrite must be
921 // delayed, otherwise we may miss an opportunity for a pair.
922 //
923 // Also rewrite frequent pairs
924 // aload_0, aload_1
925 // aload_0, iload_1
926 // These bytecodes with a small amount of code are most profitable
927 // to rewrite
928 if (RewriteFrequentPairs && rc == may_rewrite) {
929 Label rewrite, done;
930
931 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
932 LP64_ONLY(assert(rbx != bc, "register damaged"));
933
934 // get next byte
935 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
936
937 // if _getfield then wait with rewrite
938 __ cmpl(rbx, Bytecodes::_getfield);
939 __ jcc(Assembler::equal, done);
940
941 // if _igetfield then rewrite to _fast_iaccess_0
942 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
943 __ cmpl(rbx, Bytecodes::_fast_igetfield);
944 __ movl(bc, Bytecodes::_fast_iaccess_0);
945 __ jccb(Assembler::equal, rewrite);
946
947 // if _agetfield then rewrite to _fast_aaccess_0
948 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
949 __ cmpl(rbx, Bytecodes::_fast_agetfield);
950 __ movl(bc, Bytecodes::_fast_aaccess_0);
951 __ jccb(Assembler::equal, rewrite);
952
953 // if _fgetfield then rewrite to _fast_faccess_0
954 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
955 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
956 __ movl(bc, Bytecodes::_fast_faccess_0);
957 __ jccb(Assembler::equal, rewrite);
958
959 // else rewrite to _fast_aload0
960 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
961 __ movl(bc, Bytecodes::_fast_aload_0);
962
963 // rewrite
964 // bc: fast bytecode
965 __ bind(rewrite);
966 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
967
968 __ bind(done);
969 }
970
971 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
972 aload(0);
973 }
974
975 void TemplateTable::istore() {
976 transition(itos, vtos);
977 locals_index(rbx);
978 __ movl(iaddress(rbx), rax);
979 }
980
981
982 void TemplateTable::lstore() {
983 transition(ltos, vtos);
984 locals_index(rbx);
985 __ movptr(laddress(rbx), rax);
986 NOT_LP64(__ movptr(haddress(rbx), rdx));
987 }
988
989 void TemplateTable::fstore() {
990 transition(ftos, vtos);
991 locals_index(rbx);
992 __ store_float(faddress(rbx));
993 }
994
995 void TemplateTable::dstore() {
996 transition(dtos, vtos);
997 locals_index(rbx);
998 __ store_double(daddress(rbx));
999 }
1000
1001 void TemplateTable::astore() {
1002 transition(vtos, vtos);
1003 __ pop_ptr(rax);
1004 locals_index(rbx);
1005 __ movptr(aaddress(rbx), rax);
1006 }
1007
1008 void TemplateTable::wide_istore() {
1009 transition(vtos, vtos);
1010 __ pop_i();
1011 locals_index_wide(rbx);
1012 __ movl(iaddress(rbx), rax);
1013 }
1014
1015 void TemplateTable::wide_lstore() {
1016 transition(vtos, vtos);
1017 NOT_LP64(__ pop_l(rax, rdx));
1018 LP64_ONLY(__ pop_l());
1019 locals_index_wide(rbx);
1020 __ movptr(laddress(rbx), rax);
1021 NOT_LP64(__ movl(haddress(rbx), rdx));
1022 }
1023
1024 void TemplateTable::wide_fstore() {
1025 #ifdef _LP64
1026 transition(vtos, vtos);
1027 __ pop_f(xmm0);
1028 locals_index_wide(rbx);
1029 __ movflt(faddress(rbx), xmm0);
1030 #else
1031 wide_istore();
1032 #endif
1033 }
1034
1035 void TemplateTable::wide_dstore() {
1036 #ifdef _LP64
1037 transition(vtos, vtos);
1038 __ pop_d(xmm0);
1039 locals_index_wide(rbx);
1040 __ movdbl(daddress(rbx), xmm0);
1041 #else
1042 wide_lstore();
1043 #endif
1044 }
1045
1046 void TemplateTable::wide_astore() {
1047 transition(vtos, vtos);
1048 __ pop_ptr(rax);
1049 locals_index_wide(rbx);
1050 __ movptr(aaddress(rbx), rax);
1051 }
1052
1053 void TemplateTable::iastore() {
1054 transition(itos, vtos);
1055 __ pop_i(rbx);
1056 // rax: value
1057 // rbx: index
1058 // rdx: array
1059 index_check(rdx, rbx); // prefer index in rbx
1060 __ resolve_for_write(OOP_NOT_NULL, rdx);
1061 __ movl(Address(rdx, rbx,
1062 Address::times_4,
1063 arrayOopDesc::base_offset_in_bytes(T_INT)),
1064 rax);
1065 }
1066
1067 void TemplateTable::lastore() {
1068 transition(ltos, vtos);
1069 __ pop_i(rbx);
1070 // rax,: low(value)
1071 // rcx: array
1072 // rdx: high(value)
1073 index_check(rcx, rbx); // prefer index in rbx,
1074 // rbx,: index
1075 __ resolve_for_write(OOP_NOT_NULL, rcx);
1076 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1077 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1078 }
1079
1080
1081 void TemplateTable::fastore() {
1082 transition(ftos, vtos);
1083 __ pop_i(rbx);
1084 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1085 // rbx: index
1086 // rdx: array
1087 index_check(rdx, rbx); // prefer index in rbx
1088 __ resolve_for_write(OOP_NOT_NULL, rdx);
1089 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1090 }
1091
1092 void TemplateTable::dastore() {
1093 transition(dtos, vtos);
1094 __ pop_i(rbx);
1095 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1096 // rbx: index
1097 // rdx: array
1098 index_check(rdx, rbx); // prefer index in rbx
1099 __ resolve_for_write(OOP_NOT_NULL, rdx);
1100 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1101 }
1102
1103 void TemplateTable::aastore() {
1104 Label is_null, ok_is_subtype, done;
1105 transition(vtos, vtos);
1106 // stack: ..., array, index, value
1107 __ movptr(rax, at_tos()); // value
1108 __ movl(rcx, at_tos_p1()); // index
1109 __ movptr(rdx, at_tos_p2()); // array
1110
1111 Address element_address(rdx, rcx,
1112 UseCompressedOops? Address::times_4 : Address::times_ptr,
1113 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1114
1115 index_check_without_pop(rdx, rcx); // kills rbx
1116 __ resolve_for_write(OOP_NOT_NULL, rdx);
1117 __ testptr(rax, rax);
1118 __ jcc(Assembler::zero, is_null);
1119
1120 // Move subklass into rbx
1121 __ load_klass(rbx, rax);
1122 // Move superklass into rax
1123 __ load_klass(rax, rdx);
1124 __ movptr(rax, Address(rax,
1125 ObjArrayKlass::element_klass_offset()));
1126 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1127 __ lea(rdx, element_address);
1128
1129 // Generate subtype check. Blows rcx, rdi
1130 // Superklass in rax. Subklass in rbx.
1131 __ gen_subtype_check(rbx, ok_is_subtype);
1132
1133 // Come here on failure
1134 // object is at TOS
1135 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1136
1137 // Come here on success
1138 __ bind(ok_is_subtype);
1139
1140 // Get the value we will store
1141 __ movptr(rax, at_tos());
1142 // Now store using the appropriate barrier
1143 do_oop_store(_masm, Address(rdx, 0), rax, IN_HEAP_ARRAY);
1144 __ jmp(done);
1145
1146 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1147 __ bind(is_null);
1148 __ profile_null_seen(rbx);
1149
1150 // Store a NULL
1151 do_oop_store(_masm, element_address, noreg, IN_HEAP_ARRAY);
1152
1153 // Pop stack arguments
1154 __ bind(done);
1155 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1156 }
1157
1158 void TemplateTable::bastore() {
1159 transition(itos, vtos);
1160 __ pop_i(rbx);
1161 // rax: value
1162 // rbx: index
1163 // rdx: array
1164 index_check(rdx, rbx); // prefer index in rbx
1165 __ resolve_for_write(OOP_NOT_NULL, rdx);
1166 // Need to check whether array is boolean or byte
1167 // since both types share the bastore bytecode.
1168 __ load_klass(rcx, rdx);
1169 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1170 int diffbit = Klass::layout_helper_boolean_diffbit();
1171 __ testl(rcx, diffbit);
1172 Label L_skip;
1173 __ jccb(Assembler::zero, L_skip);
1174 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1175 __ bind(L_skip);
1176 __ movb(Address(rdx, rbx,
1177 Address::times_1,
1178 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1179 rax);
1180 }
1181
1182 void TemplateTable::castore() {
1183 transition(itos, vtos);
1184 __ pop_i(rbx);
1185 // rax: value
1186 // rbx: index
1187 // rdx: array
1188 index_check(rdx, rbx); // prefer index in rbx
1189 __ resolve_for_write(OOP_NOT_NULL, rdx);
1190 __ movw(Address(rdx, rbx,
1191 Address::times_2,
1192 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1193 rax);
1194 }
1195
1196
1197 void TemplateTable::sastore() {
1198 castore();
1199 }
1200
1201 void TemplateTable::istore(int n) {
1202 transition(itos, vtos);
1203 __ movl(iaddress(n), rax);
1204 }
1205
1206 void TemplateTable::lstore(int n) {
1207 transition(ltos, vtos);
1208 __ movptr(laddress(n), rax);
1209 NOT_LP64(__ movptr(haddress(n), rdx));
1210 }
1211
1212 void TemplateTable::fstore(int n) {
1213 transition(ftos, vtos);
1214 __ store_float(faddress(n));
1215 }
1216
1217 void TemplateTable::dstore(int n) {
1218 transition(dtos, vtos);
1219 __ store_double(daddress(n));
1220 }
1221
1222
1223 void TemplateTable::astore(int n) {
1224 transition(vtos, vtos);
1225 __ pop_ptr(rax);
1226 __ movptr(aaddress(n), rax);
1227 }
1228
1229 void TemplateTable::pop() {
1230 transition(vtos, vtos);
1231 __ addptr(rsp, Interpreter::stackElementSize);
1232 }
1233
1234 void TemplateTable::pop2() {
1235 transition(vtos, vtos);
1236 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1237 }
1238
1239
1240 void TemplateTable::dup() {
1241 transition(vtos, vtos);
1242 __ load_ptr(0, rax);
1243 __ push_ptr(rax);
1244 // stack: ..., a, a
1245 }
1246
1247 void TemplateTable::dup_x1() {
1248 transition(vtos, vtos);
1249 // stack: ..., a, b
1250 __ load_ptr( 0, rax); // load b
1251 __ load_ptr( 1, rcx); // load a
1252 __ store_ptr(1, rax); // store b
1253 __ store_ptr(0, rcx); // store a
1254 __ push_ptr(rax); // push b
1255 // stack: ..., b, a, b
1256 }
1257
1258 void TemplateTable::dup_x2() {
1259 transition(vtos, vtos);
1260 // stack: ..., a, b, c
1261 __ load_ptr( 0, rax); // load c
1262 __ load_ptr( 2, rcx); // load a
1263 __ store_ptr(2, rax); // store c in a
1264 __ push_ptr(rax); // push c
1265 // stack: ..., c, b, c, c
1266 __ load_ptr( 2, rax); // load b
1267 __ store_ptr(2, rcx); // store a in b
1268 // stack: ..., c, a, c, c
1269 __ store_ptr(1, rax); // store b in c
1270 // stack: ..., c, a, b, c
1271 }
1272
1273 void TemplateTable::dup2() {
1274 transition(vtos, vtos);
1275 // stack: ..., a, b
1276 __ load_ptr(1, rax); // load a
1277 __ push_ptr(rax); // push a
1278 __ load_ptr(1, rax); // load b
1279 __ push_ptr(rax); // push b
1280 // stack: ..., a, b, a, b
1281 }
1282
1283
1284 void TemplateTable::dup2_x1() {
1285 transition(vtos, vtos);
1286 // stack: ..., a, b, c
1287 __ load_ptr( 0, rcx); // load c
1288 __ load_ptr( 1, rax); // load b
1289 __ push_ptr(rax); // push b
1290 __ push_ptr(rcx); // push c
1291 // stack: ..., a, b, c, b, c
1292 __ store_ptr(3, rcx); // store c in b
1293 // stack: ..., a, c, c, b, c
1294 __ load_ptr( 4, rcx); // load a
1295 __ store_ptr(2, rcx); // store a in 2nd c
1296 // stack: ..., a, c, a, b, c
1297 __ store_ptr(4, rax); // store b in a
1298 // stack: ..., b, c, a, b, c
1299 }
1300
1301 void TemplateTable::dup2_x2() {
1302 transition(vtos, vtos);
1303 // stack: ..., a, b, c, d
1304 __ load_ptr( 0, rcx); // load d
1305 __ load_ptr( 1, rax); // load c
1306 __ push_ptr(rax); // push c
1307 __ push_ptr(rcx); // push d
1308 // stack: ..., a, b, c, d, c, d
1309 __ load_ptr( 4, rax); // load b
1310 __ store_ptr(2, rax); // store b in d
1311 __ store_ptr(4, rcx); // store d in b
1312 // stack: ..., a, d, c, b, c, d
1313 __ load_ptr( 5, rcx); // load a
1314 __ load_ptr( 3, rax); // load c
1315 __ store_ptr(3, rcx); // store a in c
1316 __ store_ptr(5, rax); // store c in a
1317 // stack: ..., c, d, a, b, c, d
1318 }
1319
1320 void TemplateTable::swap() {
1321 transition(vtos, vtos);
1322 // stack: ..., a, b
1323 __ load_ptr( 1, rcx); // load a
1324 __ load_ptr( 0, rax); // load b
1325 __ store_ptr(0, rcx); // store a in b
1326 __ store_ptr(1, rax); // store b in a
1327 // stack: ..., b, a
1328 }
1329
1330 void TemplateTable::iop2(Operation op) {
1331 transition(itos, itos);
1332 switch (op) {
1333 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1334 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1335 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1336 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1337 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1338 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1339 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1340 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1341 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1342 default : ShouldNotReachHere();
1343 }
1344 }
1345
1346 void TemplateTable::lop2(Operation op) {
1347 transition(ltos, ltos);
1348 #ifdef _LP64
1349 switch (op) {
1350 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1351 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1352 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1353 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1354 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1355 default : ShouldNotReachHere();
1356 }
1357 #else
1358 __ pop_l(rbx, rcx);
1359 switch (op) {
1360 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1361 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1362 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1363 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1364 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1365 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1366 default : ShouldNotReachHere();
1367 }
1368 #endif
1369 }
1370
1371 void TemplateTable::idiv() {
1372 transition(itos, itos);
1373 __ movl(rcx, rax);
1374 __ pop_i(rax);
1375 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1376 // they are not equal, one could do a normal division (no correction
1377 // needed), which may speed up this implementation for the common case.
1378 // (see also JVM spec., p.243 & p.271)
1379 __ corrected_idivl(rcx);
1380 }
1381
1382 void TemplateTable::irem() {
1383 transition(itos, itos);
1384 __ movl(rcx, rax);
1385 __ pop_i(rax);
1386 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1387 // they are not equal, one could do a normal division (no correction
1388 // needed), which may speed up this implementation for the common case.
1389 // (see also JVM spec., p.243 & p.271)
1390 __ corrected_idivl(rcx);
1391 __ movl(rax, rdx);
1392 }
1393
1394 void TemplateTable::lmul() {
1395 transition(ltos, ltos);
1396 #ifdef _LP64
1397 __ pop_l(rdx);
1398 __ imulq(rax, rdx);
1399 #else
1400 __ pop_l(rbx, rcx);
1401 __ push(rcx); __ push(rbx);
1402 __ push(rdx); __ push(rax);
1403 __ lmul(2 * wordSize, 0);
1404 __ addptr(rsp, 4 * wordSize); // take off temporaries
1405 #endif
1406 }
1407
1408 void TemplateTable::ldiv() {
1409 transition(ltos, ltos);
1410 #ifdef _LP64
1411 __ mov(rcx, rax);
1412 __ pop_l(rax);
1413 // generate explicit div0 check
1414 __ testq(rcx, rcx);
1415 __ jump_cc(Assembler::zero,
1416 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1417 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1418 // they are not equal, one could do a normal division (no correction
1419 // needed), which may speed up this implementation for the common case.
1420 // (see also JVM spec., p.243 & p.271)
1421 __ corrected_idivq(rcx); // kills rbx
1422 #else
1423 __ pop_l(rbx, rcx);
1424 __ push(rcx); __ push(rbx);
1425 __ push(rdx); __ push(rax);
1426 // check if y = 0
1427 __ orl(rax, rdx);
1428 __ jump_cc(Assembler::zero,
1429 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1430 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1431 __ addptr(rsp, 4 * wordSize); // take off temporaries
1432 #endif
1433 }
1434
1435 void TemplateTable::lrem() {
1436 transition(ltos, ltos);
1437 #ifdef _LP64
1438 __ mov(rcx, rax);
1439 __ pop_l(rax);
1440 __ testq(rcx, rcx);
1441 __ jump_cc(Assembler::zero,
1442 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1443 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1444 // they are not equal, one could do a normal division (no correction
1445 // needed), which may speed up this implementation for the common case.
1446 // (see also JVM spec., p.243 & p.271)
1447 __ corrected_idivq(rcx); // kills rbx
1448 __ mov(rax, rdx);
1449 #else
1450 __ pop_l(rbx, rcx);
1451 __ push(rcx); __ push(rbx);
1452 __ push(rdx); __ push(rax);
1453 // check if y = 0
1454 __ orl(rax, rdx);
1455 __ jump_cc(Assembler::zero,
1456 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1457 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1458 __ addptr(rsp, 4 * wordSize);
1459 #endif
1460 }
1461
1462 void TemplateTable::lshl() {
1463 transition(itos, ltos);
1464 __ movl(rcx, rax); // get shift count
1465 #ifdef _LP64
1466 __ pop_l(rax); // get shift value
1467 __ shlq(rax);
1468 #else
1469 __ pop_l(rax, rdx); // get shift value
1470 __ lshl(rdx, rax);
1471 #endif
1472 }
1473
1474 void TemplateTable::lshr() {
1475 #ifdef _LP64
1476 transition(itos, ltos);
1477 __ movl(rcx, rax); // get shift count
1478 __ pop_l(rax); // get shift value
1479 __ sarq(rax);
1480 #else
1481 transition(itos, ltos);
1482 __ mov(rcx, rax); // get shift count
1483 __ pop_l(rax, rdx); // get shift value
1484 __ lshr(rdx, rax, true);
1485 #endif
1486 }
1487
1488 void TemplateTable::lushr() {
1489 transition(itos, ltos);
1490 #ifdef _LP64
1491 __ movl(rcx, rax); // get shift count
1492 __ pop_l(rax); // get shift value
1493 __ shrq(rax);
1494 #else
1495 __ mov(rcx, rax); // get shift count
1496 __ pop_l(rax, rdx); // get shift value
1497 __ lshr(rdx, rax);
1498 #endif
1499 }
1500
1501 void TemplateTable::fop2(Operation op) {
1502 transition(ftos, ftos);
1503
1504 if (UseSSE >= 1) {
1505 switch (op) {
1506 case add:
1507 __ addss(xmm0, at_rsp());
1508 __ addptr(rsp, Interpreter::stackElementSize);
1509 break;
1510 case sub:
1511 __ movflt(xmm1, xmm0);
1512 __ pop_f(xmm0);
1513 __ subss(xmm0, xmm1);
1514 break;
1515 case mul:
1516 __ mulss(xmm0, at_rsp());
1517 __ addptr(rsp, Interpreter::stackElementSize);
1518 break;
1519 case div:
1520 __ movflt(xmm1, xmm0);
1521 __ pop_f(xmm0);
1522 __ divss(xmm0, xmm1);
1523 break;
1524 case rem:
1525 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1526 // modulo operation. The frem method calls the function
1527 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1528 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1529 // (signalling or quiet) is returned.
1530 //
1531 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1532 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1533 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1534 // The fprem instruction used on x86_32 is functionally equivalent to
1535 // SharedRuntime::frem in that it returns a NaN.
1536 #ifdef _LP64
1537 __ movflt(xmm1, xmm0);
1538 __ pop_f(xmm0);
1539 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1540 #else
1541 __ push_f(xmm0);
1542 __ pop_f();
1543 __ fld_s(at_rsp());
1544 __ fremr(rax);
1545 __ f2ieee();
1546 __ pop(rax); // pop second operand off the stack
1547 __ push_f();
1548 __ pop_f(xmm0);
1549 #endif
1550 break;
1551 default:
1552 ShouldNotReachHere();
1553 break;
1554 }
1555 } else {
1556 #ifdef _LP64
1557 ShouldNotReachHere();
1558 #else
1559 switch (op) {
1560 case add: __ fadd_s (at_rsp()); break;
1561 case sub: __ fsubr_s(at_rsp()); break;
1562 case mul: __ fmul_s (at_rsp()); break;
1563 case div: __ fdivr_s(at_rsp()); break;
1564 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1565 default : ShouldNotReachHere();
1566 }
1567 __ f2ieee();
1568 __ pop(rax); // pop second operand off the stack
1569 #endif // _LP64
1570 }
1571 }
1572
1573 void TemplateTable::dop2(Operation op) {
1574 transition(dtos, dtos);
1575 if (UseSSE >= 2) {
1576 switch (op) {
1577 case add:
1578 __ addsd(xmm0, at_rsp());
1579 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1580 break;
1581 case sub:
1582 __ movdbl(xmm1, xmm0);
1583 __ pop_d(xmm0);
1584 __ subsd(xmm0, xmm1);
1585 break;
1586 case mul:
1587 __ mulsd(xmm0, at_rsp());
1588 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1589 break;
1590 case div:
1591 __ movdbl(xmm1, xmm0);
1592 __ pop_d(xmm0);
1593 __ divsd(xmm0, xmm1);
1594 break;
1595 case rem:
1596 // Similar to fop2(), the modulo operation is performed using the
1597 // SharedRuntime::drem method (on x86_64 platforms) or using the
1598 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1599 #ifdef _LP64
1600 __ movdbl(xmm1, xmm0);
1601 __ pop_d(xmm0);
1602 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1603 #else
1604 __ push_d(xmm0);
1605 __ pop_d();
1606 __ fld_d(at_rsp());
1607 __ fremr(rax);
1608 __ d2ieee();
1609 __ pop(rax);
1610 __ pop(rdx);
1611 __ push_d();
1612 __ pop_d(xmm0);
1613 #endif
1614 break;
1615 default:
1616 ShouldNotReachHere();
1617 break;
1618 }
1619 } else {
1620 #ifdef _LP64
1621 ShouldNotReachHere();
1622 #else
1623 switch (op) {
1624 case add: __ fadd_d (at_rsp()); break;
1625 case sub: __ fsubr_d(at_rsp()); break;
1626 case mul: {
1627 Label L_strict;
1628 Label L_join;
1629 const Address access_flags (rcx, Method::access_flags_offset());
1630 __ get_method(rcx);
1631 __ movl(rcx, access_flags);
1632 __ testl(rcx, JVM_ACC_STRICT);
1633 __ jccb(Assembler::notZero, L_strict);
1634 __ fmul_d (at_rsp());
1635 __ jmpb(L_join);
1636 __ bind(L_strict);
1637 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1638 __ fmulp();
1639 __ fmul_d (at_rsp());
1640 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1641 __ fmulp();
1642 __ bind(L_join);
1643 break;
1644 }
1645 case div: {
1646 Label L_strict;
1647 Label L_join;
1648 const Address access_flags (rcx, Method::access_flags_offset());
1649 __ get_method(rcx);
1650 __ movl(rcx, access_flags);
1651 __ testl(rcx, JVM_ACC_STRICT);
1652 __ jccb(Assembler::notZero, L_strict);
1653 __ fdivr_d(at_rsp());
1654 __ jmp(L_join);
1655 __ bind(L_strict);
1656 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1657 __ fmul_d (at_rsp());
1658 __ fdivrp();
1659 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1660 __ fmulp();
1661 __ bind(L_join);
1662 break;
1663 }
1664 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1665 default : ShouldNotReachHere();
1666 }
1667 __ d2ieee();
1668 // Pop double precision number from rsp.
1669 __ pop(rax);
1670 __ pop(rdx);
1671 #endif
1672 }
1673 }
1674
1675 void TemplateTable::ineg() {
1676 transition(itos, itos);
1677 __ negl(rax);
1678 }
1679
1680 void TemplateTable::lneg() {
1681 transition(ltos, ltos);
1682 LP64_ONLY(__ negq(rax));
1683 NOT_LP64(__ lneg(rdx, rax));
1684 }
1685
1686 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1687 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1688 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1689 // of 128-bits operands for SSE instructions.
1690 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1691 // Store the value to a 128-bits operand.
1692 operand[0] = lo;
1693 operand[1] = hi;
1694 return operand;
1695 }
1696
1697 // Buffer for 128-bits masks used by SSE instructions.
1698 static jlong float_signflip_pool[2*2];
1699 static jlong double_signflip_pool[2*2];
1700
1701 void TemplateTable::fneg() {
1702 transition(ftos, ftos);
1703 if (UseSSE >= 1) {
1704 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1705 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1706 } else {
1707 LP64_ONLY(ShouldNotReachHere());
1708 NOT_LP64(__ fchs());
1709 }
1710 }
1711
1712 void TemplateTable::dneg() {
1713 transition(dtos, dtos);
1714 if (UseSSE >= 2) {
1715 static jlong *double_signflip =
1716 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1717 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1718 } else {
1719 #ifdef _LP64
1720 ShouldNotReachHere();
1721 #else
1722 __ fchs();
1723 #endif
1724 }
1725 }
1726
1727 void TemplateTable::iinc() {
1728 transition(vtos, vtos);
1729 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1730 locals_index(rbx);
1731 __ addl(iaddress(rbx), rdx);
1732 }
1733
1734 void TemplateTable::wide_iinc() {
1735 transition(vtos, vtos);
1736 __ movl(rdx, at_bcp(4)); // get constant
1737 locals_index_wide(rbx);
1738 __ bswapl(rdx); // swap bytes & sign-extend constant
1739 __ sarl(rdx, 16);
1740 __ addl(iaddress(rbx), rdx);
1741 // Note: should probably use only one movl to get both
1742 // the index and the constant -> fix this
1743 }
1744
1745 void TemplateTable::convert() {
1746 #ifdef _LP64
1747 // Checking
1748 #ifdef ASSERT
1749 {
1750 TosState tos_in = ilgl;
1751 TosState tos_out = ilgl;
1752 switch (bytecode()) {
1753 case Bytecodes::_i2l: // fall through
1754 case Bytecodes::_i2f: // fall through
1755 case Bytecodes::_i2d: // fall through
1756 case Bytecodes::_i2b: // fall through
1757 case Bytecodes::_i2c: // fall through
1758 case Bytecodes::_i2s: tos_in = itos; break;
1759 case Bytecodes::_l2i: // fall through
1760 case Bytecodes::_l2f: // fall through
1761 case Bytecodes::_l2d: tos_in = ltos; break;
1762 case Bytecodes::_f2i: // fall through
1763 case Bytecodes::_f2l: // fall through
1764 case Bytecodes::_f2d: tos_in = ftos; break;
1765 case Bytecodes::_d2i: // fall through
1766 case Bytecodes::_d2l: // fall through
1767 case Bytecodes::_d2f: tos_in = dtos; break;
1768 default : ShouldNotReachHere();
1769 }
1770 switch (bytecode()) {
1771 case Bytecodes::_l2i: // fall through
1772 case Bytecodes::_f2i: // fall through
1773 case Bytecodes::_d2i: // fall through
1774 case Bytecodes::_i2b: // fall through
1775 case Bytecodes::_i2c: // fall through
1776 case Bytecodes::_i2s: tos_out = itos; break;
1777 case Bytecodes::_i2l: // fall through
1778 case Bytecodes::_f2l: // fall through
1779 case Bytecodes::_d2l: tos_out = ltos; break;
1780 case Bytecodes::_i2f: // fall through
1781 case Bytecodes::_l2f: // fall through
1782 case Bytecodes::_d2f: tos_out = ftos; break;
1783 case Bytecodes::_i2d: // fall through
1784 case Bytecodes::_l2d: // fall through
1785 case Bytecodes::_f2d: tos_out = dtos; break;
1786 default : ShouldNotReachHere();
1787 }
1788 transition(tos_in, tos_out);
1789 }
1790 #endif // ASSERT
1791
1792 static const int64_t is_nan = 0x8000000000000000L;
1793
1794 // Conversion
1795 switch (bytecode()) {
1796 case Bytecodes::_i2l:
1797 __ movslq(rax, rax);
1798 break;
1799 case Bytecodes::_i2f:
1800 __ cvtsi2ssl(xmm0, rax);
1801 break;
1802 case Bytecodes::_i2d:
1803 __ cvtsi2sdl(xmm0, rax);
1804 break;
1805 case Bytecodes::_i2b:
1806 __ movsbl(rax, rax);
1807 break;
1808 case Bytecodes::_i2c:
1809 __ movzwl(rax, rax);
1810 break;
1811 case Bytecodes::_i2s:
1812 __ movswl(rax, rax);
1813 break;
1814 case Bytecodes::_l2i:
1815 __ movl(rax, rax);
1816 break;
1817 case Bytecodes::_l2f:
1818 __ cvtsi2ssq(xmm0, rax);
1819 break;
1820 case Bytecodes::_l2d:
1821 __ cvtsi2sdq(xmm0, rax);
1822 break;
1823 case Bytecodes::_f2i:
1824 {
1825 Label L;
1826 __ cvttss2sil(rax, xmm0);
1827 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1828 __ jcc(Assembler::notEqual, L);
1829 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1830 __ bind(L);
1831 }
1832 break;
1833 case Bytecodes::_f2l:
1834 {
1835 Label L;
1836 __ cvttss2siq(rax, xmm0);
1837 // NaN or overflow/underflow?
1838 __ cmp64(rax, ExternalAddress((address) &is_nan));
1839 __ jcc(Assembler::notEqual, L);
1840 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1841 __ bind(L);
1842 }
1843 break;
1844 case Bytecodes::_f2d:
1845 __ cvtss2sd(xmm0, xmm0);
1846 break;
1847 case Bytecodes::_d2i:
1848 {
1849 Label L;
1850 __ cvttsd2sil(rax, xmm0);
1851 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1852 __ jcc(Assembler::notEqual, L);
1853 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1854 __ bind(L);
1855 }
1856 break;
1857 case Bytecodes::_d2l:
1858 {
1859 Label L;
1860 __ cvttsd2siq(rax, xmm0);
1861 // NaN or overflow/underflow?
1862 __ cmp64(rax, ExternalAddress((address) &is_nan));
1863 __ jcc(Assembler::notEqual, L);
1864 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1865 __ bind(L);
1866 }
1867 break;
1868 case Bytecodes::_d2f:
1869 __ cvtsd2ss(xmm0, xmm0);
1870 break;
1871 default:
1872 ShouldNotReachHere();
1873 }
1874 #else
1875 // Checking
1876 #ifdef ASSERT
1877 { TosState tos_in = ilgl;
1878 TosState tos_out = ilgl;
1879 switch (bytecode()) {
1880 case Bytecodes::_i2l: // fall through
1881 case Bytecodes::_i2f: // fall through
1882 case Bytecodes::_i2d: // fall through
1883 case Bytecodes::_i2b: // fall through
1884 case Bytecodes::_i2c: // fall through
1885 case Bytecodes::_i2s: tos_in = itos; break;
1886 case Bytecodes::_l2i: // fall through
1887 case Bytecodes::_l2f: // fall through
1888 case Bytecodes::_l2d: tos_in = ltos; break;
1889 case Bytecodes::_f2i: // fall through
1890 case Bytecodes::_f2l: // fall through
1891 case Bytecodes::_f2d: tos_in = ftos; break;
1892 case Bytecodes::_d2i: // fall through
1893 case Bytecodes::_d2l: // fall through
1894 case Bytecodes::_d2f: tos_in = dtos; break;
1895 default : ShouldNotReachHere();
1896 }
1897 switch (bytecode()) {
1898 case Bytecodes::_l2i: // fall through
1899 case Bytecodes::_f2i: // fall through
1900 case Bytecodes::_d2i: // fall through
1901 case Bytecodes::_i2b: // fall through
1902 case Bytecodes::_i2c: // fall through
1903 case Bytecodes::_i2s: tos_out = itos; break;
1904 case Bytecodes::_i2l: // fall through
1905 case Bytecodes::_f2l: // fall through
1906 case Bytecodes::_d2l: tos_out = ltos; break;
1907 case Bytecodes::_i2f: // fall through
1908 case Bytecodes::_l2f: // fall through
1909 case Bytecodes::_d2f: tos_out = ftos; break;
1910 case Bytecodes::_i2d: // fall through
1911 case Bytecodes::_l2d: // fall through
1912 case Bytecodes::_f2d: tos_out = dtos; break;
1913 default : ShouldNotReachHere();
1914 }
1915 transition(tos_in, tos_out);
1916 }
1917 #endif // ASSERT
1918
1919 // Conversion
1920 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1921 switch (bytecode()) {
1922 case Bytecodes::_i2l:
1923 __ extend_sign(rdx, rax);
1924 break;
1925 case Bytecodes::_i2f:
1926 if (UseSSE >= 1) {
1927 __ cvtsi2ssl(xmm0, rax);
1928 } else {
1929 __ push(rax); // store int on tos
1930 __ fild_s(at_rsp()); // load int to ST0
1931 __ f2ieee(); // truncate to float size
1932 __ pop(rcx); // adjust rsp
1933 }
1934 break;
1935 case Bytecodes::_i2d:
1936 if (UseSSE >= 2) {
1937 __ cvtsi2sdl(xmm0, rax);
1938 } else {
1939 __ push(rax); // add one slot for d2ieee()
1940 __ push(rax); // store int on tos
1941 __ fild_s(at_rsp()); // load int to ST0
1942 __ d2ieee(); // truncate to double size
1943 __ pop(rcx); // adjust rsp
1944 __ pop(rcx);
1945 }
1946 break;
1947 case Bytecodes::_i2b:
1948 __ shll(rax, 24); // truncate upper 24 bits
1949 __ sarl(rax, 24); // and sign-extend byte
1950 LP64_ONLY(__ movsbl(rax, rax));
1951 break;
1952 case Bytecodes::_i2c:
1953 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1954 LP64_ONLY(__ movzwl(rax, rax));
1955 break;
1956 case Bytecodes::_i2s:
1957 __ shll(rax, 16); // truncate upper 16 bits
1958 __ sarl(rax, 16); // and sign-extend short
1959 LP64_ONLY(__ movswl(rax, rax));
1960 break;
1961 case Bytecodes::_l2i:
1962 /* nothing to do */
1963 break;
1964 case Bytecodes::_l2f:
1965 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1966 // 64-bit long values to floats. On 32-bit platforms it is not possible
1967 // to use that instruction with 64-bit operands, therefore the FPU is
1968 // used to perform the conversion.
1969 __ push(rdx); // store long on tos
1970 __ push(rax);
1971 __ fild_d(at_rsp()); // load long to ST0
1972 __ f2ieee(); // truncate to float size
1973 __ pop(rcx); // adjust rsp
1974 __ pop(rcx);
1975 if (UseSSE >= 1) {
1976 __ push_f();
1977 __ pop_f(xmm0);
1978 }
1979 break;
1980 case Bytecodes::_l2d:
1981 // On 32-bit platforms the FPU is used for conversion because on
1982 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1983 // instruction with 64-bit operands.
1984 __ push(rdx); // store long on tos
1985 __ push(rax);
1986 __ fild_d(at_rsp()); // load long to ST0
1987 __ d2ieee(); // truncate to double size
1988 __ pop(rcx); // adjust rsp
1989 __ pop(rcx);
1990 if (UseSSE >= 2) {
1991 __ push_d();
1992 __ pop_d(xmm0);
1993 }
1994 break;
1995 case Bytecodes::_f2i:
1996 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1997 // as it returns 0 for any NaN.
1998 if (UseSSE >= 1) {
1999 __ push_f(xmm0);
2000 } else {
2001 __ push(rcx); // reserve space for argument
2002 __ fstp_s(at_rsp()); // pass float argument on stack
2003 }
2004 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2005 break;
2006 case Bytecodes::_f2l:
2007 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2008 // as it returns 0 for any NaN.
2009 if (UseSSE >= 1) {
2010 __ push_f(xmm0);
2011 } else {
2012 __ push(rcx); // reserve space for argument
2013 __ fstp_s(at_rsp()); // pass float argument on stack
2014 }
2015 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2016 break;
2017 case Bytecodes::_f2d:
2018 if (UseSSE < 1) {
2019 /* nothing to do */
2020 } else if (UseSSE == 1) {
2021 __ push_f(xmm0);
2022 __ pop_f();
2023 } else { // UseSSE >= 2
2024 __ cvtss2sd(xmm0, xmm0);
2025 }
2026 break;
2027 case Bytecodes::_d2i:
2028 if (UseSSE >= 2) {
2029 __ push_d(xmm0);
2030 } else {
2031 __ push(rcx); // reserve space for argument
2032 __ push(rcx);
2033 __ fstp_d(at_rsp()); // pass double argument on stack
2034 }
2035 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2036 break;
2037 case Bytecodes::_d2l:
2038 if (UseSSE >= 2) {
2039 __ push_d(xmm0);
2040 } else {
2041 __ push(rcx); // reserve space for argument
2042 __ push(rcx);
2043 __ fstp_d(at_rsp()); // pass double argument on stack
2044 }
2045 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2046 break;
2047 case Bytecodes::_d2f:
2048 if (UseSSE <= 1) {
2049 __ push(rcx); // reserve space for f2ieee()
2050 __ f2ieee(); // truncate to float size
2051 __ pop(rcx); // adjust rsp
2052 if (UseSSE == 1) {
2053 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2054 // the conversion is performed using the FPU in this case.
2055 __ push_f();
2056 __ pop_f(xmm0);
2057 }
2058 } else { // UseSSE >= 2
2059 __ cvtsd2ss(xmm0, xmm0);
2060 }
2061 break;
2062 default :
2063 ShouldNotReachHere();
2064 }
2065 #endif
2066 }
2067
2068 void TemplateTable::lcmp() {
2069 transition(ltos, itos);
2070 #ifdef _LP64
2071 Label done;
2072 __ pop_l(rdx);
2073 __ cmpq(rdx, rax);
2074 __ movl(rax, -1);
2075 __ jccb(Assembler::less, done);
2076 __ setb(Assembler::notEqual, rax);
2077 __ movzbl(rax, rax);
2078 __ bind(done);
2079 #else
2080
2081 // y = rdx:rax
2082 __ pop_l(rbx, rcx); // get x = rcx:rbx
2083 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2084 __ mov(rax, rcx);
2085 #endif
2086 }
2087
2088 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2089 if ((is_float && UseSSE >= 1) ||
2090 (!is_float && UseSSE >= 2)) {
2091 Label done;
2092 if (is_float) {
2093 // XXX get rid of pop here, use ... reg, mem32
2094 __ pop_f(xmm1);
2095 __ ucomiss(xmm1, xmm0);
2096 } else {
2097 // XXX get rid of pop here, use ... reg, mem64
2098 __ pop_d(xmm1);
2099 __ ucomisd(xmm1, xmm0);
2100 }
2101 if (unordered_result < 0) {
2102 __ movl(rax, -1);
2103 __ jccb(Assembler::parity, done);
2104 __ jccb(Assembler::below, done);
2105 __ setb(Assembler::notEqual, rdx);
2106 __ movzbl(rax, rdx);
2107 } else {
2108 __ movl(rax, 1);
2109 __ jccb(Assembler::parity, done);
2110 __ jccb(Assembler::above, done);
2111 __ movl(rax, 0);
2112 __ jccb(Assembler::equal, done);
2113 __ decrementl(rax);
2114 }
2115 __ bind(done);
2116 } else {
2117 #ifdef _LP64
2118 ShouldNotReachHere();
2119 #else
2120 if (is_float) {
2121 __ fld_s(at_rsp());
2122 } else {
2123 __ fld_d(at_rsp());
2124 __ pop(rdx);
2125 }
2126 __ pop(rcx);
2127 __ fcmp2int(rax, unordered_result < 0);
2128 #endif // _LP64
2129 }
2130 }
2131
2132 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2133 __ get_method(rcx); // rcx holds method
2134 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2135 // holds bumped taken count
2136
2137 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2138 InvocationCounter::counter_offset();
2139 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2140 InvocationCounter::counter_offset();
2141
2142 // Load up edx with the branch displacement
2143 if (is_wide) {
2144 __ movl(rdx, at_bcp(1));
2145 } else {
2146 __ load_signed_short(rdx, at_bcp(1));
2147 }
2148 __ bswapl(rdx);
2149
2150 if (!is_wide) {
2151 __ sarl(rdx, 16);
2152 }
2153 LP64_ONLY(__ movl2ptr(rdx, rdx));
2154
2155 // Handle all the JSR stuff here, then exit.
2156 // It's much shorter and cleaner than intermingling with the non-JSR
2157 // normal-branch stuff occurring below.
2158 if (is_jsr) {
2159 // Pre-load the next target bytecode into rbx
2160 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2161
2162 // compute return address as bci in rax
2163 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2164 in_bytes(ConstMethod::codes_offset())));
2165 __ subptr(rax, Address(rcx, Method::const_offset()));
2166 // Adjust the bcp in r13 by the displacement in rdx
2167 __ addptr(rbcp, rdx);
2168 // jsr returns atos that is not an oop
2169 __ push_i(rax);
2170 __ dispatch_only(vtos, true);
2171 return;
2172 }
2173
2174 // Normal (non-jsr) branch handling
2175
2176 // Adjust the bcp in r13 by the displacement in rdx
2177 __ addptr(rbcp, rdx);
2178
2179 assert(UseLoopCounter || !UseOnStackReplacement,
2180 "on-stack-replacement requires loop counters");
2181 Label backedge_counter_overflow;
2182 Label profile_method;
2183 Label dispatch;
2184 if (UseLoopCounter) {
2185 // increment backedge counter for backward branches
2186 // rax: MDO
2187 // rbx: MDO bumped taken-count
2188 // rcx: method
2189 // rdx: target offset
2190 // r13: target bcp
2191 // r14: locals pointer
2192 __ testl(rdx, rdx); // check if forward or backward branch
2193 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2194
2195 // check if MethodCounters exists
2196 Label has_counters;
2197 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2198 __ testptr(rax, rax);
2199 __ jcc(Assembler::notZero, has_counters);
2200 __ push(rdx);
2201 __ push(rcx);
2202 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2203 rcx);
2204 __ pop(rcx);
2205 __ pop(rdx);
2206 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2207 __ testptr(rax, rax);
2208 __ jcc(Assembler::zero, dispatch);
2209 __ bind(has_counters);
2210
2211 if (TieredCompilation) {
2212 Label no_mdo;
2213 int increment = InvocationCounter::count_increment;
2214 if (ProfileInterpreter) {
2215 // Are we profiling?
2216 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2217 __ testptr(rbx, rbx);
2218 __ jccb(Assembler::zero, no_mdo);
2219 // Increment the MDO backedge counter
2220 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2221 in_bytes(InvocationCounter::counter_offset()));
2222 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2223 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2224 rax, false, Assembler::zero, &backedge_counter_overflow);
2225 __ jmp(dispatch);
2226 }
2227 __ bind(no_mdo);
2228 // Increment backedge counter in MethodCounters*
2229 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2230 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2231 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2232 rax, false, Assembler::zero, &backedge_counter_overflow);
2233 } else { // not TieredCompilation
2234 // increment counter
2235 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2236 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2237 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2238 __ movl(Address(rcx, be_offset), rax); // store counter
2239
2240 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2241
2242 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2243 __ addl(rax, Address(rcx, be_offset)); // add both counters
2244
2245 if (ProfileInterpreter) {
2246 // Test to see if we should create a method data oop
2247 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2248 __ jcc(Assembler::less, dispatch);
2249
2250 // if no method data exists, go to profile method
2251 __ test_method_data_pointer(rax, profile_method);
2252
2253 if (UseOnStackReplacement) {
2254 // check for overflow against rbx which is the MDO taken count
2255 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2256 __ jcc(Assembler::below, dispatch);
2257
2258 // When ProfileInterpreter is on, the backedge_count comes
2259 // from the MethodData*, which value does not get reset on
2260 // the call to frequency_counter_overflow(). To avoid
2261 // excessive calls to the overflow routine while the method is
2262 // being compiled, add a second test to make sure the overflow
2263 // function is called only once every overflow_frequency.
2264 const int overflow_frequency = 1024;
2265 __ andl(rbx, overflow_frequency - 1);
2266 __ jcc(Assembler::zero, backedge_counter_overflow);
2267
2268 }
2269 } else {
2270 if (UseOnStackReplacement) {
2271 // check for overflow against rax, which is the sum of the
2272 // counters
2273 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2274 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2275
2276 }
2277 }
2278 }
2279 __ bind(dispatch);
2280 }
2281
2282 // Pre-load the next target bytecode into rbx
2283 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2284
2285 // continue with the bytecode @ target
2286 // rax: return bci for jsr's, unused otherwise
2287 // rbx: target bytecode
2288 // r13: target bcp
2289 __ dispatch_only(vtos, true);
2290
2291 if (UseLoopCounter) {
2292 if (ProfileInterpreter) {
2293 // Out-of-line code to allocate method data oop.
2294 __ bind(profile_method);
2295 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2296 __ set_method_data_pointer_for_bcp();
2297 __ jmp(dispatch);
2298 }
2299
2300 if (UseOnStackReplacement) {
2301 // invocation counter overflow
2302 __ bind(backedge_counter_overflow);
2303 __ negptr(rdx);
2304 __ addptr(rdx, rbcp); // branch bcp
2305 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2306 __ call_VM(noreg,
2307 CAST_FROM_FN_PTR(address,
2308 InterpreterRuntime::frequency_counter_overflow),
2309 rdx);
2310
2311 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2312 // rdx: scratch
2313 // r14: locals pointer
2314 // r13: bcp
2315 __ testptr(rax, rax); // test result
2316 __ jcc(Assembler::zero, dispatch); // no osr if null
2317 // nmethod may have been invalidated (VM may block upon call_VM return)
2318 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2319 __ jcc(Assembler::notEqual, dispatch);
2320
2321 // We have the address of an on stack replacement routine in rax.
2322 // In preparation of invoking it, first we must migrate the locals
2323 // and monitors from off the interpreter frame on the stack.
2324 // Ensure to save the osr nmethod over the migration call,
2325 // it will be preserved in rbx.
2326 __ mov(rbx, rax);
2327
2328 NOT_LP64(__ get_thread(rcx));
2329
2330 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2331
2332 // rax is OSR buffer, move it to expected parameter location
2333 LP64_ONLY(__ mov(j_rarg0, rax));
2334 NOT_LP64(__ mov(rcx, rax));
2335 // We use j_rarg definitions here so that registers don't conflict as parameter
2336 // registers change across platforms as we are in the midst of a calling
2337 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2338
2339 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2340 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2341
2342 // pop the interpreter frame
2343 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2344 __ leave(); // remove frame anchor
2345 __ pop(retaddr); // get return address
2346 __ mov(rsp, sender_sp); // set sp to sender sp
2347 // Ensure compiled code always sees stack at proper alignment
2348 __ andptr(rsp, -(StackAlignmentInBytes));
2349
2350 // unlike x86 we need no specialized return from compiled code
2351 // to the interpreter or the call stub.
2352
2353 // push the return address
2354 __ push(retaddr);
2355
2356 // and begin the OSR nmethod
2357 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2358 }
2359 }
2360 }
2361
2362 void TemplateTable::if_0cmp(Condition cc) {
2363 transition(itos, vtos);
2364 // assume branch is more often taken than not (loops use backward branches)
2365 Label not_taken;
2366 __ testl(rax, rax);
2367 __ jcc(j_not(cc), not_taken);
2368 branch(false, false);
2369 __ bind(not_taken);
2370 __ profile_not_taken_branch(rax);
2371 }
2372
2373 void TemplateTable::if_icmp(Condition cc) {
2374 transition(itos, vtos);
2375 // assume branch is more often taken than not (loops use backward branches)
2376 Label not_taken;
2377 __ pop_i(rdx);
2378 __ cmpl(rdx, rax);
2379 __ jcc(j_not(cc), not_taken);
2380 branch(false, false);
2381 __ bind(not_taken);
2382 __ profile_not_taken_branch(rax);
2383 }
2384
2385 void TemplateTable::if_nullcmp(Condition cc) {
2386 transition(atos, vtos);
2387 // assume branch is more often taken than not (loops use backward branches)
2388 Label not_taken;
2389 __ testptr(rax, rax);
2390 __ jcc(j_not(cc), not_taken);
2391 branch(false, false);
2392 __ bind(not_taken);
2393 __ profile_not_taken_branch(rax);
2394 }
2395
2396 void TemplateTable::if_acmp(Condition cc) {
2397 transition(atos, vtos);
2398 // assume branch is more often taken than not (loops use backward branches)
2399 Label not_taken;
2400 __ pop_ptr(rdx);
2401 __ cmpoop(rdx, rax);
2402 __ jcc(j_not(cc), not_taken);
2403 branch(false, false);
2404 __ bind(not_taken);
2405 __ profile_not_taken_branch(rax);
2406 }
2407
2408 void TemplateTable::ret() {
2409 transition(vtos, vtos);
2410 locals_index(rbx);
2411 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2412 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2413 __ profile_ret(rbx, rcx);
2414 __ get_method(rax);
2415 __ movptr(rbcp, Address(rax, Method::const_offset()));
2416 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2417 ConstMethod::codes_offset()));
2418 __ dispatch_next(vtos, 0, true);
2419 }
2420
2421 void TemplateTable::wide_ret() {
2422 transition(vtos, vtos);
2423 locals_index_wide(rbx);
2424 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2425 __ profile_ret(rbx, rcx);
2426 __ get_method(rax);
2427 __ movptr(rbcp, Address(rax, Method::const_offset()));
2428 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2429 __ dispatch_next(vtos, 0, true);
2430 }
2431
2432 void TemplateTable::tableswitch() {
2433 Label default_case, continue_execution;
2434 transition(itos, vtos);
2435
2436 // align r13/rsi
2437 __ lea(rbx, at_bcp(BytesPerInt));
2438 __ andptr(rbx, -BytesPerInt);
2439 // load lo & hi
2440 __ movl(rcx, Address(rbx, BytesPerInt));
2441 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2442 __ bswapl(rcx);
2443 __ bswapl(rdx);
2444 // check against lo & hi
2445 __ cmpl(rax, rcx);
2446 __ jcc(Assembler::less, default_case);
2447 __ cmpl(rax, rdx);
2448 __ jcc(Assembler::greater, default_case);
2449 // lookup dispatch offset
2450 __ subl(rax, rcx);
2451 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2452 __ profile_switch_case(rax, rbx, rcx);
2453 // continue execution
2454 __ bind(continue_execution);
2455 __ bswapl(rdx);
2456 LP64_ONLY(__ movl2ptr(rdx, rdx));
2457 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2458 __ addptr(rbcp, rdx);
2459 __ dispatch_only(vtos, true);
2460 // handle default
2461 __ bind(default_case);
2462 __ profile_switch_default(rax);
2463 __ movl(rdx, Address(rbx, 0));
2464 __ jmp(continue_execution);
2465 }
2466
2467 void TemplateTable::lookupswitch() {
2468 transition(itos, itos);
2469 __ stop("lookupswitch bytecode should have been rewritten");
2470 }
2471
2472 void TemplateTable::fast_linearswitch() {
2473 transition(itos, vtos);
2474 Label loop_entry, loop, found, continue_execution;
2475 // bswap rax so we can avoid bswapping the table entries
2476 __ bswapl(rax);
2477 // align r13
2478 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2479 // this instruction (change offsets
2480 // below)
2481 __ andptr(rbx, -BytesPerInt);
2482 // set counter
2483 __ movl(rcx, Address(rbx, BytesPerInt));
2484 __ bswapl(rcx);
2485 __ jmpb(loop_entry);
2486 // table search
2487 __ bind(loop);
2488 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2489 __ jcc(Assembler::equal, found);
2490 __ bind(loop_entry);
2491 __ decrementl(rcx);
2492 __ jcc(Assembler::greaterEqual, loop);
2493 // default case
2494 __ profile_switch_default(rax);
2495 __ movl(rdx, Address(rbx, 0));
2496 __ jmp(continue_execution);
2497 // entry found -> get offset
2498 __ bind(found);
2499 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2500 __ profile_switch_case(rcx, rax, rbx);
2501 // continue execution
2502 __ bind(continue_execution);
2503 __ bswapl(rdx);
2504 __ movl2ptr(rdx, rdx);
2505 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2506 __ addptr(rbcp, rdx);
2507 __ dispatch_only(vtos, true);
2508 }
2509
2510 void TemplateTable::fast_binaryswitch() {
2511 transition(itos, vtos);
2512 // Implementation using the following core algorithm:
2513 //
2514 // int binary_search(int key, LookupswitchPair* array, int n) {
2515 // // Binary search according to "Methodik des Programmierens" by
2516 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2517 // int i = 0;
2518 // int j = n;
2519 // while (i+1 < j) {
2520 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2521 // // with Q: for all i: 0 <= i < n: key < a[i]
2522 // // where a stands for the array and assuming that the (inexisting)
2523 // // element a[n] is infinitely big.
2524 // int h = (i + j) >> 1;
2525 // // i < h < j
2526 // if (key < array[h].fast_match()) {
2527 // j = h;
2528 // } else {
2529 // i = h;
2530 // }
2531 // }
2532 // // R: a[i] <= key < a[i+1] or Q
2533 // // (i.e., if key is within array, i is the correct index)
2534 // return i;
2535 // }
2536
2537 // Register allocation
2538 const Register key = rax; // already set (tosca)
2539 const Register array = rbx;
2540 const Register i = rcx;
2541 const Register j = rdx;
2542 const Register h = rdi;
2543 const Register temp = rsi;
2544
2545 // Find array start
2546 NOT_LP64(__ save_bcp());
2547
2548 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2549 // get rid of this
2550 // instruction (change
2551 // offsets below)
2552 __ andptr(array, -BytesPerInt);
2553
2554 // Initialize i & j
2555 __ xorl(i, i); // i = 0;
2556 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2557
2558 // Convert j into native byteordering
2559 __ bswapl(j);
2560
2561 // And start
2562 Label entry;
2563 __ jmp(entry);
2564
2565 // binary search loop
2566 {
2567 Label loop;
2568 __ bind(loop);
2569 // int h = (i + j) >> 1;
2570 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2571 __ sarl(h, 1); // h = (i + j) >> 1;
2572 // if (key < array[h].fast_match()) {
2573 // j = h;
2574 // } else {
2575 // i = h;
2576 // }
2577 // Convert array[h].match to native byte-ordering before compare
2578 __ movl(temp, Address(array, h, Address::times_8));
2579 __ bswapl(temp);
2580 __ cmpl(key, temp);
2581 // j = h if (key < array[h].fast_match())
2582 __ cmov32(Assembler::less, j, h);
2583 // i = h if (key >= array[h].fast_match())
2584 __ cmov32(Assembler::greaterEqual, i, h);
2585 // while (i+1 < j)
2586 __ bind(entry);
2587 __ leal(h, Address(i, 1)); // i+1
2588 __ cmpl(h, j); // i+1 < j
2589 __ jcc(Assembler::less, loop);
2590 }
2591
2592 // end of binary search, result index is i (must check again!)
2593 Label default_case;
2594 // Convert array[i].match to native byte-ordering before compare
2595 __ movl(temp, Address(array, i, Address::times_8));
2596 __ bswapl(temp);
2597 __ cmpl(key, temp);
2598 __ jcc(Assembler::notEqual, default_case);
2599
2600 // entry found -> j = offset
2601 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2602 __ profile_switch_case(i, key, array);
2603 __ bswapl(j);
2604 LP64_ONLY(__ movslq(j, j));
2605
2606 NOT_LP64(__ restore_bcp());
2607 NOT_LP64(__ restore_locals()); // restore rdi
2608
2609 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2610 __ addptr(rbcp, j);
2611 __ dispatch_only(vtos, true);
2612
2613 // default case -> j = default offset
2614 __ bind(default_case);
2615 __ profile_switch_default(i);
2616 __ movl(j, Address(array, -2 * BytesPerInt));
2617 __ bswapl(j);
2618 LP64_ONLY(__ movslq(j, j));
2619
2620 NOT_LP64(__ restore_bcp());
2621 NOT_LP64(__ restore_locals());
2622
2623 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2624 __ addptr(rbcp, j);
2625 __ dispatch_only(vtos, true);
2626 }
2627
2628 void TemplateTable::_return(TosState state) {
2629 transition(state, state);
2630
2631 assert(_desc->calls_vm(),
2632 "inconsistent calls_vm information"); // call in remove_activation
2633
2634 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2635 assert(state == vtos, "only valid state");
2636 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2637 __ movptr(robj, aaddress(0));
2638 __ load_klass(rdi, robj);
2639 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2640 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2641 Label skip_register_finalizer;
2642 __ jcc(Assembler::zero, skip_register_finalizer);
2643
2644 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2645
2646 __ bind(skip_register_finalizer);
2647 }
2648
2649 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2650 Label no_safepoint;
2651 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2652 #ifdef _LP64
2653 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2654 #else
2655 const Register thread = rdi;
2656 __ get_thread(thread);
2657 __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2658 #endif
2659 __ jcc(Assembler::zero, no_safepoint);
2660 __ push(state);
2661 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2662 InterpreterRuntime::at_safepoint));
2663 __ pop(state);
2664 __ bind(no_safepoint);
2665 }
2666
2667 // Narrow result if state is itos but result type is smaller.
2668 // Need to narrow in the return bytecode rather than in generate_return_entry
2669 // since compiled code callers expect the result to already be narrowed.
2670 if (state == itos) {
2671 __ narrow(rax);
2672 }
2673 __ remove_activation(state, rbcp);
2674
2675 __ jmp(rbcp);
2676 }
2677
2678 // ----------------------------------------------------------------------------
2679 // Volatile variables demand their effects be made known to all CPU's
2680 // in order. Store buffers on most chips allow reads & writes to
2681 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2682 // without some kind of memory barrier (i.e., it's not sufficient that
2683 // the interpreter does not reorder volatile references, the hardware
2684 // also must not reorder them).
2685 //
2686 // According to the new Java Memory Model (JMM):
2687 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2688 // writes act as aquire & release, so:
2689 // (2) A read cannot let unrelated NON-volatile memory refs that
2690 // happen after the read float up to before the read. It's OK for
2691 // non-volatile memory refs that happen before the volatile read to
2692 // float down below it.
2693 // (3) Similar a volatile write cannot let unrelated NON-volatile
2694 // memory refs that happen BEFORE the write float down to after the
2695 // write. It's OK for non-volatile memory refs that happen after the
2696 // volatile write to float up before it.
2697 //
2698 // We only put in barriers around volatile refs (they are expensive),
2699 // not _between_ memory refs (that would require us to track the
2700 // flavor of the previous memory refs). Requirements (2) and (3)
2701 // require some barriers before volatile stores and after volatile
2702 // loads. These nearly cover requirement (1) but miss the
2703 // volatile-store-volatile-load case. This final case is placed after
2704 // volatile-stores although it could just as well go before
2705 // volatile-loads.
2706
2707 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2708 // Helper function to insert a is-volatile test and memory barrier
2709 if(!os::is_MP()) return; // Not needed on single CPU
2710 __ membar(order_constraint);
2711 }
2712
2713 void TemplateTable::resolve_cache_and_index(int byte_no,
2714 Register Rcache,
2715 Register index,
2716 size_t index_size) {
2717 const Register temp = rbx;
2718 assert_different_registers(Rcache, index, temp);
2719
2720 Label resolved;
2721
2722 Bytecodes::Code code = bytecode();
2723 switch (code) {
2724 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2725 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2726 default: break;
2727 }
2728
2729 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2730 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2731 __ cmpl(temp, code); // have we resolved this bytecode?
2732 __ jcc(Assembler::equal, resolved);
2733
2734 // resolve first time through
2735 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2736 __ movl(temp, code);
2737 __ call_VM(noreg, entry, temp);
2738 // Update registers with resolved info
2739 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2740 __ bind(resolved);
2741 }
2742
2743 // The cache and index registers must be set before call
2744 void TemplateTable::load_field_cp_cache_entry(Register obj,
2745 Register cache,
2746 Register index,
2747 Register off,
2748 Register flags,
2749 bool is_static = false) {
2750 assert_different_registers(cache, index, flags, off);
2751
2752 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2753 // Field offset
2754 __ movptr(off, Address(cache, index, Address::times_ptr,
2755 in_bytes(cp_base_offset +
2756 ConstantPoolCacheEntry::f2_offset())));
2757 // Flags
2758 __ movl(flags, Address(cache, index, Address::times_ptr,
2759 in_bytes(cp_base_offset +
2760 ConstantPoolCacheEntry::flags_offset())));
2761
2762 // klass overwrite register
2763 if (is_static) {
2764 __ movptr(obj, Address(cache, index, Address::times_ptr,
2765 in_bytes(cp_base_offset +
2766 ConstantPoolCacheEntry::f1_offset())));
2767 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2768 __ movptr(obj, Address(obj, mirror_offset));
2769 __ resolve_oop_handle(obj);
2770 }
2771 }
2772
2773 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2774 Register method,
2775 Register itable_index,
2776 Register flags,
2777 bool is_invokevirtual,
2778 bool is_invokevfinal, /*unused*/
2779 bool is_invokedynamic) {
2780 // setup registers
2781 const Register cache = rcx;
2782 const Register index = rdx;
2783 assert_different_registers(method, flags);
2784 assert_different_registers(method, cache, index);
2785 assert_different_registers(itable_index, flags);
2786 assert_different_registers(itable_index, cache, index);
2787 // determine constant pool cache field offsets
2788 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2789 const int method_offset = in_bytes(
2790 ConstantPoolCache::base_offset() +
2791 ((byte_no == f2_byte)
2792 ? ConstantPoolCacheEntry::f2_offset()
2793 : ConstantPoolCacheEntry::f1_offset()));
2794 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2795 ConstantPoolCacheEntry::flags_offset());
2796 // access constant pool cache fields
2797 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2798 ConstantPoolCacheEntry::f2_offset());
2799
2800 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2801 resolve_cache_and_index(byte_no, cache, index, index_size);
2802 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2803
2804 if (itable_index != noreg) {
2805 // pick up itable or appendix index from f2 also:
2806 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2807 }
2808 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2809 }
2810
2811 // The registers cache and index expected to be set before call.
2812 // Correct values of the cache and index registers are preserved.
2813 void TemplateTable::jvmti_post_field_access(Register cache,
2814 Register index,
2815 bool is_static,
2816 bool has_tos) {
2817 if (JvmtiExport::can_post_field_access()) {
2818 // Check to see if a field access watch has been set before we take
2819 // the time to call into the VM.
2820 Label L1;
2821 assert_different_registers(cache, index, rax);
2822 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2823 __ testl(rax,rax);
2824 __ jcc(Assembler::zero, L1);
2825
2826 // cache entry pointer
2827 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2828 __ shll(index, LogBytesPerWord);
2829 __ addptr(cache, index);
2830 if (is_static) {
2831 __ xorptr(rax, rax); // NULL object reference
2832 } else {
2833 __ pop(atos); // Get the object
2834 __ verify_oop(rax);
2835 __ push(atos); // Restore stack state
2836 }
2837 // rax,: object pointer or NULL
2838 // cache: cache entry pointer
2839 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2840 rax, cache);
2841 __ get_cache_and_index_at_bcp(cache, index, 1);
2842 __ bind(L1);
2843 }
2844 }
2845
2846 void TemplateTable::pop_and_check_object(Register r) {
2847 __ pop_ptr(r);
2848 __ null_check(r); // for field access must check obj.
2849 __ verify_oop(r);
2850 }
2851
2852 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2853 transition(vtos, vtos);
2854
2855 const Register cache = rcx;
2856 const Register index = rdx;
2857 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2858 const Register off = rbx;
2859 const Register flags = rax;
2860 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2861
2862 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2863 jvmti_post_field_access(cache, index, is_static, false);
2864 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2865
2866 if (!is_static) pop_and_check_object(obj);
2867 __ resolve_for_read(OOP_NOT_NULL, obj);
2868
2869 const Address field(obj, off, Address::times_1, 0*wordSize);
2870 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2871
2872 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2873
2874 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2875 // Make sure we don't need to mask edx after the above shift
2876 assert(btos == 0, "change code, btos != 0");
2877
2878 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2879
2880 __ jcc(Assembler::notZero, notByte);
2881 // btos
2882 __ load_signed_byte(rax, field);
2883 __ push(btos);
2884 // Rewrite bytecode to be faster
2885 if (!is_static && rc == may_rewrite) {
2886 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2887 }
2888 __ jmp(Done);
2889
2890 __ bind(notByte);
2891 __ cmpl(flags, ztos);
2892 __ jcc(Assembler::notEqual, notBool);
2893
2894 // ztos (same code as btos)
2895 __ load_signed_byte(rax, field);
2896 __ push(ztos);
2897 // Rewrite bytecode to be faster
2898 if (!is_static && rc == may_rewrite) {
2899 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2900 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2901 }
2902 __ jmp(Done);
2903
2904 __ bind(notBool);
2905 __ cmpl(flags, atos);
2906 __ jcc(Assembler::notEqual, notObj);
2907 // atos
2908 do_oop_load(_masm, field, rax);
2909 __ push(atos);
2910 if (!is_static && rc == may_rewrite) {
2911 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2912 }
2913 __ jmp(Done);
2914
2915 __ bind(notObj);
2916 __ cmpl(flags, itos);
2917 __ jcc(Assembler::notEqual, notInt);
2918 // itos
2919 __ movl(rax, field);
2920 __ push(itos);
2921 // Rewrite bytecode to be faster
2922 if (!is_static && rc == may_rewrite) {
2923 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2924 }
2925 __ jmp(Done);
2926
2927 __ bind(notInt);
2928 __ cmpl(flags, ctos);
2929 __ jcc(Assembler::notEqual, notChar);
2930 // ctos
2931 __ load_unsigned_short(rax, field);
2932 __ push(ctos);
2933 // Rewrite bytecode to be faster
2934 if (!is_static && rc == may_rewrite) {
2935 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2936 }
2937 __ jmp(Done);
2938
2939 __ bind(notChar);
2940 __ cmpl(flags, stos);
2941 __ jcc(Assembler::notEqual, notShort);
2942 // stos
2943 __ load_signed_short(rax, field);
2944 __ push(stos);
2945 // Rewrite bytecode to be faster
2946 if (!is_static && rc == may_rewrite) {
2947 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2948 }
2949 __ jmp(Done);
2950
2951 __ bind(notShort);
2952 __ cmpl(flags, ltos);
2953 __ jcc(Assembler::notEqual, notLong);
2954 // ltos
2955
2956 #ifndef _LP64
2957 // Generate code as if volatile. There just aren't enough registers to
2958 // save that information and this code is faster than the test.
2959 __ fild_d(field); // Must load atomically
2960 __ subptr(rsp,2*wordSize); // Make space for store
2961 __ fistp_d(Address(rsp,0));
2962 __ pop(rax);
2963 __ pop(rdx);
2964 #else
2965 __ movq(rax, field);
2966 #endif
2967
2968 __ push(ltos);
2969 // Rewrite bytecode to be faster
2970 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2971 __ jmp(Done);
2972
2973 __ bind(notLong);
2974 __ cmpl(flags, ftos);
2975 __ jcc(Assembler::notEqual, notFloat);
2976 // ftos
2977
2978 __ load_float(field);
2979 __ push(ftos);
2980 // Rewrite bytecode to be faster
2981 if (!is_static && rc == may_rewrite) {
2982 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2983 }
2984 __ jmp(Done);
2985
2986 __ bind(notFloat);
2987 #ifdef ASSERT
2988 __ cmpl(flags, dtos);
2989 __ jcc(Assembler::notEqual, notDouble);
2990 #endif
2991 // dtos
2992 __ load_double(field);
2993 __ push(dtos);
2994 // Rewrite bytecode to be faster
2995 if (!is_static && rc == may_rewrite) {
2996 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2997 }
2998 #ifdef ASSERT
2999 __ jmp(Done);
3000
3001
3002 __ bind(notDouble);
3003 __ stop("Bad state");
3004 #endif
3005
3006 __ bind(Done);
3007 // [jk] not needed currently
3008 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3009 // Assembler::LoadStore));
3010 }
3011
3012 void TemplateTable::getfield(int byte_no) {
3013 getfield_or_static(byte_no, false);
3014 }
3015
3016 void TemplateTable::nofast_getfield(int byte_no) {
3017 getfield_or_static(byte_no, false, may_not_rewrite);
3018 }
3019
3020 void TemplateTable::getstatic(int byte_no) {
3021 getfield_or_static(byte_no, true);
3022 }
3023
3024
3025 // The registers cache and index expected to be set before call.
3026 // The function may destroy various registers, just not the cache and index registers.
3027 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3028
3029 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3030 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3031 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3032 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3033
3034 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3035
3036 if (JvmtiExport::can_post_field_modification()) {
3037 // Check to see if a field modification watch has been set before
3038 // we take the time to call into the VM.
3039 Label L1;
3040 assert_different_registers(cache, index, rax);
3041 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3042 __ testl(rax, rax);
3043 __ jcc(Assembler::zero, L1);
3044
3045 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3046
3047
3048 if (is_static) {
3049 // Life is simple. Null out the object pointer.
3050 __ xorl(RBX, RBX);
3051
3052 } else {
3053 // Life is harder. The stack holds the value on top, followed by
3054 // the object. We don't know the size of the value, though; it
3055 // could be one or two words depending on its type. As a result,
3056 // we must find the type to determine where the object is.
3057 #ifndef _LP64
3058 Label two_word, valsize_known;
3059 #endif
3060 __ movl(RCX, Address(robj, RDX,
3061 Address::times_ptr,
3062 in_bytes(cp_base_offset +
3063 ConstantPoolCacheEntry::flags_offset())));
3064 NOT_LP64(__ mov(rbx, rsp));
3065 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3066
3067 // Make sure we don't need to mask rcx after the above shift
3068 ConstantPoolCacheEntry::verify_tos_state_shift();
3069 #ifdef _LP64
3070 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3071 __ cmpl(c_rarg3, ltos);
3072 __ cmovptr(Assembler::equal,
3073 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3074 __ cmpl(c_rarg3, dtos);
3075 __ cmovptr(Assembler::equal,
3076 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3077 #else
3078 __ cmpl(rcx, ltos);
3079 __ jccb(Assembler::equal, two_word);
3080 __ cmpl(rcx, dtos);
3081 __ jccb(Assembler::equal, two_word);
3082 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3083 __ jmpb(valsize_known);
3084
3085 __ bind(two_word);
3086 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3087
3088 __ bind(valsize_known);
3089 // setup object pointer
3090 __ movptr(rbx, Address(rbx, 0));
3091 #endif
3092 }
3093 // cache entry pointer
3094 __ addptr(robj, in_bytes(cp_base_offset));
3095 __ shll(RDX, LogBytesPerWord);
3096 __ addptr(robj, RDX);
3097 // object (tos)
3098 __ mov(RCX, rsp);
3099 // c_rarg1: object pointer set up above (NULL if static)
3100 // c_rarg2: cache entry pointer
3101 // c_rarg3: jvalue object on the stack
3102 __ call_VM(noreg,
3103 CAST_FROM_FN_PTR(address,
3104 InterpreterRuntime::post_field_modification),
3105 RBX, robj, RCX);
3106 __ get_cache_and_index_at_bcp(cache, index, 1);
3107 __ bind(L1);
3108 }
3109 }
3110
3111 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3112 transition(vtos, vtos);
3113
3114 const Register cache = rcx;
3115 const Register index = rdx;
3116 const Register obj = rcx;
3117 const Register off = rbx;
3118 const Register flags = rax;
3119 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3120
3121 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3122 jvmti_post_field_mod(cache, index, is_static);
3123 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3124
3125 // [jk] not needed currently
3126 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3127 // Assembler::StoreStore));
3128
3129 Label notVolatile, Done;
3130 __ movl(rdx, flags);
3131 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3132 __ andl(rdx, 0x1);
3133
3134 // field addresses
3135 const Address field(obj, off, Address::times_1, 0*wordSize);
3136 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3137
3138 Label notByte, notBool, notInt, notShort, notChar,
3139 notLong, notFloat, notObj, notDouble;
3140
3141 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3142
3143 assert(btos == 0, "change code, btos != 0");
3144 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3145 __ jcc(Assembler::notZero, notByte);
3146
3147 // btos
3148 {
3149 __ pop(btos);
3150 if (!is_static) pop_and_check_object(obj);
3151 __ resolve_for_write(OOP_NOT_NULL, obj);
3152 __ movb(field, rax);
3153 if (!is_static && rc == may_rewrite) {
3154 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3155 }
3156 __ jmp(Done);
3157 }
3158
3159 __ bind(notByte);
3160 __ cmpl(flags, ztos);
3161 __ jcc(Assembler::notEqual, notBool);
3162
3163 // ztos
3164 {
3165 __ pop(ztos);
3166 if (!is_static) pop_and_check_object(obj);
3167 __ resolve_for_write(OOP_NOT_NULL, obj);
3168 __ andl(rax, 0x1);
3169 __ movb(field, rax);
3170 if (!is_static && rc == may_rewrite) {
3171 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3172 }
3173 __ jmp(Done);
3174 }
3175
3176 __ bind(notBool);
3177 __ cmpl(flags, atos);
3178 __ jcc(Assembler::notEqual, notObj);
3179
3180 // atos
3181 {
3182 __ pop(atos);
3183 if (!is_static) pop_and_check_object(obj);
3184 __ resolve_for_write(OOP_NOT_NULL, obj);
3185 // Store into the field
3186 do_oop_store(_masm, field, rax);
3187 if (!is_static && rc == may_rewrite) {
3188 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3189 }
3190 __ jmp(Done);
3191 }
3192
3193 __ bind(notObj);
3194 __ cmpl(flags, itos);
3195 __ jcc(Assembler::notEqual, notInt);
3196
3197 // itos
3198 {
3199 __ pop(itos);
3200 if (!is_static) pop_and_check_object(obj);
3201 __ resolve_for_write(OOP_NOT_NULL, obj);
3202 __ movl(field, rax);
3203 if (!is_static && rc == may_rewrite) {
3204 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3205 }
3206 __ jmp(Done);
3207 }
3208
3209 __ bind(notInt);
3210 __ cmpl(flags, ctos);
3211 __ jcc(Assembler::notEqual, notChar);
3212
3213 // ctos
3214 {
3215 __ pop(ctos);
3216 if (!is_static) pop_and_check_object(obj);
3217 __ resolve_for_write(OOP_NOT_NULL, obj);
3218 __ movw(field, rax);
3219 if (!is_static && rc == may_rewrite) {
3220 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3221 }
3222 __ jmp(Done);
3223 }
3224
3225 __ bind(notChar);
3226 __ cmpl(flags, stos);
3227 __ jcc(Assembler::notEqual, notShort);
3228
3229 // stos
3230 {
3231 __ pop(stos);
3232 if (!is_static) pop_and_check_object(obj);
3233 __ resolve_for_write(OOP_NOT_NULL, obj);
3234 __ movw(field, rax);
3235 if (!is_static && rc == may_rewrite) {
3236 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3237 }
3238 __ jmp(Done);
3239 }
3240
3241 __ bind(notShort);
3242 __ cmpl(flags, ltos);
3243 __ jcc(Assembler::notEqual, notLong);
3244
3245 // ltos
3246 #ifdef _LP64
3247 {
3248 __ pop(ltos);
3249 if (!is_static) pop_and_check_object(obj);
3250 __ resolve_for_write(OOP_NOT_NULL, obj);
3251 __ movq(field, rax);
3252 if (!is_static && rc == may_rewrite) {
3253 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3254 }
3255 __ jmp(Done);
3256 }
3257 #else
3258 {
3259 Label notVolatileLong;
3260 __ testl(rdx, rdx);
3261 __ jcc(Assembler::zero, notVolatileLong);
3262
3263 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3264 if (!is_static) pop_and_check_object(obj);
3265
3266 // Replace with real volatile test
3267 __ push(rdx);
3268 __ push(rax); // Must update atomically with FIST
3269 __ fild_d(Address(rsp,0)); // So load into FPU register
3270 __ fistp_d(field); // and put into memory atomically
3271 __ addptr(rsp, 2*wordSize);
3272 // volatile_barrier();
3273 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3274 Assembler::StoreStore));
3275 // Don't rewrite volatile version
3276 __ jmp(notVolatile);
3277
3278 __ bind(notVolatileLong);
3279
3280 __ pop(ltos); // overwrites rdx
3281 if (!is_static) pop_and_check_object(obj);
3282 __ movptr(hi, rdx);
3283 __ movptr(field, rax);
3284 // Don't rewrite to _fast_lputfield for potential volatile case.
3285 __ jmp(notVolatile);
3286 }
3287 #endif // _LP64
3288
3289 __ bind(notLong);
3290 __ cmpl(flags, ftos);
3291 __ jcc(Assembler::notEqual, notFloat);
3292
3293 // ftos
3294 {
3295 __ pop(ftos);
3296 if (!is_static) pop_and_check_object(obj);
3297 __ resolve_for_write(OOP_NOT_NULL, obj);
3298 __ store_float(field);
3299 if (!is_static && rc == may_rewrite) {
3300 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3301 }
3302 __ jmp(Done);
3303 }
3304
3305 __ bind(notFloat);
3306 #ifdef ASSERT
3307 __ cmpl(flags, dtos);
3308 __ jcc(Assembler::notEqual, notDouble);
3309 #endif
3310
3311 // dtos
3312 {
3313 __ pop(dtos);
3314 if (!is_static) pop_and_check_object(obj);
3315 __ resolve_for_write(OOP_NOT_NULL, obj);
3316 __ store_double(field);
3317 if (!is_static && rc == may_rewrite) {
3318 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3319 }
3320 }
3321
3322 #ifdef ASSERT
3323 __ jmp(Done);
3324
3325 __ bind(notDouble);
3326 __ stop("Bad state");
3327 #endif
3328
3329 __ bind(Done);
3330
3331 // Check for volatile store
3332 __ testl(rdx, rdx);
3333 __ jcc(Assembler::zero, notVolatile);
3334 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3335 Assembler::StoreStore));
3336 __ bind(notVolatile);
3337 }
3338
3339 void TemplateTable::putfield(int byte_no) {
3340 putfield_or_static(byte_no, false);
3341 }
3342
3343 void TemplateTable::nofast_putfield(int byte_no) {
3344 putfield_or_static(byte_no, false, may_not_rewrite);
3345 }
3346
3347 void TemplateTable::putstatic(int byte_no) {
3348 putfield_or_static(byte_no, true);
3349 }
3350
3351 void TemplateTable::jvmti_post_fast_field_mod() {
3352
3353 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3354
3355 if (JvmtiExport::can_post_field_modification()) {
3356 // Check to see if a field modification watch has been set before
3357 // we take the time to call into the VM.
3358 Label L2;
3359 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3360 __ testl(scratch, scratch);
3361 __ jcc(Assembler::zero, L2);
3362 __ pop_ptr(rbx); // copy the object pointer from tos
3363 __ verify_oop(rbx);
3364 __ push_ptr(rbx); // put the object pointer back on tos
3365 // Save tos values before call_VM() clobbers them. Since we have
3366 // to do it for every data type, we use the saved values as the
3367 // jvalue object.
3368 switch (bytecode()) { // load values into the jvalue object
3369 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3370 case Bytecodes::_fast_bputfield: // fall through
3371 case Bytecodes::_fast_zputfield: // fall through
3372 case Bytecodes::_fast_sputfield: // fall through
3373 case Bytecodes::_fast_cputfield: // fall through
3374 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3375 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3376 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3377 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3378
3379 default:
3380 ShouldNotReachHere();
3381 }
3382 __ mov(scratch, rsp); // points to jvalue on the stack
3383 // access constant pool cache entry
3384 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3385 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3386 __ verify_oop(rbx);
3387 // rbx: object pointer copied above
3388 // c_rarg2: cache entry pointer
3389 // c_rarg3: jvalue object on the stack
3390 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3391 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3392
3393 switch (bytecode()) { // restore tos values
3394 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3395 case Bytecodes::_fast_bputfield: // fall through
3396 case Bytecodes::_fast_zputfield: // fall through
3397 case Bytecodes::_fast_sputfield: // fall through
3398 case Bytecodes::_fast_cputfield: // fall through
3399 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3400 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3401 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3402 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3403 default: break;
3404 }
3405 __ bind(L2);
3406 }
3407 }
3408
3409 void TemplateTable::fast_storefield(TosState state) {
3410 transition(state, vtos);
3411
3412 ByteSize base = ConstantPoolCache::base_offset();
3413
3414 jvmti_post_fast_field_mod();
3415
3416 // access constant pool cache
3417 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3418
3419 // test for volatile with rdx but rdx is tos register for lputfield.
3420 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3421 in_bytes(base +
3422 ConstantPoolCacheEntry::flags_offset())));
3423
3424 // replace index with field offset from cache entry
3425 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3426 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3427
3428 // [jk] not needed currently
3429 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3430 // Assembler::StoreStore));
3431
3432 Label notVolatile;
3433 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3434 __ andl(rdx, 0x1);
3435
3436 // Get object from stack
3437 pop_and_check_object(rcx);
3438 __ resolve_for_write(OOP_NOT_NULL, rcx);
3439
3440 // field address
3441 const Address field(rcx, rbx, Address::times_1);
3442
3443 // access field
3444 switch (bytecode()) {
3445 case Bytecodes::_fast_aputfield:
3446 do_oop_store(_masm, field, rax);
3447 break;
3448 case Bytecodes::_fast_lputfield:
3449 #ifdef _LP64
3450 __ movq(field, rax);
3451 #else
3452 __ stop("should not be rewritten");
3453 #endif
3454 break;
3455 case Bytecodes::_fast_iputfield:
3456 __ movl(field, rax);
3457 break;
3458 case Bytecodes::_fast_zputfield:
3459 __ andl(rax, 0x1); // boolean is true if LSB is 1
3460 // fall through to bputfield
3461 case Bytecodes::_fast_bputfield:
3462 __ movb(field, rax);
3463 break;
3464 case Bytecodes::_fast_sputfield:
3465 // fall through
3466 case Bytecodes::_fast_cputfield:
3467 __ movw(field, rax);
3468 break;
3469 case Bytecodes::_fast_fputfield:
3470 __ store_float(field);
3471 break;
3472 case Bytecodes::_fast_dputfield:
3473 __ store_double(field);
3474 break;
3475 default:
3476 ShouldNotReachHere();
3477 }
3478
3479 // Check for volatile store
3480 __ testl(rdx, rdx);
3481 __ jcc(Assembler::zero, notVolatile);
3482 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3483 Assembler::StoreStore));
3484 __ bind(notVolatile);
3485 }
3486
3487 void TemplateTable::fast_accessfield(TosState state) {
3488 transition(atos, state);
3489
3490 // Do the JVMTI work here to avoid disturbing the register state below
3491 if (JvmtiExport::can_post_field_access()) {
3492 // Check to see if a field access watch has been set before we
3493 // take the time to call into the VM.
3494 Label L1;
3495 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3496 __ testl(rcx, rcx);
3497 __ jcc(Assembler::zero, L1);
3498 // access constant pool cache entry
3499 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3500 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3501 __ verify_oop(rax);
3502 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3503 LP64_ONLY(__ mov(c_rarg1, rax));
3504 // c_rarg1: object pointer copied above
3505 // c_rarg2: cache entry pointer
3506 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3507 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3508 __ pop_ptr(rax); // restore object pointer
3509 __ bind(L1);
3510 }
3511
3512 // access constant pool cache
3513 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3514 // replace index with field offset from cache entry
3515 // [jk] not needed currently
3516 // if (os::is_MP()) {
3517 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3518 // in_bytes(ConstantPoolCache::base_offset() +
3519 // ConstantPoolCacheEntry::flags_offset())));
3520 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3521 // __ andl(rdx, 0x1);
3522 // }
3523 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3524 in_bytes(ConstantPoolCache::base_offset() +
3525 ConstantPoolCacheEntry::f2_offset())));
3526
3527 // rax: object
3528 __ verify_oop(rax);
3529 __ null_check(rax);
3530 __ resolve_for_read(OOP_NOT_NULL, rax);
3531 Address field(rax, rbx, Address::times_1);
3532
3533 // access field
3534 switch (bytecode()) {
3535 case Bytecodes::_fast_agetfield:
3536 do_oop_load(_masm, field, rax);
3537 __ verify_oop(rax);
3538 break;
3539 case Bytecodes::_fast_lgetfield:
3540 #ifdef _LP64
3541 __ movq(rax, field);
3542 #else
3543 __ stop("should not be rewritten");
3544 #endif
3545 break;
3546 case Bytecodes::_fast_igetfield:
3547 __ movl(rax, field);
3548 break;
3549 case Bytecodes::_fast_bgetfield:
3550 __ movsbl(rax, field);
3551 break;
3552 case Bytecodes::_fast_sgetfield:
3553 __ load_signed_short(rax, field);
3554 break;
3555 case Bytecodes::_fast_cgetfield:
3556 __ load_unsigned_short(rax, field);
3557 break;
3558 case Bytecodes::_fast_fgetfield:
3559 __ load_float(field);
3560 break;
3561 case Bytecodes::_fast_dgetfield:
3562 __ load_double(field);
3563 break;
3564 default:
3565 ShouldNotReachHere();
3566 }
3567 // [jk] not needed currently
3568 // if (os::is_MP()) {
3569 // Label notVolatile;
3570 // __ testl(rdx, rdx);
3571 // __ jcc(Assembler::zero, notVolatile);
3572 // __ membar(Assembler::LoadLoad);
3573 // __ bind(notVolatile);
3574 //};
3575 }
3576
3577 void TemplateTable::fast_xaccess(TosState state) {
3578 transition(vtos, state);
3579
3580 // get receiver
3581 __ movptr(rax, aaddress(0));
3582 // access constant pool cache
3583 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3584 __ movptr(rbx,
3585 Address(rcx, rdx, Address::times_ptr,
3586 in_bytes(ConstantPoolCache::base_offset() +
3587 ConstantPoolCacheEntry::f2_offset())));
3588 // make sure exception is reported in correct bcp range (getfield is
3589 // next instruction)
3590 __ increment(rbcp);
3591 __ null_check(rax);
3592 __ resolve_for_read(OOP_NOT_NULL, rax);
3593 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3594 switch (state) {
3595 case itos:
3596 __ movl(rax, field);
3597 break;
3598 case atos:
3599 do_oop_load(_masm, field, rax);
3600 __ verify_oop(rax);
3601 break;
3602 case ftos:
3603 __ load_float(field);
3604 break;
3605 default:
3606 ShouldNotReachHere();
3607 }
3608
3609 // [jk] not needed currently
3610 // if (os::is_MP()) {
3611 // Label notVolatile;
3612 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3613 // in_bytes(ConstantPoolCache::base_offset() +
3614 // ConstantPoolCacheEntry::flags_offset())));
3615 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3616 // __ testl(rdx, 0x1);
3617 // __ jcc(Assembler::zero, notVolatile);
3618 // __ membar(Assembler::LoadLoad);
3619 // __ bind(notVolatile);
3620 // }
3621
3622 __ decrement(rbcp);
3623 }
3624
3625 //-----------------------------------------------------------------------------
3626 // Calls
3627
3628 void TemplateTable::count_calls(Register method, Register temp) {
3629 // implemented elsewhere
3630 ShouldNotReachHere();
3631 }
3632
3633 void TemplateTable::prepare_invoke(int byte_no,
3634 Register method, // linked method (or i-klass)
3635 Register index, // itable index, MethodType, etc.
3636 Register recv, // if caller wants to see it
3637 Register flags // if caller wants to test it
3638 ) {
3639 // determine flags
3640 const Bytecodes::Code code = bytecode();
3641 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3642 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3643 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3644 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3645 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3646 const bool load_receiver = (recv != noreg);
3647 const bool save_flags = (flags != noreg);
3648 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3649 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3650 assert(flags == noreg || flags == rdx, "");
3651 assert(recv == noreg || recv == rcx, "");
3652
3653 // setup registers & access constant pool cache
3654 if (recv == noreg) recv = rcx;
3655 if (flags == noreg) flags = rdx;
3656 assert_different_registers(method, index, recv, flags);
3657
3658 // save 'interpreter return address'
3659 __ save_bcp();
3660
3661 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3662
3663 // maybe push appendix to arguments (just before return address)
3664 if (is_invokedynamic || is_invokehandle) {
3665 Label L_no_push;
3666 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3667 __ jcc(Assembler::zero, L_no_push);
3668 // Push the appendix as a trailing parameter.
3669 // This must be done before we get the receiver,
3670 // since the parameter_size includes it.
3671 __ push(rbx);
3672 __ mov(rbx, index);
3673 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3674 __ load_resolved_reference_at_index(index, rbx);
3675 __ pop(rbx);
3676 __ push(index); // push appendix (MethodType, CallSite, etc.)
3677 __ bind(L_no_push);
3678 }
3679
3680 // load receiver if needed (after appendix is pushed so parameter size is correct)
3681 // Note: no return address pushed yet
3682 if (load_receiver) {
3683 __ movl(recv, flags);
3684 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3685 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3686 const int receiver_is_at_end = -1; // back off one slot to get receiver
3687 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3688 __ movptr(recv, recv_addr);
3689 __ verify_oop(recv);
3690 }
3691
3692 if (save_flags) {
3693 __ movl(rbcp, flags);
3694 }
3695
3696 // compute return type
3697 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3698 // Make sure we don't need to mask flags after the above shift
3699 ConstantPoolCacheEntry::verify_tos_state_shift();
3700 // load return address
3701 {
3702 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3703 ExternalAddress table(table_addr);
3704 LP64_ONLY(__ lea(rscratch1, table));
3705 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3706 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3707 }
3708
3709 // push return address
3710 __ push(flags);
3711
3712 // Restore flags value from the constant pool cache, and restore rsi
3713 // for later null checks. r13 is the bytecode pointer
3714 if (save_flags) {
3715 __ movl(flags, rbcp);
3716 __ restore_bcp();
3717 }
3718 }
3719
3720 void TemplateTable::invokevirtual_helper(Register index,
3721 Register recv,
3722 Register flags) {
3723 // Uses temporary registers rax, rdx
3724 assert_different_registers(index, recv, rax, rdx);
3725 assert(index == rbx, "");
3726 assert(recv == rcx, "");
3727
3728 // Test for an invoke of a final method
3729 Label notFinal;
3730 __ movl(rax, flags);
3731 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3732 __ jcc(Assembler::zero, notFinal);
3733
3734 const Register method = index; // method must be rbx
3735 assert(method == rbx,
3736 "Method* must be rbx for interpreter calling convention");
3737
3738 // do the call - the index is actually the method to call
3739 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3740
3741 // It's final, need a null check here!
3742 __ null_check(recv);
3743
3744 // profile this call
3745 __ profile_final_call(rax);
3746 __ profile_arguments_type(rax, method, rbcp, true);
3747
3748 __ jump_from_interpreted(method, rax);
3749
3750 __ bind(notFinal);
3751
3752 // get receiver klass
3753 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3754 __ load_klass(rax, recv);
3755
3756 // profile this call
3757 __ profile_virtual_call(rax, rlocals, rdx);
3758 // get target Method* & entry point
3759 __ lookup_virtual_method(rax, index, method);
3760 __ profile_called_method(method, rdx, rbcp);
3761
3762 __ profile_arguments_type(rdx, method, rbcp, true);
3763 __ jump_from_interpreted(method, rdx);
3764 }
3765
3766 void TemplateTable::invokevirtual(int byte_no) {
3767 transition(vtos, vtos);
3768 assert(byte_no == f2_byte, "use this argument");
3769 prepare_invoke(byte_no,
3770 rbx, // method or vtable index
3771 noreg, // unused itable index
3772 rcx, rdx); // recv, flags
3773
3774 // rbx: index
3775 // rcx: receiver
3776 // rdx: flags
3777
3778 invokevirtual_helper(rbx, rcx, rdx);
3779 }
3780
3781 void TemplateTable::invokespecial(int byte_no) {
3782 transition(vtos, vtos);
3783 assert(byte_no == f1_byte, "use this argument");
3784 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3785 rcx); // get receiver also for null check
3786 __ verify_oop(rcx);
3787 __ null_check(rcx);
3788 // do the call
3789 __ profile_call(rax);
3790 __ profile_arguments_type(rax, rbx, rbcp, false);
3791 __ jump_from_interpreted(rbx, rax);
3792 }
3793
3794 void TemplateTable::invokestatic(int byte_no) {
3795 transition(vtos, vtos);
3796 assert(byte_no == f1_byte, "use this argument");
3797 prepare_invoke(byte_no, rbx); // get f1 Method*
3798 // do the call
3799 __ profile_call(rax);
3800 __ profile_arguments_type(rax, rbx, rbcp, false);
3801 __ jump_from_interpreted(rbx, rax);
3802 }
3803
3804
3805 void TemplateTable::fast_invokevfinal(int byte_no) {
3806 transition(vtos, vtos);
3807 assert(byte_no == f2_byte, "use this argument");
3808 __ stop("fast_invokevfinal not used on x86");
3809 }
3810
3811
3812 void TemplateTable::invokeinterface(int byte_no) {
3813 transition(vtos, vtos);
3814 assert(byte_no == f1_byte, "use this argument");
3815 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
3816 rcx, rdx); // recv, flags
3817
3818 // rax: reference klass (from f1)
3819 // rbx: method (from f2)
3820 // rcx: receiver
3821 // rdx: flags
3822
3823 // Special case of invokeinterface called for virtual method of
3824 // java.lang.Object. See cpCacheOop.cpp for details.
3825 // This code isn't produced by javac, but could be produced by
3826 // another compliant java compiler.
3827 Label notMethod;
3828 __ movl(rlocals, rdx);
3829 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3830
3831 __ jcc(Assembler::zero, notMethod);
3832
3833 invokevirtual_helper(rbx, rcx, rdx);
3834 __ bind(notMethod);
3835
3836 // Get receiver klass into rdx - also a null check
3837 __ restore_locals(); // restore r14
3838 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3839 __ load_klass(rdx, rcx);
3840
3841 Label no_such_interface, no_such_method;
3842
3843 // Preserve method for throw_AbstractMethodErrorVerbose.
3844 __ mov(rcx, rbx);
3845 // Receiver subtype check against REFC.
3846 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3847 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3848 rdx, rax, noreg,
3849 // outputs: scan temp. reg, scan temp. reg
3850 rbcp, rlocals,
3851 no_such_interface,
3852 /*return_method=*/false);
3853
3854 // profile this call
3855 __ restore_bcp(); // rbcp was destroyed by receiver type check
3856 __ profile_virtual_call(rdx, rbcp, rlocals);
3857
3858 // Get declaring interface class from method, and itable index
3859 __ movptr(rax, Address(rbx, Method::const_offset()));
3860 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3861 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3862 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3863 __ subl(rbx, Method::itable_index_max);
3864 __ negl(rbx);
3865
3866 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3867 __ mov(rlocals, rdx);
3868 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3869 rlocals, rax, rbx,
3870 // outputs: method, scan temp. reg
3871 rbx, rbcp,
3872 no_such_interface);
3873
3874 // rbx: Method* to call
3875 // rcx: receiver
3876 // Check for abstract method error
3877 // Note: This should be done more efficiently via a throw_abstract_method_error
3878 // interpreter entry point and a conditional jump to it in case of a null
3879 // method.
3880 __ testptr(rbx, rbx);
3881 __ jcc(Assembler::zero, no_such_method);
3882
3883 __ profile_called_method(rbx, rbcp, rdx);
3884 __ profile_arguments_type(rdx, rbx, rbcp, true);
3885
3886 // do the call
3887 // rcx: receiver
3888 // rbx,: Method*
3889 __ jump_from_interpreted(rbx, rdx);
3890 __ should_not_reach_here();
3891
3892 // exception handling code follows...
3893 // note: must restore interpreter registers to canonical
3894 // state for exception handling to work correctly!
3895
3896 __ bind(no_such_method);
3897 // throw exception
3898 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3899 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3900 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3901 // Pass arguments for generating a verbose error message.
3902 #ifdef _LP64
3903 Register recvKlass = c_rarg1;
3904 Register method = c_rarg2;
3905 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3906 if (method != rcx) { __ movq(method, rcx); }
3907 #else
3908 Register recvKlass = rdx;
3909 Register method = rcx;
3910 #endif
3911 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3912 recvKlass, method);
3913 // The call_VM checks for exception, so we should never return here.
3914 __ should_not_reach_here();
3915
3916 __ bind(no_such_interface);
3917 // throw exception
3918 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3919 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3920 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3921 // Pass arguments for generating a verbose error message.
3922 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3923 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3924 recvKlass, rax);
3925 // the call_VM checks for exception, so we should never return here.
3926 __ should_not_reach_here();
3927 }
3928
3929 void TemplateTable::invokehandle(int byte_no) {
3930 transition(vtos, vtos);
3931 assert(byte_no == f1_byte, "use this argument");
3932 const Register rbx_method = rbx;
3933 const Register rax_mtype = rax;
3934 const Register rcx_recv = rcx;
3935 const Register rdx_flags = rdx;
3936
3937 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3938 __ verify_method_ptr(rbx_method);
3939 __ verify_oop(rcx_recv);
3940 __ null_check(rcx_recv);
3941
3942 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3943 // rbx: MH.invokeExact_MT method (from f2)
3944
3945 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3946
3947 // FIXME: profile the LambdaForm also
3948 __ profile_final_call(rax);
3949 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3950
3951 __ jump_from_interpreted(rbx_method, rdx);
3952 }
3953
3954 void TemplateTable::invokedynamic(int byte_no) {
3955 transition(vtos, vtos);
3956 assert(byte_no == f1_byte, "use this argument");
3957
3958 const Register rbx_method = rbx;
3959 const Register rax_callsite = rax;
3960
3961 prepare_invoke(byte_no, rbx_method, rax_callsite);
3962
3963 // rax: CallSite object (from cpool->resolved_references[f1])
3964 // rbx: MH.linkToCallSite method (from f2)
3965
3966 // Note: rax_callsite is already pushed by prepare_invoke
3967
3968 // %%% should make a type profile for any invokedynamic that takes a ref argument
3969 // profile this call
3970 __ profile_call(rbcp);
3971 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3972
3973 __ verify_oop(rax_callsite);
3974
3975 __ jump_from_interpreted(rbx_method, rdx);
3976 }
3977
3978 //-----------------------------------------------------------------------------
3979 // Allocation
3980
3981 void TemplateTable::_new() {
3982 transition(vtos, atos);
3983 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3984 Label slow_case;
3985 Label slow_case_no_pop;
3986 Label done;
3987 Label initialize_header;
3988 Label initialize_object; // including clearing the fields
3989
3990 __ get_cpool_and_tags(rcx, rax);
3991
3992 // Make sure the class we're about to instantiate has been resolved.
3993 // This is done before loading InstanceKlass to be consistent with the order
3994 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3995 const int tags_offset = Array<u1>::base_offset_in_bytes();
3996 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3997 __ jcc(Assembler::notEqual, slow_case_no_pop);
3998
3999 // get InstanceKlass
4000 __ load_resolved_klass_at_index(rcx, rdx, rcx);
4001 __ push(rcx); // save the contexts of klass for initializing the header
4002
4003 // make sure klass is initialized & doesn't have finalizer
4004 // make sure klass is fully initialized
4005 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4006 __ jcc(Assembler::notEqual, slow_case);
4007
4008 // get instance_size in InstanceKlass (scaled to a count of bytes)
4009 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4010 // test to see if it has a finalizer or is malformed in some way
4011 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4012 __ jcc(Assembler::notZero, slow_case);
4013
4014 // Allocate the instance:
4015 // If TLAB is enabled:
4016 // Try to allocate in the TLAB.
4017 // If fails, go to the slow path.
4018 // Else If inline contiguous allocations are enabled:
4019 // Try to allocate in eden.
4020 // If fails due to heap end, go to slow path.
4021 //
4022 // If TLAB is enabled OR inline contiguous is enabled:
4023 // Initialize the allocation.
4024 // Exit.
4025 //
4026 // Go to slow path.
4027
4028 const bool allow_shared_alloc =
4029 Universe::heap()->supports_inline_contig_alloc();
4030
4031 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4032 #ifndef _LP64
4033 if (UseTLAB || allow_shared_alloc) {
4034 __ get_thread(thread);
4035 }
4036 #endif // _LP64
4037
4038 if (UseTLAB) {
4039 uint oop_extra_words = Universe::heap()->oop_extra_words();
4040 if (oop_extra_words > 0) {
4041 __ addptr(rdx, oop_extra_words * HeapWordSize);
4042 }
4043
4044 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4045 __ lea(rbx, Address(rax, rdx, Address::times_1));
4046 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4047 __ jcc(Assembler::above, slow_case);
4048 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4049 Universe::heap()->compile_prepare_oop(_masm, rax);
4050 if (ZeroTLAB) {
4051 // the fields have been already cleared
4052 __ jmp(initialize_header);
4053 } else {
4054 // initialize both the header and fields
4055 __ jmp(initialize_object);
4056 }
4057 } else {
4058 // Allocation in the shared Eden, if allowed.
4059 //
4060 // rdx: instance size in bytes
4061 if (allow_shared_alloc) {
4062 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4063 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4064
4065 Label retry;
4066 __ bind(retry);
4067 __ movptr(rax, heap_top);
4068 __ lea(rbx, Address(rax, rdx, Address::times_1));
4069 __ cmpptr(rbx, heap_end);
4070 __ jcc(Assembler::above, slow_case);
4071
4072 // Compare rax, with the top addr, and if still equal, store the new
4073 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4074 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4075 //
4076 // rax,: object begin
4077 // rbx,: object end
4078 // rdx: instance size in bytes
4079 __ locked_cmpxchgptr(rbx, heap_top);
4080
4081 // if someone beat us on the allocation, try again, otherwise continue
4082 __ jcc(Assembler::notEqual, retry);
4083
4084 __ incr_allocated_bytes(thread, rdx, 0);
4085 }
4086 }
4087
4088 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4089 // there is an initialize need. Otherwise, skip and go to the slow path.
4090 if (UseTLAB || allow_shared_alloc) {
4091 // The object is initialized before the header. If the object size is
4092 // zero, go directly to the header initialization.
4093 __ bind(initialize_object);
4094 __ decrement(rdx, sizeof(oopDesc));
4095 __ jcc(Assembler::zero, initialize_header);
4096
4097 // Initialize topmost object field, divide rdx by 8, check if odd and
4098 // test if zero.
4099 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4100 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4101
4102 // rdx must have been multiple of 8
4103 #ifdef ASSERT
4104 // make sure rdx was multiple of 8
4105 Label L;
4106 // Ignore partial flag stall after shrl() since it is debug VM
4107 __ jccb(Assembler::carryClear, L);
4108 __ stop("object size is not multiple of 2 - adjust this code");
4109 __ bind(L);
4110 // rdx must be > 0, no extra check needed here
4111 #endif
4112
4113 // initialize remaining object fields: rdx was a multiple of 8
4114 { Label loop;
4115 __ bind(loop);
4116 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4117 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4118 __ decrement(rdx);
4119 __ jcc(Assembler::notZero, loop);
4120 }
4121
4122 // initialize object header only.
4123 __ bind(initialize_header);
4124 if (UseBiasedLocking) {
4125 __ pop(rcx); // get saved klass back in the register.
4126 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4127 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4128 } else {
4129 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4130 (intptr_t)markOopDesc::prototype()); // header
4131 __ pop(rcx); // get saved klass back in the register.
4132 }
4133 #ifdef _LP64
4134 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4135 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4136 #endif
4137 __ store_klass(rax, rcx); // klass
4138
4139 {
4140 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4141 // Trigger dtrace event for fastpath
4142 __ push(atos);
4143 __ call_VM_leaf(
4144 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4145 __ pop(atos);
4146 }
4147
4148 __ jmp(done);
4149 }
4150
4151 // slow case
4152 __ bind(slow_case);
4153 __ pop(rcx); // restore stack pointer to what it was when we came in.
4154 __ bind(slow_case_no_pop);
4155
4156 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4157 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4158
4159 __ get_constant_pool(rarg1);
4160 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4161 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4162 __ verify_oop(rax);
4163
4164 // continue
4165 __ bind(done);
4166 }
4167
4168 void TemplateTable::newarray() {
4169 transition(itos, atos);
4170 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4171 __ load_unsigned_byte(rarg1, at_bcp(1));
4172 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4173 rarg1, rax);
4174 }
4175
4176 void TemplateTable::anewarray() {
4177 transition(itos, atos);
4178
4179 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4180 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4181
4182 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4183 __ get_constant_pool(rarg1);
4184 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4185 rarg1, rarg2, rax);
4186 }
4187
4188 void TemplateTable::arraylength() {
4189 transition(atos, itos);
4190 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4191 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4192 }
4193
4194 void TemplateTable::checkcast() {
4195 transition(atos, atos);
4196 Label done, is_null, ok_is_subtype, quicked, resolved;
4197 __ testptr(rax, rax); // object is in rax
4198 __ jcc(Assembler::zero, is_null);
4199
4200 // Get cpool & tags index
4201 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4202 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4203 // See if bytecode has already been quicked
4204 __ cmpb(Address(rdx, rbx,
4205 Address::times_1,
4206 Array<u1>::base_offset_in_bytes()),
4207 JVM_CONSTANT_Class);
4208 __ jcc(Assembler::equal, quicked);
4209 __ push(atos); // save receiver for result, and for GC
4210 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4211
4212 // vm_result_2 has metadata result
4213 #ifndef _LP64
4214 // borrow rdi from locals
4215 __ get_thread(rdi);
4216 __ get_vm_result_2(rax, rdi);
4217 __ restore_locals();
4218 #else
4219 __ get_vm_result_2(rax, r15_thread);
4220 #endif
4221
4222 __ pop_ptr(rdx); // restore receiver
4223 __ jmpb(resolved);
4224
4225 // Get superklass in rax and subklass in rbx
4226 __ bind(quicked);
4227 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4228 __ load_resolved_klass_at_index(rcx, rbx, rax);
4229
4230 __ bind(resolved);
4231 __ load_klass(rbx, rdx);
4232
4233 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4234 // Superklass in rax. Subklass in rbx.
4235 __ gen_subtype_check(rbx, ok_is_subtype);
4236
4237 // Come here on failure
4238 __ push_ptr(rdx);
4239 // object is at TOS
4240 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4241
4242 // Come here on success
4243 __ bind(ok_is_subtype);
4244 __ mov(rax, rdx); // Restore object in rdx
4245
4246 // Collect counts on whether this check-cast sees NULLs a lot or not.
4247 if (ProfileInterpreter) {
4248 __ jmp(done);
4249 __ bind(is_null);
4250 __ profile_null_seen(rcx);
4251 } else {
4252 __ bind(is_null); // same as 'done'
4253 }
4254 __ bind(done);
4255 }
4256
4257 void TemplateTable::instanceof() {
4258 transition(atos, itos);
4259 Label done, is_null, ok_is_subtype, quicked, resolved;
4260 __ testptr(rax, rax);
4261 __ jcc(Assembler::zero, is_null);
4262
4263 // Get cpool & tags index
4264 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4265 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4266 // See if bytecode has already been quicked
4267 __ cmpb(Address(rdx, rbx,
4268 Address::times_1,
4269 Array<u1>::base_offset_in_bytes()),
4270 JVM_CONSTANT_Class);
4271 __ jcc(Assembler::equal, quicked);
4272
4273 __ push(atos); // save receiver for result, and for GC
4274 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4275 // vm_result_2 has metadata result
4276
4277 #ifndef _LP64
4278 // borrow rdi from locals
4279 __ get_thread(rdi);
4280 __ get_vm_result_2(rax, rdi);
4281 __ restore_locals();
4282 #else
4283 __ get_vm_result_2(rax, r15_thread);
4284 #endif
4285
4286 __ pop_ptr(rdx); // restore receiver
4287 __ verify_oop(rdx);
4288 __ load_klass(rdx, rdx);
4289 __ jmpb(resolved);
4290
4291 // Get superklass in rax and subklass in rdx
4292 __ bind(quicked);
4293 __ load_klass(rdx, rax);
4294 __ load_resolved_klass_at_index(rcx, rbx, rax);
4295
4296 __ bind(resolved);
4297
4298 // Generate subtype check. Blows rcx, rdi
4299 // Superklass in rax. Subklass in rdx.
4300 __ gen_subtype_check(rdx, ok_is_subtype);
4301
4302 // Come here on failure
4303 __ xorl(rax, rax);
4304 __ jmpb(done);
4305 // Come here on success
4306 __ bind(ok_is_subtype);
4307 __ movl(rax, 1);
4308
4309 // Collect counts on whether this test sees NULLs a lot or not.
4310 if (ProfileInterpreter) {
4311 __ jmp(done);
4312 __ bind(is_null);
4313 __ profile_null_seen(rcx);
4314 } else {
4315 __ bind(is_null); // same as 'done'
4316 }
4317 __ bind(done);
4318 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4319 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4320 }
4321
4322
4323 //----------------------------------------------------------------------------------------------------
4324 // Breakpoints
4325 void TemplateTable::_breakpoint() {
4326 // Note: We get here even if we are single stepping..
4327 // jbug insists on setting breakpoints at every bytecode
4328 // even if we are in single step mode.
4329
4330 transition(vtos, vtos);
4331
4332 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4333
4334 // get the unpatched byte code
4335 __ get_method(rarg);
4336 __ call_VM(noreg,
4337 CAST_FROM_FN_PTR(address,
4338 InterpreterRuntime::get_original_bytecode_at),
4339 rarg, rbcp);
4340 __ mov(rbx, rax); // why?
4341
4342 // post the breakpoint event
4343 __ get_method(rarg);
4344 __ call_VM(noreg,
4345 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4346 rarg, rbcp);
4347
4348 // complete the execution of original bytecode
4349 __ dispatch_only_normal(vtos);
4350 }
4351
4352 //-----------------------------------------------------------------------------
4353 // Exceptions
4354
4355 void TemplateTable::athrow() {
4356 transition(atos, vtos);
4357 __ null_check(rax);
4358 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4359 }
4360
4361 //-----------------------------------------------------------------------------
4362 // Synchronization
4363 //
4364 // Note: monitorenter & exit are symmetric routines; which is reflected
4365 // in the assembly code structure as well
4366 //
4367 // Stack layout:
4368 //
4369 // [expressions ] <--- rsp = expression stack top
4370 // ..
4371 // [expressions ]
4372 // [monitor entry] <--- monitor block top = expression stack bot
4373 // ..
4374 // [monitor entry]
4375 // [frame data ] <--- monitor block bot
4376 // ...
4377 // [saved rbp ] <--- rbp
4378 void TemplateTable::monitorenter() {
4379 transition(atos, vtos);
4380
4381 // check for NULL object
4382 __ null_check(rax);
4383
4384 __ resolve_for_write(OOP_NOT_NULL, rax);
4385
4386 const Address monitor_block_top(
4387 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4388 const Address monitor_block_bot(
4389 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4390 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4391
4392 Label allocated;
4393
4394 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4395 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4396 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4397
4398 // initialize entry pointer
4399 __ xorl(rmon, rmon); // points to free slot or NULL
4400
4401 // find a free slot in the monitor block (result in rmon)
4402 {
4403 Label entry, loop, exit;
4404 __ movptr(rtop, monitor_block_top); // points to current entry,
4405 // starting with top-most entry
4406 __ lea(rbot, monitor_block_bot); // points to word before bottom
4407 // of monitor block
4408 __ jmpb_if_possible(entry);
4409
4410 __ bind(loop);
4411 // check if current entry is used
4412 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4413 // if not used then remember entry in rmon
4414 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4415 // check if current entry is for same object
4416 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4417 // if same object then stop searching
4418 __ jccb(Assembler::equal, exit);
4419 // otherwise advance to next entry
4420 __ addptr(rtop, entry_size);
4421 __ bind(entry);
4422 // check if bottom reached
4423 __ cmpptr(rtop, rbot);
4424 // if not at bottom then check this entry
4425 __ jcc(Assembler::notEqual, loop);
4426 __ bind(exit);
4427 }
4428
4429 __ testptr(rmon, rmon); // check if a slot has been found
4430 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4431
4432 // allocate one if there's no free slot
4433 {
4434 Label entry, loop;
4435 // 1. compute new pointers // rsp: old expression stack top
4436 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4437 __ subptr(rsp, entry_size); // move expression stack top
4438 __ subptr(rmon, entry_size); // move expression stack bottom
4439 __ mov(rtop, rsp); // set start value for copy loop
4440 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4441 __ jmp(entry);
4442 // 2. move expression stack contents
4443 __ bind(loop);
4444 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4445 // word from old location
4446 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4447 __ addptr(rtop, wordSize); // advance to next word
4448 __ bind(entry);
4449 __ cmpptr(rtop, rmon); // check if bottom reached
4450 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4451 // copy next word
4452 }
4453
4454 // call run-time routine
4455 // rmon: points to monitor entry
4456 __ bind(allocated);
4457
4458 // Increment bcp to point to the next bytecode, so exception
4459 // handling for async. exceptions work correctly.
4460 // The object has already been poped from the stack, so the
4461 // expression stack looks correct.
4462 __ increment(rbcp);
4463
4464 // store object
4465 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4466 __ lock_object(rmon);
4467
4468 // check to make sure this monitor doesn't cause stack overflow after locking
4469 __ save_bcp(); // in case of exception
4470 __ generate_stack_overflow_check(0);
4471
4472 // The bcp has already been incremented. Just need to dispatch to
4473 // next instruction.
4474 __ dispatch_next(vtos);
4475 }
4476
4477 void TemplateTable::monitorexit() {
4478 transition(atos, vtos);
4479
4480 // check for NULL object
4481 __ null_check(rax);
4482
4483 __ resolve_for_write(OOP_NOT_NULL, rax);
4484
4485 const Address monitor_block_top(
4486 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4487 const Address monitor_block_bot(
4488 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4489 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4490
4491 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4492 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4493
4494 Label found;
4495
4496 // find matching slot
4497 {
4498 Label entry, loop;
4499 __ movptr(rtop, monitor_block_top); // points to current entry,
4500 // starting with top-most entry
4501 __ lea(rbot, monitor_block_bot); // points to word before bottom
4502 // of monitor block
4503 __ jmpb_if_possible(entry);
4504
4505 __ bind(loop);
4506 // check if current entry is for same object
4507 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4508 // if same object then stop searching
4509 __ jcc(Assembler::equal, found);
4510 // otherwise advance to next entry
4511 __ addptr(rtop, entry_size);
4512 __ bind(entry);
4513 // check if bottom reached
4514 __ cmpptr(rtop, rbot);
4515 // if not at bottom then check this entry
4516 __ jcc(Assembler::notEqual, loop);
4517 }
4518
4519 // error handling. Unlocking was not block-structured
4520 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4521 InterpreterRuntime::throw_illegal_monitor_state_exception));
4522 __ should_not_reach_here();
4523
4524 // call run-time routine
4525 __ bind(found);
4526 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4527 __ unlock_object(rtop);
4528 __ pop_ptr(rax); // discard object
4529 }
4530
4531 // Wide instructions
4532 void TemplateTable::wide() {
4533 transition(vtos, vtos);
4534 __ load_unsigned_byte(rbx, at_bcp(1));
4535 ExternalAddress wtable((address)Interpreter::_wentry_point);
4536 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4537 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4538 }
4539
4540 // Multi arrays
4541 void TemplateTable::multianewarray() {
4542 transition(vtos, atos);
4543
4544 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4545 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4546 // last dim is on top of stack; we want address of first one:
4547 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4548 // the latter wordSize to point to the beginning of the array.
4549 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4550 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4551 __ load_unsigned_byte(rbx, at_bcp(3));
4552 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4553 }