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