1 /*
2 * Copyright (c) 1997, 2025, 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 "asm/macroAssembler.hpp"
26 #include "compiler/disassembler.hpp"
27 #include "gc/shared/collectedHeap.hpp"
28 #include "gc/shared/gc_globals.hpp"
29 #include "gc/shared/tlab_globals.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "memory/universe.hpp"
35 #include "oops/methodCounters.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "oops/inlineKlass.hpp"
40 #include "oops/resolvedFieldEntry.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/methodHandles.hpp"
45 #include "runtime/frame.inline.hpp"
46 #include "runtime/safepointMechanism.hpp"
47 #include "runtime/sharedRuntime.hpp"
48 #include "runtime/stubRoutines.hpp"
49 #include "runtime/synchronizer.hpp"
50 #include "utilities/macros.hpp"
51
52 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
53
54 // Global Register Names
55 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
56 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
57
58 // Address Computation: local variables
59 static inline Address iaddress(int n) {
60 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
61 }
62
63 static inline Address laddress(int n) {
64 return iaddress(n + 1);
65 }
66
67 #ifndef _LP64
68 static inline Address haddress(int n) {
69 return iaddress(n + 0);
70 }
71 #endif
72
73 static inline Address faddress(int n) {
74 return iaddress(n);
75 }
76
77 static inline Address daddress(int n) {
78 return laddress(n);
79 }
80
81 static inline Address aaddress(int n) {
82 return iaddress(n);
83 }
84
85 static inline Address iaddress(Register r) {
86 return Address(rlocals, r, Address::times_ptr);
87 }
88
89 static inline Address laddress(Register r) {
90 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
91 }
92
93 #ifndef _LP64
94 static inline Address haddress(Register r) {
95 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
96 }
97 #endif
98
99 static inline Address faddress(Register r) {
100 return iaddress(r);
101 }
102
103 static inline Address daddress(Register r) {
104 return laddress(r);
105 }
106
107 static inline Address aaddress(Register r) {
108 return iaddress(r);
109 }
110
111
112 // expression stack
113 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
114 // data beyond the rsp which is potentially unsafe in an MT environment;
115 // an interrupt may overwrite that data.)
116 static inline Address at_rsp () {
117 return Address(rsp, 0);
118 }
119
120 // At top of Java expression stack which may be different than esp(). It
121 // isn't for category 1 objects.
122 static inline Address at_tos () {
123 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
124 }
125
126 static inline Address at_tos_p1() {
127 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
128 }
129
130 static inline Address at_tos_p2() {
131 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
132 }
133
134 // Condition conversion
135 static Assembler::Condition j_not(TemplateTable::Condition cc) {
136 switch (cc) {
137 case TemplateTable::equal : return Assembler::notEqual;
138 case TemplateTable::not_equal : return Assembler::equal;
139 case TemplateTable::less : return Assembler::greaterEqual;
140 case TemplateTable::less_equal : return Assembler::greater;
141 case TemplateTable::greater : return Assembler::lessEqual;
142 case TemplateTable::greater_equal: return Assembler::less;
143 }
144 ShouldNotReachHere();
145 return Assembler::zero;
146 }
147
148
149
150 // Miscellaneous helper routines
151 // Store an oop (or null) at the address described by obj.
152 // If val == noreg this means store a null
153
154
155 static void do_oop_store(InterpreterMacroAssembler* _masm,
156 Address dst,
157 Register val,
158 DecoratorSet decorators = 0) {
159 assert(val == noreg || val == rax, "parameter is just for looks");
160 __ store_heap_oop(dst, val,
161 NOT_LP64(rdx) LP64_ONLY(rscratch2),
162 NOT_LP64(rbx) LP64_ONLY(r9),
163 NOT_LP64(rsi) LP64_ONLY(r8), decorators);
164 }
165
166 static void do_oop_load(InterpreterMacroAssembler* _masm,
167 Address src,
168 Register dst,
169 DecoratorSet decorators = 0) {
170 __ load_heap_oop(dst, src, rdx, rbx, decorators);
171 }
172
173 Address TemplateTable::at_bcp(int offset) {
174 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
175 return Address(rbcp, offset);
176 }
177
178
179 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
180 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
181 int byte_no) {
182 if (!RewriteBytecodes) return;
183 Label L_patch_done;
184
185 switch (bc) {
186 case Bytecodes::_fast_vputfield:
187 case Bytecodes::_fast_aputfield:
188 case Bytecodes::_fast_bputfield:
189 case Bytecodes::_fast_zputfield:
190 case Bytecodes::_fast_cputfield:
191 case Bytecodes::_fast_dputfield:
192 case Bytecodes::_fast_fputfield:
193 case Bytecodes::_fast_iputfield:
194 case Bytecodes::_fast_lputfield:
195 case Bytecodes::_fast_sputfield:
196 {
197 // We skip bytecode quickening for putfield instructions when
198 // the put_code written to the constant pool cache is zero.
199 // This is required so that every execution of this instruction
200 // calls out to InterpreterRuntime::resolve_get_put to do
201 // additional, required work.
202 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
203 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
204 __ load_field_entry(temp_reg, bc_reg);
205 if (byte_no == f1_byte) {
206 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::get_code_offset())));
207 } else {
208 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::put_code_offset())));
209 }
210
211 __ movl(bc_reg, bc);
212 __ cmpl(temp_reg, (int) 0);
213 __ jcc(Assembler::zero, L_patch_done); // don't patch
214 }
215 break;
216 default:
217 assert(byte_no == -1, "sanity");
218 // the pair bytecodes have already done the load.
219 if (load_bc_into_bc_reg) {
220 __ movl(bc_reg, bc);
221 }
222 }
223
224 if (JvmtiExport::can_post_breakpoint()) {
225 Label L_fast_patch;
226 // if a breakpoint is present we can't rewrite the stream directly
227 __ movzbl(temp_reg, at_bcp(0));
228 __ cmpl(temp_reg, Bytecodes::_breakpoint);
229 __ jcc(Assembler::notEqual, L_fast_patch);
230 __ get_method(temp_reg);
231 // Let breakpoint table handling rewrite to quicker bytecode
232 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
233 #ifndef ASSERT
234 __ jmpb(L_patch_done);
235 #else
236 __ jmp(L_patch_done);
237 #endif
238 __ bind(L_fast_patch);
239 }
240
241 #ifdef ASSERT
242 Label L_okay;
243 __ load_unsigned_byte(temp_reg, at_bcp(0));
244 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
245 __ jcc(Assembler::equal, L_okay);
246 __ cmpl(temp_reg, bc_reg);
247 __ jcc(Assembler::equal, L_okay);
248 __ stop("patching the wrong bytecode");
249 __ bind(L_okay);
250 #endif
251
252 // patch bytecode
253 __ movb(at_bcp(0), bc_reg);
254 __ bind(L_patch_done);
255 }
256 // Individual instructions
257
258
259 void TemplateTable::nop() {
260 transition(vtos, vtos);
261 // nothing to do
262 }
263
264 void TemplateTable::shouldnotreachhere() {
265 transition(vtos, vtos);
266 __ stop("shouldnotreachhere bytecode");
267 }
268
269 void TemplateTable::aconst_null() {
270 transition(vtos, atos);
271 __ xorl(rax, rax);
272 }
273
274 void TemplateTable::iconst(int value) {
275 transition(vtos, itos);
276 if (value == 0) {
277 __ xorl(rax, rax);
278 } else {
279 __ movl(rax, value);
280 }
281 }
282
283 void TemplateTable::lconst(int value) {
284 transition(vtos, ltos);
285 if (value == 0) {
286 __ xorl(rax, rax);
287 } else {
288 __ movl(rax, value);
289 }
290 #ifndef _LP64
291 assert(value >= 0, "check this code");
292 __ xorptr(rdx, rdx);
293 #endif
294 }
295
296
297
298 void TemplateTable::fconst(int value) {
299 transition(vtos, ftos);
300 if (UseSSE >= 1) {
301 static float one = 1.0f, two = 2.0f;
302 switch (value) {
303 case 0:
304 __ xorps(xmm0, xmm0);
305 break;
306 case 1:
307 __ movflt(xmm0, ExternalAddress((address) &one), rscratch1);
308 break;
309 case 2:
310 __ movflt(xmm0, ExternalAddress((address) &two), rscratch1);
311 break;
312 default:
313 ShouldNotReachHere();
314 break;
315 }
316 } else {
317 #ifdef _LP64
318 ShouldNotReachHere();
319 #else
320 if (value == 0) { __ fldz();
321 } else if (value == 1) { __ fld1();
322 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
323 } else { ShouldNotReachHere();
324 }
325 #endif // _LP64
326 }
327 }
328
329 void TemplateTable::dconst(int value) {
330 transition(vtos, dtos);
331 if (UseSSE >= 2) {
332 static double one = 1.0;
333 switch (value) {
334 case 0:
335 __ xorpd(xmm0, xmm0);
336 break;
337 case 1:
338 __ movdbl(xmm0, ExternalAddress((address) &one), rscratch1);
339 break;
340 default:
341 ShouldNotReachHere();
342 break;
343 }
344 } else {
345 #ifdef _LP64
346 ShouldNotReachHere();
347 #else
348 if (value == 0) { __ fldz();
349 } else if (value == 1) { __ fld1();
350 } else { ShouldNotReachHere();
351 }
352 #endif
353 }
354 }
355
356 void TemplateTable::bipush() {
357 transition(vtos, itos);
358 __ load_signed_byte(rax, at_bcp(1));
359 }
360
361 void TemplateTable::sipush() {
362 transition(vtos, itos);
363 __ load_unsigned_short(rax, at_bcp(1));
364 __ bswapl(rax);
365 __ sarl(rax, 16);
366 }
367
368 void TemplateTable::ldc(LdcType type) {
369 transition(vtos, vtos);
370 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
371 Label call_ldc, notFloat, notClass, notInt, Done;
372
373 if (is_ldc_wide(type)) {
374 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
375 } else {
376 __ load_unsigned_byte(rbx, at_bcp(1));
377 }
378
379 __ get_cpool_and_tags(rcx, rax);
380 const int base_offset = ConstantPool::header_size() * wordSize;
381 const int tags_offset = Array<u1>::base_offset_in_bytes();
382
383 // get type
384 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
385
386 // unresolved class - get the resolved class
387 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
388 __ jccb(Assembler::equal, call_ldc);
389
390 // unresolved class in error state - call into runtime to throw the error
391 // from the first resolution attempt
392 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
393 __ jccb(Assembler::equal, call_ldc);
394
395 // resolved class - need to call vm to get java mirror of the class
396 __ cmpl(rdx, JVM_CONSTANT_Class);
397 __ jcc(Assembler::notEqual, notClass);
398
399 __ bind(call_ldc);
400
401 __ movl(rarg, is_ldc_wide(type) ? 1 : 0);
402 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
403
404 __ push(atos);
405 __ jmp(Done);
406
407 __ bind(notClass);
408 __ cmpl(rdx, JVM_CONSTANT_Float);
409 __ jccb(Assembler::notEqual, notFloat);
410
411 // ftos
412 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
413 __ push(ftos);
414 __ jmp(Done);
415
416 __ bind(notFloat);
417 __ cmpl(rdx, JVM_CONSTANT_Integer);
418 __ jccb(Assembler::notEqual, notInt);
419
420 // itos
421 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
422 __ push(itos);
423 __ jmp(Done);
424
425 // assume the tag is for condy; if not, the VM runtime will tell us
426 __ bind(notInt);
427 condy_helper(Done);
428
429 __ bind(Done);
430 }
431
432 // Fast path for caching oop constants.
433 void TemplateTable::fast_aldc(LdcType type) {
434 transition(vtos, atos);
435
436 Register result = rax;
437 Register tmp = rdx;
438 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
439 int index_size = is_ldc_wide(type) ? sizeof(u2) : sizeof(u1);
440
441 Label resolved;
442
443 // We are resolved if the resolved reference cache entry contains a
444 // non-null object (String, MethodType, etc.)
445 assert_different_registers(result, tmp);
446 __ get_cache_index_at_bcp(tmp, 1, index_size);
447 __ load_resolved_reference_at_index(result, tmp);
448 __ testptr(result, result);
449 __ jcc(Assembler::notZero, resolved);
450
451 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
452
453 // first time invocation - must resolve first
454 __ movl(rarg, (int)bytecode());
455 __ call_VM(result, entry, rarg);
456 __ bind(resolved);
457
458 { // Check for the null sentinel.
459 // If we just called the VM, it already did the mapping for us,
460 // but it's harmless to retry.
461 Label notNull;
462 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
463 __ movptr(tmp, null_sentinel);
464 __ resolve_oop_handle(tmp, rscratch2);
465 __ cmpoop(tmp, result);
466 __ jccb(Assembler::notEqual, notNull);
467 __ xorptr(result, result); // null object reference
468 __ bind(notNull);
469 }
470
471 if (VerifyOops) {
472 __ verify_oop(result);
473 }
474 }
475
476 void TemplateTable::ldc2_w() {
477 transition(vtos, vtos);
478 Label notDouble, notLong, Done;
479 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
480
481 __ get_cpool_and_tags(rcx, rax);
482 const int base_offset = ConstantPool::header_size() * wordSize;
483 const int tags_offset = Array<u1>::base_offset_in_bytes();
484
485 // get type
486 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
487 __ cmpl(rdx, JVM_CONSTANT_Double);
488 __ jccb(Assembler::notEqual, notDouble);
489
490 // dtos
491 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
492 __ push(dtos);
493
494 __ jmp(Done);
495 __ bind(notDouble);
496 __ cmpl(rdx, JVM_CONSTANT_Long);
497 __ jccb(Assembler::notEqual, notLong);
498
499 // ltos
500 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
501 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
502 __ push(ltos);
503 __ jmp(Done);
504
505 __ bind(notLong);
506 condy_helper(Done);
507
508 __ bind(Done);
509 }
510
511 void TemplateTable::condy_helper(Label& Done) {
512 const Register obj = rax;
513 const Register off = rbx;
514 const Register flags = rcx;
515 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
516 __ movl(rarg, (int)bytecode());
517 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
518 #ifndef _LP64
519 // borrow rdi from locals
520 __ get_thread(rdi);
521 __ get_vm_result_2(flags, rdi);
522 __ restore_locals();
523 #else
524 __ get_vm_result_2(flags, r15_thread);
525 #endif
526 // VMr = obj = base address to find primitive value to push
527 // VMr2 = flags = (tos, off) using format of CPCE::_flags
528 __ movl(off, flags);
529 __ andl(off, ConstantPoolCache::field_index_mask);
530 const Address field(obj, off, Address::times_1, 0*wordSize);
531
532 // What sort of thing are we loading?
533 __ shrl(flags, ConstantPoolCache::tos_state_shift);
534 __ andl(flags, ConstantPoolCache::tos_state_mask);
535
536 switch (bytecode()) {
537 case Bytecodes::_ldc:
538 case Bytecodes::_ldc_w:
539 {
540 // tos in (itos, ftos, stos, btos, ctos, ztos)
541 Label notInt, notFloat, notShort, notByte, notChar, notBool;
542 __ cmpl(flags, itos);
543 __ jccb(Assembler::notEqual, notInt);
544 // itos
545 __ movl(rax, field);
546 __ push(itos);
547 __ jmp(Done);
548
549 __ bind(notInt);
550 __ cmpl(flags, ftos);
551 __ jccb(Assembler::notEqual, notFloat);
552 // ftos
553 __ load_float(field);
554 __ push(ftos);
555 __ jmp(Done);
556
557 __ bind(notFloat);
558 __ cmpl(flags, stos);
559 __ jccb(Assembler::notEqual, notShort);
560 // stos
561 __ load_signed_short(rax, field);
562 __ push(stos);
563 __ jmp(Done);
564
565 __ bind(notShort);
566 __ cmpl(flags, btos);
567 __ jccb(Assembler::notEqual, notByte);
568 // btos
569 __ load_signed_byte(rax, field);
570 __ push(btos);
571 __ jmp(Done);
572
573 __ bind(notByte);
574 __ cmpl(flags, ctos);
575 __ jccb(Assembler::notEqual, notChar);
576 // ctos
577 __ load_unsigned_short(rax, field);
578 __ push(ctos);
579 __ jmp(Done);
580
581 __ bind(notChar);
582 __ cmpl(flags, ztos);
583 __ jccb(Assembler::notEqual, notBool);
584 // ztos
585 __ load_signed_byte(rax, field);
586 __ push(ztos);
587 __ jmp(Done);
588
589 __ bind(notBool);
590 break;
591 }
592
593 case Bytecodes::_ldc2_w:
594 {
595 Label notLong, notDouble;
596 __ cmpl(flags, ltos);
597 __ jccb(Assembler::notEqual, notLong);
598 // ltos
599 // Loading high word first because movptr clobbers rax
600 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
601 __ movptr(rax, field);
602 __ push(ltos);
603 __ jmp(Done);
604
605 __ bind(notLong);
606 __ cmpl(flags, dtos);
607 __ jccb(Assembler::notEqual, notDouble);
608 // dtos
609 __ load_double(field);
610 __ push(dtos);
611 __ jmp(Done);
612
613 __ bind(notDouble);
614 break;
615 }
616
617 default:
618 ShouldNotReachHere();
619 }
620
621 __ stop("bad ldc/condy");
622 }
623
624 void TemplateTable::locals_index(Register reg, int offset) {
625 __ load_unsigned_byte(reg, at_bcp(offset));
626 __ negptr(reg);
627 }
628
629 void TemplateTable::iload() {
630 iload_internal();
631 }
632
633 void TemplateTable::nofast_iload() {
634 iload_internal(may_not_rewrite);
635 }
636
637 void TemplateTable::iload_internal(RewriteControl rc) {
638 transition(vtos, itos);
639 if (RewriteFrequentPairs && rc == may_rewrite) {
640 Label rewrite, done;
641 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
642 LP64_ONLY(assert(rbx != bc, "register damaged"));
643
644 // get next byte
645 __ load_unsigned_byte(rbx,
646 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
647 // if _iload, wait to rewrite to iload2. We only want to rewrite the
648 // last two iloads in a pair. Comparing against fast_iload means that
649 // the next bytecode is neither an iload or a caload, and therefore
650 // an iload pair.
651 __ cmpl(rbx, Bytecodes::_iload);
652 __ jcc(Assembler::equal, done);
653
654 __ cmpl(rbx, Bytecodes::_fast_iload);
655 __ movl(bc, Bytecodes::_fast_iload2);
656
657 __ jccb(Assembler::equal, rewrite);
658
659 // if _caload, rewrite to fast_icaload
660 __ cmpl(rbx, Bytecodes::_caload);
661 __ movl(bc, Bytecodes::_fast_icaload);
662 __ jccb(Assembler::equal, rewrite);
663
664 // rewrite so iload doesn't check again.
665 __ movl(bc, Bytecodes::_fast_iload);
666
667 // rewrite
668 // bc: fast bytecode
669 __ bind(rewrite);
670 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
671 __ bind(done);
672 }
673
674 // Get the local value into tos
675 locals_index(rbx);
676 __ movl(rax, iaddress(rbx));
677 }
678
679 void TemplateTable::fast_iload2() {
680 transition(vtos, itos);
681 locals_index(rbx);
682 __ movl(rax, iaddress(rbx));
683 __ push(itos);
684 locals_index(rbx, 3);
685 __ movl(rax, iaddress(rbx));
686 }
687
688 void TemplateTable::fast_iload() {
689 transition(vtos, itos);
690 locals_index(rbx);
691 __ movl(rax, iaddress(rbx));
692 }
693
694 void TemplateTable::lload() {
695 transition(vtos, ltos);
696 locals_index(rbx);
697 __ movptr(rax, laddress(rbx));
698 NOT_LP64(__ movl(rdx, haddress(rbx)));
699 }
700
701 void TemplateTable::fload() {
702 transition(vtos, ftos);
703 locals_index(rbx);
704 __ load_float(faddress(rbx));
705 }
706
707 void TemplateTable::dload() {
708 transition(vtos, dtos);
709 locals_index(rbx);
710 __ load_double(daddress(rbx));
711 }
712
713 void TemplateTable::aload() {
714 transition(vtos, atos);
715 locals_index(rbx);
716 __ movptr(rax, aaddress(rbx));
717 }
718
719 void TemplateTable::locals_index_wide(Register reg) {
720 __ load_unsigned_short(reg, at_bcp(2));
721 __ bswapl(reg);
722 __ shrl(reg, 16);
723 __ negptr(reg);
724 }
725
726 void TemplateTable::wide_iload() {
727 transition(vtos, itos);
728 locals_index_wide(rbx);
729 __ movl(rax, iaddress(rbx));
730 }
731
732 void TemplateTable::wide_lload() {
733 transition(vtos, ltos);
734 locals_index_wide(rbx);
735 __ movptr(rax, laddress(rbx));
736 NOT_LP64(__ movl(rdx, haddress(rbx)));
737 }
738
739 void TemplateTable::wide_fload() {
740 transition(vtos, ftos);
741 locals_index_wide(rbx);
742 __ load_float(faddress(rbx));
743 }
744
745 void TemplateTable::wide_dload() {
746 transition(vtos, dtos);
747 locals_index_wide(rbx);
748 __ load_double(daddress(rbx));
749 }
750
751 void TemplateTable::wide_aload() {
752 transition(vtos, atos);
753 locals_index_wide(rbx);
754 __ movptr(rax, aaddress(rbx));
755 }
756
757 void TemplateTable::index_check(Register array, Register index) {
758 // Pop ptr into array
759 __ pop_ptr(array);
760 index_check_without_pop(array, index);
761 }
762
763 void TemplateTable::index_check_without_pop(Register array, Register index) {
764 // destroys rbx
765 // sign extend index for use by indexed load
766 __ movl2ptr(index, index);
767 // check index
768 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
769 if (index != rbx) {
770 // ??? convention: move aberrant index into rbx for exception message
771 assert(rbx != array, "different registers");
772 __ movl(rbx, index);
773 }
774 Label skip;
775 __ jccb(Assembler::below, skip);
776 // Pass array to create more detailed exceptions.
777 __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
778 __ jump(RuntimeAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
779 __ bind(skip);
780 }
781
782 void TemplateTable::iaload() {
783 transition(itos, itos);
784 // rax: index
785 // rdx: array
786 index_check(rdx, rax); // kills rbx
787 __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
788 Address(rdx, rax, Address::times_4,
789 arrayOopDesc::base_offset_in_bytes(T_INT)),
790 noreg, noreg);
791 }
792
793 void TemplateTable::laload() {
794 transition(itos, ltos);
795 // rax: index
796 // rdx: array
797 index_check(rdx, rax); // kills rbx
798 NOT_LP64(__ mov(rbx, rax));
799 // rbx,: index
800 __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
801 Address(rdx, rbx, Address::times_8,
802 arrayOopDesc::base_offset_in_bytes(T_LONG)),
803 noreg, noreg);
804 }
805
806
807
808 void TemplateTable::faload() {
809 transition(itos, ftos);
810 // rax: index
811 // rdx: array
812 index_check(rdx, rax); // kills rbx
813 __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
814 Address(rdx, rax,
815 Address::times_4,
816 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
817 noreg, noreg);
818 }
819
820 void TemplateTable::daload() {
821 transition(itos, dtos);
822 // rax: index
823 // rdx: array
824 index_check(rdx, rax); // kills rbx
825 __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
826 Address(rdx, rax,
827 Address::times_8,
828 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
829 noreg, noreg);
830 }
831
832 void TemplateTable::aaload() {
833 transition(itos, atos);
834 Register array = rdx;
835 Register index = rax;
836
837 index_check(array, index); // kills rbx
838 __ profile_array_type<ArrayLoadData>(rbx, array, rcx);
839 if (UseArrayFlattening) {
840 Label is_flat_array, done;
841 __ test_flat_array_oop(array, rbx, is_flat_array);
842 do_oop_load(_masm,
843 Address(array, index,
844 UseCompressedOops ? Address::times_4 : Address::times_ptr,
845 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
846 rax,
847 IS_ARRAY);
848 __ jmp(done);
849 __ bind(is_flat_array);
850 __ movptr(rcx, array);
851 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::flat_array_load), rcx, index);
852 __ bind(done);
853 } else {
854 do_oop_load(_masm,
855 Address(array, index,
856 UseCompressedOops ? Address::times_4 : Address::times_ptr,
857 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
858 rax,
859 IS_ARRAY);
860 }
861 __ profile_element_type(rbx, rax, rcx);
862 }
863
864 void TemplateTable::baload() {
865 transition(itos, itos);
866 // rax: index
867 // rdx: array
868 index_check(rdx, rax); // kills rbx
869 __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
870 Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
871 noreg, noreg);
872 }
873
874 void TemplateTable::caload() {
875 transition(itos, itos);
876 // rax: index
877 // rdx: array
878 index_check(rdx, rax); // kills rbx
879 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
880 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
881 noreg, noreg);
882 }
883
884 // iload followed by caload frequent pair
885 void TemplateTable::fast_icaload() {
886 transition(vtos, itos);
887 // load index out of locals
888 locals_index(rbx);
889 __ movl(rax, iaddress(rbx));
890
891 // rax: index
892 // rdx: array
893 index_check(rdx, rax); // kills rbx
894 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
895 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
896 noreg, noreg);
897 }
898
899
900 void TemplateTable::saload() {
901 transition(itos, itos);
902 // rax: index
903 // rdx: array
904 index_check(rdx, rax); // kills rbx
905 __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
906 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
907 noreg, noreg);
908 }
909
910 void TemplateTable::iload(int n) {
911 transition(vtos, itos);
912 __ movl(rax, iaddress(n));
913 }
914
915 void TemplateTable::lload(int n) {
916 transition(vtos, ltos);
917 __ movptr(rax, laddress(n));
918 NOT_LP64(__ movptr(rdx, haddress(n)));
919 }
920
921 void TemplateTable::fload(int n) {
922 transition(vtos, ftos);
923 __ load_float(faddress(n));
924 }
925
926 void TemplateTable::dload(int n) {
927 transition(vtos, dtos);
928 __ load_double(daddress(n));
929 }
930
931 void TemplateTable::aload(int n) {
932 transition(vtos, atos);
933 __ movptr(rax, aaddress(n));
934 }
935
936 void TemplateTable::aload_0() {
937 aload_0_internal();
938 }
939
940 void TemplateTable::nofast_aload_0() {
941 aload_0_internal(may_not_rewrite);
942 }
943
944 void TemplateTable::aload_0_internal(RewriteControl rc) {
945 transition(vtos, atos);
946 // According to bytecode histograms, the pairs:
947 //
948 // _aload_0, _fast_igetfield
949 // _aload_0, _fast_agetfield
950 // _aload_0, _fast_fgetfield
951 //
952 // occur frequently. If RewriteFrequentPairs is set, the (slow)
953 // _aload_0 bytecode checks if the next bytecode is either
954 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
955 // rewrites the current bytecode into a pair bytecode; otherwise it
956 // rewrites the current bytecode into _fast_aload_0 that doesn't do
957 // the pair check anymore.
958 //
959 // Note: If the next bytecode is _getfield, the rewrite must be
960 // delayed, otherwise we may miss an opportunity for a pair.
961 //
962 // Also rewrite frequent pairs
963 // aload_0, aload_1
964 // aload_0, iload_1
965 // These bytecodes with a small amount of code are most profitable
966 // to rewrite
967 if (RewriteFrequentPairs && rc == may_rewrite) {
968 Label rewrite, done;
969
970 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
971 LP64_ONLY(assert(rbx != bc, "register damaged"));
972
973 // get next byte
974 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
975
976 // if _getfield then wait with rewrite
977 __ cmpl(rbx, Bytecodes::_getfield);
978 __ jcc(Assembler::equal, done);
979
980 // if _igetfield then rewrite to _fast_iaccess_0
981 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
982 __ cmpl(rbx, Bytecodes::_fast_igetfield);
983 __ movl(bc, Bytecodes::_fast_iaccess_0);
984 __ jccb(Assembler::equal, rewrite);
985
986 // if _agetfield then rewrite to _fast_aaccess_0
987 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
988 __ cmpl(rbx, Bytecodes::_fast_agetfield);
989 __ movl(bc, Bytecodes::_fast_aaccess_0);
990 __ jccb(Assembler::equal, rewrite);
991
992 // if _fgetfield then rewrite to _fast_faccess_0
993 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
994 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
995 __ movl(bc, Bytecodes::_fast_faccess_0);
996 __ jccb(Assembler::equal, rewrite);
997
998 // else rewrite to _fast_aload0
999 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
1000 __ movl(bc, Bytecodes::_fast_aload_0);
1001
1002 // rewrite
1003 // bc: fast bytecode
1004 __ bind(rewrite);
1005 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
1006
1007 __ bind(done);
1008 }
1009
1010 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
1011 aload(0);
1012 }
1013
1014 void TemplateTable::istore() {
1015 transition(itos, vtos);
1016 locals_index(rbx);
1017 __ movl(iaddress(rbx), rax);
1018 }
1019
1020
1021 void TemplateTable::lstore() {
1022 transition(ltos, vtos);
1023 locals_index(rbx);
1024 __ movptr(laddress(rbx), rax);
1025 NOT_LP64(__ movptr(haddress(rbx), rdx));
1026 }
1027
1028 void TemplateTable::fstore() {
1029 transition(ftos, vtos);
1030 locals_index(rbx);
1031 __ store_float(faddress(rbx));
1032 }
1033
1034 void TemplateTable::dstore() {
1035 transition(dtos, vtos);
1036 locals_index(rbx);
1037 __ store_double(daddress(rbx));
1038 }
1039
1040 void TemplateTable::astore() {
1041 transition(vtos, vtos);
1042 __ pop_ptr(rax);
1043 locals_index(rbx);
1044 __ movptr(aaddress(rbx), rax);
1045 }
1046
1047 void TemplateTable::wide_istore() {
1048 transition(vtos, vtos);
1049 __ pop_i();
1050 locals_index_wide(rbx);
1051 __ movl(iaddress(rbx), rax);
1052 }
1053
1054 void TemplateTable::wide_lstore() {
1055 transition(vtos, vtos);
1056 NOT_LP64(__ pop_l(rax, rdx));
1057 LP64_ONLY(__ pop_l());
1058 locals_index_wide(rbx);
1059 __ movptr(laddress(rbx), rax);
1060 NOT_LP64(__ movl(haddress(rbx), rdx));
1061 }
1062
1063 void TemplateTable::wide_fstore() {
1064 #ifdef _LP64
1065 transition(vtos, vtos);
1066 __ pop_f(xmm0);
1067 locals_index_wide(rbx);
1068 __ movflt(faddress(rbx), xmm0);
1069 #else
1070 wide_istore();
1071 #endif
1072 }
1073
1074 void TemplateTable::wide_dstore() {
1075 #ifdef _LP64
1076 transition(vtos, vtos);
1077 __ pop_d(xmm0);
1078 locals_index_wide(rbx);
1079 __ movdbl(daddress(rbx), xmm0);
1080 #else
1081 wide_lstore();
1082 #endif
1083 }
1084
1085 void TemplateTable::wide_astore() {
1086 transition(vtos, vtos);
1087 __ pop_ptr(rax);
1088 locals_index_wide(rbx);
1089 __ movptr(aaddress(rbx), rax);
1090 }
1091
1092 void TemplateTable::iastore() {
1093 transition(itos, vtos);
1094 __ pop_i(rbx);
1095 // rax: value
1096 // rbx: index
1097 // rdx: array
1098 index_check(rdx, rbx); // prefer index in rbx
1099 __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1100 Address(rdx, rbx, Address::times_4,
1101 arrayOopDesc::base_offset_in_bytes(T_INT)),
1102 rax, noreg, noreg, noreg);
1103 }
1104
1105 void TemplateTable::lastore() {
1106 transition(ltos, vtos);
1107 __ pop_i(rbx);
1108 // rax,: low(value)
1109 // rcx: array
1110 // rdx: high(value)
1111 index_check(rcx, rbx); // prefer index in rbx,
1112 // rbx,: index
1113 __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1114 Address(rcx, rbx, Address::times_8,
1115 arrayOopDesc::base_offset_in_bytes(T_LONG)),
1116 noreg /* ltos */, noreg, noreg, noreg);
1117 }
1118
1119
1120 void TemplateTable::fastore() {
1121 transition(ftos, vtos);
1122 __ pop_i(rbx);
1123 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1124 // rbx: index
1125 // rdx: array
1126 index_check(rdx, rbx); // prefer index in rbx
1127 __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1128 Address(rdx, rbx, Address::times_4,
1129 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1130 noreg /* ftos */, noreg, noreg, noreg);
1131 }
1132
1133 void TemplateTable::dastore() {
1134 transition(dtos, vtos);
1135 __ pop_i(rbx);
1136 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1137 // rbx: index
1138 // rdx: array
1139 index_check(rdx, rbx); // prefer index in rbx
1140 __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1141 Address(rdx, rbx, Address::times_8,
1142 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1143 noreg /* dtos */, noreg, noreg, noreg);
1144 }
1145
1146 void TemplateTable::aastore() {
1147 Label is_null, is_flat_array, ok_is_subtype, done;
1148 transition(vtos, vtos);
1149 // stack: ..., array, index, value
1150 __ movptr(rax, at_tos()); // value
1151 __ movl(rcx, at_tos_p1()); // index
1152 __ movptr(rdx, at_tos_p2()); // array
1153
1154 Address element_address(rdx, rcx,
1155 UseCompressedOops? Address::times_4 : Address::times_ptr,
1156 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1157
1158 index_check_without_pop(rdx, rcx); // kills rbx
1159
1160 __ profile_array_type<ArrayStoreData>(rdi, rdx, rbx);
1161 __ profile_multiple_element_types(rdi, rax, rbx, rcx);
1162
1163 __ testptr(rax, rax);
1164 __ jcc(Assembler::zero, is_null);
1165
1166 // Move array class to rdi
1167 __ load_klass(rdi, rdx, rscratch1);
1168 if (UseArrayFlattening) {
1169 __ movl(rbx, Address(rdi, Klass::layout_helper_offset()));
1170 __ test_flat_array_layout(rbx, is_flat_array);
1171 }
1172
1173 // Move subklass into rbx
1174 __ load_klass(rbx, rax, rscratch1);
1175 // Move array element superklass into rax
1176 __ movptr(rax, Address(rdi,
1177 ObjArrayKlass::element_klass_offset()));
1178
1179 // Generate subtype check. Blows rcx, rdi
1180 // Superklass in rax. Subklass in rbx.
1181 // is "rbx <: rax" ? (value subclass <: array element superclass)
1182 __ gen_subtype_check(rbx, ok_is_subtype, false);
1183
1184 // Come here on failure
1185 // object is at TOS
1186 __ jump(RuntimeAddress(Interpreter::_throw_ArrayStoreException_entry));
1187
1188 // Come here on success
1189 __ bind(ok_is_subtype);
1190
1191 // Get the value we will store
1192 __ movptr(rax, at_tos());
1193 __ movl(rcx, at_tos_p1()); // index
1194 // Now store using the appropriate barrier
1195 do_oop_store(_masm, element_address, rax, IS_ARRAY);
1196 __ jmp(done);
1197
1198 // Have a null in rax, rdx=array, ecx=index. Store null at ary[idx]
1199 __ bind(is_null);
1200 if (EnableValhalla) {
1201 Label is_null_into_value_array_npe, store_null;
1202
1203 // Move array class to rdi
1204 __ load_klass(rdi, rdx, rscratch1);
1205 if (UseArrayFlattening) {
1206 __ movl(rbx, Address(rdi, Klass::layout_helper_offset()));
1207 __ test_flat_array_layout(rbx, is_flat_array);
1208 }
1209
1210 // No way to store null in null-free array
1211 __ test_null_free_array_oop(rdx, rbx, is_null_into_value_array_npe);
1212 __ jmp(store_null);
1213
1214 __ bind(is_null_into_value_array_npe);
1215 __ jump(RuntimeAddress(Interpreter::_throw_NullPointerException_entry));
1216
1217 __ bind(store_null);
1218 }
1219 // Store a null
1220 do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1221 __ jmp(done);
1222
1223 if (UseArrayFlattening) {
1224 Label is_type_ok;
1225 __ bind(is_flat_array); // Store non-null value to flat
1226
1227 __ movptr(rax, at_tos());
1228 __ movl(rcx, at_tos_p1()); // index
1229 __ movptr(rdx, at_tos_p2()); // array
1230
1231 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::flat_array_store), rax, rdx, rcx);
1232 }
1233 // Pop stack arguments
1234 __ bind(done);
1235 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1236 }
1237
1238 void TemplateTable::bastore() {
1239 transition(itos, vtos);
1240 __ pop_i(rbx);
1241 // rax: value
1242 // rbx: index
1243 // rdx: array
1244 index_check(rdx, rbx); // prefer index in rbx
1245 // Need to check whether array is boolean or byte
1246 // since both types share the bastore bytecode.
1247 __ load_klass(rcx, rdx, rscratch1);
1248 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1249 int diffbit = Klass::layout_helper_boolean_diffbit();
1250 __ testl(rcx, diffbit);
1251 Label L_skip;
1252 __ jccb(Assembler::zero, L_skip);
1253 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1254 __ bind(L_skip);
1255 __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1256 Address(rdx, rbx,Address::times_1,
1257 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1258 rax, noreg, noreg, noreg);
1259 }
1260
1261 void TemplateTable::castore() {
1262 transition(itos, vtos);
1263 __ pop_i(rbx);
1264 // rax: value
1265 // rbx: index
1266 // rdx: array
1267 index_check(rdx, rbx); // prefer index in rbx
1268 __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1269 Address(rdx, rbx, Address::times_2,
1270 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1271 rax, noreg, noreg, noreg);
1272 }
1273
1274
1275 void TemplateTable::sastore() {
1276 castore();
1277 }
1278
1279 void TemplateTable::istore(int n) {
1280 transition(itos, vtos);
1281 __ movl(iaddress(n), rax);
1282 }
1283
1284 void TemplateTable::lstore(int n) {
1285 transition(ltos, vtos);
1286 __ movptr(laddress(n), rax);
1287 NOT_LP64(__ movptr(haddress(n), rdx));
1288 }
1289
1290 void TemplateTable::fstore(int n) {
1291 transition(ftos, vtos);
1292 __ store_float(faddress(n));
1293 }
1294
1295 void TemplateTable::dstore(int n) {
1296 transition(dtos, vtos);
1297 __ store_double(daddress(n));
1298 }
1299
1300
1301 void TemplateTable::astore(int n) {
1302 transition(vtos, vtos);
1303 __ pop_ptr(rax);
1304 __ movptr(aaddress(n), rax);
1305 }
1306
1307 void TemplateTable::pop() {
1308 transition(vtos, vtos);
1309 __ addptr(rsp, Interpreter::stackElementSize);
1310 }
1311
1312 void TemplateTable::pop2() {
1313 transition(vtos, vtos);
1314 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1315 }
1316
1317
1318 void TemplateTable::dup() {
1319 transition(vtos, vtos);
1320 __ load_ptr(0, rax);
1321 __ push_ptr(rax);
1322 // stack: ..., a, a
1323 }
1324
1325 void TemplateTable::dup_x1() {
1326 transition(vtos, vtos);
1327 // stack: ..., a, b
1328 __ load_ptr( 0, rax); // load b
1329 __ load_ptr( 1, rcx); // load a
1330 __ store_ptr(1, rax); // store b
1331 __ store_ptr(0, rcx); // store a
1332 __ push_ptr(rax); // push b
1333 // stack: ..., b, a, b
1334 }
1335
1336 void TemplateTable::dup_x2() {
1337 transition(vtos, vtos);
1338 // stack: ..., a, b, c
1339 __ load_ptr( 0, rax); // load c
1340 __ load_ptr( 2, rcx); // load a
1341 __ store_ptr(2, rax); // store c in a
1342 __ push_ptr(rax); // push c
1343 // stack: ..., c, b, c, c
1344 __ load_ptr( 2, rax); // load b
1345 __ store_ptr(2, rcx); // store a in b
1346 // stack: ..., c, a, c, c
1347 __ store_ptr(1, rax); // store b in c
1348 // stack: ..., c, a, b, c
1349 }
1350
1351 void TemplateTable::dup2() {
1352 transition(vtos, vtos);
1353 // stack: ..., a, b
1354 __ load_ptr(1, rax); // load a
1355 __ push_ptr(rax); // push a
1356 __ load_ptr(1, rax); // load b
1357 __ push_ptr(rax); // push b
1358 // stack: ..., a, b, a, b
1359 }
1360
1361
1362 void TemplateTable::dup2_x1() {
1363 transition(vtos, vtos);
1364 // stack: ..., a, b, c
1365 __ load_ptr( 0, rcx); // load c
1366 __ load_ptr( 1, rax); // load b
1367 __ push_ptr(rax); // push b
1368 __ push_ptr(rcx); // push c
1369 // stack: ..., a, b, c, b, c
1370 __ store_ptr(3, rcx); // store c in b
1371 // stack: ..., a, c, c, b, c
1372 __ load_ptr( 4, rcx); // load a
1373 __ store_ptr(2, rcx); // store a in 2nd c
1374 // stack: ..., a, c, a, b, c
1375 __ store_ptr(4, rax); // store b in a
1376 // stack: ..., b, c, a, b, c
1377 }
1378
1379 void TemplateTable::dup2_x2() {
1380 transition(vtos, vtos);
1381 // stack: ..., a, b, c, d
1382 __ load_ptr( 0, rcx); // load d
1383 __ load_ptr( 1, rax); // load c
1384 __ push_ptr(rax); // push c
1385 __ push_ptr(rcx); // push d
1386 // stack: ..., a, b, c, d, c, d
1387 __ load_ptr( 4, rax); // load b
1388 __ store_ptr(2, rax); // store b in d
1389 __ store_ptr(4, rcx); // store d in b
1390 // stack: ..., a, d, c, b, c, d
1391 __ load_ptr( 5, rcx); // load a
1392 __ load_ptr( 3, rax); // load c
1393 __ store_ptr(3, rcx); // store a in c
1394 __ store_ptr(5, rax); // store c in a
1395 // stack: ..., c, d, a, b, c, d
1396 }
1397
1398 void TemplateTable::swap() {
1399 transition(vtos, vtos);
1400 // stack: ..., a, b
1401 __ load_ptr( 1, rcx); // load a
1402 __ load_ptr( 0, rax); // load b
1403 __ store_ptr(0, rcx); // store a in b
1404 __ store_ptr(1, rax); // store b in a
1405 // stack: ..., b, a
1406 }
1407
1408 void TemplateTable::iop2(Operation op) {
1409 transition(itos, itos);
1410 switch (op) {
1411 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1412 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1413 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1414 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1415 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1416 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1417 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1418 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1419 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1420 default : ShouldNotReachHere();
1421 }
1422 }
1423
1424 void TemplateTable::lop2(Operation op) {
1425 transition(ltos, ltos);
1426 #ifdef _LP64
1427 switch (op) {
1428 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1429 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1430 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1431 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1432 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1433 default : ShouldNotReachHere();
1434 }
1435 #else
1436 __ pop_l(rbx, rcx);
1437 switch (op) {
1438 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1439 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1440 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1441 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1442 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1443 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1444 default : ShouldNotReachHere();
1445 }
1446 #endif
1447 }
1448
1449 void TemplateTable::idiv() {
1450 transition(itos, itos);
1451 __ movl(rcx, rax);
1452 __ pop_i(rax);
1453 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1454 // they are not equal, one could do a normal division (no correction
1455 // needed), which may speed up this implementation for the common case.
1456 // (see also JVM spec., p.243 & p.271)
1457 __ corrected_idivl(rcx);
1458 }
1459
1460 void TemplateTable::irem() {
1461 transition(itos, itos);
1462 __ movl(rcx, rax);
1463 __ pop_i(rax);
1464 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1465 // they are not equal, one could do a normal division (no correction
1466 // needed), which may speed up this implementation for the common case.
1467 // (see also JVM spec., p.243 & p.271)
1468 __ corrected_idivl(rcx);
1469 __ movl(rax, rdx);
1470 }
1471
1472 void TemplateTable::lmul() {
1473 transition(ltos, ltos);
1474 #ifdef _LP64
1475 __ pop_l(rdx);
1476 __ imulq(rax, rdx);
1477 #else
1478 __ pop_l(rbx, rcx);
1479 __ push(rcx); __ push(rbx);
1480 __ push(rdx); __ push(rax);
1481 __ lmul(2 * wordSize, 0);
1482 __ addptr(rsp, 4 * wordSize); // take off temporaries
1483 #endif
1484 }
1485
1486 void TemplateTable::ldiv() {
1487 transition(ltos, ltos);
1488 #ifdef _LP64
1489 __ mov(rcx, rax);
1490 __ pop_l(rax);
1491 // generate explicit div0 check
1492 __ testq(rcx, rcx);
1493 __ jump_cc(Assembler::zero,
1494 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1495 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1496 // they are not equal, one could do a normal division (no correction
1497 // needed), which may speed up this implementation for the common case.
1498 // (see also JVM spec., p.243 & p.271)
1499 __ corrected_idivq(rcx); // kills rbx
1500 #else
1501 __ pop_l(rbx, rcx);
1502 __ push(rcx); __ push(rbx);
1503 __ push(rdx); __ push(rax);
1504 // check if y = 0
1505 __ orl(rax, rdx);
1506 __ jump_cc(Assembler::zero,
1507 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1508 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1509 __ addptr(rsp, 4 * wordSize); // take off temporaries
1510 #endif
1511 }
1512
1513 void TemplateTable::lrem() {
1514 transition(ltos, ltos);
1515 #ifdef _LP64
1516 __ mov(rcx, rax);
1517 __ pop_l(rax);
1518 __ testq(rcx, rcx);
1519 __ jump_cc(Assembler::zero,
1520 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1521 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1522 // they are not equal, one could do a normal division (no correction
1523 // needed), which may speed up this implementation for the common case.
1524 // (see also JVM spec., p.243 & p.271)
1525 __ corrected_idivq(rcx); // kills rbx
1526 __ mov(rax, rdx);
1527 #else
1528 __ pop_l(rbx, rcx);
1529 __ push(rcx); __ push(rbx);
1530 __ push(rdx); __ push(rax);
1531 // check if y = 0
1532 __ orl(rax, rdx);
1533 __ jump_cc(Assembler::zero,
1534 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1535 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1536 __ addptr(rsp, 4 * wordSize);
1537 #endif
1538 }
1539
1540 void TemplateTable::lshl() {
1541 transition(itos, ltos);
1542 __ movl(rcx, rax); // get shift count
1543 #ifdef _LP64
1544 __ pop_l(rax); // get shift value
1545 __ shlq(rax);
1546 #else
1547 __ pop_l(rax, rdx); // get shift value
1548 __ lshl(rdx, rax);
1549 #endif
1550 }
1551
1552 void TemplateTable::lshr() {
1553 #ifdef _LP64
1554 transition(itos, ltos);
1555 __ movl(rcx, rax); // get shift count
1556 __ pop_l(rax); // get shift value
1557 __ sarq(rax);
1558 #else
1559 transition(itos, ltos);
1560 __ mov(rcx, rax); // get shift count
1561 __ pop_l(rax, rdx); // get shift value
1562 __ lshr(rdx, rax, true);
1563 #endif
1564 }
1565
1566 void TemplateTable::lushr() {
1567 transition(itos, ltos);
1568 #ifdef _LP64
1569 __ movl(rcx, rax); // get shift count
1570 __ pop_l(rax); // get shift value
1571 __ shrq(rax);
1572 #else
1573 __ mov(rcx, rax); // get shift count
1574 __ pop_l(rax, rdx); // get shift value
1575 __ lshr(rdx, rax);
1576 #endif
1577 }
1578
1579 void TemplateTable::fop2(Operation op) {
1580 transition(ftos, ftos);
1581
1582 if (UseSSE >= 1) {
1583 switch (op) {
1584 case add:
1585 __ addss(xmm0, at_rsp());
1586 __ addptr(rsp, Interpreter::stackElementSize);
1587 break;
1588 case sub:
1589 __ movflt(xmm1, xmm0);
1590 __ pop_f(xmm0);
1591 __ subss(xmm0, xmm1);
1592 break;
1593 case mul:
1594 __ mulss(xmm0, at_rsp());
1595 __ addptr(rsp, Interpreter::stackElementSize);
1596 break;
1597 case div:
1598 __ movflt(xmm1, xmm0);
1599 __ pop_f(xmm0);
1600 __ divss(xmm0, xmm1);
1601 break;
1602 case rem:
1603 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1604 // modulo operation. The frem method calls the function
1605 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1606 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1607 // (signalling or quiet) is returned.
1608 //
1609 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1610 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1611 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1612 // The fprem instruction used on x86_32 is functionally equivalent to
1613 // SharedRuntime::frem in that it returns a NaN.
1614 #ifdef _LP64
1615 __ movflt(xmm1, xmm0);
1616 __ pop_f(xmm0);
1617 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1618 #else // !_LP64
1619 __ push_f(xmm0);
1620 __ pop_f();
1621 __ fld_s(at_rsp());
1622 __ fremr(rax);
1623 __ f2ieee();
1624 __ pop(rax); // pop second operand off the stack
1625 __ push_f();
1626 __ pop_f(xmm0);
1627 #endif // _LP64
1628 break;
1629 default:
1630 ShouldNotReachHere();
1631 break;
1632 }
1633 } else {
1634 #ifdef _LP64
1635 ShouldNotReachHere();
1636 #else // !_LP64
1637 switch (op) {
1638 case add: __ fadd_s (at_rsp()); break;
1639 case sub: __ fsubr_s(at_rsp()); break;
1640 case mul: __ fmul_s (at_rsp()); break;
1641 case div: __ fdivr_s(at_rsp()); break;
1642 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1643 default : ShouldNotReachHere();
1644 }
1645 __ f2ieee();
1646 __ pop(rax); // pop second operand off the stack
1647 #endif // _LP64
1648 }
1649 }
1650
1651 void TemplateTable::dop2(Operation op) {
1652 transition(dtos, dtos);
1653 if (UseSSE >= 2) {
1654 switch (op) {
1655 case add:
1656 __ addsd(xmm0, at_rsp());
1657 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1658 break;
1659 case sub:
1660 __ movdbl(xmm1, xmm0);
1661 __ pop_d(xmm0);
1662 __ subsd(xmm0, xmm1);
1663 break;
1664 case mul:
1665 __ mulsd(xmm0, at_rsp());
1666 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1667 break;
1668 case div:
1669 __ movdbl(xmm1, xmm0);
1670 __ pop_d(xmm0);
1671 __ divsd(xmm0, xmm1);
1672 break;
1673 case rem:
1674 // Similar to fop2(), the modulo operation is performed using the
1675 // SharedRuntime::drem method (on x86_64 platforms) or using the
1676 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1677 #ifdef _LP64
1678 __ movdbl(xmm1, xmm0);
1679 __ pop_d(xmm0);
1680 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1681 #else // !_LP64
1682 __ push_d(xmm0);
1683 __ pop_d();
1684 __ fld_d(at_rsp());
1685 __ fremr(rax);
1686 __ d2ieee();
1687 __ pop(rax);
1688 __ pop(rdx);
1689 __ push_d();
1690 __ pop_d(xmm0);
1691 #endif // _LP64
1692 break;
1693 default:
1694 ShouldNotReachHere();
1695 break;
1696 }
1697 } else {
1698 #ifdef _LP64
1699 ShouldNotReachHere();
1700 #else // !_LP64
1701 switch (op) {
1702 case add: __ fadd_d (at_rsp()); break;
1703 case sub: __ fsubr_d(at_rsp()); break;
1704 case mul: {
1705 // strict semantics
1706 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1707 __ fmulp();
1708 __ fmul_d (at_rsp());
1709 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1710 __ fmulp();
1711 break;
1712 }
1713 case div: {
1714 // strict semantics
1715 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1716 __ fmul_d (at_rsp());
1717 __ fdivrp();
1718 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1719 __ fmulp();
1720 break;
1721 }
1722 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1723 default : ShouldNotReachHere();
1724 }
1725 __ d2ieee();
1726 // Pop double precision number from rsp.
1727 __ pop(rax);
1728 __ pop(rdx);
1729 #endif // _LP64
1730 }
1731 }
1732
1733 void TemplateTable::ineg() {
1734 transition(itos, itos);
1735 __ negl(rax);
1736 }
1737
1738 void TemplateTable::lneg() {
1739 transition(ltos, ltos);
1740 LP64_ONLY(__ negq(rax));
1741 NOT_LP64(__ lneg(rdx, rax));
1742 }
1743
1744 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1745 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1746 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1747 // of 128-bits operands for SSE instructions.
1748 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1749 // Store the value to a 128-bits operand.
1750 operand[0] = lo;
1751 operand[1] = hi;
1752 return operand;
1753 }
1754
1755 // Buffer for 128-bits masks used by SSE instructions.
1756 static jlong float_signflip_pool[2*2];
1757 static jlong double_signflip_pool[2*2];
1758
1759 void TemplateTable::fneg() {
1760 transition(ftos, ftos);
1761 if (UseSSE >= 1) {
1762 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1763 __ xorps(xmm0, ExternalAddress((address) float_signflip), rscratch1);
1764 } else {
1765 LP64_ONLY(ShouldNotReachHere());
1766 NOT_LP64(__ fchs());
1767 }
1768 }
1769
1770 void TemplateTable::dneg() {
1771 transition(dtos, dtos);
1772 if (UseSSE >= 2) {
1773 static jlong *double_signflip =
1774 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1775 __ xorpd(xmm0, ExternalAddress((address) double_signflip), rscratch1);
1776 } else {
1777 #ifdef _LP64
1778 ShouldNotReachHere();
1779 #else
1780 __ fchs();
1781 #endif
1782 }
1783 }
1784
1785 void TemplateTable::iinc() {
1786 transition(vtos, vtos);
1787 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1788 locals_index(rbx);
1789 __ addl(iaddress(rbx), rdx);
1790 }
1791
1792 void TemplateTable::wide_iinc() {
1793 transition(vtos, vtos);
1794 __ movl(rdx, at_bcp(4)); // get constant
1795 locals_index_wide(rbx);
1796 __ bswapl(rdx); // swap bytes & sign-extend constant
1797 __ sarl(rdx, 16);
1798 __ addl(iaddress(rbx), rdx);
1799 // Note: should probably use only one movl to get both
1800 // the index and the constant -> fix this
1801 }
1802
1803 void TemplateTable::convert() {
1804 #ifdef _LP64
1805 // Checking
1806 #ifdef ASSERT
1807 {
1808 TosState tos_in = ilgl;
1809 TosState tos_out = ilgl;
1810 switch (bytecode()) {
1811 case Bytecodes::_i2l: // fall through
1812 case Bytecodes::_i2f: // fall through
1813 case Bytecodes::_i2d: // fall through
1814 case Bytecodes::_i2b: // fall through
1815 case Bytecodes::_i2c: // fall through
1816 case Bytecodes::_i2s: tos_in = itos; break;
1817 case Bytecodes::_l2i: // fall through
1818 case Bytecodes::_l2f: // fall through
1819 case Bytecodes::_l2d: tos_in = ltos; break;
1820 case Bytecodes::_f2i: // fall through
1821 case Bytecodes::_f2l: // fall through
1822 case Bytecodes::_f2d: tos_in = ftos; break;
1823 case Bytecodes::_d2i: // fall through
1824 case Bytecodes::_d2l: // fall through
1825 case Bytecodes::_d2f: tos_in = dtos; break;
1826 default : ShouldNotReachHere();
1827 }
1828 switch (bytecode()) {
1829 case Bytecodes::_l2i: // fall through
1830 case Bytecodes::_f2i: // fall through
1831 case Bytecodes::_d2i: // fall through
1832 case Bytecodes::_i2b: // fall through
1833 case Bytecodes::_i2c: // fall through
1834 case Bytecodes::_i2s: tos_out = itos; break;
1835 case Bytecodes::_i2l: // fall through
1836 case Bytecodes::_f2l: // fall through
1837 case Bytecodes::_d2l: tos_out = ltos; break;
1838 case Bytecodes::_i2f: // fall through
1839 case Bytecodes::_l2f: // fall through
1840 case Bytecodes::_d2f: tos_out = ftos; break;
1841 case Bytecodes::_i2d: // fall through
1842 case Bytecodes::_l2d: // fall through
1843 case Bytecodes::_f2d: tos_out = dtos; break;
1844 default : ShouldNotReachHere();
1845 }
1846 transition(tos_in, tos_out);
1847 }
1848 #endif // ASSERT
1849
1850 static const int64_t is_nan = 0x8000000000000000L;
1851
1852 // Conversion
1853 switch (bytecode()) {
1854 case Bytecodes::_i2l:
1855 __ movslq(rax, rax);
1856 break;
1857 case Bytecodes::_i2f:
1858 __ cvtsi2ssl(xmm0, rax);
1859 break;
1860 case Bytecodes::_i2d:
1861 __ cvtsi2sdl(xmm0, rax);
1862 break;
1863 case Bytecodes::_i2b:
1864 __ movsbl(rax, rax);
1865 break;
1866 case Bytecodes::_i2c:
1867 __ movzwl(rax, rax);
1868 break;
1869 case Bytecodes::_i2s:
1870 __ movswl(rax, rax);
1871 break;
1872 case Bytecodes::_l2i:
1873 __ movl(rax, rax);
1874 break;
1875 case Bytecodes::_l2f:
1876 __ cvtsi2ssq(xmm0, rax);
1877 break;
1878 case Bytecodes::_l2d:
1879 __ cvtsi2sdq(xmm0, rax);
1880 break;
1881 case Bytecodes::_f2i:
1882 {
1883 Label L;
1884 __ cvttss2sil(rax, xmm0);
1885 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1886 __ jcc(Assembler::notEqual, L);
1887 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1888 __ bind(L);
1889 }
1890 break;
1891 case Bytecodes::_f2l:
1892 {
1893 Label L;
1894 __ cvttss2siq(rax, xmm0);
1895 // NaN or overflow/underflow?
1896 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1897 __ jcc(Assembler::notEqual, L);
1898 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1899 __ bind(L);
1900 }
1901 break;
1902 case Bytecodes::_f2d:
1903 __ cvtss2sd(xmm0, xmm0);
1904 break;
1905 case Bytecodes::_d2i:
1906 {
1907 Label L;
1908 __ cvttsd2sil(rax, xmm0);
1909 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1910 __ jcc(Assembler::notEqual, L);
1911 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1912 __ bind(L);
1913 }
1914 break;
1915 case Bytecodes::_d2l:
1916 {
1917 Label L;
1918 __ cvttsd2siq(rax, xmm0);
1919 // NaN or overflow/underflow?
1920 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1921 __ jcc(Assembler::notEqual, L);
1922 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1923 __ bind(L);
1924 }
1925 break;
1926 case Bytecodes::_d2f:
1927 __ cvtsd2ss(xmm0, xmm0);
1928 break;
1929 default:
1930 ShouldNotReachHere();
1931 }
1932 #else // !_LP64
1933 // Checking
1934 #ifdef ASSERT
1935 { TosState tos_in = ilgl;
1936 TosState tos_out = ilgl;
1937 switch (bytecode()) {
1938 case Bytecodes::_i2l: // fall through
1939 case Bytecodes::_i2f: // fall through
1940 case Bytecodes::_i2d: // fall through
1941 case Bytecodes::_i2b: // fall through
1942 case Bytecodes::_i2c: // fall through
1943 case Bytecodes::_i2s: tos_in = itos; break;
1944 case Bytecodes::_l2i: // fall through
1945 case Bytecodes::_l2f: // fall through
1946 case Bytecodes::_l2d: tos_in = ltos; break;
1947 case Bytecodes::_f2i: // fall through
1948 case Bytecodes::_f2l: // fall through
1949 case Bytecodes::_f2d: tos_in = ftos; break;
1950 case Bytecodes::_d2i: // fall through
1951 case Bytecodes::_d2l: // fall through
1952 case Bytecodes::_d2f: tos_in = dtos; break;
1953 default : ShouldNotReachHere();
1954 }
1955 switch (bytecode()) {
1956 case Bytecodes::_l2i: // fall through
1957 case Bytecodes::_f2i: // fall through
1958 case Bytecodes::_d2i: // fall through
1959 case Bytecodes::_i2b: // fall through
1960 case Bytecodes::_i2c: // fall through
1961 case Bytecodes::_i2s: tos_out = itos; break;
1962 case Bytecodes::_i2l: // fall through
1963 case Bytecodes::_f2l: // fall through
1964 case Bytecodes::_d2l: tos_out = ltos; break;
1965 case Bytecodes::_i2f: // fall through
1966 case Bytecodes::_l2f: // fall through
1967 case Bytecodes::_d2f: tos_out = ftos; break;
1968 case Bytecodes::_i2d: // fall through
1969 case Bytecodes::_l2d: // fall through
1970 case Bytecodes::_f2d: tos_out = dtos; break;
1971 default : ShouldNotReachHere();
1972 }
1973 transition(tos_in, tos_out);
1974 }
1975 #endif // ASSERT
1976
1977 // Conversion
1978 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1979 switch (bytecode()) {
1980 case Bytecodes::_i2l:
1981 __ extend_sign(rdx, rax);
1982 break;
1983 case Bytecodes::_i2f:
1984 if (UseSSE >= 1) {
1985 __ cvtsi2ssl(xmm0, rax);
1986 } else {
1987 __ push(rax); // store int on tos
1988 __ fild_s(at_rsp()); // load int to ST0
1989 __ f2ieee(); // truncate to float size
1990 __ pop(rcx); // adjust rsp
1991 }
1992 break;
1993 case Bytecodes::_i2d:
1994 if (UseSSE >= 2) {
1995 __ cvtsi2sdl(xmm0, rax);
1996 } else {
1997 __ push(rax); // add one slot for d2ieee()
1998 __ push(rax); // store int on tos
1999 __ fild_s(at_rsp()); // load int to ST0
2000 __ d2ieee(); // truncate to double size
2001 __ pop(rcx); // adjust rsp
2002 __ pop(rcx);
2003 }
2004 break;
2005 case Bytecodes::_i2b:
2006 __ shll(rax, 24); // truncate upper 24 bits
2007 __ sarl(rax, 24); // and sign-extend byte
2008 LP64_ONLY(__ movsbl(rax, rax));
2009 break;
2010 case Bytecodes::_i2c:
2011 __ andl(rax, 0xFFFF); // truncate upper 16 bits
2012 LP64_ONLY(__ movzwl(rax, rax));
2013 break;
2014 case Bytecodes::_i2s:
2015 __ shll(rax, 16); // truncate upper 16 bits
2016 __ sarl(rax, 16); // and sign-extend short
2017 LP64_ONLY(__ movswl(rax, rax));
2018 break;
2019 case Bytecodes::_l2i:
2020 /* nothing to do */
2021 break;
2022 case Bytecodes::_l2f:
2023 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
2024 // 64-bit long values to floats. On 32-bit platforms it is not possible
2025 // to use that instruction with 64-bit operands, therefore the FPU is
2026 // used to perform the conversion.
2027 __ push(rdx); // store long on tos
2028 __ push(rax);
2029 __ fild_d(at_rsp()); // load long to ST0
2030 __ f2ieee(); // truncate to float size
2031 __ pop(rcx); // adjust rsp
2032 __ pop(rcx);
2033 if (UseSSE >= 1) {
2034 __ push_f();
2035 __ pop_f(xmm0);
2036 }
2037 break;
2038 case Bytecodes::_l2d:
2039 // On 32-bit platforms the FPU is used for conversion because on
2040 // 32-bit platforms it is not not possible to use the cvtsi2sdq
2041 // instruction with 64-bit operands.
2042 __ push(rdx); // store long on tos
2043 __ push(rax);
2044 __ fild_d(at_rsp()); // load long to ST0
2045 __ d2ieee(); // truncate to double size
2046 __ pop(rcx); // adjust rsp
2047 __ pop(rcx);
2048 if (UseSSE >= 2) {
2049 __ push_d();
2050 __ pop_d(xmm0);
2051 }
2052 break;
2053 case Bytecodes::_f2i:
2054 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2055 // as it returns 0 for any NaN.
2056 if (UseSSE >= 1) {
2057 __ push_f(xmm0);
2058 } else {
2059 __ push(rcx); // reserve space for argument
2060 __ fstp_s(at_rsp()); // pass float argument on stack
2061 }
2062 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2063 break;
2064 case Bytecodes::_f2l:
2065 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2066 // as it returns 0 for any NaN.
2067 if (UseSSE >= 1) {
2068 __ push_f(xmm0);
2069 } else {
2070 __ push(rcx); // reserve space for argument
2071 __ fstp_s(at_rsp()); // pass float argument on stack
2072 }
2073 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2074 break;
2075 case Bytecodes::_f2d:
2076 if (UseSSE < 1) {
2077 /* nothing to do */
2078 } else if (UseSSE == 1) {
2079 __ push_f(xmm0);
2080 __ pop_f();
2081 } else { // UseSSE >= 2
2082 __ cvtss2sd(xmm0, xmm0);
2083 }
2084 break;
2085 case Bytecodes::_d2i:
2086 if (UseSSE >= 2) {
2087 __ push_d(xmm0);
2088 } else {
2089 __ push(rcx); // reserve space for argument
2090 __ push(rcx);
2091 __ fstp_d(at_rsp()); // pass double argument on stack
2092 }
2093 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2094 break;
2095 case Bytecodes::_d2l:
2096 if (UseSSE >= 2) {
2097 __ push_d(xmm0);
2098 } else {
2099 __ push(rcx); // reserve space for argument
2100 __ push(rcx);
2101 __ fstp_d(at_rsp()); // pass double argument on stack
2102 }
2103 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2104 break;
2105 case Bytecodes::_d2f:
2106 if (UseSSE <= 1) {
2107 __ push(rcx); // reserve space for f2ieee()
2108 __ f2ieee(); // truncate to float size
2109 __ pop(rcx); // adjust rsp
2110 if (UseSSE == 1) {
2111 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2112 // the conversion is performed using the FPU in this case.
2113 __ push_f();
2114 __ pop_f(xmm0);
2115 }
2116 } else { // UseSSE >= 2
2117 __ cvtsd2ss(xmm0, xmm0);
2118 }
2119 break;
2120 default :
2121 ShouldNotReachHere();
2122 }
2123 #endif // _LP64
2124 }
2125
2126 void TemplateTable::lcmp() {
2127 transition(ltos, itos);
2128 #ifdef _LP64
2129 Label done;
2130 __ pop_l(rdx);
2131 __ cmpq(rdx, rax);
2132 __ movl(rax, -1);
2133 __ jccb(Assembler::less, done);
2134 __ setb(Assembler::notEqual, rax);
2135 __ movzbl(rax, rax);
2136 __ bind(done);
2137 #else
2138
2139 // y = rdx:rax
2140 __ pop_l(rbx, rcx); // get x = rcx:rbx
2141 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2142 __ mov(rax, rcx);
2143 #endif
2144 }
2145
2146 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2147 if ((is_float && UseSSE >= 1) ||
2148 (!is_float && UseSSE >= 2)) {
2149 Label done;
2150 if (is_float) {
2151 // XXX get rid of pop here, use ... reg, mem32
2152 __ pop_f(xmm1);
2153 __ ucomiss(xmm1, xmm0);
2154 } else {
2155 // XXX get rid of pop here, use ... reg, mem64
2156 __ pop_d(xmm1);
2157 __ ucomisd(xmm1, xmm0);
2158 }
2159 if (unordered_result < 0) {
2160 __ movl(rax, -1);
2161 __ jccb(Assembler::parity, done);
2162 __ jccb(Assembler::below, done);
2163 __ setb(Assembler::notEqual, rdx);
2164 __ movzbl(rax, rdx);
2165 } else {
2166 __ movl(rax, 1);
2167 __ jccb(Assembler::parity, done);
2168 __ jccb(Assembler::above, done);
2169 __ movl(rax, 0);
2170 __ jccb(Assembler::equal, done);
2171 __ decrementl(rax);
2172 }
2173 __ bind(done);
2174 } else {
2175 #ifdef _LP64
2176 ShouldNotReachHere();
2177 #else // !_LP64
2178 if (is_float) {
2179 __ fld_s(at_rsp());
2180 } else {
2181 __ fld_d(at_rsp());
2182 __ pop(rdx);
2183 }
2184 __ pop(rcx);
2185 __ fcmp2int(rax, unordered_result < 0);
2186 #endif // _LP64
2187 }
2188 }
2189
2190 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2191 __ get_method(rcx); // rcx holds method
2192 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2193 // holds bumped taken count
2194
2195 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2196 InvocationCounter::counter_offset();
2197 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2198 InvocationCounter::counter_offset();
2199
2200 // Load up edx with the branch displacement
2201 if (is_wide) {
2202 __ movl(rdx, at_bcp(1));
2203 } else {
2204 __ load_signed_short(rdx, at_bcp(1));
2205 }
2206 __ bswapl(rdx);
2207
2208 if (!is_wide) {
2209 __ sarl(rdx, 16);
2210 }
2211 LP64_ONLY(__ movl2ptr(rdx, rdx));
2212
2213 // Handle all the JSR stuff here, then exit.
2214 // It's much shorter and cleaner than intermingling with the non-JSR
2215 // normal-branch stuff occurring below.
2216 if (is_jsr) {
2217 // Pre-load the next target bytecode into rbx
2218 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2219
2220 // compute return address as bci in rax
2221 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2222 in_bytes(ConstMethod::codes_offset())));
2223 __ subptr(rax, Address(rcx, Method::const_offset()));
2224 // Adjust the bcp in r13 by the displacement in rdx
2225 __ addptr(rbcp, rdx);
2226 // jsr returns atos that is not an oop
2227 __ push_i(rax);
2228 __ dispatch_only(vtos, true);
2229 return;
2230 }
2231
2232 // Normal (non-jsr) branch handling
2233
2234 // Adjust the bcp in r13 by the displacement in rdx
2235 __ addptr(rbcp, rdx);
2236
2237 assert(UseLoopCounter || !UseOnStackReplacement,
2238 "on-stack-replacement requires loop counters");
2239 Label backedge_counter_overflow;
2240 Label dispatch;
2241 if (UseLoopCounter) {
2242 // increment backedge counter for backward branches
2243 // rax: MDO
2244 // rbx: MDO bumped taken-count
2245 // rcx: method
2246 // rdx: target offset
2247 // r13: target bcp
2248 // r14: locals pointer
2249 __ testl(rdx, rdx); // check if forward or backward branch
2250 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2251
2252 // check if MethodCounters exists
2253 Label has_counters;
2254 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2255 __ testptr(rax, rax);
2256 __ jcc(Assembler::notZero, has_counters);
2257 __ push(rdx);
2258 __ push(rcx);
2259 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2260 rcx);
2261 __ pop(rcx);
2262 __ pop(rdx);
2263 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2264 __ testptr(rax, rax);
2265 __ jcc(Assembler::zero, dispatch);
2266 __ bind(has_counters);
2267
2268 Label no_mdo;
2269 if (ProfileInterpreter) {
2270 // Are we profiling?
2271 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2272 __ testptr(rbx, rbx);
2273 __ jccb(Assembler::zero, no_mdo);
2274 // Increment the MDO backedge counter
2275 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2276 in_bytes(InvocationCounter::counter_offset()));
2277 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2278 __ increment_mask_and_jump(mdo_backedge_counter, mask, rax,
2279 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2280 __ jmp(dispatch);
2281 }
2282 __ bind(no_mdo);
2283 // Increment backedge counter in MethodCounters*
2284 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2285 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2286 __ increment_mask_and_jump(Address(rcx, be_offset), mask, rax,
2287 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2288 __ bind(dispatch);
2289 }
2290
2291 // Pre-load the next target bytecode into rbx
2292 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2293
2294 // continue with the bytecode @ target
2295 // rax: return bci for jsr's, unused otherwise
2296 // rbx: target bytecode
2297 // r13: target bcp
2298 __ dispatch_only(vtos, true);
2299
2300 if (UseLoopCounter) {
2301 if (UseOnStackReplacement) {
2302 Label set_mdp;
2303 // invocation counter overflow
2304 __ bind(backedge_counter_overflow);
2305 __ negptr(rdx);
2306 __ addptr(rdx, rbcp); // branch bcp
2307 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2308 __ call_VM(noreg,
2309 CAST_FROM_FN_PTR(address,
2310 InterpreterRuntime::frequency_counter_overflow),
2311 rdx);
2312
2313 // rax: osr nmethod (osr ok) or null (osr not possible)
2314 // rdx: scratch
2315 // r14: locals pointer
2316 // r13: bcp
2317 __ testptr(rax, rax); // test result
2318 __ jcc(Assembler::zero, dispatch); // no osr if null
2319 // nmethod may have been invalidated (VM may block upon call_VM return)
2320 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2321 __ jcc(Assembler::notEqual, dispatch);
2322
2323 // We have the address of an on stack replacement routine in rax.
2324 // In preparation of invoking it, first we must migrate the locals
2325 // and monitors from off the interpreter frame on the stack.
2326 // Ensure to save the osr nmethod over the migration call,
2327 // it will be preserved in rbx.
2328 __ mov(rbx, rax);
2329
2330 NOT_LP64(__ get_thread(rcx));
2331
2332 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2333
2334 // rax is OSR buffer, move it to expected parameter location
2335 LP64_ONLY(__ mov(j_rarg0, rax));
2336 NOT_LP64(__ mov(rcx, rax));
2337 // We use j_rarg definitions here so that registers don't conflict as parameter
2338 // registers change across platforms as we are in the midst of a calling
2339 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2340
2341 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2342 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2343
2344 // pop the interpreter frame
2345 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2346 __ leave(); // remove frame anchor
2347 __ pop(retaddr); // get return address
2348 __ mov(rsp, sender_sp); // set sp to sender sp
2349 // Ensure compiled code always sees stack at proper alignment
2350 __ andptr(rsp, -(StackAlignmentInBytes));
2351
2352 // unlike x86 we need no specialized return from compiled code
2353 // to the interpreter or the call stub.
2354
2355 // push the return address
2356 __ push(retaddr);
2357
2358 // and begin the OSR nmethod
2359 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2360 }
2361 }
2362 }
2363
2364 void TemplateTable::if_0cmp(Condition cc) {
2365 transition(itos, vtos);
2366 // assume branch is more often taken than not (loops use backward branches)
2367 Label not_taken;
2368 __ testl(rax, rax);
2369 __ jcc(j_not(cc), not_taken);
2370 branch(false, false);
2371 __ bind(not_taken);
2372 __ profile_not_taken_branch(rax);
2373 }
2374
2375 void TemplateTable::if_icmp(Condition cc) {
2376 transition(itos, vtos);
2377 // assume branch is more often taken than not (loops use backward branches)
2378 Label not_taken;
2379 __ pop_i(rdx);
2380 __ cmpl(rdx, rax);
2381 __ jcc(j_not(cc), not_taken);
2382 branch(false, false);
2383 __ bind(not_taken);
2384 __ profile_not_taken_branch(rax);
2385 }
2386
2387 void TemplateTable::if_nullcmp(Condition cc) {
2388 transition(atos, vtos);
2389 // assume branch is more often taken than not (loops use backward branches)
2390 Label not_taken;
2391 __ testptr(rax, rax);
2392 __ jcc(j_not(cc), not_taken);
2393 branch(false, false);
2394 __ bind(not_taken);
2395 __ profile_not_taken_branch(rax);
2396 }
2397
2398 void TemplateTable::if_acmp(Condition cc) {
2399 transition(atos, vtos);
2400 // assume branch is more often taken than not (loops use backward branches)
2401 Label taken, not_taken;
2402 __ pop_ptr(rdx);
2403
2404 __ profile_acmp(rbx, rdx, rax, rcx);
2405
2406 const int is_inline_type_mask = markWord::inline_type_pattern;
2407 if (EnableValhalla) {
2408 __ cmpoop(rdx, rax);
2409 __ jcc(Assembler::equal, (cc == equal) ? taken : not_taken);
2410
2411 // might be substitutable, test if either rax or rdx is null
2412 __ testptr(rax, rax);
2413 __ jcc(Assembler::zero, (cc == equal) ? not_taken : taken);
2414 __ testptr(rdx, rdx);
2415 __ jcc(Assembler::zero, (cc == equal) ? not_taken : taken);
2416
2417 // and both are values ?
2418 __ movptr(rbx, Address(rdx, oopDesc::mark_offset_in_bytes()));
2419 __ andptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
2420 __ andptr(rbx, is_inline_type_mask);
2421 __ cmpptr(rbx, is_inline_type_mask);
2422 __ jcc(Assembler::notEqual, (cc == equal) ? not_taken : taken);
2423
2424 // same value klass ?
2425 __ load_metadata(rbx, rdx);
2426 __ load_metadata(rcx, rax);
2427 __ cmpptr(rbx, rcx);
2428 __ jcc(Assembler::notEqual, (cc == equal) ? not_taken : taken);
2429
2430 // Know both are the same type, let's test for substitutability...
2431 if (cc == equal) {
2432 invoke_is_substitutable(rax, rdx, taken, not_taken);
2433 } else {
2434 invoke_is_substitutable(rax, rdx, not_taken, taken);
2435 }
2436 __ stop("Not reachable");
2437 }
2438
2439 __ cmpoop(rdx, rax);
2440 __ jcc(j_not(cc), not_taken);
2441 __ bind(taken);
2442 branch(false, false);
2443 __ bind(not_taken);
2444 __ profile_not_taken_branch(rax, true);
2445 }
2446
2447 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2448 Label& is_subst, Label& not_subst) {
2449 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2450 // Restored...rax answer, jmp to outcome...
2451 __ testl(rax, rax);
2452 __ jcc(Assembler::zero, not_subst);
2453 __ jmp(is_subst);
2454 }
2455
2456 void TemplateTable::ret() {
2457 transition(vtos, vtos);
2458 locals_index(rbx);
2459 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2460 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2461 __ profile_ret(rbx, rcx);
2462 __ get_method(rax);
2463 __ movptr(rbcp, Address(rax, Method::const_offset()));
2464 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2465 ConstMethod::codes_offset()));
2466 __ dispatch_next(vtos, 0, true);
2467 }
2468
2469 void TemplateTable::wide_ret() {
2470 transition(vtos, vtos);
2471 locals_index_wide(rbx);
2472 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2473 __ profile_ret(rbx, rcx);
2474 __ get_method(rax);
2475 __ movptr(rbcp, Address(rax, Method::const_offset()));
2476 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2477 __ dispatch_next(vtos, 0, true);
2478 }
2479
2480 void TemplateTable::tableswitch() {
2481 Label default_case, continue_execution;
2482 transition(itos, vtos);
2483
2484 // align r13/rsi
2485 __ lea(rbx, at_bcp(BytesPerInt));
2486 __ andptr(rbx, -BytesPerInt);
2487 // load lo & hi
2488 __ movl(rcx, Address(rbx, BytesPerInt));
2489 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2490 __ bswapl(rcx);
2491 __ bswapl(rdx);
2492 // check against lo & hi
2493 __ cmpl(rax, rcx);
2494 __ jcc(Assembler::less, default_case);
2495 __ cmpl(rax, rdx);
2496 __ jcc(Assembler::greater, default_case);
2497 // lookup dispatch offset
2498 __ subl(rax, rcx);
2499 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2500 __ profile_switch_case(rax, rbx, rcx);
2501 // continue execution
2502 __ bind(continue_execution);
2503 __ bswapl(rdx);
2504 LP64_ONLY(__ movl2ptr(rdx, rdx));
2505 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2506 __ addptr(rbcp, rdx);
2507 __ dispatch_only(vtos, true);
2508 // handle default
2509 __ bind(default_case);
2510 __ profile_switch_default(rax);
2511 __ movl(rdx, Address(rbx, 0));
2512 __ jmp(continue_execution);
2513 }
2514
2515 void TemplateTable::lookupswitch() {
2516 transition(itos, itos);
2517 __ stop("lookupswitch bytecode should have been rewritten");
2518 }
2519
2520 void TemplateTable::fast_linearswitch() {
2521 transition(itos, vtos);
2522 Label loop_entry, loop, found, continue_execution;
2523 // bswap rax so we can avoid bswapping the table entries
2524 __ bswapl(rax);
2525 // align r13
2526 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2527 // this instruction (change offsets
2528 // below)
2529 __ andptr(rbx, -BytesPerInt);
2530 // set counter
2531 __ movl(rcx, Address(rbx, BytesPerInt));
2532 __ bswapl(rcx);
2533 __ jmpb(loop_entry);
2534 // table search
2535 __ bind(loop);
2536 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2537 __ jcc(Assembler::equal, found);
2538 __ bind(loop_entry);
2539 __ decrementl(rcx);
2540 __ jcc(Assembler::greaterEqual, loop);
2541 // default case
2542 __ profile_switch_default(rax);
2543 __ movl(rdx, Address(rbx, 0));
2544 __ jmp(continue_execution);
2545 // entry found -> get offset
2546 __ bind(found);
2547 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2548 __ profile_switch_case(rcx, rax, rbx);
2549 // continue execution
2550 __ bind(continue_execution);
2551 __ bswapl(rdx);
2552 __ movl2ptr(rdx, rdx);
2553 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2554 __ addptr(rbcp, rdx);
2555 __ dispatch_only(vtos, true);
2556 }
2557
2558 void TemplateTable::fast_binaryswitch() {
2559 transition(itos, vtos);
2560 // Implementation using the following core algorithm:
2561 //
2562 // int binary_search(int key, LookupswitchPair* array, int n) {
2563 // // Binary search according to "Methodik des Programmierens" by
2564 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2565 // int i = 0;
2566 // int j = n;
2567 // while (i+1 < j) {
2568 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2569 // // with Q: for all i: 0 <= i < n: key < a[i]
2570 // // where a stands for the array and assuming that the (inexisting)
2571 // // element a[n] is infinitely big.
2572 // int h = (i + j) >> 1;
2573 // // i < h < j
2574 // if (key < array[h].fast_match()) {
2575 // j = h;
2576 // } else {
2577 // i = h;
2578 // }
2579 // }
2580 // // R: a[i] <= key < a[i+1] or Q
2581 // // (i.e., if key is within array, i is the correct index)
2582 // return i;
2583 // }
2584
2585 // Register allocation
2586 const Register key = rax; // already set (tosca)
2587 const Register array = rbx;
2588 const Register i = rcx;
2589 const Register j = rdx;
2590 const Register h = rdi;
2591 const Register temp = rsi;
2592
2593 // Find array start
2594 NOT_LP64(__ save_bcp());
2595
2596 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2597 // get rid of this
2598 // instruction (change
2599 // offsets below)
2600 __ andptr(array, -BytesPerInt);
2601
2602 // Initialize i & j
2603 __ xorl(i, i); // i = 0;
2604 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2605
2606 // Convert j into native byteordering
2607 __ bswapl(j);
2608
2609 // And start
2610 Label entry;
2611 __ jmp(entry);
2612
2613 // binary search loop
2614 {
2615 Label loop;
2616 __ bind(loop);
2617 // int h = (i + j) >> 1;
2618 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2619 __ sarl(h, 1); // h = (i + j) >> 1;
2620 // if (key < array[h].fast_match()) {
2621 // j = h;
2622 // } else {
2623 // i = h;
2624 // }
2625 // Convert array[h].match to native byte-ordering before compare
2626 __ movl(temp, Address(array, h, Address::times_8));
2627 __ bswapl(temp);
2628 __ cmpl(key, temp);
2629 // j = h if (key < array[h].fast_match())
2630 __ cmov32(Assembler::less, j, h);
2631 // i = h if (key >= array[h].fast_match())
2632 __ cmov32(Assembler::greaterEqual, i, h);
2633 // while (i+1 < j)
2634 __ bind(entry);
2635 __ leal(h, Address(i, 1)); // i+1
2636 __ cmpl(h, j); // i+1 < j
2637 __ jcc(Assembler::less, loop);
2638 }
2639
2640 // end of binary search, result index is i (must check again!)
2641 Label default_case;
2642 // Convert array[i].match to native byte-ordering before compare
2643 __ movl(temp, Address(array, i, Address::times_8));
2644 __ bswapl(temp);
2645 __ cmpl(key, temp);
2646 __ jcc(Assembler::notEqual, default_case);
2647
2648 // entry found -> j = offset
2649 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2650 __ profile_switch_case(i, key, array);
2651 __ bswapl(j);
2652 LP64_ONLY(__ movslq(j, j));
2653
2654 NOT_LP64(__ restore_bcp());
2655 NOT_LP64(__ restore_locals()); // restore rdi
2656
2657 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2658 __ addptr(rbcp, j);
2659 __ dispatch_only(vtos, true);
2660
2661 // default case -> j = default offset
2662 __ bind(default_case);
2663 __ profile_switch_default(i);
2664 __ movl(j, Address(array, -2 * BytesPerInt));
2665 __ bswapl(j);
2666 LP64_ONLY(__ movslq(j, j));
2667
2668 NOT_LP64(__ restore_bcp());
2669 NOT_LP64(__ restore_locals());
2670
2671 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2672 __ addptr(rbcp, j);
2673 __ dispatch_only(vtos, true);
2674 }
2675
2676 void TemplateTable::_return(TosState state) {
2677 transition(state, state);
2678
2679 assert(_desc->calls_vm(),
2680 "inconsistent calls_vm information"); // call in remove_activation
2681
2682 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2683 assert(state == vtos, "only valid state");
2684 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2685 __ movptr(robj, aaddress(0));
2686 __ load_klass(rdi, robj, rscratch1);
2687 __ testb(Address(rdi, Klass::misc_flags_offset()), KlassFlags::_misc_has_finalizer);
2688 Label skip_register_finalizer;
2689 __ jcc(Assembler::zero, skip_register_finalizer);
2690
2691 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2692
2693 __ bind(skip_register_finalizer);
2694 }
2695
2696 if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2697 Label no_safepoint;
2698 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2699 #ifdef _LP64
2700 __ testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2701 #else
2702 const Register thread = rdi;
2703 __ get_thread(thread);
2704 __ testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2705 #endif
2706 __ jcc(Assembler::zero, no_safepoint);
2707 __ push(state);
2708 __ push_cont_fastpath();
2709 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2710 InterpreterRuntime::at_safepoint));
2711 __ pop_cont_fastpath();
2712 __ pop(state);
2713 __ bind(no_safepoint);
2714 }
2715
2716 // Narrow result if state is itos but result type is smaller.
2717 // Need to narrow in the return bytecode rather than in generate_return_entry
2718 // since compiled code callers expect the result to already be narrowed.
2719 if (state == itos) {
2720 __ narrow(rax);
2721 }
2722
2723 __ remove_activation(state, rbcp, true, true, true);
2724
2725 __ jmp(rbcp);
2726 }
2727
2728 // ----------------------------------------------------------------------------
2729 // Volatile variables demand their effects be made known to all CPU's
2730 // in order. Store buffers on most chips allow reads & writes to
2731 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2732 // without some kind of memory barrier (i.e., it's not sufficient that
2733 // the interpreter does not reorder volatile references, the hardware
2734 // also must not reorder them).
2735 //
2736 // According to the new Java Memory Model (JMM):
2737 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2738 // writes act as acquire & release, so:
2739 // (2) A read cannot let unrelated NON-volatile memory refs that
2740 // happen after the read float up to before the read. It's OK for
2741 // non-volatile memory refs that happen before the volatile read to
2742 // float down below it.
2743 // (3) Similar a volatile write cannot let unrelated NON-volatile
2744 // memory refs that happen BEFORE the write float down to after the
2745 // write. It's OK for non-volatile memory refs that happen after the
2746 // volatile write to float up before it.
2747 //
2748 // We only put in barriers around volatile refs (they are expensive),
2749 // not _between_ memory refs (that would require us to track the
2750 // flavor of the previous memory refs). Requirements (2) and (3)
2751 // require some barriers before volatile stores and after volatile
2752 // loads. These nearly cover requirement (1) but miss the
2753 // volatile-store-volatile-load case. This final case is placed after
2754 // volatile-stores although it could just as well go before
2755 // volatile-loads.
2756
2757 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2758 // Helper function to insert a is-volatile test and memory barrier
2759 __ membar(order_constraint);
2760 }
2761
2762 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2763 Register cache,
2764 Register index) {
2765 const Register temp = rbx;
2766 assert_different_registers(cache, index, temp);
2767
2768 Label L_clinit_barrier_slow;
2769 Label resolved;
2770
2771 Bytecodes::Code code = bytecode();
2772
2773 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2774
2775 __ load_method_entry(cache, index);
2776 switch(byte_no) {
2777 case f1_byte:
2778 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2779 break;
2780 case f2_byte:
2781 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2782 break;
2783 default:
2784 ShouldNotReachHere();
2785 }
2786 __ cmpl(temp, code); // have we resolved this bytecode?
2787 __ jcc(Assembler::equal, resolved);
2788
2789 // resolve first time through
2790 // Class initialization barrier slow path lands here as well.
2791 __ bind(L_clinit_barrier_slow);
2792 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2793 __ movl(temp, code);
2794 __ call_VM(noreg, entry, temp);
2795 // Update registers with resolved info
2796 __ load_method_entry(cache, index);
2797
2798 __ bind(resolved);
2799
2800 // Class initialization barrier for static methods
2801 if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2802 const Register method = temp;
2803 const Register klass = temp;
2804 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2805 assert(thread != noreg, "x86_32 not supported");
2806
2807 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2808 __ load_method_holder(klass, method);
2809 __ clinit_barrier(klass, thread, nullptr /*L_fast_path*/, &L_clinit_barrier_slow);
2810 }
2811 }
2812
2813 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2814 Register cache,
2815 Register index) {
2816 const Register temp = rbx;
2817 assert_different_registers(cache, index, temp);
2818
2819 Label resolved;
2820
2821 Bytecodes::Code code = bytecode();
2822 switch (code) {
2823 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2824 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2825 default: break;
2826 }
2827
2828 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2829 __ load_field_entry(cache, index);
2830 if (byte_no == f1_byte) {
2831 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2832 } else {
2833 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2834 }
2835 __ cmpl(temp, code); // have we resolved this bytecode?
2836 __ jcc(Assembler::equal, resolved);
2837
2838 // resolve first time through
2839 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2840 __ movl(temp, code);
2841 __ call_VM(noreg, entry, temp);
2842 // Update registers with resolved info
2843 __ load_field_entry(cache, index);
2844
2845 __ bind(resolved);
2846 }
2847
2848 void TemplateTable::load_resolved_field_entry(Register obj,
2849 Register cache,
2850 Register tos_state,
2851 Register offset,
2852 Register flags,
2853 bool is_static = false) {
2854 assert_different_registers(cache, tos_state, flags, offset);
2855
2856 // Field offset
2857 __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2858
2859 // Flags
2860 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2861
2862 // TOS state
2863 __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2864
2865 // Klass overwrite register
2866 if (is_static) {
2867 __ movptr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2868 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2869 __ movptr(obj, Address(obj, mirror_offset));
2870 __ resolve_oop_handle(obj, rscratch2);
2871 }
2872
2873 }
2874
2875 void TemplateTable::load_invokedynamic_entry(Register method) {
2876 // setup registers
2877 const Register appendix = rax;
2878 const Register cache = rcx;
2879 const Register index = rdx;
2880 assert_different_registers(method, appendix, cache, index);
2881
2882 __ save_bcp();
2883
2884 Label resolved;
2885
2886 __ load_resolved_indy_entry(cache, index);
2887 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2888
2889 // Compare the method to zero
2890 __ testptr(method, method);
2891 __ jcc(Assembler::notZero, resolved);
2892
2893 Bytecodes::Code code = bytecode();
2894
2895 // Call to the interpreter runtime to resolve invokedynamic
2896 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2897 __ movl(method, code); // this is essentially Bytecodes::_invokedynamic
2898 __ call_VM(noreg, entry, method);
2899 // Update registers with resolved info
2900 __ load_resolved_indy_entry(cache, index);
2901 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2902
2903 #ifdef ASSERT
2904 __ testptr(method, method);
2905 __ jcc(Assembler::notZero, resolved);
2906 __ stop("Should be resolved by now");
2907 #endif // ASSERT
2908 __ bind(resolved);
2909
2910 Label L_no_push;
2911 // Check if there is an appendix
2912 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2913 __ testl(index, (1 << ResolvedIndyEntry::has_appendix_shift));
2914 __ jcc(Assembler::zero, L_no_push);
2915
2916 // Get appendix
2917 __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2918 // Push the appendix as a trailing parameter
2919 // since the parameter_size includes it.
2920 __ load_resolved_reference_at_index(appendix, index);
2921 __ verify_oop(appendix);
2922 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2923 __ bind(L_no_push);
2924
2925 // compute return type
2926 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2927 // load return address
2928 {
2929 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2930 ExternalAddress table(table_addr);
2931 #ifdef _LP64
2932 __ lea(rscratch1, table);
2933 __ movptr(index, Address(rscratch1, index, Address::times_ptr));
2934 #else
2935 __ movptr(index, ArrayAddress(table, Address(noreg, index, Address::times_ptr)));
2936 #endif // _LP64
2937 }
2938
2939 // push return address
2940 __ push(index);
2941 }
2942
2943 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2944 Register method,
2945 Register flags) {
2946 // setup registers
2947 const Register index = rdx;
2948 assert_different_registers(cache, index);
2949 assert_different_registers(method, cache, flags);
2950
2951 // determine constant pool cache field offsets
2952 resolve_cache_and_index_for_method(f1_byte, cache, index);
2953 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2954 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2955 }
2956
2957 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2958 Register method,
2959 Register ref_index,
2960 Register flags) {
2961 // setup registers
2962 const Register index = rdx;
2963 assert_different_registers(cache, index);
2964 assert_different_registers(cache, method, ref_index, flags);
2965
2966 // determine constant pool cache field offsets
2967 resolve_cache_and_index_for_method(f1_byte, cache, index);
2968 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2969
2970 // Maybe push appendix
2971 Label L_no_push;
2972 __ testl(flags, (1 << ResolvedMethodEntry::has_appendix_shift));
2973 __ jcc(Assembler::zero, L_no_push);
2974 // invokehandle uses an index into the resolved references array
2975 __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2976 // Push the appendix as a trailing parameter.
2977 // This must be done before we get the receiver,
2978 // since the parameter_size includes it.
2979 Register appendix = method;
2980 __ load_resolved_reference_at_index(appendix, ref_index);
2981 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2982 __ bind(L_no_push);
2983
2984 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2985 }
2986
2987 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2988 Register klass,
2989 Register method_or_table_index,
2990 Register flags) {
2991 // setup registers
2992 const Register index = rdx;
2993 assert_different_registers(cache, klass, method_or_table_index, flags);
2994
2995 // determine constant pool cache field offsets
2996 resolve_cache_and_index_for_method(f1_byte, cache, index);
2997 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2998
2999 // Invokeinterface can behave in different ways:
3000 // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
3001 // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
3002 // vtable index is placed in the register.
3003 // Otherwise, the registers will be populated with the klass and method.
3004
3005 Label NotVirtual; Label NotVFinal; Label Done;
3006 __ testl(flags, 1 << ResolvedMethodEntry::is_forced_virtual_shift);
3007 __ jcc(Assembler::zero, NotVirtual);
3008 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
3009 __ jcc(Assembler::zero, NotVFinal);
3010 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3011 __ jmp(Done);
3012
3013 __ bind(NotVFinal);
3014 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
3015 __ jmp(Done);
3016
3017 __ bind(NotVirtual);
3018 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3019 __ movptr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
3020 __ bind(Done);
3021 }
3022
3023 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
3024 Register method_or_table_index,
3025 Register flags) {
3026 // setup registers
3027 const Register index = rdx;
3028 assert_different_registers(index, cache);
3029 assert_different_registers(method_or_table_index, cache, flags);
3030
3031 // determine constant pool cache field offsets
3032 resolve_cache_and_index_for_method(f2_byte, cache, index);
3033 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
3034
3035 // method_or_table_index can either be an itable index or a method depending on the virtual final flag
3036 Label isVFinal; Label Done;
3037 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
3038 __ jcc(Assembler::notZero, isVFinal);
3039 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
3040 __ jmp(Done);
3041 __ bind(isVFinal);
3042 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3043 __ bind(Done);
3044 }
3045
3046 // The registers cache and index expected to be set before call.
3047 // Correct values of the cache and index registers are preserved.
3048 void TemplateTable::jvmti_post_field_access(Register cache,
3049 Register index,
3050 bool is_static,
3051 bool has_tos) {
3052 if (JvmtiExport::can_post_field_access()) {
3053 // Check to see if a field access watch has been set before we take
3054 // the time to call into the VM.
3055 Label L1;
3056 assert_different_registers(cache, index, rax);
3057 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3058 __ testl(rax,rax);
3059 __ jcc(Assembler::zero, L1);
3060
3061 // cache entry pointer
3062 __ load_field_entry(cache, index);
3063 if (is_static) {
3064 __ xorptr(rax, rax); // null object reference
3065 } else {
3066 __ pop(atos); // Get the object
3067 __ verify_oop(rax);
3068 __ push(atos); // Restore stack state
3069 }
3070 // rax,: object pointer or null
3071 // cache: cache entry pointer
3072 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
3073 rax, cache);
3074
3075 __ load_field_entry(cache, index);
3076 __ bind(L1);
3077 }
3078 }
3079
3080 void TemplateTable::pop_and_check_object(Register r) {
3081 __ pop_ptr(r);
3082 __ null_check(r); // for field access must check obj.
3083 __ verify_oop(r);
3084 }
3085
3086 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3087 transition(vtos, vtos);
3088
3089 const Register obj = LP64_ONLY(r9) NOT_LP64(rcx);
3090 const Register cache = rcx;
3091 const Register index = rdx;
3092 const Register off = rbx;
3093 const Register tos_state = rax;
3094 const Register flags = rdx;
3095 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
3096
3097 resolve_cache_and_index_for_field(byte_no, cache, index);
3098 jvmti_post_field_access(cache, index, is_static, false);
3099 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3100
3101 const Address field(obj, off, Address::times_1, 0*wordSize);
3102
3103 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notInlineType;
3104
3105 // Make sure we don't need to mask edx after the above shift
3106 assert(btos == 0, "change code, btos != 0");
3107 __ testl(tos_state, tos_state);
3108 __ jcc(Assembler::notZero, notByte);
3109
3110 // btos
3111 if (!is_static) pop_and_check_object(obj);
3112 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3113 __ push(btos);
3114 // Rewrite bytecode to be faster
3115 if (!is_static && rc == may_rewrite) {
3116 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3117 }
3118 __ jmp(Done);
3119
3120 __ bind(notByte);
3121 __ cmpl(tos_state, ztos);
3122 __ jcc(Assembler::notEqual, notBool);
3123 if (!is_static) pop_and_check_object(obj);
3124 // ztos (same code as btos)
3125 __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
3126 __ push(ztos);
3127 // Rewrite bytecode to be faster
3128 if (!is_static && rc == may_rewrite) {
3129 // use btos rewriting, no truncating to t/f bit is needed for getfield.
3130 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3131 }
3132 __ jmp(Done);
3133
3134 __ bind(notBool);
3135 __ cmpl(tos_state, atos);
3136 __ jcc(Assembler::notEqual, notObj);
3137 // atos
3138 if (!EnableValhalla) {
3139 if (!is_static) pop_and_check_object(obj);
3140 do_oop_load(_masm, field, rax);
3141 __ push(atos);
3142 if (!is_static && rc == may_rewrite) {
3143 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3144 }
3145 __ jmp(Done);
3146 } else {
3147 if (is_static) {
3148 __ load_heap_oop(rax, field);
3149 Label is_null_free_inline_type, uninitialized;
3150 // Issue below if the static field has not been initialized yet
3151 __ test_field_is_null_free_inline_type(flags, rscratch1, is_null_free_inline_type);
3152 // field is not a null free inline type
3153 __ push(atos);
3154 __ jmp(Done);
3155 // field is a null free inline type, must not return null even if uninitialized
3156 __ bind(is_null_free_inline_type);
3157 __ testptr(rax, rax);
3158 __ jcc(Assembler::zero, uninitialized);
3159 __ push(atos);
3160 __ jmp(Done);
3161 __ bind(uninitialized);
3162 #ifdef _LP64
3163 Label slow_case, finish;
3164 __ movptr(rbx, Address(obj, java_lang_Class::klass_offset()));
3165 __ cmpb(Address(rbx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3166 __ jcc(Assembler::notEqual, slow_case);
3167 __ get_default_value_oop(rbx, rscratch1, rax);
3168 __ jmp(finish);
3169 __ bind(slow_case);
3170 #endif // LP64
3171 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_inline_type_field),
3172 obj, cache);
3173 #ifdef _LP64
3174 __ bind(finish);
3175 #endif // _LP64
3176 __ verify_oop(rax);
3177 __ push(atos);
3178 __ jmp(Done);
3179 } else {
3180 Label is_flat, nonnull, is_inline_type, rewrite_inline, has_null_marker;
3181 __ test_field_is_null_free_inline_type(flags, rscratch1, is_inline_type);
3182 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3183 // field is not a null free inline type
3184 pop_and_check_object(obj);
3185 __ load_heap_oop(rax, field);
3186 __ push(atos);
3187 if (rc == may_rewrite) {
3188 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3189 }
3190 __ jmp(Done);
3191 __ bind(is_inline_type);
3192 __ test_field_is_flat(flags, rscratch1, is_flat);
3193 // field is not flat
3194 pop_and_check_object(obj);
3195 __ load_heap_oop(rax, field);
3196 __ testptr(rax, rax);
3197 __ jcc(Assembler::notZero, nonnull);
3198 __ load_unsigned_short(flags, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
3199 __ movptr(rcx, Address(cache, ResolvedFieldEntry::field_holder_offset()));
3200 __ get_inline_type_field_klass(rcx, flags, rbx);
3201 __ get_default_value_oop(rbx, rcx, rax);
3202 __ bind(nonnull);
3203 __ verify_oop(rax);
3204 __ push(atos);
3205 __ jmp(rewrite_inline);
3206 __ bind(is_flat);
3207 pop_and_check_object(rax);
3208 __ read_flat_field(rcx, rdx, rbx, rax);
3209 __ verify_oop(rax);
3210 __ push(atos);
3211 __ jmp(rewrite_inline);
3212 __ bind(has_null_marker);
3213 pop_and_check_object(rax);
3214 __ load_field_entry(rcx, rbx);
3215 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), rax, rcx);
3216 __ get_vm_result(rax, r15_thread);
3217 __ push(atos);
3218 __ bind(rewrite_inline);
3219 if (rc == may_rewrite) {
3220 patch_bytecode(Bytecodes::_fast_vgetfield, bc, rbx);
3221 }
3222 __ jmp(Done);
3223 }
3224 }
3225
3226 __ bind(notObj);
3227
3228 if (!is_static) pop_and_check_object(obj);
3229
3230 __ cmpl(tos_state, itos);
3231 __ jcc(Assembler::notEqual, notInt);
3232 // itos
3233 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3234 __ push(itos);
3235 // Rewrite bytecode to be faster
3236 if (!is_static && rc == may_rewrite) {
3237 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
3238 }
3239 __ jmp(Done);
3240
3241 __ bind(notInt);
3242 __ cmpl(tos_state, ctos);
3243 __ jcc(Assembler::notEqual, notChar);
3244 // ctos
3245 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3246 __ push(ctos);
3247 // Rewrite bytecode to be faster
3248 if (!is_static && rc == may_rewrite) {
3249 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3250 }
3251 __ jmp(Done);
3252
3253 __ bind(notChar);
3254 __ cmpl(tos_state, stos);
3255 __ jcc(Assembler::notEqual, notShort);
3256 // stos
3257 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3258 __ push(stos);
3259 // Rewrite bytecode to be faster
3260 if (!is_static && rc == may_rewrite) {
3261 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3262 }
3263 __ jmp(Done);
3264
3265 __ bind(notShort);
3266 __ cmpl(tos_state, ltos);
3267 __ jcc(Assembler::notEqual, notLong);
3268 // ltos
3269 // Generate code as if volatile (x86_32). There just aren't enough registers to
3270 // save that information and this code is faster than the test.
3271 __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
3272 __ push(ltos);
3273 // Rewrite bytecode to be faster
3274 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3275 __ jmp(Done);
3276
3277 __ bind(notLong);
3278 __ cmpl(tos_state, ftos);
3279 __ jcc(Assembler::notEqual, notFloat);
3280 // ftos
3281
3282 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3283 __ push(ftos);
3284 // Rewrite bytecode to be faster
3285 if (!is_static && rc == may_rewrite) {
3286 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3287 }
3288 __ jmp(Done);
3289
3290 __ bind(notFloat);
3291 #ifdef ASSERT
3292 Label notDouble;
3293 __ cmpl(tos_state, dtos);
3294 __ jcc(Assembler::notEqual, notDouble);
3295 #endif
3296 // dtos
3297 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3298 __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
3299 __ push(dtos);
3300 // Rewrite bytecode to be faster
3301 if (!is_static && rc == may_rewrite) {
3302 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3303 }
3304 #ifdef ASSERT
3305 __ jmp(Done);
3306
3307 __ bind(notDouble);
3308 __ stop("Bad state");
3309 #endif
3310
3311 __ bind(Done);
3312 // [jk] not needed currently
3313 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3314 // Assembler::LoadStore));
3315 }
3316
3317 void TemplateTable::getfield(int byte_no) {
3318 getfield_or_static(byte_no, false);
3319 }
3320
3321 void TemplateTable::nofast_getfield(int byte_no) {
3322 getfield_or_static(byte_no, false, may_not_rewrite);
3323 }
3324
3325 void TemplateTable::getstatic(int byte_no) {
3326 getfield_or_static(byte_no, true);
3327 }
3328
3329 // The registers cache and index expected to be set before call.
3330 // The function may destroy various registers, just not the cache and index registers.
3331 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3332 // Cache is rcx and index is rdx
3333 const Register entry = LP64_ONLY(c_rarg2) NOT_LP64(rax); // ResolvedFieldEntry
3334 const Register obj = LP64_ONLY(c_rarg1) NOT_LP64(rbx); // Object pointer
3335 const Register value = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // JValue object
3336
3337 if (JvmtiExport::can_post_field_modification()) {
3338 // Check to see if a field modification watch has been set before
3339 // we take the time to call into the VM.
3340 Label L1;
3341 assert_different_registers(cache, obj, rax);
3342 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3343 __ testl(rax, rax);
3344 __ jcc(Assembler::zero, L1);
3345
3346 __ mov(entry, cache);
3347
3348 if (is_static) {
3349 // Life is simple. Null out the object pointer.
3350 __ xorl(obj, obj);
3351
3352 } else {
3353 // Life is harder. The stack holds the value on top, followed by
3354 // the object. We don't know the size of the value, though; it
3355 // could be one or two words depending on its type. As a result,
3356 // we must find the type to determine where the object is.
3357 #ifndef _LP64
3358 Label two_word, valsize_known;
3359 #endif
3360 __ load_unsigned_byte(value, Address(entry, in_bytes(ResolvedFieldEntry::type_offset())));
3361 #ifdef _LP64
3362 __ movptr(obj, at_tos_p1()); // initially assume a one word jvalue
3363 __ cmpl(value, ltos);
3364 __ cmovptr(Assembler::equal,
3365 obj, at_tos_p2()); // ltos (two word jvalue)
3366 __ cmpl(value, dtos);
3367 __ cmovptr(Assembler::equal,
3368 obj, at_tos_p2()); // dtos (two word jvalue)
3369 #else
3370 __ mov(obj, rsp);
3371 __ cmpl(value, ltos);
3372 __ jccb(Assembler::equal, two_word);
3373 __ cmpl(value, dtos);
3374 __ jccb(Assembler::equal, two_word);
3375 __ addptr(obj, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3376 __ jmpb(valsize_known);
3377
3378 __ bind(two_word);
3379 __ addptr(obj, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3380
3381 __ bind(valsize_known);
3382 // setup object pointer
3383 __ movptr(obj, Address(obj, 0));
3384 #endif
3385 }
3386
3387 // object (tos)
3388 __ mov(value, rsp);
3389 // obj: object pointer set up above (null if static)
3390 // cache: field entry pointer
3391 // value: jvalue object on the stack
3392 __ call_VM(noreg,
3393 CAST_FROM_FN_PTR(address,
3394 InterpreterRuntime::post_field_modification),
3395 obj, entry, value);
3396 // Reload field entry
3397 __ load_field_entry(cache, index);
3398 __ bind(L1);
3399 }
3400 }
3401
3402 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3403 transition(vtos, vtos);
3404
3405 const Register obj = rcx;
3406 const Register cache = rcx;
3407 const Register index = rdx;
3408 const Register tos_state = rdx;
3409 const Register off = rbx;
3410 const Register flags = r9;
3411
3412 resolve_cache_and_index_for_field(byte_no, cache, index);
3413 jvmti_post_field_mod(cache, index, is_static);
3414 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3415
3416 // [jk] not needed currently
3417 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3418 // Assembler::StoreStore));
3419
3420 Label notVolatile, Done;
3421
3422 // Check for volatile store
3423 __ movl(rscratch1, flags);
3424 __ andl(rscratch1, (1 << ResolvedFieldEntry::is_volatile_shift));
3425 __ testl(rscratch1, rscratch1);
3426 __ jcc(Assembler::zero, notVolatile);
3427
3428 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state, flags);
3429 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3430 Assembler::StoreStore));
3431 __ jmp(Done);
3432 __ bind(notVolatile);
3433
3434 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state, flags);
3435
3436 __ bind(Done);
3437 }
3438
3439 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3440 Register obj, Register off, Register tos_state, Register flags) {
3441
3442 // field addresses
3443 const Address field(obj, off, Address::times_1, 0*wordSize);
3444 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3445
3446 Label notByte, notBool, notInt, notShort, notChar,
3447 notLong, notFloat, notObj, notInlineType;
3448 Label Done;
3449
3450 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3451
3452 // Test TOS state
3453 __ testl(tos_state, tos_state);
3454 __ jcc(Assembler::notZero, notByte);
3455
3456 // btos
3457 {
3458 __ pop(btos);
3459 if (!is_static) pop_and_check_object(obj);
3460 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3461 if (!is_static && rc == may_rewrite) {
3462 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3463 }
3464 __ jmp(Done);
3465 }
3466
3467 __ bind(notByte);
3468 __ cmpl(tos_state, ztos);
3469 __ jcc(Assembler::notEqual, notBool);
3470
3471 // ztos
3472 {
3473 __ pop(ztos);
3474 if (!is_static) pop_and_check_object(obj);
3475 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3476 if (!is_static && rc == may_rewrite) {
3477 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3478 }
3479 __ jmp(Done);
3480 }
3481
3482 __ bind(notBool);
3483 __ cmpl(tos_state, atos);
3484 __ jcc(Assembler::notEqual, notObj);
3485
3486 // atos
3487 {
3488 if (!EnableValhalla) {
3489 __ pop(atos);
3490 if (!is_static) pop_and_check_object(obj);
3491 // Store into the field
3492 do_oop_store(_masm, field, rax);
3493 if (!is_static && rc == may_rewrite) {
3494 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3495 }
3496 __ jmp(Done);
3497 } else {
3498 __ pop(atos);
3499 if (is_static) {
3500 Label is_inline_type;
3501 __ test_field_is_not_null_free_inline_type(flags, rscratch1, is_inline_type);
3502 __ null_check(rax);
3503 __ bind(is_inline_type);
3504 do_oop_store(_masm, field, rax);
3505 __ jmp(Done);
3506 } else {
3507 Label is_null_free_inline_type, is_flat, has_null_marker,
3508 write_null, rewrite_not_inline, rewrite_inline;
3509 __ test_field_is_null_free_inline_type(flags, rscratch1, is_null_free_inline_type);
3510 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3511 // Not an inline type
3512 pop_and_check_object(obj);
3513 // Store into the field
3514 do_oop_store(_masm, field, rax);
3515 __ bind(rewrite_not_inline);
3516 if (rc == may_rewrite) {
3517 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3518 }
3519 __ jmp(Done);
3520 // Implementation of the inline type semantic
3521 __ bind(is_null_free_inline_type);
3522 __ null_check(rax);
3523 __ test_field_is_flat(flags, rscratch1, is_flat);
3524 // field is not flat
3525 pop_and_check_object(obj);
3526 // Store into the field
3527 do_oop_store(_masm, field, rax);
3528 __ jmp(rewrite_inline);
3529 __ bind(is_flat);
3530 // field is flat
3531 __ load_unsigned_short(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_index_offset())));
3532 __ movptr(r9, Address(rcx, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3533 pop_and_check_object(obj); // obj = rcx
3534 __ load_klass(r8, rax, rscratch1);
3535 __ payload_addr(rax, rax, r8);
3536 __ addptr(obj, off);
3537 __ inline_layout_info(r9, rdx, rbx);
3538 // because we use InlineLayoutInfo, we need special value access code specialized for fields (arrays will need a different API)
3539 __ flat_field_copy(IN_HEAP, rax, obj, rbx);
3540 __ jmp(rewrite_inline);
3541 __ bind(has_null_marker); // has null marker means the field is flat with a null marker
3542 pop_and_check_object(rbx);
3543 __ load_field_entry(rcx, rdx);
3544 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), rbx, rax, rcx);
3545 __ bind(rewrite_inline);
3546 if (rc == may_rewrite) {
3547 patch_bytecode(Bytecodes::_fast_vputfield, bc, rbx, true, byte_no);
3548 }
3549 __ jmp(Done);
3550 }
3551 }
3552 }
3553
3554 __ bind(notObj);
3555 __ cmpl(tos_state, itos);
3556 __ jcc(Assembler::notEqual, notInt);
3557
3558 // itos
3559 {
3560 __ pop(itos);
3561 if (!is_static) pop_and_check_object(obj);
3562 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3563 if (!is_static && rc == may_rewrite) {
3564 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3565 }
3566 __ jmp(Done);
3567 }
3568
3569 __ bind(notInt);
3570 __ cmpl(tos_state, ctos);
3571 __ jcc(Assembler::notEqual, notChar);
3572
3573 // ctos
3574 {
3575 __ pop(ctos);
3576 if (!is_static) pop_and_check_object(obj);
3577 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3578 if (!is_static && rc == may_rewrite) {
3579 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3580 }
3581 __ jmp(Done);
3582 }
3583
3584 __ bind(notChar);
3585 __ cmpl(tos_state, stos);
3586 __ jcc(Assembler::notEqual, notShort);
3587
3588 // stos
3589 {
3590 __ pop(stos);
3591 if (!is_static) pop_and_check_object(obj);
3592 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3593 if (!is_static && rc == may_rewrite) {
3594 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3595 }
3596 __ jmp(Done);
3597 }
3598
3599 __ bind(notShort);
3600 __ cmpl(tos_state, ltos);
3601 __ jcc(Assembler::notEqual, notLong);
3602
3603 // ltos
3604 {
3605 __ pop(ltos);
3606 if (!is_static) pop_and_check_object(obj);
3607 // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3608 __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg, noreg);
3609 #ifdef _LP64
3610 if (!is_static && rc == may_rewrite) {
3611 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3612 }
3613 #endif // _LP64
3614 __ jmp(Done);
3615 }
3616
3617 __ bind(notLong);
3618 __ cmpl(tos_state, ftos);
3619 __ jcc(Assembler::notEqual, notFloat);
3620
3621 // ftos
3622 {
3623 __ pop(ftos);
3624 if (!is_static) pop_and_check_object(obj);
3625 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3626 if (!is_static && rc == may_rewrite) {
3627 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3628 }
3629 __ jmp(Done);
3630 }
3631
3632 __ bind(notFloat);
3633 #ifdef ASSERT
3634 Label notDouble;
3635 __ cmpl(tos_state, dtos);
3636 __ jcc(Assembler::notEqual, notDouble);
3637 #endif
3638
3639 // dtos
3640 {
3641 __ pop(dtos);
3642 if (!is_static) pop_and_check_object(obj);
3643 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3644 __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg, noreg);
3645 if (!is_static && rc == may_rewrite) {
3646 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3647 }
3648 }
3649
3650 #ifdef ASSERT
3651 __ jmp(Done);
3652
3653 __ bind(notDouble);
3654 __ stop("Bad state");
3655 #endif
3656
3657 __ bind(Done);
3658 }
3659
3660 void TemplateTable::putfield(int byte_no) {
3661 putfield_or_static(byte_no, false);
3662 }
3663
3664 void TemplateTable::nofast_putfield(int byte_no) {
3665 putfield_or_static(byte_no, false, may_not_rewrite);
3666 }
3667
3668 void TemplateTable::putstatic(int byte_no) {
3669 putfield_or_static(byte_no, true);
3670 }
3671
3672 void TemplateTable::jvmti_post_fast_field_mod() {
3673
3674 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3675
3676 if (JvmtiExport::can_post_field_modification()) {
3677 // Check to see if a field modification watch has been set before
3678 // we take the time to call into the VM.
3679 Label L2;
3680 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3681 __ testl(scratch, scratch);
3682 __ jcc(Assembler::zero, L2);
3683 __ pop_ptr(rbx); // copy the object pointer from tos
3684 __ verify_oop(rbx);
3685 __ push_ptr(rbx); // put the object pointer back on tos
3686 // Save tos values before call_VM() clobbers them. Since we have
3687 // to do it for every data type, we use the saved values as the
3688 // jvalue object.
3689 switch (bytecode()) { // load values into the jvalue object
3690 case Bytecodes::_fast_vputfield: //fall through
3691 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3692 case Bytecodes::_fast_bputfield: // fall through
3693 case Bytecodes::_fast_zputfield: // fall through
3694 case Bytecodes::_fast_sputfield: // fall through
3695 case Bytecodes::_fast_cputfield: // fall through
3696 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3697 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3698 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3699 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3700
3701 default:
3702 ShouldNotReachHere();
3703 }
3704 __ mov(scratch, rsp); // points to jvalue on the stack
3705 // access constant pool cache entry
3706 LP64_ONLY(__ load_field_entry(c_rarg2, rax));
3707 NOT_LP64(__ load_field_entry(rax, rdx));
3708 __ verify_oop(rbx);
3709 // rbx: object pointer copied above
3710 // c_rarg2: cache entry pointer
3711 // c_rarg3: jvalue object on the stack
3712 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3713 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3714
3715 switch (bytecode()) { // restore tos values
3716 case Bytecodes::_fast_vputfield: // fall through
3717 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3718 case Bytecodes::_fast_bputfield: // fall through
3719 case Bytecodes::_fast_zputfield: // fall through
3720 case Bytecodes::_fast_sputfield: // fall through
3721 case Bytecodes::_fast_cputfield: // fall through
3722 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3723 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3724 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3725 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3726 default: break;
3727 }
3728 __ bind(L2);
3729 }
3730 }
3731
3732 void TemplateTable::fast_storefield(TosState state) {
3733 transition(state, vtos);
3734
3735 Label notVolatile, Done;
3736
3737 jvmti_post_fast_field_mod();
3738
3739 __ push(rax);
3740 __ load_field_entry(rcx, rax);
3741 load_resolved_field_entry(noreg, rcx, rax, rbx, rdx);
3742 __ pop(rax);
3743 // RBX: field offset, RCX: RAX: TOS, RDX: flags
3744
3745 // Get object from stack
3746 pop_and_check_object(rcx);
3747
3748 // field address
3749 const Address field(rcx, rbx, Address::times_1);
3750
3751 // Check for volatile store
3752 __ movl(rscratch2, rdx); // saving flags for is_flat test
3753 __ andl(rscratch2, (1 << ResolvedFieldEntry::is_volatile_shift));
3754 __ testl(rscratch2, rscratch2);
3755 __ jcc(Assembler::zero, notVolatile);
3756
3757 fast_storefield_helper(field, rax, rdx);
3758 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3759 Assembler::StoreStore));
3760 __ jmp(Done);
3761 __ bind(notVolatile);
3762
3763 fast_storefield_helper(field, rax, rdx);
3764
3765 __ bind(Done);
3766 }
3767
3768 void TemplateTable::fast_storefield_helper(Address field, Register rax, Register flags) {
3769
3770 // DANGER: 'field' argument depends on rcx and rbx
3771
3772 // access field
3773 switch (bytecode()) {
3774 case Bytecodes::_fast_vputfield:
3775 {
3776 Label is_flat, has_null_marker, write_null, done;
3777 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3778 // Null free field cases: flat or not flat
3779 __ null_check(rax);
3780 __ test_field_is_flat(flags, rscratch1, is_flat);
3781 // field is not flat
3782 do_oop_store(_masm, field, rax);
3783 __ jmp(done);
3784 __ bind(is_flat);
3785 __ load_field_entry(r8, r9);
3786 __ load_unsigned_short(r9, Address(r8, in_bytes(ResolvedFieldEntry::field_index_offset())));
3787 __ movptr(r8, Address(r8, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3788 __ inline_layout_info(r8, r9, r8);
3789 __ load_klass(rdx, rax, rscratch1);
3790 __ payload_addr(rax, rax, rdx);
3791 __ lea(rcx, field);
3792 __ flat_field_copy(IN_HEAP, rax, rcx, r8);
3793 __ jmp(done);
3794 __ bind(has_null_marker); // has null marker means the field is flat with a null marker
3795 __ movptr(rbx, rcx);
3796 __ load_field_entry(rcx, rdx);
3797 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), rbx, rax, rcx);
3798 __ bind(done);
3799 }
3800 break;
3801 case Bytecodes::_fast_aputfield:
3802 {
3803 do_oop_store(_masm, field, rax);
3804 }
3805 break;
3806 case Bytecodes::_fast_lputfield:
3807 #ifdef _LP64
3808 __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg, noreg);
3809 #else
3810 __ stop("should not be rewritten");
3811 #endif
3812 break;
3813 case Bytecodes::_fast_iputfield:
3814 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3815 break;
3816 case Bytecodes::_fast_zputfield:
3817 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3818 break;
3819 case Bytecodes::_fast_bputfield:
3820 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3821 break;
3822 case Bytecodes::_fast_sputfield:
3823 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3824 break;
3825 case Bytecodes::_fast_cputfield:
3826 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3827 break;
3828 case Bytecodes::_fast_fputfield:
3829 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg, noreg);
3830 break;
3831 case Bytecodes::_fast_dputfield:
3832 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg, noreg);
3833 break;
3834 default:
3835 ShouldNotReachHere();
3836 }
3837 }
3838
3839 void TemplateTable::fast_accessfield(TosState state) {
3840 transition(atos, state);
3841
3842 // Do the JVMTI work here to avoid disturbing the register state below
3843 if (JvmtiExport::can_post_field_access()) {
3844 // Check to see if a field access watch has been set before we
3845 // take the time to call into the VM.
3846 Label L1;
3847 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3848 __ testl(rcx, rcx);
3849 __ jcc(Assembler::zero, L1);
3850 // access constant pool cache entry
3851 LP64_ONLY(__ load_field_entry(c_rarg2, rcx));
3852 NOT_LP64(__ load_field_entry(rcx, rdx));
3853 __ verify_oop(rax);
3854 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3855 LP64_ONLY(__ mov(c_rarg1, rax));
3856 // c_rarg1: object pointer copied above
3857 // c_rarg2: cache entry pointer
3858 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3859 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3860 __ pop_ptr(rax); // restore object pointer
3861 __ bind(L1);
3862 }
3863
3864 // access constant pool cache
3865 __ load_field_entry(rcx, rbx);
3866 __ load_sized_value(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3867
3868 // rax: object
3869 __ verify_oop(rax);
3870 __ null_check(rax);
3871 Address field(rax, rdx, Address::times_1);
3872
3873 // access field
3874 switch (bytecode()) {
3875 case Bytecodes::_fast_vgetfield:
3876 {
3877 Label is_flat, nonnull, Done, has_null_marker;
3878 __ load_unsigned_byte(rscratch1, Address(rcx, in_bytes(ResolvedFieldEntry::flags_offset())));
3879 __ test_field_has_null_marker(rscratch1, rscratch2, has_null_marker);
3880 __ test_field_is_flat(rscratch1, rscratch2, is_flat);
3881 // field is not flat
3882 __ load_heap_oop(rax, field);
3883 __ testptr(rax, rax);
3884 __ jcc(Assembler::notZero, nonnull);
3885 __ load_unsigned_short(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_index_offset())));
3886 __ movptr(rcx, Address(rcx, ResolvedFieldEntry::field_holder_offset()));
3887 __ get_inline_type_field_klass(rcx, rdx, rbx);
3888 __ get_default_value_oop(rbx, rcx, rax);
3889 __ bind(nonnull);
3890 __ verify_oop(rax);
3891 __ jmp(Done);
3892 __ bind(is_flat);
3893 // field is flat
3894 __ read_flat_field(rcx, rdx, rbx, rax);
3895 __ jmp(Done);
3896 __ bind(has_null_marker);
3897 // rax = instance, rcx = resolved entry
3898 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), rax, rcx);
3899 __ get_vm_result(rax, r15_thread);
3900 __ bind(Done);
3901 __ verify_oop(rax);
3902 }
3903 break;
3904 case Bytecodes::_fast_agetfield:
3905 do_oop_load(_masm, field, rax);
3906 __ verify_oop(rax);
3907 break;
3908 case Bytecodes::_fast_lgetfield:
3909 #ifdef _LP64
3910 __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3911 #else
3912 __ stop("should not be rewritten");
3913 #endif
3914 break;
3915 case Bytecodes::_fast_igetfield:
3916 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3917 break;
3918 case Bytecodes::_fast_bgetfield:
3919 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3920 break;
3921 case Bytecodes::_fast_sgetfield:
3922 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3923 break;
3924 case Bytecodes::_fast_cgetfield:
3925 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3926 break;
3927 case Bytecodes::_fast_fgetfield:
3928 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3929 break;
3930 case Bytecodes::_fast_dgetfield:
3931 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3932 break;
3933 default:
3934 ShouldNotReachHere();
3935 }
3936 // [jk] not needed currently
3937 // Label notVolatile;
3938 // __ testl(rdx, rdx);
3939 // __ jcc(Assembler::zero, notVolatile);
3940 // __ membar(Assembler::LoadLoad);
3941 // __ bind(notVolatile);
3942 }
3943
3944 void TemplateTable::fast_xaccess(TosState state) {
3945 transition(vtos, state);
3946
3947 // get receiver
3948 __ movptr(rax, aaddress(0));
3949 // access constant pool cache
3950 __ load_field_entry(rcx, rdx, 2);
3951 __ load_sized_value(rbx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3952
3953 // make sure exception is reported in correct bcp range (getfield is
3954 // next instruction)
3955 __ increment(rbcp);
3956 __ null_check(rax);
3957 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3958 switch (state) {
3959 case itos:
3960 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3961 break;
3962 case atos:
3963 do_oop_load(_masm, field, rax);
3964 __ verify_oop(rax);
3965 break;
3966 case ftos:
3967 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3968 break;
3969 default:
3970 ShouldNotReachHere();
3971 }
3972
3973 // [jk] not needed currently
3974 // Label notVolatile;
3975 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3976 // in_bytes(ConstantPoolCache::base_offset() +
3977 // ConstantPoolCacheEntry::flags_offset())));
3978 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3979 // __ testl(rdx, 0x1);
3980 // __ jcc(Assembler::zero, notVolatile);
3981 // __ membar(Assembler::LoadLoad);
3982 // __ bind(notVolatile);
3983
3984 __ decrement(rbcp);
3985 }
3986
3987 //-----------------------------------------------------------------------------
3988 // Calls
3989
3990 void TemplateTable::prepare_invoke(Register cache, Register recv, Register flags) {
3991 // determine flags
3992 const Bytecodes::Code code = bytecode();
3993 const bool load_receiver = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3994 assert_different_registers(recv, flags);
3995
3996 // save 'interpreter return address'
3997 __ save_bcp();
3998
3999 // Save flags and load TOS
4000 __ movl(rbcp, flags);
4001 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
4002
4003 // load receiver if needed (after appendix is pushed so parameter size is correct)
4004 // Note: no return address pushed yet
4005 if (load_receiver) {
4006 __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
4007 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
4008 const int receiver_is_at_end = -1; // back off one slot to get receiver
4009 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
4010 __ movptr(recv, recv_addr);
4011 __ verify_oop(recv);
4012 }
4013
4014 // load return address
4015 {
4016 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
4017 ExternalAddress table(table_addr);
4018 #ifdef _LP64
4019 __ lea(rscratch1, table);
4020 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
4021 #else
4022 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
4023 #endif // _LP64
4024 }
4025
4026 // push return address
4027 __ push(flags);
4028
4029 // Restore flags value from the constant pool cache entry, and restore rsi
4030 // for later null checks. r13 is the bytecode pointer
4031 __ movl(flags, rbcp);
4032 __ restore_bcp();
4033 }
4034
4035 void TemplateTable::invokevirtual_helper(Register index,
4036 Register recv,
4037 Register flags) {
4038 // Uses temporary registers rax, rdx
4039 assert_different_registers(index, recv, rax, rdx);
4040 assert(index == rbx, "");
4041 assert(recv == rcx, "");
4042
4043 // Test for an invoke of a final method
4044 Label notFinal;
4045 __ movl(rax, flags);
4046 __ andl(rax, (1 << ResolvedMethodEntry::is_vfinal_shift));
4047 __ jcc(Assembler::zero, notFinal);
4048
4049 const Register method = index; // method must be rbx
4050 assert(method == rbx,
4051 "Method* must be rbx for interpreter calling convention");
4052
4053 // do the call - the index is actually the method to call
4054 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
4055
4056 // It's final, need a null check here!
4057 __ null_check(recv);
4058
4059 // profile this call
4060 __ profile_final_call(rax);
4061 __ profile_arguments_type(rax, method, rbcp, true);
4062
4063 __ jump_from_interpreted(method, rax);
4064
4065 __ bind(notFinal);
4066
4067 // get receiver klass
4068 __ load_klass(rax, recv, rscratch1);
4069
4070 // profile this call
4071 __ profile_virtual_call(rax, rlocals, rdx);
4072 // get target Method* & entry point
4073 __ lookup_virtual_method(rax, index, method);
4074
4075 __ profile_arguments_type(rdx, method, rbcp, true);
4076 __ jump_from_interpreted(method, rdx);
4077 }
4078
4079 void TemplateTable::invokevirtual(int byte_no) {
4080 transition(vtos, vtos);
4081 assert(byte_no == f2_byte, "use this argument");
4082
4083 load_resolved_method_entry_virtual(rcx, // ResolvedMethodEntry*
4084 rbx, // Method or itable index
4085 rdx); // Flags
4086 prepare_invoke(rcx, // ResolvedMethodEntry*
4087 rcx, // Receiver
4088 rdx); // flags
4089
4090 // rbx: index
4091 // rcx: receiver
4092 // rdx: flags
4093 invokevirtual_helper(rbx, rcx, rdx);
4094 }
4095
4096 void TemplateTable::invokespecial(int byte_no) {
4097 transition(vtos, vtos);
4098 assert(byte_no == f1_byte, "use this argument");
4099
4100 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
4101 rbx, // Method*
4102 rdx); // flags
4103 prepare_invoke(rcx,
4104 rcx, // get receiver also for null check
4105 rdx); // flags
4106
4107 __ verify_oop(rcx);
4108 __ null_check(rcx);
4109 // do the call
4110 __ profile_call(rax);
4111 __ profile_arguments_type(rax, rbx, rbcp, false);
4112 __ jump_from_interpreted(rbx, rax);
4113 }
4114
4115 void TemplateTable::invokestatic(int byte_no) {
4116 transition(vtos, vtos);
4117 assert(byte_no == f1_byte, "use this argument");
4118
4119 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
4120 rbx, // Method*
4121 rdx // flags
4122 );
4123 prepare_invoke(rcx, rcx, rdx); // cache and flags
4124
4125 // do the call
4126 __ profile_call(rax);
4127 __ profile_arguments_type(rax, rbx, rbcp, false);
4128 __ jump_from_interpreted(rbx, rax);
4129 }
4130
4131
4132 void TemplateTable::fast_invokevfinal(int byte_no) {
4133 transition(vtos, vtos);
4134 assert(byte_no == f2_byte, "use this argument");
4135 __ stop("fast_invokevfinal not used on x86");
4136 }
4137
4138
4139 void TemplateTable::invokeinterface(int byte_no) {
4140 transition(vtos, vtos);
4141 assert(byte_no == f1_byte, "use this argument");
4142
4143 load_resolved_method_entry_interface(rcx, // ResolvedMethodEntry*
4144 rax, // Klass*
4145 rbx, // Method* or itable/vtable index
4146 rdx); // flags
4147 prepare_invoke(rcx, rcx, rdx); // receiver, flags
4148
4149 // First check for Object case, then private interface method,
4150 // then regular interface method.
4151
4152 // Special case of invokeinterface called for virtual method of
4153 // java.lang.Object. See cpCache.cpp for details.
4154 Label notObjectMethod;
4155 __ movl(rlocals, rdx);
4156 __ andl(rlocals, (1 << ResolvedMethodEntry::is_forced_virtual_shift));
4157 __ jcc(Assembler::zero, notObjectMethod);
4158
4159 invokevirtual_helper(rbx, rcx, rdx);
4160 // no return from above
4161 __ bind(notObjectMethod);
4162
4163 Label no_such_interface; // for receiver subtype check
4164 Register recvKlass; // used for exception processing
4165
4166 // Check for private method invocation - indicated by vfinal
4167 Label notVFinal;
4168 __ movl(rlocals, rdx);
4169 __ andl(rlocals, (1 << ResolvedMethodEntry::is_vfinal_shift));
4170 __ jcc(Assembler::zero, notVFinal);
4171
4172 // Get receiver klass into rlocals - also a null check
4173 __ load_klass(rlocals, rcx, rscratch1);
4174
4175 Label subtype;
4176 __ check_klass_subtype(rlocals, rax, rbcp, subtype);
4177 // If we get here the typecheck failed
4178 recvKlass = rdx;
4179 __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
4180 __ jmp(no_such_interface);
4181
4182 __ bind(subtype);
4183
4184 // do the call - rbx is actually the method to call
4185
4186 __ profile_final_call(rdx);
4187 __ profile_arguments_type(rdx, rbx, rbcp, true);
4188
4189 __ jump_from_interpreted(rbx, rdx);
4190 // no return from above
4191 __ bind(notVFinal);
4192
4193 // Get receiver klass into rdx - also a null check
4194 __ restore_locals(); // restore r14
4195 __ load_klass(rdx, rcx, rscratch1);
4196
4197 Label no_such_method;
4198
4199 // Preserve method for throw_AbstractMethodErrorVerbose.
4200 __ mov(rcx, rbx);
4201 // Receiver subtype check against REFC.
4202 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
4203 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4204 rdx, rax, noreg,
4205 // outputs: scan temp. reg, scan temp. reg
4206 rbcp, rlocals,
4207 no_such_interface,
4208 /*return_method=*/false);
4209
4210 // profile this call
4211 __ restore_bcp(); // rbcp was destroyed by receiver type check
4212 __ profile_virtual_call(rdx, rbcp, rlocals);
4213
4214 // Get declaring interface class from method, and itable index
4215 __ load_method_holder(rax, rbx);
4216 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
4217 __ subl(rbx, Method::itable_index_max);
4218 __ negl(rbx);
4219
4220 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
4221 __ mov(rlocals, rdx);
4222 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4223 rlocals, rax, rbx,
4224 // outputs: method, scan temp. reg
4225 rbx, rbcp,
4226 no_such_interface);
4227
4228 // rbx: Method* to call
4229 // rcx: receiver
4230 // Check for abstract method error
4231 // Note: This should be done more efficiently via a throw_abstract_method_error
4232 // interpreter entry point and a conditional jump to it in case of a null
4233 // method.
4234 __ testptr(rbx, rbx);
4235 __ jcc(Assembler::zero, no_such_method);
4236
4237 __ profile_arguments_type(rdx, rbx, rbcp, true);
4238
4239 // do the call
4240 // rcx: receiver
4241 // rbx,: Method*
4242 __ jump_from_interpreted(rbx, rdx);
4243 __ should_not_reach_here();
4244
4245 // exception handling code follows...
4246 // note: must restore interpreter registers to canonical
4247 // state for exception handling to work correctly!
4248
4249 __ bind(no_such_method);
4250 // throw exception
4251 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4252 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4253 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4254 // Pass arguments for generating a verbose error message.
4255 #ifdef _LP64
4256 recvKlass = c_rarg1;
4257 Register method = c_rarg2;
4258 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
4259 if (method != rcx) { __ movq(method, rcx); }
4260 #else
4261 recvKlass = rdx;
4262 Register method = rcx;
4263 #endif
4264 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
4265 recvKlass, method);
4266 // The call_VM checks for exception, so we should never return here.
4267 __ should_not_reach_here();
4268
4269 __ bind(no_such_interface);
4270 // throw exception
4271 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4272 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4273 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4274 // Pass arguments for generating a verbose error message.
4275 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
4276 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
4277 recvKlass, rax);
4278 // the call_VM checks for exception, so we should never return here.
4279 __ should_not_reach_here();
4280 }
4281
4282 void TemplateTable::invokehandle(int byte_no) {
4283 transition(vtos, vtos);
4284 assert(byte_no == f1_byte, "use this argument");
4285 const Register rbx_method = rbx;
4286 const Register rax_mtype = rax;
4287 const Register rcx_recv = rcx;
4288 const Register rdx_flags = rdx;
4289
4290 load_resolved_method_entry_handle(rcx, rbx_method, rax_mtype, rdx_flags);
4291 prepare_invoke(rcx, rcx_recv, rdx_flags);
4292
4293 __ verify_method_ptr(rbx_method);
4294 __ verify_oop(rcx_recv);
4295 __ null_check(rcx_recv);
4296
4297 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4298 // rbx: MH.invokeExact_MT method
4299
4300 // Note: rax_mtype is already pushed (if necessary)
4301
4302 // FIXME: profile the LambdaForm also
4303 __ profile_final_call(rax);
4304 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4305
4306 __ jump_from_interpreted(rbx_method, rdx);
4307 }
4308
4309 void TemplateTable::invokedynamic(int byte_no) {
4310 transition(vtos, vtos);
4311 assert(byte_no == f1_byte, "use this argument");
4312
4313 const Register rbx_method = rbx;
4314 const Register rax_callsite = rax;
4315
4316 load_invokedynamic_entry(rbx_method);
4317 // rax: CallSite object (from cpool->resolved_references[])
4318 // rbx: MH.linkToCallSite method
4319
4320 // Note: rax_callsite is already pushed
4321
4322 // %%% should make a type profile for any invokedynamic that takes a ref argument
4323 // profile this call
4324 __ profile_call(rbcp);
4325 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4326
4327 __ verify_oop(rax_callsite);
4328
4329 __ jump_from_interpreted(rbx_method, rdx);
4330 }
4331
4332 //-----------------------------------------------------------------------------
4333 // Allocation
4334
4335 void TemplateTable::_new() {
4336 transition(vtos, atos);
4337 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4338 Label slow_case;
4339 Label done;
4340
4341 __ get_cpool_and_tags(rcx, rax);
4342
4343 // Make sure the class we're about to instantiate has been resolved.
4344 // This is done before loading InstanceKlass to be consistent with the order
4345 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4346 const int tags_offset = Array<u1>::base_offset_in_bytes();
4347 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4348 __ jcc(Assembler::notEqual, slow_case);
4349
4350 // get InstanceKlass
4351 __ load_resolved_klass_at_index(rcx, rcx, rdx);
4352
4353 // make sure klass is initialized
4354 // init_state needs acquire, but x86 is TSO, and so we are already good.
4355 #ifdef _LP64
4356 assert(VM_Version::supports_fast_class_init_checks(), "must support fast class initialization checks");
4357 __ clinit_barrier(rcx, r15_thread, nullptr /*L_fast_path*/, &slow_case);
4358 #else
4359 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4360 __ jcc(Assembler::notEqual, slow_case);
4361 #endif
4362
4363 __ allocate_instance(rcx, rax, rdx, rbx, true, slow_case);
4364 if (DTraceAllocProbes) {
4365 // Trigger dtrace event for fastpath
4366 __ push(atos);
4367 __ call_VM_leaf(
4368 CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), rax);
4369 __ pop(atos);
4370 }
4371 __ jmp(done);
4372
4373 // slow case
4374 __ bind(slow_case);
4375
4376 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4377 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4378
4379 __ get_constant_pool(rarg1);
4380 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4381 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4382 __ verify_oop(rax);
4383
4384 // continue
4385 __ bind(done);
4386 }
4387
4388 void TemplateTable::newarray() {
4389 transition(itos, atos);
4390 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4391 __ load_unsigned_byte(rarg1, at_bcp(1));
4392 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4393 rarg1, rax);
4394 }
4395
4396 void TemplateTable::anewarray() {
4397 transition(itos, atos);
4398
4399 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4400 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4401
4402 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4403 __ get_constant_pool(rarg1);
4404 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4405 rarg1, rarg2, rax);
4406 }
4407
4408 void TemplateTable::arraylength() {
4409 transition(atos, itos);
4410 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4411 }
4412
4413 void TemplateTable::checkcast() {
4414 transition(atos, atos);
4415 Label done, is_null, ok_is_subtype, quicked, resolved;
4416 __ testptr(rax, rax); // object is in rax
4417 __ jcc(Assembler::zero, is_null);
4418
4419 // Get cpool & tags index
4420 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4421 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4422 // See if bytecode has already been quicked
4423 __ movzbl(rdx, Address(rdx, rbx,
4424 Address::times_1,
4425 Array<u1>::base_offset_in_bytes()));
4426 __ cmpl(rdx, JVM_CONSTANT_Class);
4427 __ jcc(Assembler::equal, quicked);
4428 __ push(atos); // save receiver for result, and for GC
4429 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4430
4431 // vm_result_2 has metadata result
4432 #ifndef _LP64
4433 // borrow rdi from locals
4434 __ get_thread(rdi);
4435 __ get_vm_result_2(rax, rdi);
4436 __ restore_locals();
4437 #else
4438 __ get_vm_result_2(rax, r15_thread);
4439 #endif
4440
4441 __ pop_ptr(rdx); // restore receiver
4442 __ jmpb(resolved);
4443
4444 // Get superklass in rax and subklass in rbx
4445 __ bind(quicked);
4446 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4447 __ load_resolved_klass_at_index(rax, rcx, rbx);
4448
4449 __ bind(resolved);
4450 __ load_klass(rbx, rdx, rscratch1);
4451
4452 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4453 // Superklass in rax. Subklass in rbx.
4454 __ gen_subtype_check(rbx, ok_is_subtype);
4455
4456 // Come here on failure
4457 __ push_ptr(rdx);
4458 // object is at TOS
4459 __ jump(RuntimeAddress(Interpreter::_throw_ClassCastException_entry));
4460
4461 // Come here on success
4462 __ bind(ok_is_subtype);
4463 __ mov(rax, rdx); // Restore object in rdx
4464 __ jmp(done);
4465
4466 __ bind(is_null);
4467
4468 // Collect counts on whether this check-cast sees nulls a lot or not.
4469 if (ProfileInterpreter) {
4470 __ profile_null_seen(rcx);
4471 }
4472
4473 __ bind(done);
4474 }
4475
4476 void TemplateTable::instanceof() {
4477 transition(atos, itos);
4478 Label done, is_null, ok_is_subtype, quicked, resolved;
4479 __ testptr(rax, rax);
4480 __ jcc(Assembler::zero, is_null);
4481
4482 // Get cpool & tags index
4483 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4484 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4485 // See if bytecode has already been quicked
4486 __ movzbl(rdx, Address(rdx, rbx,
4487 Address::times_1,
4488 Array<u1>::base_offset_in_bytes()));
4489 __ cmpl(rdx, JVM_CONSTANT_Class);
4490 __ jcc(Assembler::equal, quicked);
4491
4492 __ push(atos); // save receiver for result, and for GC
4493 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4494 // vm_result_2 has metadata result
4495
4496 #ifndef _LP64
4497 // borrow rdi from locals
4498 __ get_thread(rdi);
4499 __ get_vm_result_2(rax, rdi);
4500 __ restore_locals();
4501 #else
4502 __ get_vm_result_2(rax, r15_thread);
4503 #endif
4504
4505 __ pop_ptr(rdx); // restore receiver
4506 __ verify_oop(rdx);
4507 __ load_klass(rdx, rdx, rscratch1);
4508 __ jmpb(resolved);
4509
4510 // Get superklass in rax and subklass in rdx
4511 __ bind(quicked);
4512 __ load_klass(rdx, rax, rscratch1);
4513 __ load_resolved_klass_at_index(rax, rcx, rbx);
4514
4515 __ bind(resolved);
4516
4517 // Generate subtype check. Blows rcx, rdi
4518 // Superklass in rax. Subklass in rdx.
4519 __ gen_subtype_check(rdx, ok_is_subtype);
4520
4521 // Come here on failure
4522 __ xorl(rax, rax);
4523 __ jmpb(done);
4524 // Come here on success
4525 __ bind(ok_is_subtype);
4526 __ movl(rax, 1);
4527
4528 // Collect counts on whether this test sees nulls a lot or not.
4529 if (ProfileInterpreter) {
4530 __ jmp(done);
4531 __ bind(is_null);
4532 __ profile_null_seen(rcx);
4533 } else {
4534 __ bind(is_null); // same as 'done'
4535 }
4536 __ bind(done);
4537 // rax = 0: obj == nullptr or obj is not an instanceof the specified klass
4538 // rax = 1: obj != nullptr and obj is an instanceof the specified klass
4539 }
4540
4541 //----------------------------------------------------------------------------------------------------
4542 // Breakpoints
4543 void TemplateTable::_breakpoint() {
4544 // Note: We get here even if we are single stepping..
4545 // jbug insists on setting breakpoints at every bytecode
4546 // even if we are in single step mode.
4547
4548 transition(vtos, vtos);
4549
4550 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4551
4552 // get the unpatched byte code
4553 __ get_method(rarg);
4554 __ call_VM(noreg,
4555 CAST_FROM_FN_PTR(address,
4556 InterpreterRuntime::get_original_bytecode_at),
4557 rarg, rbcp);
4558 __ mov(rbx, rax); // why?
4559
4560 // post the breakpoint event
4561 __ get_method(rarg);
4562 __ call_VM(noreg,
4563 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4564 rarg, rbcp);
4565
4566 // complete the execution of original bytecode
4567 __ dispatch_only_normal(vtos);
4568 }
4569
4570 //-----------------------------------------------------------------------------
4571 // Exceptions
4572
4573 void TemplateTable::athrow() {
4574 transition(atos, vtos);
4575 __ null_check(rax);
4576 __ jump(RuntimeAddress(Interpreter::throw_exception_entry()));
4577 }
4578
4579 //-----------------------------------------------------------------------------
4580 // Synchronization
4581 //
4582 // Note: monitorenter & exit are symmetric routines; which is reflected
4583 // in the assembly code structure as well
4584 //
4585 // Stack layout:
4586 //
4587 // [expressions ] <--- rsp = expression stack top
4588 // ..
4589 // [expressions ]
4590 // [monitor entry] <--- monitor block top = expression stack bot
4591 // ..
4592 // [monitor entry]
4593 // [frame data ] <--- monitor block bot
4594 // ...
4595 // [saved rbp ] <--- rbp
4596 void TemplateTable::monitorenter() {
4597 transition(atos, vtos);
4598
4599 // check for null object
4600 __ null_check(rax);
4601
4602 Label is_inline_type;
4603 __ movptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
4604 __ test_markword_is_inline_type(rbx, is_inline_type);
4605
4606 const Address monitor_block_top(
4607 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4608 const Address monitor_block_bot(
4609 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4610 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4611
4612 Label allocated;
4613
4614 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4615 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4616 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4617
4618 // initialize entry pointer
4619 __ xorl(rmon, rmon); // points to free slot or null
4620
4621 // find a free slot in the monitor block (result in rmon)
4622 {
4623 Label entry, loop, exit;
4624 __ movptr(rtop, monitor_block_top); // derelativize pointer
4625 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4626 // rtop points to current entry, starting with top-most entry
4627
4628 __ lea(rbot, monitor_block_bot); // points to word before bottom
4629 // of monitor block
4630 __ jmpb(entry);
4631
4632 __ bind(loop);
4633 // check if current entry is used
4634 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset()), NULL_WORD);
4635 // if not used then remember entry in rmon
4636 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4637 // check if current entry is for same object
4638 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4639 // if same object then stop searching
4640 __ jccb(Assembler::equal, exit);
4641 // otherwise advance to next entry
4642 __ addptr(rtop, entry_size);
4643 __ bind(entry);
4644 // check if bottom reached
4645 __ cmpptr(rtop, rbot);
4646 // if not at bottom then check this entry
4647 __ jcc(Assembler::notEqual, loop);
4648 __ bind(exit);
4649 }
4650
4651 __ testptr(rmon, rmon); // check if a slot has been found
4652 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4653
4654 // allocate one if there's no free slot
4655 {
4656 Label entry, loop;
4657 // 1. compute new pointers // rsp: old expression stack top
4658 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4659 __ lea(rmon, Address(rbp, rmon, Address::times_ptr));
4660 __ subptr(rsp, entry_size); // move expression stack top
4661 __ subptr(rmon, entry_size); // move expression stack bottom
4662 __ mov(rtop, rsp); // set start value for copy loop
4663 __ subptr(monitor_block_bot, entry_size / wordSize); // set new monitor block bottom
4664 __ jmp(entry);
4665 // 2. move expression stack contents
4666 __ bind(loop);
4667 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4668 // word from old location
4669 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4670 __ addptr(rtop, wordSize); // advance to next word
4671 __ bind(entry);
4672 __ cmpptr(rtop, rmon); // check if bottom reached
4673 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4674 // copy next word
4675 }
4676
4677 // call run-time routine
4678 // rmon: points to monitor entry
4679 __ bind(allocated);
4680
4681 // Increment bcp to point to the next bytecode, so exception
4682 // handling for async. exceptions work correctly.
4683 // The object has already been popped from the stack, so the
4684 // expression stack looks correct.
4685 __ increment(rbcp);
4686
4687 // store object
4688 __ movptr(Address(rmon, BasicObjectLock::obj_offset()), rax);
4689 __ lock_object(rmon);
4690
4691 // check to make sure this monitor doesn't cause stack overflow after locking
4692 __ save_bcp(); // in case of exception
4693 __ generate_stack_overflow_check(0);
4694
4695 // The bcp has already been incremented. Just need to dispatch to
4696 // next instruction.
4697 __ dispatch_next(vtos);
4698
4699 __ bind(is_inline_type);
4700 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4701 InterpreterRuntime::throw_identity_exception), rax);
4702 __ should_not_reach_here();
4703 }
4704
4705 void TemplateTable::monitorexit() {
4706 transition(atos, vtos);
4707
4708 // check for null object
4709 __ null_check(rax);
4710
4711 const int is_inline_type_mask = markWord::inline_type_pattern;
4712 Label has_identity;
4713 __ movptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
4714 __ andptr(rbx, is_inline_type_mask);
4715 __ cmpl(rbx, is_inline_type_mask);
4716 __ jcc(Assembler::notEqual, has_identity);
4717 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4718 InterpreterRuntime::throw_illegal_monitor_state_exception));
4719 __ should_not_reach_here();
4720 __ bind(has_identity);
4721
4722 const Address monitor_block_top(
4723 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4724 const Address monitor_block_bot(
4725 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4726 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4727
4728 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4729 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4730
4731 Label found;
4732
4733 // find matching slot
4734 {
4735 Label entry, loop;
4736 __ movptr(rtop, monitor_block_top); // derelativize pointer
4737 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4738 // rtop points to current entry, starting with top-most entry
4739
4740 __ lea(rbot, monitor_block_bot); // points to word before bottom
4741 // of monitor block
4742 __ jmpb(entry);
4743
4744 __ bind(loop);
4745 // check if current entry is for same object
4746 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4747 // if same object then stop searching
4748 __ jcc(Assembler::equal, found);
4749 // otherwise advance to next entry
4750 __ addptr(rtop, entry_size);
4751 __ bind(entry);
4752 // check if bottom reached
4753 __ cmpptr(rtop, rbot);
4754 // if not at bottom then check this entry
4755 __ jcc(Assembler::notEqual, loop);
4756 }
4757
4758 // error handling. Unlocking was not block-structured
4759 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4760 InterpreterRuntime::throw_illegal_monitor_state_exception));
4761 __ should_not_reach_here();
4762
4763 // call run-time routine
4764 __ bind(found);
4765 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4766 __ unlock_object(rtop);
4767 __ pop_ptr(rax); // discard object
4768 }
4769
4770 // Wide instructions
4771 void TemplateTable::wide() {
4772 transition(vtos, vtos);
4773 __ load_unsigned_byte(rbx, at_bcp(1));
4774 ExternalAddress wtable((address)Interpreter::_wentry_point);
4775 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)), rscratch1);
4776 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4777 }
4778
4779 // Multi arrays
4780 void TemplateTable::multianewarray() {
4781 transition(vtos, atos);
4782
4783 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4784 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4785 // last dim is on top of stack; we want address of first one:
4786 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4787 // the latter wordSize to point to the beginning of the array.
4788 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4789 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4790 __ load_unsigned_byte(rbx, at_bcp(3));
4791 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4792 }