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 "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "compiler/disassembler.hpp"
28 #include "gc/shared/collectedHeap.hpp"
29 #include "gc/shared/gc_globals.hpp"
30 #include "gc/shared/tlab_globals.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "memory/universe.hpp"
36 #include "oops/methodCounters.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/objArrayKlass.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "oops/inlineKlass.hpp"
41 #include "oops/resolvedFieldEntry.hpp"
42 #include "oops/resolvedIndyEntry.hpp"
43 #include "oops/resolvedMethodEntry.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/methodHandles.hpp"
46 #include "runtime/frame.inline.hpp"
47 #include "runtime/safepointMechanism.hpp"
48 #include "runtime/sharedRuntime.hpp"
49 #include "runtime/stubRoutines.hpp"
50 #include "runtime/synchronizer.hpp"
51 #include "utilities/macros.hpp"
52
53 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
54
55 // Global Register Names
56 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
57 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
58
59 // Address Computation: local variables
60 static inline Address iaddress(int n) {
61 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
62 }
63
64 static inline Address laddress(int n) {
65 return iaddress(n + 1);
66 }
67
68 #ifndef _LP64
69 static inline Address haddress(int n) {
70 return iaddress(n + 0);
71 }
72 #endif
73
74 static inline Address faddress(int n) {
75 return iaddress(n);
76 }
77
78 static inline Address daddress(int n) {
79 return laddress(n);
80 }
81
82 static inline Address aaddress(int n) {
83 return iaddress(n);
84 }
85
86 static inline Address iaddress(Register r) {
87 return Address(rlocals, r, Address::times_ptr);
88 }
89
90 static inline Address laddress(Register r) {
91 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
92 }
93
94 #ifndef _LP64
95 static inline Address haddress(Register r) {
96 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
97 }
98 #endif
99
100 static inline Address faddress(Register r) {
101 return iaddress(r);
102 }
103
104 static inline Address daddress(Register r) {
105 return laddress(r);
106 }
107
108 static inline Address aaddress(Register r) {
109 return iaddress(r);
110 }
111
112
113 // expression stack
114 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
115 // data beyond the rsp which is potentially unsafe in an MT environment;
116 // an interrupt may overwrite that data.)
117 static inline Address at_rsp () {
118 return Address(rsp, 0);
119 }
120
121 // At top of Java expression stack which may be different than esp(). It
122 // isn't for category 1 objects.
123 static inline Address at_tos () {
124 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
125 }
126
127 static inline Address at_tos_p1() {
128 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
129 }
130
131 static inline Address at_tos_p2() {
132 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
133 }
134
135 // Condition conversion
136 static Assembler::Condition j_not(TemplateTable::Condition cc) {
137 switch (cc) {
138 case TemplateTable::equal : return Assembler::notEqual;
139 case TemplateTable::not_equal : return Assembler::equal;
140 case TemplateTable::less : return Assembler::greaterEqual;
141 case TemplateTable::less_equal : return Assembler::greater;
142 case TemplateTable::greater : return Assembler::lessEqual;
143 case TemplateTable::greater_equal: return Assembler::less;
144 }
145 ShouldNotReachHere();
146 return Assembler::zero;
147 }
148
149
150
151 // Miscellaneous helper routines
152 // Store an oop (or null) at the address described by obj.
153 // If val == noreg this means store a null
154
155
156 static void do_oop_store(InterpreterMacroAssembler* _masm,
157 Address dst,
158 Register val,
159 DecoratorSet decorators = 0) {
160 assert(val == noreg || val == rax, "parameter is just for looks");
161 __ store_heap_oop(dst, val,
162 NOT_LP64(rdx) LP64_ONLY(rscratch2),
163 NOT_LP64(rbx) LP64_ONLY(r9),
164 NOT_LP64(rsi) LP64_ONLY(r8), decorators);
165 }
166
167 static void do_oop_load(InterpreterMacroAssembler* _masm,
168 Address src,
169 Register dst,
170 DecoratorSet decorators = 0) {
171 __ load_heap_oop(dst, src, rdx, rbx, decorators);
172 }
173
174 Address TemplateTable::at_bcp(int offset) {
175 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
176 return Address(rbcp, offset);
177 }
178
179
180 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
181 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
182 int byte_no) {
183 if (!RewriteBytecodes) return;
184 Label L_patch_done;
185
186 switch (bc) {
187 case Bytecodes::_fast_vputfield:
188 case Bytecodes::_fast_aputfield:
189 case Bytecodes::_fast_bputfield:
190 case Bytecodes::_fast_zputfield:
191 case Bytecodes::_fast_cputfield:
192 case Bytecodes::_fast_dputfield:
193 case Bytecodes::_fast_fputfield:
194 case Bytecodes::_fast_iputfield:
195 case Bytecodes::_fast_lputfield:
196 case Bytecodes::_fast_sputfield:
197 {
198 // We skip bytecode quickening for putfield instructions when
199 // the put_code written to the constant pool cache is zero.
200 // This is required so that every execution of this instruction
201 // calls out to InterpreterRuntime::resolve_get_put to do
202 // additional, required work.
203 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
204 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
205 __ load_field_entry(temp_reg, bc_reg);
206 if (byte_no == f1_byte) {
207 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::get_code_offset())));
208 } else {
209 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::put_code_offset())));
210 }
211
212 __ movl(bc_reg, bc);
213 __ cmpl(temp_reg, (int) 0);
214 __ jcc(Assembler::zero, L_patch_done); // don't patch
215 }
216 break;
217 default:
218 assert(byte_no == -1, "sanity");
219 // the pair bytecodes have already done the load.
220 if (load_bc_into_bc_reg) {
221 __ movl(bc_reg, bc);
222 }
223 }
224
225 if (JvmtiExport::can_post_breakpoint()) {
226 Label L_fast_patch;
227 // if a breakpoint is present we can't rewrite the stream directly
228 __ movzbl(temp_reg, at_bcp(0));
229 __ cmpl(temp_reg, Bytecodes::_breakpoint);
230 __ jcc(Assembler::notEqual, L_fast_patch);
231 __ get_method(temp_reg);
232 // Let breakpoint table handling rewrite to quicker bytecode
233 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
234 #ifndef ASSERT
235 __ jmpb(L_patch_done);
236 #else
237 __ jmp(L_patch_done);
238 #endif
239 __ bind(L_fast_patch);
240 }
241
242 #ifdef ASSERT
243 Label L_okay;
244 __ load_unsigned_byte(temp_reg, at_bcp(0));
245 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
246 __ jcc(Assembler::equal, L_okay);
247 __ cmpl(temp_reg, bc_reg);
248 __ jcc(Assembler::equal, L_okay);
249 __ stop("patching the wrong bytecode");
250 __ bind(L_okay);
251 #endif
252
253 // patch bytecode
254 __ movb(at_bcp(0), bc_reg);
255 __ bind(L_patch_done);
256 }
257 // Individual instructions
258
259
260 void TemplateTable::nop() {
261 transition(vtos, vtos);
262 // nothing to do
263 }
264
265 void TemplateTable::shouldnotreachhere() {
266 transition(vtos, vtos);
267 __ stop("shouldnotreachhere bytecode");
268 }
269
270 void TemplateTable::aconst_null() {
271 transition(vtos, atos);
272 __ xorl(rax, rax);
273 }
274
275 void TemplateTable::iconst(int value) {
276 transition(vtos, itos);
277 if (value == 0) {
278 __ xorl(rax, rax);
279 } else {
280 __ movl(rax, value);
281 }
282 }
283
284 void TemplateTable::lconst(int value) {
285 transition(vtos, ltos);
286 if (value == 0) {
287 __ xorl(rax, rax);
288 } else {
289 __ movl(rax, value);
290 }
291 #ifndef _LP64
292 assert(value >= 0, "check this code");
293 __ xorptr(rdx, rdx);
294 #endif
295 }
296
297
298
299 void TemplateTable::fconst(int value) {
300 transition(vtos, ftos);
301 if (UseSSE >= 1) {
302 static float one = 1.0f, two = 2.0f;
303 switch (value) {
304 case 0:
305 __ xorps(xmm0, xmm0);
306 break;
307 case 1:
308 __ movflt(xmm0, ExternalAddress((address) &one), rscratch1);
309 break;
310 case 2:
311 __ movflt(xmm0, ExternalAddress((address) &two), rscratch1);
312 break;
313 default:
314 ShouldNotReachHere();
315 break;
316 }
317 } else {
318 #ifdef _LP64
319 ShouldNotReachHere();
320 #else
321 if (value == 0) { __ fldz();
322 } else if (value == 1) { __ fld1();
323 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
324 } else { ShouldNotReachHere();
325 }
326 #endif // _LP64
327 }
328 }
329
330 void TemplateTable::dconst(int value) {
331 transition(vtos, dtos);
332 if (UseSSE >= 2) {
333 static double one = 1.0;
334 switch (value) {
335 case 0:
336 __ xorpd(xmm0, xmm0);
337 break;
338 case 1:
339 __ movdbl(xmm0, ExternalAddress((address) &one), rscratch1);
340 break;
341 default:
342 ShouldNotReachHere();
343 break;
344 }
345 } else {
346 #ifdef _LP64
347 ShouldNotReachHere();
348 #else
349 if (value == 0) { __ fldz();
350 } else if (value == 1) { __ fld1();
351 } else { ShouldNotReachHere();
352 }
353 #endif
354 }
355 }
356
357 void TemplateTable::bipush() {
358 transition(vtos, itos);
359 __ load_signed_byte(rax, at_bcp(1));
360 }
361
362 void TemplateTable::sipush() {
363 transition(vtos, itos);
364 __ load_unsigned_short(rax, at_bcp(1));
365 __ bswapl(rax);
366 __ sarl(rax, 16);
367 }
368
369 void TemplateTable::ldc(LdcType type) {
370 transition(vtos, vtos);
371 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
372 Label call_ldc, notFloat, notClass, notInt, Done;
373
374 if (is_ldc_wide(type)) {
375 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
376 } else {
377 __ load_unsigned_byte(rbx, at_bcp(1));
378 }
379
380 __ get_cpool_and_tags(rcx, rax);
381 const int base_offset = ConstantPool::header_size() * wordSize;
382 const int tags_offset = Array<u1>::base_offset_in_bytes();
383
384 // get type
385 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
386
387 // unresolved class - get the resolved class
388 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
389 __ jccb(Assembler::equal, call_ldc);
390
391 // unresolved class in error state - call into runtime to throw the error
392 // from the first resolution attempt
393 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
394 __ jccb(Assembler::equal, call_ldc);
395
396 // resolved class - need to call vm to get java mirror of the class
397 __ cmpl(rdx, JVM_CONSTANT_Class);
398 __ jcc(Assembler::notEqual, notClass);
399
400 __ bind(call_ldc);
401
402 __ movl(rarg, is_ldc_wide(type) ? 1 : 0);
403 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
404
405 __ push(atos);
406 __ jmp(Done);
407
408 __ bind(notClass);
409 __ cmpl(rdx, JVM_CONSTANT_Float);
410 __ jccb(Assembler::notEqual, notFloat);
411
412 // ftos
413 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
414 __ push(ftos);
415 __ jmp(Done);
416
417 __ bind(notFloat);
418 __ cmpl(rdx, JVM_CONSTANT_Integer);
419 __ jccb(Assembler::notEqual, notInt);
420
421 // itos
422 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
423 __ push(itos);
424 __ jmp(Done);
425
426 // assume the tag is for condy; if not, the VM runtime will tell us
427 __ bind(notInt);
428 condy_helper(Done);
429
430 __ bind(Done);
431 }
432
433 // Fast path for caching oop constants.
434 void TemplateTable::fast_aldc(LdcType type) {
435 transition(vtos, atos);
436
437 Register result = rax;
438 Register tmp = rdx;
439 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
440 int index_size = is_ldc_wide(type) ? sizeof(u2) : sizeof(u1);
441
442 Label resolved;
443
444 // We are resolved if the resolved reference cache entry contains a
445 // non-null object (String, MethodType, etc.)
446 assert_different_registers(result, tmp);
447 __ get_cache_index_at_bcp(tmp, 1, index_size);
448 __ load_resolved_reference_at_index(result, tmp);
449 __ testptr(result, result);
450 __ jcc(Assembler::notZero, resolved);
451
452 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
453
454 // first time invocation - must resolve first
455 __ movl(rarg, (int)bytecode());
456 __ call_VM(result, entry, rarg);
457 __ bind(resolved);
458
459 { // Check for the null sentinel.
460 // If we just called the VM, it already did the mapping for us,
461 // but it's harmless to retry.
462 Label notNull;
463 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
464 __ movptr(tmp, null_sentinel);
465 __ resolve_oop_handle(tmp, rscratch2);
466 __ cmpoop(tmp, result);
467 __ jccb(Assembler::notEqual, notNull);
468 __ xorptr(result, result); // null object reference
469 __ bind(notNull);
470 }
471
472 if (VerifyOops) {
473 __ verify_oop(result);
474 }
475 }
476
477 void TemplateTable::ldc2_w() {
478 transition(vtos, vtos);
479 Label notDouble, notLong, Done;
480 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
481
482 __ get_cpool_and_tags(rcx, rax);
483 const int base_offset = ConstantPool::header_size() * wordSize;
484 const int tags_offset = Array<u1>::base_offset_in_bytes();
485
486 // get type
487 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
488 __ cmpl(rdx, JVM_CONSTANT_Double);
489 __ jccb(Assembler::notEqual, notDouble);
490
491 // dtos
492 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
493 __ push(dtos);
494
495 __ jmp(Done);
496 __ bind(notDouble);
497 __ cmpl(rdx, JVM_CONSTANT_Long);
498 __ jccb(Assembler::notEqual, notLong);
499
500 // ltos
501 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
502 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
503 __ push(ltos);
504 __ jmp(Done);
505
506 __ bind(notLong);
507 condy_helper(Done);
508
509 __ bind(Done);
510 }
511
512 void TemplateTable::condy_helper(Label& Done) {
513 const Register obj = rax;
514 const Register off = rbx;
515 const Register flags = rcx;
516 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
517 __ movl(rarg, (int)bytecode());
518 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
519 #ifndef _LP64
520 // borrow rdi from locals
521 __ get_thread(rdi);
522 __ get_vm_result_2(flags, rdi);
523 __ restore_locals();
524 #else
525 __ get_vm_result_2(flags, r15_thread);
526 #endif
527 // VMr = obj = base address to find primitive value to push
528 // VMr2 = flags = (tos, off) using format of CPCE::_flags
529 __ movl(off, flags);
530 __ andl(off, ConstantPoolCache::field_index_mask);
531 const Address field(obj, off, Address::times_1, 0*wordSize);
532
533 // What sort of thing are we loading?
534 __ shrl(flags, ConstantPoolCache::tos_state_shift);
535 __ andl(flags, ConstantPoolCache::tos_state_mask);
536
537 switch (bytecode()) {
538 case Bytecodes::_ldc:
539 case Bytecodes::_ldc_w:
540 {
541 // tos in (itos, ftos, stos, btos, ctos, ztos)
542 Label notInt, notFloat, notShort, notByte, notChar, notBool;
543 __ cmpl(flags, itos);
544 __ jccb(Assembler::notEqual, notInt);
545 // itos
546 __ movl(rax, field);
547 __ push(itos);
548 __ jmp(Done);
549
550 __ bind(notInt);
551 __ cmpl(flags, ftos);
552 __ jccb(Assembler::notEqual, notFloat);
553 // ftos
554 __ load_float(field);
555 __ push(ftos);
556 __ jmp(Done);
557
558 __ bind(notFloat);
559 __ cmpl(flags, stos);
560 __ jccb(Assembler::notEqual, notShort);
561 // stos
562 __ load_signed_short(rax, field);
563 __ push(stos);
564 __ jmp(Done);
565
566 __ bind(notShort);
567 __ cmpl(flags, btos);
568 __ jccb(Assembler::notEqual, notByte);
569 // btos
570 __ load_signed_byte(rax, field);
571 __ push(btos);
572 __ jmp(Done);
573
574 __ bind(notByte);
575 __ cmpl(flags, ctos);
576 __ jccb(Assembler::notEqual, notChar);
577 // ctos
578 __ load_unsigned_short(rax, field);
579 __ push(ctos);
580 __ jmp(Done);
581
582 __ bind(notChar);
583 __ cmpl(flags, ztos);
584 __ jccb(Assembler::notEqual, notBool);
585 // ztos
586 __ load_signed_byte(rax, field);
587 __ push(ztos);
588 __ jmp(Done);
589
590 __ bind(notBool);
591 break;
592 }
593
594 case Bytecodes::_ldc2_w:
595 {
596 Label notLong, notDouble;
597 __ cmpl(flags, ltos);
598 __ jccb(Assembler::notEqual, notLong);
599 // ltos
600 // Loading high word first because movptr clobbers rax
601 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
602 __ movptr(rax, field);
603 __ push(ltos);
604 __ jmp(Done);
605
606 __ bind(notLong);
607 __ cmpl(flags, dtos);
608 __ jccb(Assembler::notEqual, notDouble);
609 // dtos
610 __ load_double(field);
611 __ push(dtos);
612 __ jmp(Done);
613
614 __ bind(notDouble);
615 break;
616 }
617
618 default:
619 ShouldNotReachHere();
620 }
621
622 __ stop("bad ldc/condy");
623 }
624
625 void TemplateTable::locals_index(Register reg, int offset) {
626 __ load_unsigned_byte(reg, at_bcp(offset));
627 __ negptr(reg);
628 }
629
630 void TemplateTable::iload() {
631 iload_internal();
632 }
633
634 void TemplateTable::nofast_iload() {
635 iload_internal(may_not_rewrite);
636 }
637
638 void TemplateTable::iload_internal(RewriteControl rc) {
639 transition(vtos, itos);
640 if (RewriteFrequentPairs && rc == may_rewrite) {
641 Label rewrite, done;
642 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
643 LP64_ONLY(assert(rbx != bc, "register damaged"));
644
645 // get next byte
646 __ load_unsigned_byte(rbx,
647 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
648 // if _iload, wait to rewrite to iload2. We only want to rewrite the
649 // last two iloads in a pair. Comparing against fast_iload means that
650 // the next bytecode is neither an iload or a caload, and therefore
651 // an iload pair.
652 __ cmpl(rbx, Bytecodes::_iload);
653 __ jcc(Assembler::equal, done);
654
655 __ cmpl(rbx, Bytecodes::_fast_iload);
656 __ movl(bc, Bytecodes::_fast_iload2);
657
658 __ jccb(Assembler::equal, rewrite);
659
660 // if _caload, rewrite to fast_icaload
661 __ cmpl(rbx, Bytecodes::_caload);
662 __ movl(bc, Bytecodes::_fast_icaload);
663 __ jccb(Assembler::equal, rewrite);
664
665 // rewrite so iload doesn't check again.
666 __ movl(bc, Bytecodes::_fast_iload);
667
668 // rewrite
669 // bc: fast bytecode
670 __ bind(rewrite);
671 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
672 __ bind(done);
673 }
674
675 // Get the local value into tos
676 locals_index(rbx);
677 __ movl(rax, iaddress(rbx));
678 }
679
680 void TemplateTable::fast_iload2() {
681 transition(vtos, itos);
682 locals_index(rbx);
683 __ movl(rax, iaddress(rbx));
684 __ push(itos);
685 locals_index(rbx, 3);
686 __ movl(rax, iaddress(rbx));
687 }
688
689 void TemplateTable::fast_iload() {
690 transition(vtos, itos);
691 locals_index(rbx);
692 __ movl(rax, iaddress(rbx));
693 }
694
695 void TemplateTable::lload() {
696 transition(vtos, ltos);
697 locals_index(rbx);
698 __ movptr(rax, laddress(rbx));
699 NOT_LP64(__ movl(rdx, haddress(rbx)));
700 }
701
702 void TemplateTable::fload() {
703 transition(vtos, ftos);
704 locals_index(rbx);
705 __ load_float(faddress(rbx));
706 }
707
708 void TemplateTable::dload() {
709 transition(vtos, dtos);
710 locals_index(rbx);
711 __ load_double(daddress(rbx));
712 }
713
714 void TemplateTable::aload() {
715 transition(vtos, atos);
716 locals_index(rbx);
717 __ movptr(rax, aaddress(rbx));
718 }
719
720 void TemplateTable::locals_index_wide(Register reg) {
721 __ load_unsigned_short(reg, at_bcp(2));
722 __ bswapl(reg);
723 __ shrl(reg, 16);
724 __ negptr(reg);
725 }
726
727 void TemplateTable::wide_iload() {
728 transition(vtos, itos);
729 locals_index_wide(rbx);
730 __ movl(rax, iaddress(rbx));
731 }
732
733 void TemplateTable::wide_lload() {
734 transition(vtos, ltos);
735 locals_index_wide(rbx);
736 __ movptr(rax, laddress(rbx));
737 NOT_LP64(__ movl(rdx, haddress(rbx)));
738 }
739
740 void TemplateTable::wide_fload() {
741 transition(vtos, ftos);
742 locals_index_wide(rbx);
743 __ load_float(faddress(rbx));
744 }
745
746 void TemplateTable::wide_dload() {
747 transition(vtos, dtos);
748 locals_index_wide(rbx);
749 __ load_double(daddress(rbx));
750 }
751
752 void TemplateTable::wide_aload() {
753 transition(vtos, atos);
754 locals_index_wide(rbx);
755 __ movptr(rax, aaddress(rbx));
756 }
757
758 void TemplateTable::index_check(Register array, Register index) {
759 // Pop ptr into array
760 __ pop_ptr(array);
761 index_check_without_pop(array, index);
762 }
763
764 void TemplateTable::index_check_without_pop(Register array, Register index) {
765 // destroys rbx
766 // sign extend index for use by indexed load
767 __ movl2ptr(index, index);
768 // check index
769 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
770 if (index != rbx) {
771 // ??? convention: move aberrant index into rbx for exception message
772 assert(rbx != array, "different registers");
773 __ movl(rbx, index);
774 }
775 Label skip;
776 __ jccb(Assembler::below, skip);
777 // Pass array to create more detailed exceptions.
778 __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
779 __ jump(RuntimeAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
780 __ bind(skip);
781 }
782
783 void TemplateTable::iaload() {
784 transition(itos, itos);
785 // rax: index
786 // rdx: array
787 index_check(rdx, rax); // kills rbx
788 __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
789 Address(rdx, rax, Address::times_4,
790 arrayOopDesc::base_offset_in_bytes(T_INT)),
791 noreg, noreg);
792 }
793
794 void TemplateTable::laload() {
795 transition(itos, ltos);
796 // rax: index
797 // rdx: array
798 index_check(rdx, rax); // kills rbx
799 NOT_LP64(__ mov(rbx, rax));
800 // rbx,: index
801 __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
802 Address(rdx, rbx, Address::times_8,
803 arrayOopDesc::base_offset_in_bytes(T_LONG)),
804 noreg, noreg);
805 }
806
807
808
809 void TemplateTable::faload() {
810 transition(itos, ftos);
811 // rax: index
812 // rdx: array
813 index_check(rdx, rax); // kills rbx
814 __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
815 Address(rdx, rax,
816 Address::times_4,
817 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
818 noreg, noreg);
819 }
820
821 void TemplateTable::daload() {
822 transition(itos, dtos);
823 // rax: index
824 // rdx: array
825 index_check(rdx, rax); // kills rbx
826 __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
827 Address(rdx, rax,
828 Address::times_8,
829 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
830 noreg, noreg);
831 }
832
833 void TemplateTable::aaload() {
834 transition(itos, atos);
835 Register array = rdx;
836 Register index = rax;
837
838 index_check(array, index); // kills rbx
839 __ profile_array_type<ArrayLoadData>(rbx, array, rcx);
840 if (UseFlatArray) {
841 Label is_flat_array, done;
842 __ test_flat_array_oop(array, rbx, is_flat_array);
843 do_oop_load(_masm,
844 Address(array, index,
845 UseCompressedOops ? Address::times_4 : Address::times_ptr,
846 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
847 rax,
848 IS_ARRAY);
849 __ jmp(done);
850 __ bind(is_flat_array);
851 __ movptr(rcx, array);
852 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::flat_array_load), rcx, index);
853 __ bind(done);
854 } else {
855 do_oop_load(_masm,
856 Address(array, index,
857 UseCompressedOops ? Address::times_4 : Address::times_ptr,
858 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
859 rax,
860 IS_ARRAY);
861 }
862 __ profile_element_type(rbx, rax, rcx);
863 }
864
865 void TemplateTable::baload() {
866 transition(itos, itos);
867 // rax: index
868 // rdx: array
869 index_check(rdx, rax); // kills rbx
870 __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
871 Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
872 noreg, noreg);
873 }
874
875 void TemplateTable::caload() {
876 transition(itos, itos);
877 // rax: index
878 // rdx: array
879 index_check(rdx, rax); // kills rbx
880 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
881 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
882 noreg, noreg);
883 }
884
885 // iload followed by caload frequent pair
886 void TemplateTable::fast_icaload() {
887 transition(vtos, itos);
888 // load index out of locals
889 locals_index(rbx);
890 __ movl(rax, iaddress(rbx));
891
892 // rax: index
893 // rdx: array
894 index_check(rdx, rax); // kills rbx
895 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
896 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
897 noreg, noreg);
898 }
899
900
901 void TemplateTable::saload() {
902 transition(itos, itos);
903 // rax: index
904 // rdx: array
905 index_check(rdx, rax); // kills rbx
906 __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
907 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
908 noreg, noreg);
909 }
910
911 void TemplateTable::iload(int n) {
912 transition(vtos, itos);
913 __ movl(rax, iaddress(n));
914 }
915
916 void TemplateTable::lload(int n) {
917 transition(vtos, ltos);
918 __ movptr(rax, laddress(n));
919 NOT_LP64(__ movptr(rdx, haddress(n)));
920 }
921
922 void TemplateTable::fload(int n) {
923 transition(vtos, ftos);
924 __ load_float(faddress(n));
925 }
926
927 void TemplateTable::dload(int n) {
928 transition(vtos, dtos);
929 __ load_double(daddress(n));
930 }
931
932 void TemplateTable::aload(int n) {
933 transition(vtos, atos);
934 __ movptr(rax, aaddress(n));
935 }
936
937 void TemplateTable::aload_0() {
938 aload_0_internal();
939 }
940
941 void TemplateTable::nofast_aload_0() {
942 aload_0_internal(may_not_rewrite);
943 }
944
945 void TemplateTable::aload_0_internal(RewriteControl rc) {
946 transition(vtos, atos);
947 // According to bytecode histograms, the pairs:
948 //
949 // _aload_0, _fast_igetfield
950 // _aload_0, _fast_agetfield
951 // _aload_0, _fast_fgetfield
952 //
953 // occur frequently. If RewriteFrequentPairs is set, the (slow)
954 // _aload_0 bytecode checks if the next bytecode is either
955 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
956 // rewrites the current bytecode into a pair bytecode; otherwise it
957 // rewrites the current bytecode into _fast_aload_0 that doesn't do
958 // the pair check anymore.
959 //
960 // Note: If the next bytecode is _getfield, the rewrite must be
961 // delayed, otherwise we may miss an opportunity for a pair.
962 //
963 // Also rewrite frequent pairs
964 // aload_0, aload_1
965 // aload_0, iload_1
966 // These bytecodes with a small amount of code are most profitable
967 // to rewrite
968 if (RewriteFrequentPairs && rc == may_rewrite) {
969 Label rewrite, done;
970
971 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
972 LP64_ONLY(assert(rbx != bc, "register damaged"));
973
974 // get next byte
975 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
976
977 // if _getfield then wait with rewrite
978 __ cmpl(rbx, Bytecodes::_getfield);
979 __ jcc(Assembler::equal, done);
980
981 // if _igetfield then rewrite to _fast_iaccess_0
982 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
983 __ cmpl(rbx, Bytecodes::_fast_igetfield);
984 __ movl(bc, Bytecodes::_fast_iaccess_0);
985 __ jccb(Assembler::equal, rewrite);
986
987 // if _agetfield then rewrite to _fast_aaccess_0
988 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
989 __ cmpl(rbx, Bytecodes::_fast_agetfield);
990 __ movl(bc, Bytecodes::_fast_aaccess_0);
991 __ jccb(Assembler::equal, rewrite);
992
993 // if _fgetfield then rewrite to _fast_faccess_0
994 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
995 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
996 __ movl(bc, Bytecodes::_fast_faccess_0);
997 __ jccb(Assembler::equal, rewrite);
998
999 // else rewrite to _fast_aload0
1000 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
1001 __ movl(bc, Bytecodes::_fast_aload_0);
1002
1003 // rewrite
1004 // bc: fast bytecode
1005 __ bind(rewrite);
1006 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
1007
1008 __ bind(done);
1009 }
1010
1011 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
1012 aload(0);
1013 }
1014
1015 void TemplateTable::istore() {
1016 transition(itos, vtos);
1017 locals_index(rbx);
1018 __ movl(iaddress(rbx), rax);
1019 }
1020
1021
1022 void TemplateTable::lstore() {
1023 transition(ltos, vtos);
1024 locals_index(rbx);
1025 __ movptr(laddress(rbx), rax);
1026 NOT_LP64(__ movptr(haddress(rbx), rdx));
1027 }
1028
1029 void TemplateTable::fstore() {
1030 transition(ftos, vtos);
1031 locals_index(rbx);
1032 __ store_float(faddress(rbx));
1033 }
1034
1035 void TemplateTable::dstore() {
1036 transition(dtos, vtos);
1037 locals_index(rbx);
1038 __ store_double(daddress(rbx));
1039 }
1040
1041 void TemplateTable::astore() {
1042 transition(vtos, vtos);
1043 __ pop_ptr(rax);
1044 locals_index(rbx);
1045 __ movptr(aaddress(rbx), rax);
1046 }
1047
1048 void TemplateTable::wide_istore() {
1049 transition(vtos, vtos);
1050 __ pop_i();
1051 locals_index_wide(rbx);
1052 __ movl(iaddress(rbx), rax);
1053 }
1054
1055 void TemplateTable::wide_lstore() {
1056 transition(vtos, vtos);
1057 NOT_LP64(__ pop_l(rax, rdx));
1058 LP64_ONLY(__ pop_l());
1059 locals_index_wide(rbx);
1060 __ movptr(laddress(rbx), rax);
1061 NOT_LP64(__ movl(haddress(rbx), rdx));
1062 }
1063
1064 void TemplateTable::wide_fstore() {
1065 #ifdef _LP64
1066 transition(vtos, vtos);
1067 __ pop_f(xmm0);
1068 locals_index_wide(rbx);
1069 __ movflt(faddress(rbx), xmm0);
1070 #else
1071 wide_istore();
1072 #endif
1073 }
1074
1075 void TemplateTable::wide_dstore() {
1076 #ifdef _LP64
1077 transition(vtos, vtos);
1078 __ pop_d(xmm0);
1079 locals_index_wide(rbx);
1080 __ movdbl(daddress(rbx), xmm0);
1081 #else
1082 wide_lstore();
1083 #endif
1084 }
1085
1086 void TemplateTable::wide_astore() {
1087 transition(vtos, vtos);
1088 __ pop_ptr(rax);
1089 locals_index_wide(rbx);
1090 __ movptr(aaddress(rbx), rax);
1091 }
1092
1093 void TemplateTable::iastore() {
1094 transition(itos, vtos);
1095 __ pop_i(rbx);
1096 // rax: value
1097 // rbx: index
1098 // rdx: array
1099 index_check(rdx, rbx); // prefer index in rbx
1100 __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1101 Address(rdx, rbx, Address::times_4,
1102 arrayOopDesc::base_offset_in_bytes(T_INT)),
1103 rax, noreg, noreg, noreg);
1104 }
1105
1106 void TemplateTable::lastore() {
1107 transition(ltos, vtos);
1108 __ pop_i(rbx);
1109 // rax,: low(value)
1110 // rcx: array
1111 // rdx: high(value)
1112 index_check(rcx, rbx); // prefer index in rbx,
1113 // rbx,: index
1114 __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1115 Address(rcx, rbx, Address::times_8,
1116 arrayOopDesc::base_offset_in_bytes(T_LONG)),
1117 noreg /* ltos */, noreg, noreg, noreg);
1118 }
1119
1120
1121 void TemplateTable::fastore() {
1122 transition(ftos, vtos);
1123 __ pop_i(rbx);
1124 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1125 // rbx: index
1126 // rdx: array
1127 index_check(rdx, rbx); // prefer index in rbx
1128 __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1129 Address(rdx, rbx, Address::times_4,
1130 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1131 noreg /* ftos */, noreg, noreg, noreg);
1132 }
1133
1134 void TemplateTable::dastore() {
1135 transition(dtos, vtos);
1136 __ pop_i(rbx);
1137 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1138 // rbx: index
1139 // rdx: array
1140 index_check(rdx, rbx); // prefer index in rbx
1141 __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1142 Address(rdx, rbx, Address::times_8,
1143 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1144 noreg /* dtos */, noreg, noreg, noreg);
1145 }
1146
1147 void TemplateTable::aastore() {
1148 Label is_null, is_flat_array, ok_is_subtype, done;
1149 transition(vtos, vtos);
1150 // stack: ..., array, index, value
1151 __ movptr(rax, at_tos()); // value
1152 __ movl(rcx, at_tos_p1()); // index
1153 __ movptr(rdx, at_tos_p2()); // array
1154
1155 Address element_address(rdx, rcx,
1156 UseCompressedOops? Address::times_4 : Address::times_ptr,
1157 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1158
1159 index_check_without_pop(rdx, rcx); // kills rbx
1160
1161 __ profile_array_type<ArrayStoreData>(rdi, rdx, rbx);
1162 __ profile_multiple_element_types(rdi, rax, rbx, rcx);
1163
1164 __ testptr(rax, rax);
1165 __ jcc(Assembler::zero, is_null);
1166
1167 // Move array class to rdi
1168 __ load_klass(rdi, rdx, rscratch1);
1169 if (UseFlatArray) {
1170 __ movl(rbx, Address(rdi, Klass::layout_helper_offset()));
1171 __ test_flat_array_layout(rbx, is_flat_array);
1172 }
1173
1174 // Move subklass into rbx
1175 __ load_klass(rbx, rax, rscratch1);
1176 // Move array element superklass into rax
1177 __ movptr(rax, Address(rdi,
1178 ObjArrayKlass::element_klass_offset()));
1179
1180 // Generate subtype check. Blows rcx, rdi
1181 // Superklass in rax. Subklass in rbx.
1182 // is "rbx <: rax" ? (value subclass <: array element superclass)
1183 __ gen_subtype_check(rbx, ok_is_subtype, false);
1184
1185 // Come here on failure
1186 // object is at TOS
1187 __ jump(RuntimeAddress(Interpreter::_throw_ArrayStoreException_entry));
1188
1189 // Come here on success
1190 __ bind(ok_is_subtype);
1191
1192 // Get the value we will store
1193 __ movptr(rax, at_tos());
1194 __ movl(rcx, at_tos_p1()); // index
1195 // Now store using the appropriate barrier
1196 do_oop_store(_masm, element_address, rax, IS_ARRAY);
1197 __ jmp(done);
1198
1199 // Have a null in rax, rdx=array, ecx=index. Store null at ary[idx]
1200 __ bind(is_null);
1201 if (EnableValhalla) {
1202 Label is_null_into_value_array_npe, store_null;
1203
1204 // Move array class to rdi
1205 __ load_klass(rdi, rdx, rscratch1);
1206 if (UseFlatArray) {
1207 __ movl(rbx, Address(rdi, Klass::layout_helper_offset()));
1208 __ test_flat_array_layout(rbx, is_flat_array);
1209 }
1210
1211 // No way to store null in null-free array
1212 __ test_null_free_array_oop(rdx, rbx, is_null_into_value_array_npe);
1213 __ jmp(store_null);
1214
1215 __ bind(is_null_into_value_array_npe);
1216 __ jump(RuntimeAddress(Interpreter::_throw_NullPointerException_entry));
1217
1218 __ bind(store_null);
1219 }
1220 // Store a null
1221 do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1222 __ jmp(done);
1223
1224 if (UseFlatArray) {
1225 Label is_type_ok;
1226 __ bind(is_flat_array); // Store non-null value to flat
1227
1228 __ movptr(rax, at_tos());
1229 __ movl(rcx, at_tos_p1()); // index
1230 __ movptr(rdx, at_tos_p2()); // array
1231
1232 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::flat_array_store), rax, rdx, rcx);
1233 }
1234 // Pop stack arguments
1235 __ bind(done);
1236 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1237 }
1238
1239 void TemplateTable::bastore() {
1240 transition(itos, vtos);
1241 __ pop_i(rbx);
1242 // rax: value
1243 // rbx: index
1244 // rdx: array
1245 index_check(rdx, rbx); // prefer index in rbx
1246 // Need to check whether array is boolean or byte
1247 // since both types share the bastore bytecode.
1248 __ load_klass(rcx, rdx, rscratch1);
1249 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1250 int diffbit = Klass::layout_helper_boolean_diffbit();
1251 __ testl(rcx, diffbit);
1252 Label L_skip;
1253 __ jccb(Assembler::zero, L_skip);
1254 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1255 __ bind(L_skip);
1256 __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1257 Address(rdx, rbx,Address::times_1,
1258 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1259 rax, noreg, noreg, noreg);
1260 }
1261
1262 void TemplateTable::castore() {
1263 transition(itos, vtos);
1264 __ pop_i(rbx);
1265 // rax: value
1266 // rbx: index
1267 // rdx: array
1268 index_check(rdx, rbx); // prefer index in rbx
1269 __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1270 Address(rdx, rbx, Address::times_2,
1271 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1272 rax, noreg, noreg, noreg);
1273 }
1274
1275
1276 void TemplateTable::sastore() {
1277 castore();
1278 }
1279
1280 void TemplateTable::istore(int n) {
1281 transition(itos, vtos);
1282 __ movl(iaddress(n), rax);
1283 }
1284
1285 void TemplateTable::lstore(int n) {
1286 transition(ltos, vtos);
1287 __ movptr(laddress(n), rax);
1288 NOT_LP64(__ movptr(haddress(n), rdx));
1289 }
1290
1291 void TemplateTable::fstore(int n) {
1292 transition(ftos, vtos);
1293 __ store_float(faddress(n));
1294 }
1295
1296 void TemplateTable::dstore(int n) {
1297 transition(dtos, vtos);
1298 __ store_double(daddress(n));
1299 }
1300
1301
1302 void TemplateTable::astore(int n) {
1303 transition(vtos, vtos);
1304 __ pop_ptr(rax);
1305 __ movptr(aaddress(n), rax);
1306 }
1307
1308 void TemplateTable::pop() {
1309 transition(vtos, vtos);
1310 __ addptr(rsp, Interpreter::stackElementSize);
1311 }
1312
1313 void TemplateTable::pop2() {
1314 transition(vtos, vtos);
1315 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1316 }
1317
1318
1319 void TemplateTable::dup() {
1320 transition(vtos, vtos);
1321 __ load_ptr(0, rax);
1322 __ push_ptr(rax);
1323 // stack: ..., a, a
1324 }
1325
1326 void TemplateTable::dup_x1() {
1327 transition(vtos, vtos);
1328 // stack: ..., a, b
1329 __ load_ptr( 0, rax); // load b
1330 __ load_ptr( 1, rcx); // load a
1331 __ store_ptr(1, rax); // store b
1332 __ store_ptr(0, rcx); // store a
1333 __ push_ptr(rax); // push b
1334 // stack: ..., b, a, b
1335 }
1336
1337 void TemplateTable::dup_x2() {
1338 transition(vtos, vtos);
1339 // stack: ..., a, b, c
1340 __ load_ptr( 0, rax); // load c
1341 __ load_ptr( 2, rcx); // load a
1342 __ store_ptr(2, rax); // store c in a
1343 __ push_ptr(rax); // push c
1344 // stack: ..., c, b, c, c
1345 __ load_ptr( 2, rax); // load b
1346 __ store_ptr(2, rcx); // store a in b
1347 // stack: ..., c, a, c, c
1348 __ store_ptr(1, rax); // store b in c
1349 // stack: ..., c, a, b, c
1350 }
1351
1352 void TemplateTable::dup2() {
1353 transition(vtos, vtos);
1354 // stack: ..., a, b
1355 __ load_ptr(1, rax); // load a
1356 __ push_ptr(rax); // push a
1357 __ load_ptr(1, rax); // load b
1358 __ push_ptr(rax); // push b
1359 // stack: ..., a, b, a, b
1360 }
1361
1362
1363 void TemplateTable::dup2_x1() {
1364 transition(vtos, vtos);
1365 // stack: ..., a, b, c
1366 __ load_ptr( 0, rcx); // load c
1367 __ load_ptr( 1, rax); // load b
1368 __ push_ptr(rax); // push b
1369 __ push_ptr(rcx); // push c
1370 // stack: ..., a, b, c, b, c
1371 __ store_ptr(3, rcx); // store c in b
1372 // stack: ..., a, c, c, b, c
1373 __ load_ptr( 4, rcx); // load a
1374 __ store_ptr(2, rcx); // store a in 2nd c
1375 // stack: ..., a, c, a, b, c
1376 __ store_ptr(4, rax); // store b in a
1377 // stack: ..., b, c, a, b, c
1378 }
1379
1380 void TemplateTable::dup2_x2() {
1381 transition(vtos, vtos);
1382 // stack: ..., a, b, c, d
1383 __ load_ptr( 0, rcx); // load d
1384 __ load_ptr( 1, rax); // load c
1385 __ push_ptr(rax); // push c
1386 __ push_ptr(rcx); // push d
1387 // stack: ..., a, b, c, d, c, d
1388 __ load_ptr( 4, rax); // load b
1389 __ store_ptr(2, rax); // store b in d
1390 __ store_ptr(4, rcx); // store d in b
1391 // stack: ..., a, d, c, b, c, d
1392 __ load_ptr( 5, rcx); // load a
1393 __ load_ptr( 3, rax); // load c
1394 __ store_ptr(3, rcx); // store a in c
1395 __ store_ptr(5, rax); // store c in a
1396 // stack: ..., c, d, a, b, c, d
1397 }
1398
1399 void TemplateTable::swap() {
1400 transition(vtos, vtos);
1401 // stack: ..., a, b
1402 __ load_ptr( 1, rcx); // load a
1403 __ load_ptr( 0, rax); // load b
1404 __ store_ptr(0, rcx); // store a in b
1405 __ store_ptr(1, rax); // store b in a
1406 // stack: ..., b, a
1407 }
1408
1409 void TemplateTable::iop2(Operation op) {
1410 transition(itos, itos);
1411 switch (op) {
1412 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1413 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1414 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1415 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1416 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1417 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1418 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1419 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1420 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1421 default : ShouldNotReachHere();
1422 }
1423 }
1424
1425 void TemplateTable::lop2(Operation op) {
1426 transition(ltos, ltos);
1427 #ifdef _LP64
1428 switch (op) {
1429 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1430 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1431 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1432 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1433 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1434 default : ShouldNotReachHere();
1435 }
1436 #else
1437 __ pop_l(rbx, rcx);
1438 switch (op) {
1439 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1440 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1441 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1442 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1443 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1444 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1445 default : ShouldNotReachHere();
1446 }
1447 #endif
1448 }
1449
1450 void TemplateTable::idiv() {
1451 transition(itos, itos);
1452 __ movl(rcx, rax);
1453 __ pop_i(rax);
1454 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1455 // they are not equal, one could do a normal division (no correction
1456 // needed), which may speed up this implementation for the common case.
1457 // (see also JVM spec., p.243 & p.271)
1458 __ corrected_idivl(rcx);
1459 }
1460
1461 void TemplateTable::irem() {
1462 transition(itos, itos);
1463 __ movl(rcx, rax);
1464 __ pop_i(rax);
1465 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1466 // they are not equal, one could do a normal division (no correction
1467 // needed), which may speed up this implementation for the common case.
1468 // (see also JVM spec., p.243 & p.271)
1469 __ corrected_idivl(rcx);
1470 __ movl(rax, rdx);
1471 }
1472
1473 void TemplateTable::lmul() {
1474 transition(ltos, ltos);
1475 #ifdef _LP64
1476 __ pop_l(rdx);
1477 __ imulq(rax, rdx);
1478 #else
1479 __ pop_l(rbx, rcx);
1480 __ push(rcx); __ push(rbx);
1481 __ push(rdx); __ push(rax);
1482 __ lmul(2 * wordSize, 0);
1483 __ addptr(rsp, 4 * wordSize); // take off temporaries
1484 #endif
1485 }
1486
1487 void TemplateTable::ldiv() {
1488 transition(ltos, ltos);
1489 #ifdef _LP64
1490 __ mov(rcx, rax);
1491 __ pop_l(rax);
1492 // generate explicit div0 check
1493 __ testq(rcx, rcx);
1494 __ jump_cc(Assembler::zero,
1495 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1496 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1497 // they are not equal, one could do a normal division (no correction
1498 // needed), which may speed up this implementation for the common case.
1499 // (see also JVM spec., p.243 & p.271)
1500 __ corrected_idivq(rcx); // kills rbx
1501 #else
1502 __ pop_l(rbx, rcx);
1503 __ push(rcx); __ push(rbx);
1504 __ push(rdx); __ push(rax);
1505 // check if y = 0
1506 __ orl(rax, rdx);
1507 __ jump_cc(Assembler::zero,
1508 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1509 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1510 __ addptr(rsp, 4 * wordSize); // take off temporaries
1511 #endif
1512 }
1513
1514 void TemplateTable::lrem() {
1515 transition(ltos, ltos);
1516 #ifdef _LP64
1517 __ mov(rcx, rax);
1518 __ pop_l(rax);
1519 __ testq(rcx, rcx);
1520 __ jump_cc(Assembler::zero,
1521 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1522 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1523 // they are not equal, one could do a normal division (no correction
1524 // needed), which may speed up this implementation for the common case.
1525 // (see also JVM spec., p.243 & p.271)
1526 __ corrected_idivq(rcx); // kills rbx
1527 __ mov(rax, rdx);
1528 #else
1529 __ pop_l(rbx, rcx);
1530 __ push(rcx); __ push(rbx);
1531 __ push(rdx); __ push(rax);
1532 // check if y = 0
1533 __ orl(rax, rdx);
1534 __ jump_cc(Assembler::zero,
1535 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1536 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1537 __ addptr(rsp, 4 * wordSize);
1538 #endif
1539 }
1540
1541 void TemplateTable::lshl() {
1542 transition(itos, ltos);
1543 __ movl(rcx, rax); // get shift count
1544 #ifdef _LP64
1545 __ pop_l(rax); // get shift value
1546 __ shlq(rax);
1547 #else
1548 __ pop_l(rax, rdx); // get shift value
1549 __ lshl(rdx, rax);
1550 #endif
1551 }
1552
1553 void TemplateTable::lshr() {
1554 #ifdef _LP64
1555 transition(itos, ltos);
1556 __ movl(rcx, rax); // get shift count
1557 __ pop_l(rax); // get shift value
1558 __ sarq(rax);
1559 #else
1560 transition(itos, ltos);
1561 __ mov(rcx, rax); // get shift count
1562 __ pop_l(rax, rdx); // get shift value
1563 __ lshr(rdx, rax, true);
1564 #endif
1565 }
1566
1567 void TemplateTable::lushr() {
1568 transition(itos, ltos);
1569 #ifdef _LP64
1570 __ movl(rcx, rax); // get shift count
1571 __ pop_l(rax); // get shift value
1572 __ shrq(rax);
1573 #else
1574 __ mov(rcx, rax); // get shift count
1575 __ pop_l(rax, rdx); // get shift value
1576 __ lshr(rdx, rax);
1577 #endif
1578 }
1579
1580 void TemplateTable::fop2(Operation op) {
1581 transition(ftos, ftos);
1582
1583 if (UseSSE >= 1) {
1584 switch (op) {
1585 case add:
1586 __ addss(xmm0, at_rsp());
1587 __ addptr(rsp, Interpreter::stackElementSize);
1588 break;
1589 case sub:
1590 __ movflt(xmm1, xmm0);
1591 __ pop_f(xmm0);
1592 __ subss(xmm0, xmm1);
1593 break;
1594 case mul:
1595 __ mulss(xmm0, at_rsp());
1596 __ addptr(rsp, Interpreter::stackElementSize);
1597 break;
1598 case div:
1599 __ movflt(xmm1, xmm0);
1600 __ pop_f(xmm0);
1601 __ divss(xmm0, xmm1);
1602 break;
1603 case rem:
1604 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1605 // modulo operation. The frem method calls the function
1606 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1607 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1608 // (signalling or quiet) is returned.
1609 //
1610 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1611 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1612 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1613 // The fprem instruction used on x86_32 is functionally equivalent to
1614 // SharedRuntime::frem in that it returns a NaN.
1615 #ifdef _LP64
1616 __ movflt(xmm1, xmm0);
1617 __ pop_f(xmm0);
1618 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1619 #else // !_LP64
1620 __ push_f(xmm0);
1621 __ pop_f();
1622 __ fld_s(at_rsp());
1623 __ fremr(rax);
1624 __ f2ieee();
1625 __ pop(rax); // pop second operand off the stack
1626 __ push_f();
1627 __ pop_f(xmm0);
1628 #endif // _LP64
1629 break;
1630 default:
1631 ShouldNotReachHere();
1632 break;
1633 }
1634 } else {
1635 #ifdef _LP64
1636 ShouldNotReachHere();
1637 #else // !_LP64
1638 switch (op) {
1639 case add: __ fadd_s (at_rsp()); break;
1640 case sub: __ fsubr_s(at_rsp()); break;
1641 case mul: __ fmul_s (at_rsp()); break;
1642 case div: __ fdivr_s(at_rsp()); break;
1643 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1644 default : ShouldNotReachHere();
1645 }
1646 __ f2ieee();
1647 __ pop(rax); // pop second operand off the stack
1648 #endif // _LP64
1649 }
1650 }
1651
1652 void TemplateTable::dop2(Operation op) {
1653 transition(dtos, dtos);
1654 if (UseSSE >= 2) {
1655 switch (op) {
1656 case add:
1657 __ addsd(xmm0, at_rsp());
1658 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1659 break;
1660 case sub:
1661 __ movdbl(xmm1, xmm0);
1662 __ pop_d(xmm0);
1663 __ subsd(xmm0, xmm1);
1664 break;
1665 case mul:
1666 __ mulsd(xmm0, at_rsp());
1667 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1668 break;
1669 case div:
1670 __ movdbl(xmm1, xmm0);
1671 __ pop_d(xmm0);
1672 __ divsd(xmm0, xmm1);
1673 break;
1674 case rem:
1675 // Similar to fop2(), the modulo operation is performed using the
1676 // SharedRuntime::drem method (on x86_64 platforms) or using the
1677 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1678 #ifdef _LP64
1679 __ movdbl(xmm1, xmm0);
1680 __ pop_d(xmm0);
1681 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1682 #else // !_LP64
1683 __ push_d(xmm0);
1684 __ pop_d();
1685 __ fld_d(at_rsp());
1686 __ fremr(rax);
1687 __ d2ieee();
1688 __ pop(rax);
1689 __ pop(rdx);
1690 __ push_d();
1691 __ pop_d(xmm0);
1692 #endif // _LP64
1693 break;
1694 default:
1695 ShouldNotReachHere();
1696 break;
1697 }
1698 } else {
1699 #ifdef _LP64
1700 ShouldNotReachHere();
1701 #else // !_LP64
1702 switch (op) {
1703 case add: __ fadd_d (at_rsp()); break;
1704 case sub: __ fsubr_d(at_rsp()); break;
1705 case mul: {
1706 // strict semantics
1707 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1708 __ fmulp();
1709 __ fmul_d (at_rsp());
1710 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1711 __ fmulp();
1712 break;
1713 }
1714 case div: {
1715 // strict semantics
1716 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1717 __ fmul_d (at_rsp());
1718 __ fdivrp();
1719 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1720 __ fmulp();
1721 break;
1722 }
1723 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1724 default : ShouldNotReachHere();
1725 }
1726 __ d2ieee();
1727 // Pop double precision number from rsp.
1728 __ pop(rax);
1729 __ pop(rdx);
1730 #endif // _LP64
1731 }
1732 }
1733
1734 void TemplateTable::ineg() {
1735 transition(itos, itos);
1736 __ negl(rax);
1737 }
1738
1739 void TemplateTable::lneg() {
1740 transition(ltos, ltos);
1741 LP64_ONLY(__ negq(rax));
1742 NOT_LP64(__ lneg(rdx, rax));
1743 }
1744
1745 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1746 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1747 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1748 // of 128-bits operands for SSE instructions.
1749 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1750 // Store the value to a 128-bits operand.
1751 operand[0] = lo;
1752 operand[1] = hi;
1753 return operand;
1754 }
1755
1756 // Buffer for 128-bits masks used by SSE instructions.
1757 static jlong float_signflip_pool[2*2];
1758 static jlong double_signflip_pool[2*2];
1759
1760 void TemplateTable::fneg() {
1761 transition(ftos, ftos);
1762 if (UseSSE >= 1) {
1763 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1764 __ xorps(xmm0, ExternalAddress((address) float_signflip), rscratch1);
1765 } else {
1766 LP64_ONLY(ShouldNotReachHere());
1767 NOT_LP64(__ fchs());
1768 }
1769 }
1770
1771 void TemplateTable::dneg() {
1772 transition(dtos, dtos);
1773 if (UseSSE >= 2) {
1774 static jlong *double_signflip =
1775 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1776 __ xorpd(xmm0, ExternalAddress((address) double_signflip), rscratch1);
1777 } else {
1778 #ifdef _LP64
1779 ShouldNotReachHere();
1780 #else
1781 __ fchs();
1782 #endif
1783 }
1784 }
1785
1786 void TemplateTable::iinc() {
1787 transition(vtos, vtos);
1788 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1789 locals_index(rbx);
1790 __ addl(iaddress(rbx), rdx);
1791 }
1792
1793 void TemplateTable::wide_iinc() {
1794 transition(vtos, vtos);
1795 __ movl(rdx, at_bcp(4)); // get constant
1796 locals_index_wide(rbx);
1797 __ bswapl(rdx); // swap bytes & sign-extend constant
1798 __ sarl(rdx, 16);
1799 __ addl(iaddress(rbx), rdx);
1800 // Note: should probably use only one movl to get both
1801 // the index and the constant -> fix this
1802 }
1803
1804 void TemplateTable::convert() {
1805 #ifdef _LP64
1806 // Checking
1807 #ifdef ASSERT
1808 {
1809 TosState tos_in = ilgl;
1810 TosState tos_out = ilgl;
1811 switch (bytecode()) {
1812 case Bytecodes::_i2l: // fall through
1813 case Bytecodes::_i2f: // fall through
1814 case Bytecodes::_i2d: // fall through
1815 case Bytecodes::_i2b: // fall through
1816 case Bytecodes::_i2c: // fall through
1817 case Bytecodes::_i2s: tos_in = itos; break;
1818 case Bytecodes::_l2i: // fall through
1819 case Bytecodes::_l2f: // fall through
1820 case Bytecodes::_l2d: tos_in = ltos; break;
1821 case Bytecodes::_f2i: // fall through
1822 case Bytecodes::_f2l: // fall through
1823 case Bytecodes::_f2d: tos_in = ftos; break;
1824 case Bytecodes::_d2i: // fall through
1825 case Bytecodes::_d2l: // fall through
1826 case Bytecodes::_d2f: tos_in = dtos; break;
1827 default : ShouldNotReachHere();
1828 }
1829 switch (bytecode()) {
1830 case Bytecodes::_l2i: // fall through
1831 case Bytecodes::_f2i: // fall through
1832 case Bytecodes::_d2i: // fall through
1833 case Bytecodes::_i2b: // fall through
1834 case Bytecodes::_i2c: // fall through
1835 case Bytecodes::_i2s: tos_out = itos; break;
1836 case Bytecodes::_i2l: // fall through
1837 case Bytecodes::_f2l: // fall through
1838 case Bytecodes::_d2l: tos_out = ltos; break;
1839 case Bytecodes::_i2f: // fall through
1840 case Bytecodes::_l2f: // fall through
1841 case Bytecodes::_d2f: tos_out = ftos; break;
1842 case Bytecodes::_i2d: // fall through
1843 case Bytecodes::_l2d: // fall through
1844 case Bytecodes::_f2d: tos_out = dtos; break;
1845 default : ShouldNotReachHere();
1846 }
1847 transition(tos_in, tos_out);
1848 }
1849 #endif // ASSERT
1850
1851 static const int64_t is_nan = 0x8000000000000000L;
1852
1853 // Conversion
1854 switch (bytecode()) {
1855 case Bytecodes::_i2l:
1856 __ movslq(rax, rax);
1857 break;
1858 case Bytecodes::_i2f:
1859 __ cvtsi2ssl(xmm0, rax);
1860 break;
1861 case Bytecodes::_i2d:
1862 __ cvtsi2sdl(xmm0, rax);
1863 break;
1864 case Bytecodes::_i2b:
1865 __ movsbl(rax, rax);
1866 break;
1867 case Bytecodes::_i2c:
1868 __ movzwl(rax, rax);
1869 break;
1870 case Bytecodes::_i2s:
1871 __ movswl(rax, rax);
1872 break;
1873 case Bytecodes::_l2i:
1874 __ movl(rax, rax);
1875 break;
1876 case Bytecodes::_l2f:
1877 __ cvtsi2ssq(xmm0, rax);
1878 break;
1879 case Bytecodes::_l2d:
1880 __ cvtsi2sdq(xmm0, rax);
1881 break;
1882 case Bytecodes::_f2i:
1883 {
1884 Label L;
1885 __ cvttss2sil(rax, xmm0);
1886 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1887 __ jcc(Assembler::notEqual, L);
1888 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1889 __ bind(L);
1890 }
1891 break;
1892 case Bytecodes::_f2l:
1893 {
1894 Label L;
1895 __ cvttss2siq(rax, xmm0);
1896 // NaN or overflow/underflow?
1897 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1898 __ jcc(Assembler::notEqual, L);
1899 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1900 __ bind(L);
1901 }
1902 break;
1903 case Bytecodes::_f2d:
1904 __ cvtss2sd(xmm0, xmm0);
1905 break;
1906 case Bytecodes::_d2i:
1907 {
1908 Label L;
1909 __ cvttsd2sil(rax, xmm0);
1910 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1911 __ jcc(Assembler::notEqual, L);
1912 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1913 __ bind(L);
1914 }
1915 break;
1916 case Bytecodes::_d2l:
1917 {
1918 Label L;
1919 __ cvttsd2siq(rax, xmm0);
1920 // NaN or overflow/underflow?
1921 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1922 __ jcc(Assembler::notEqual, L);
1923 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1924 __ bind(L);
1925 }
1926 break;
1927 case Bytecodes::_d2f:
1928 __ cvtsd2ss(xmm0, xmm0);
1929 break;
1930 default:
1931 ShouldNotReachHere();
1932 }
1933 #else // !_LP64
1934 // Checking
1935 #ifdef ASSERT
1936 { TosState tos_in = ilgl;
1937 TosState tos_out = ilgl;
1938 switch (bytecode()) {
1939 case Bytecodes::_i2l: // fall through
1940 case Bytecodes::_i2f: // fall through
1941 case Bytecodes::_i2d: // fall through
1942 case Bytecodes::_i2b: // fall through
1943 case Bytecodes::_i2c: // fall through
1944 case Bytecodes::_i2s: tos_in = itos; break;
1945 case Bytecodes::_l2i: // fall through
1946 case Bytecodes::_l2f: // fall through
1947 case Bytecodes::_l2d: tos_in = ltos; break;
1948 case Bytecodes::_f2i: // fall through
1949 case Bytecodes::_f2l: // fall through
1950 case Bytecodes::_f2d: tos_in = ftos; break;
1951 case Bytecodes::_d2i: // fall through
1952 case Bytecodes::_d2l: // fall through
1953 case Bytecodes::_d2f: tos_in = dtos; break;
1954 default : ShouldNotReachHere();
1955 }
1956 switch (bytecode()) {
1957 case Bytecodes::_l2i: // fall through
1958 case Bytecodes::_f2i: // fall through
1959 case Bytecodes::_d2i: // fall through
1960 case Bytecodes::_i2b: // fall through
1961 case Bytecodes::_i2c: // fall through
1962 case Bytecodes::_i2s: tos_out = itos; break;
1963 case Bytecodes::_i2l: // fall through
1964 case Bytecodes::_f2l: // fall through
1965 case Bytecodes::_d2l: tos_out = ltos; break;
1966 case Bytecodes::_i2f: // fall through
1967 case Bytecodes::_l2f: // fall through
1968 case Bytecodes::_d2f: tos_out = ftos; break;
1969 case Bytecodes::_i2d: // fall through
1970 case Bytecodes::_l2d: // fall through
1971 case Bytecodes::_f2d: tos_out = dtos; break;
1972 default : ShouldNotReachHere();
1973 }
1974 transition(tos_in, tos_out);
1975 }
1976 #endif // ASSERT
1977
1978 // Conversion
1979 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1980 switch (bytecode()) {
1981 case Bytecodes::_i2l:
1982 __ extend_sign(rdx, rax);
1983 break;
1984 case Bytecodes::_i2f:
1985 if (UseSSE >= 1) {
1986 __ cvtsi2ssl(xmm0, rax);
1987 } else {
1988 __ push(rax); // store int on tos
1989 __ fild_s(at_rsp()); // load int to ST0
1990 __ f2ieee(); // truncate to float size
1991 __ pop(rcx); // adjust rsp
1992 }
1993 break;
1994 case Bytecodes::_i2d:
1995 if (UseSSE >= 2) {
1996 __ cvtsi2sdl(xmm0, rax);
1997 } else {
1998 __ push(rax); // add one slot for d2ieee()
1999 __ push(rax); // store int on tos
2000 __ fild_s(at_rsp()); // load int to ST0
2001 __ d2ieee(); // truncate to double size
2002 __ pop(rcx); // adjust rsp
2003 __ pop(rcx);
2004 }
2005 break;
2006 case Bytecodes::_i2b:
2007 __ shll(rax, 24); // truncate upper 24 bits
2008 __ sarl(rax, 24); // and sign-extend byte
2009 LP64_ONLY(__ movsbl(rax, rax));
2010 break;
2011 case Bytecodes::_i2c:
2012 __ andl(rax, 0xFFFF); // truncate upper 16 bits
2013 LP64_ONLY(__ movzwl(rax, rax));
2014 break;
2015 case Bytecodes::_i2s:
2016 __ shll(rax, 16); // truncate upper 16 bits
2017 __ sarl(rax, 16); // and sign-extend short
2018 LP64_ONLY(__ movswl(rax, rax));
2019 break;
2020 case Bytecodes::_l2i:
2021 /* nothing to do */
2022 break;
2023 case Bytecodes::_l2f:
2024 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
2025 // 64-bit long values to floats. On 32-bit platforms it is not possible
2026 // to use that instruction with 64-bit operands, therefore the FPU is
2027 // used to perform the conversion.
2028 __ push(rdx); // store long on tos
2029 __ push(rax);
2030 __ fild_d(at_rsp()); // load long to ST0
2031 __ f2ieee(); // truncate to float size
2032 __ pop(rcx); // adjust rsp
2033 __ pop(rcx);
2034 if (UseSSE >= 1) {
2035 __ push_f();
2036 __ pop_f(xmm0);
2037 }
2038 break;
2039 case Bytecodes::_l2d:
2040 // On 32-bit platforms the FPU is used for conversion because on
2041 // 32-bit platforms it is not not possible to use the cvtsi2sdq
2042 // instruction with 64-bit operands.
2043 __ push(rdx); // store long on tos
2044 __ push(rax);
2045 __ fild_d(at_rsp()); // load long to ST0
2046 __ d2ieee(); // truncate to double size
2047 __ pop(rcx); // adjust rsp
2048 __ pop(rcx);
2049 if (UseSSE >= 2) {
2050 __ push_d();
2051 __ pop_d(xmm0);
2052 }
2053 break;
2054 case Bytecodes::_f2i:
2055 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2056 // as it returns 0 for any NaN.
2057 if (UseSSE >= 1) {
2058 __ push_f(xmm0);
2059 } else {
2060 __ push(rcx); // reserve space for argument
2061 __ fstp_s(at_rsp()); // pass float argument on stack
2062 }
2063 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2064 break;
2065 case Bytecodes::_f2l:
2066 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2067 // as it returns 0 for any NaN.
2068 if (UseSSE >= 1) {
2069 __ push_f(xmm0);
2070 } else {
2071 __ push(rcx); // reserve space for argument
2072 __ fstp_s(at_rsp()); // pass float argument on stack
2073 }
2074 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2075 break;
2076 case Bytecodes::_f2d:
2077 if (UseSSE < 1) {
2078 /* nothing to do */
2079 } else if (UseSSE == 1) {
2080 __ push_f(xmm0);
2081 __ pop_f();
2082 } else { // UseSSE >= 2
2083 __ cvtss2sd(xmm0, xmm0);
2084 }
2085 break;
2086 case Bytecodes::_d2i:
2087 if (UseSSE >= 2) {
2088 __ push_d(xmm0);
2089 } else {
2090 __ push(rcx); // reserve space for argument
2091 __ push(rcx);
2092 __ fstp_d(at_rsp()); // pass double argument on stack
2093 }
2094 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2095 break;
2096 case Bytecodes::_d2l:
2097 if (UseSSE >= 2) {
2098 __ push_d(xmm0);
2099 } else {
2100 __ push(rcx); // reserve space for argument
2101 __ push(rcx);
2102 __ fstp_d(at_rsp()); // pass double argument on stack
2103 }
2104 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2105 break;
2106 case Bytecodes::_d2f:
2107 if (UseSSE <= 1) {
2108 __ push(rcx); // reserve space for f2ieee()
2109 __ f2ieee(); // truncate to float size
2110 __ pop(rcx); // adjust rsp
2111 if (UseSSE == 1) {
2112 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2113 // the conversion is performed using the FPU in this case.
2114 __ push_f();
2115 __ pop_f(xmm0);
2116 }
2117 } else { // UseSSE >= 2
2118 __ cvtsd2ss(xmm0, xmm0);
2119 }
2120 break;
2121 default :
2122 ShouldNotReachHere();
2123 }
2124 #endif // _LP64
2125 }
2126
2127 void TemplateTable::lcmp() {
2128 transition(ltos, itos);
2129 #ifdef _LP64
2130 Label done;
2131 __ pop_l(rdx);
2132 __ cmpq(rdx, rax);
2133 __ movl(rax, -1);
2134 __ jccb(Assembler::less, done);
2135 __ setb(Assembler::notEqual, rax);
2136 __ movzbl(rax, rax);
2137 __ bind(done);
2138 #else
2139
2140 // y = rdx:rax
2141 __ pop_l(rbx, rcx); // get x = rcx:rbx
2142 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2143 __ mov(rax, rcx);
2144 #endif
2145 }
2146
2147 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2148 if ((is_float && UseSSE >= 1) ||
2149 (!is_float && UseSSE >= 2)) {
2150 Label done;
2151 if (is_float) {
2152 // XXX get rid of pop here, use ... reg, mem32
2153 __ pop_f(xmm1);
2154 __ ucomiss(xmm1, xmm0);
2155 } else {
2156 // XXX get rid of pop here, use ... reg, mem64
2157 __ pop_d(xmm1);
2158 __ ucomisd(xmm1, xmm0);
2159 }
2160 if (unordered_result < 0) {
2161 __ movl(rax, -1);
2162 __ jccb(Assembler::parity, done);
2163 __ jccb(Assembler::below, done);
2164 __ setb(Assembler::notEqual, rdx);
2165 __ movzbl(rax, rdx);
2166 } else {
2167 __ movl(rax, 1);
2168 __ jccb(Assembler::parity, done);
2169 __ jccb(Assembler::above, done);
2170 __ movl(rax, 0);
2171 __ jccb(Assembler::equal, done);
2172 __ decrementl(rax);
2173 }
2174 __ bind(done);
2175 } else {
2176 #ifdef _LP64
2177 ShouldNotReachHere();
2178 #else // !_LP64
2179 if (is_float) {
2180 __ fld_s(at_rsp());
2181 } else {
2182 __ fld_d(at_rsp());
2183 __ pop(rdx);
2184 }
2185 __ pop(rcx);
2186 __ fcmp2int(rax, unordered_result < 0);
2187 #endif // _LP64
2188 }
2189 }
2190
2191 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2192 __ get_method(rcx); // rcx holds method
2193 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2194 // holds bumped taken count
2195
2196 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2197 InvocationCounter::counter_offset();
2198 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2199 InvocationCounter::counter_offset();
2200
2201 // Load up edx with the branch displacement
2202 if (is_wide) {
2203 __ movl(rdx, at_bcp(1));
2204 } else {
2205 __ load_signed_short(rdx, at_bcp(1));
2206 }
2207 __ bswapl(rdx);
2208
2209 if (!is_wide) {
2210 __ sarl(rdx, 16);
2211 }
2212 LP64_ONLY(__ movl2ptr(rdx, rdx));
2213
2214 // Handle all the JSR stuff here, then exit.
2215 // It's much shorter and cleaner than intermingling with the non-JSR
2216 // normal-branch stuff occurring below.
2217 if (is_jsr) {
2218 // Pre-load the next target bytecode into rbx
2219 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2220
2221 // compute return address as bci in rax
2222 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2223 in_bytes(ConstMethod::codes_offset())));
2224 __ subptr(rax, Address(rcx, Method::const_offset()));
2225 // Adjust the bcp in r13 by the displacement in rdx
2226 __ addptr(rbcp, rdx);
2227 // jsr returns atos that is not an oop
2228 __ push_i(rax);
2229 __ dispatch_only(vtos, true);
2230 return;
2231 }
2232
2233 // Normal (non-jsr) branch handling
2234
2235 // Adjust the bcp in r13 by the displacement in rdx
2236 __ addptr(rbcp, rdx);
2237
2238 assert(UseLoopCounter || !UseOnStackReplacement,
2239 "on-stack-replacement requires loop counters");
2240 Label backedge_counter_overflow;
2241 Label dispatch;
2242 if (UseLoopCounter) {
2243 // increment backedge counter for backward branches
2244 // rax: MDO
2245 // rbx: MDO bumped taken-count
2246 // rcx: method
2247 // rdx: target offset
2248 // r13: target bcp
2249 // r14: locals pointer
2250 __ testl(rdx, rdx); // check if forward or backward branch
2251 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2252
2253 // check if MethodCounters exists
2254 Label has_counters;
2255 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2256 __ testptr(rax, rax);
2257 __ jcc(Assembler::notZero, has_counters);
2258 __ push(rdx);
2259 __ push(rcx);
2260 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2261 rcx);
2262 __ pop(rcx);
2263 __ pop(rdx);
2264 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2265 __ testptr(rax, rax);
2266 __ jcc(Assembler::zero, dispatch);
2267 __ bind(has_counters);
2268
2269 Label no_mdo;
2270 if (ProfileInterpreter) {
2271 // Are we profiling?
2272 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2273 __ testptr(rbx, rbx);
2274 __ jccb(Assembler::zero, no_mdo);
2275 // Increment the MDO backedge counter
2276 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2277 in_bytes(InvocationCounter::counter_offset()));
2278 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2279 __ increment_mask_and_jump(mdo_backedge_counter, mask, rax,
2280 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2281 __ jmp(dispatch);
2282 }
2283 __ bind(no_mdo);
2284 // Increment backedge counter in MethodCounters*
2285 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2286 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2287 __ increment_mask_and_jump(Address(rcx, be_offset), mask, rax,
2288 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2289 __ bind(dispatch);
2290 }
2291
2292 // Pre-load the next target bytecode into rbx
2293 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2294
2295 // continue with the bytecode @ target
2296 // rax: return bci for jsr's, unused otherwise
2297 // rbx: target bytecode
2298 // r13: target bcp
2299 __ dispatch_only(vtos, true);
2300
2301 if (UseLoopCounter) {
2302 if (UseOnStackReplacement) {
2303 Label set_mdp;
2304 // invocation counter overflow
2305 __ bind(backedge_counter_overflow);
2306 __ negptr(rdx);
2307 __ addptr(rdx, rbcp); // branch bcp
2308 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2309 __ call_VM(noreg,
2310 CAST_FROM_FN_PTR(address,
2311 InterpreterRuntime::frequency_counter_overflow),
2312 rdx);
2313
2314 // rax: osr nmethod (osr ok) or null (osr not possible)
2315 // rdx: scratch
2316 // r14: locals pointer
2317 // r13: bcp
2318 __ testptr(rax, rax); // test result
2319 __ jcc(Assembler::zero, dispatch); // no osr if null
2320 // nmethod may have been invalidated (VM may block upon call_VM return)
2321 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2322 __ jcc(Assembler::notEqual, dispatch);
2323
2324 // We have the address of an on stack replacement routine in rax.
2325 // In preparation of invoking it, first we must migrate the locals
2326 // and monitors from off the interpreter frame on the stack.
2327 // Ensure to save the osr nmethod over the migration call,
2328 // it will be preserved in rbx.
2329 __ mov(rbx, rax);
2330
2331 NOT_LP64(__ get_thread(rcx));
2332
2333 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2334
2335 // rax is OSR buffer, move it to expected parameter location
2336 LP64_ONLY(__ mov(j_rarg0, rax));
2337 NOT_LP64(__ mov(rcx, rax));
2338 // We use j_rarg definitions here so that registers don't conflict as parameter
2339 // registers change across platforms as we are in the midst of a calling
2340 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2341
2342 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2343 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2344
2345 // pop the interpreter frame
2346 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2347 __ leave(); // remove frame anchor
2348 __ pop(retaddr); // get return address
2349 __ mov(rsp, sender_sp); // set sp to sender sp
2350 // Ensure compiled code always sees stack at proper alignment
2351 __ andptr(rsp, -(StackAlignmentInBytes));
2352
2353 // unlike x86 we need no specialized return from compiled code
2354 // to the interpreter or the call stub.
2355
2356 // push the return address
2357 __ push(retaddr);
2358
2359 // and begin the OSR nmethod
2360 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2361 }
2362 }
2363 }
2364
2365 void TemplateTable::if_0cmp(Condition cc) {
2366 transition(itos, vtos);
2367 // assume branch is more often taken than not (loops use backward branches)
2368 Label not_taken;
2369 __ testl(rax, rax);
2370 __ jcc(j_not(cc), not_taken);
2371 branch(false, false);
2372 __ bind(not_taken);
2373 __ profile_not_taken_branch(rax);
2374 }
2375
2376 void TemplateTable::if_icmp(Condition cc) {
2377 transition(itos, vtos);
2378 // assume branch is more often taken than not (loops use backward branches)
2379 Label not_taken;
2380 __ pop_i(rdx);
2381 __ cmpl(rdx, rax);
2382 __ jcc(j_not(cc), not_taken);
2383 branch(false, false);
2384 __ bind(not_taken);
2385 __ profile_not_taken_branch(rax);
2386 }
2387
2388 void TemplateTable::if_nullcmp(Condition cc) {
2389 transition(atos, vtos);
2390 // assume branch is more often taken than not (loops use backward branches)
2391 Label not_taken;
2392 __ testptr(rax, rax);
2393 __ jcc(j_not(cc), not_taken);
2394 branch(false, false);
2395 __ bind(not_taken);
2396 __ profile_not_taken_branch(rax);
2397 }
2398
2399 void TemplateTable::if_acmp(Condition cc) {
2400 transition(atos, vtos);
2401 // assume branch is more often taken than not (loops use backward branches)
2402 Label taken, not_taken;
2403 __ pop_ptr(rdx);
2404
2405 __ profile_acmp(rbx, rdx, rax, rcx);
2406
2407 const int is_inline_type_mask = markWord::inline_type_pattern;
2408 if (EnableValhalla) {
2409 __ cmpoop(rdx, rax);
2410 __ jcc(Assembler::equal, (cc == equal) ? taken : not_taken);
2411
2412 // might be substitutable, test if either rax or rdx is null
2413 __ testptr(rax, rax);
2414 __ jcc(Assembler::zero, (cc == equal) ? not_taken : taken);
2415 __ testptr(rdx, rdx);
2416 __ jcc(Assembler::zero, (cc == equal) ? not_taken : taken);
2417
2418 // and both are values ?
2419 __ movptr(rbx, Address(rdx, oopDesc::mark_offset_in_bytes()));
2420 __ andptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
2421 __ andptr(rbx, is_inline_type_mask);
2422 __ cmpptr(rbx, is_inline_type_mask);
2423 __ jcc(Assembler::notEqual, (cc == equal) ? not_taken : taken);
2424
2425 // same value klass ?
2426 __ load_metadata(rbx, rdx);
2427 __ load_metadata(rcx, rax);
2428 __ cmpptr(rbx, rcx);
2429 __ jcc(Assembler::notEqual, (cc == equal) ? not_taken : taken);
2430
2431 // Know both are the same type, let's test for substitutability...
2432 if (cc == equal) {
2433 invoke_is_substitutable(rax, rdx, taken, not_taken);
2434 } else {
2435 invoke_is_substitutable(rax, rdx, not_taken, taken);
2436 }
2437 __ stop("Not reachable");
2438 }
2439
2440 __ cmpoop(rdx, rax);
2441 __ jcc(j_not(cc), not_taken);
2442 __ bind(taken);
2443 branch(false, false);
2444 __ bind(not_taken);
2445 __ profile_not_taken_branch(rax, true);
2446 }
2447
2448 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2449 Label& is_subst, Label& not_subst) {
2450 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2451 // Restored...rax answer, jmp to outcome...
2452 __ testl(rax, rax);
2453 __ jcc(Assembler::zero, not_subst);
2454 __ jmp(is_subst);
2455 }
2456
2457 void TemplateTable::ret() {
2458 transition(vtos, vtos);
2459 locals_index(rbx);
2460 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2461 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2462 __ profile_ret(rbx, rcx);
2463 __ get_method(rax);
2464 __ movptr(rbcp, Address(rax, Method::const_offset()));
2465 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2466 ConstMethod::codes_offset()));
2467 __ dispatch_next(vtos, 0, true);
2468 }
2469
2470 void TemplateTable::wide_ret() {
2471 transition(vtos, vtos);
2472 locals_index_wide(rbx);
2473 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2474 __ profile_ret(rbx, rcx);
2475 __ get_method(rax);
2476 __ movptr(rbcp, Address(rax, Method::const_offset()));
2477 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2478 __ dispatch_next(vtos, 0, true);
2479 }
2480
2481 void TemplateTable::tableswitch() {
2482 Label default_case, continue_execution;
2483 transition(itos, vtos);
2484
2485 // align r13/rsi
2486 __ lea(rbx, at_bcp(BytesPerInt));
2487 __ andptr(rbx, -BytesPerInt);
2488 // load lo & hi
2489 __ movl(rcx, Address(rbx, BytesPerInt));
2490 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2491 __ bswapl(rcx);
2492 __ bswapl(rdx);
2493 // check against lo & hi
2494 __ cmpl(rax, rcx);
2495 __ jcc(Assembler::less, default_case);
2496 __ cmpl(rax, rdx);
2497 __ jcc(Assembler::greater, default_case);
2498 // lookup dispatch offset
2499 __ subl(rax, rcx);
2500 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2501 __ profile_switch_case(rax, rbx, rcx);
2502 // continue execution
2503 __ bind(continue_execution);
2504 __ bswapl(rdx);
2505 LP64_ONLY(__ movl2ptr(rdx, rdx));
2506 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2507 __ addptr(rbcp, rdx);
2508 __ dispatch_only(vtos, true);
2509 // handle default
2510 __ bind(default_case);
2511 __ profile_switch_default(rax);
2512 __ movl(rdx, Address(rbx, 0));
2513 __ jmp(continue_execution);
2514 }
2515
2516 void TemplateTable::lookupswitch() {
2517 transition(itos, itos);
2518 __ stop("lookupswitch bytecode should have been rewritten");
2519 }
2520
2521 void TemplateTable::fast_linearswitch() {
2522 transition(itos, vtos);
2523 Label loop_entry, loop, found, continue_execution;
2524 // bswap rax so we can avoid bswapping the table entries
2525 __ bswapl(rax);
2526 // align r13
2527 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2528 // this instruction (change offsets
2529 // below)
2530 __ andptr(rbx, -BytesPerInt);
2531 // set counter
2532 __ movl(rcx, Address(rbx, BytesPerInt));
2533 __ bswapl(rcx);
2534 __ jmpb(loop_entry);
2535 // table search
2536 __ bind(loop);
2537 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2538 __ jcc(Assembler::equal, found);
2539 __ bind(loop_entry);
2540 __ decrementl(rcx);
2541 __ jcc(Assembler::greaterEqual, loop);
2542 // default case
2543 __ profile_switch_default(rax);
2544 __ movl(rdx, Address(rbx, 0));
2545 __ jmp(continue_execution);
2546 // entry found -> get offset
2547 __ bind(found);
2548 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2549 __ profile_switch_case(rcx, rax, rbx);
2550 // continue execution
2551 __ bind(continue_execution);
2552 __ bswapl(rdx);
2553 __ movl2ptr(rdx, rdx);
2554 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2555 __ addptr(rbcp, rdx);
2556 __ dispatch_only(vtos, true);
2557 }
2558
2559 void TemplateTable::fast_binaryswitch() {
2560 transition(itos, vtos);
2561 // Implementation using the following core algorithm:
2562 //
2563 // int binary_search(int key, LookupswitchPair* array, int n) {
2564 // // Binary search according to "Methodik des Programmierens" by
2565 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2566 // int i = 0;
2567 // int j = n;
2568 // while (i+1 < j) {
2569 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2570 // // with Q: for all i: 0 <= i < n: key < a[i]
2571 // // where a stands for the array and assuming that the (inexisting)
2572 // // element a[n] is infinitely big.
2573 // int h = (i + j) >> 1;
2574 // // i < h < j
2575 // if (key < array[h].fast_match()) {
2576 // j = h;
2577 // } else {
2578 // i = h;
2579 // }
2580 // }
2581 // // R: a[i] <= key < a[i+1] or Q
2582 // // (i.e., if key is within array, i is the correct index)
2583 // return i;
2584 // }
2585
2586 // Register allocation
2587 const Register key = rax; // already set (tosca)
2588 const Register array = rbx;
2589 const Register i = rcx;
2590 const Register j = rdx;
2591 const Register h = rdi;
2592 const Register temp = rsi;
2593
2594 // Find array start
2595 NOT_LP64(__ save_bcp());
2596
2597 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2598 // get rid of this
2599 // instruction (change
2600 // offsets below)
2601 __ andptr(array, -BytesPerInt);
2602
2603 // Initialize i & j
2604 __ xorl(i, i); // i = 0;
2605 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2606
2607 // Convert j into native byteordering
2608 __ bswapl(j);
2609
2610 // And start
2611 Label entry;
2612 __ jmp(entry);
2613
2614 // binary search loop
2615 {
2616 Label loop;
2617 __ bind(loop);
2618 // int h = (i + j) >> 1;
2619 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2620 __ sarl(h, 1); // h = (i + j) >> 1;
2621 // if (key < array[h].fast_match()) {
2622 // j = h;
2623 // } else {
2624 // i = h;
2625 // }
2626 // Convert array[h].match to native byte-ordering before compare
2627 __ movl(temp, Address(array, h, Address::times_8));
2628 __ bswapl(temp);
2629 __ cmpl(key, temp);
2630 // j = h if (key < array[h].fast_match())
2631 __ cmov32(Assembler::less, j, h);
2632 // i = h if (key >= array[h].fast_match())
2633 __ cmov32(Assembler::greaterEqual, i, h);
2634 // while (i+1 < j)
2635 __ bind(entry);
2636 __ leal(h, Address(i, 1)); // i+1
2637 __ cmpl(h, j); // i+1 < j
2638 __ jcc(Assembler::less, loop);
2639 }
2640
2641 // end of binary search, result index is i (must check again!)
2642 Label default_case;
2643 // Convert array[i].match to native byte-ordering before compare
2644 __ movl(temp, Address(array, i, Address::times_8));
2645 __ bswapl(temp);
2646 __ cmpl(key, temp);
2647 __ jcc(Assembler::notEqual, default_case);
2648
2649 // entry found -> j = offset
2650 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2651 __ profile_switch_case(i, key, array);
2652 __ bswapl(j);
2653 LP64_ONLY(__ movslq(j, j));
2654
2655 NOT_LP64(__ restore_bcp());
2656 NOT_LP64(__ restore_locals()); // restore rdi
2657
2658 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2659 __ addptr(rbcp, j);
2660 __ dispatch_only(vtos, true);
2661
2662 // default case -> j = default offset
2663 __ bind(default_case);
2664 __ profile_switch_default(i);
2665 __ movl(j, Address(array, -2 * BytesPerInt));
2666 __ bswapl(j);
2667 LP64_ONLY(__ movslq(j, j));
2668
2669 NOT_LP64(__ restore_bcp());
2670 NOT_LP64(__ restore_locals());
2671
2672 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2673 __ addptr(rbcp, j);
2674 __ dispatch_only(vtos, true);
2675 }
2676
2677 void TemplateTable::_return(TosState state) {
2678 transition(state, state);
2679
2680 assert(_desc->calls_vm(),
2681 "inconsistent calls_vm information"); // call in remove_activation
2682
2683 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2684 assert(state == vtos, "only valid state");
2685 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2686 __ movptr(robj, aaddress(0));
2687 __ load_klass(rdi, robj, rscratch1);
2688 __ testb(Address(rdi, Klass::misc_flags_offset()), KlassFlags::_misc_has_finalizer);
2689 Label skip_register_finalizer;
2690 __ jcc(Assembler::zero, skip_register_finalizer);
2691
2692 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2693
2694 __ bind(skip_register_finalizer);
2695 }
2696
2697 if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2698 Label no_safepoint;
2699 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2700 #ifdef _LP64
2701 __ testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2702 #else
2703 const Register thread = rdi;
2704 __ get_thread(thread);
2705 __ testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2706 #endif
2707 __ jcc(Assembler::zero, no_safepoint);
2708 __ push(state);
2709 __ push_cont_fastpath();
2710 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2711 InterpreterRuntime::at_safepoint));
2712 __ pop_cont_fastpath();
2713 __ pop(state);
2714 __ bind(no_safepoint);
2715 }
2716
2717 // Narrow result if state is itos but result type is smaller.
2718 // Need to narrow in the return bytecode rather than in generate_return_entry
2719 // since compiled code callers expect the result to already be narrowed.
2720 if (state == itos) {
2721 __ narrow(rax);
2722 }
2723
2724 __ remove_activation(state, rbcp, true, true, true);
2725
2726 __ jmp(rbcp);
2727 }
2728
2729 // ----------------------------------------------------------------------------
2730 // Volatile variables demand their effects be made known to all CPU's
2731 // in order. Store buffers on most chips allow reads & writes to
2732 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2733 // without some kind of memory barrier (i.e., it's not sufficient that
2734 // the interpreter does not reorder volatile references, the hardware
2735 // also must not reorder them).
2736 //
2737 // According to the new Java Memory Model (JMM):
2738 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2739 // writes act as acquire & release, so:
2740 // (2) A read cannot let unrelated NON-volatile memory refs that
2741 // happen after the read float up to before the read. It's OK for
2742 // non-volatile memory refs that happen before the volatile read to
2743 // float down below it.
2744 // (3) Similar a volatile write cannot let unrelated NON-volatile
2745 // memory refs that happen BEFORE the write float down to after the
2746 // write. It's OK for non-volatile memory refs that happen after the
2747 // volatile write to float up before it.
2748 //
2749 // We only put in barriers around volatile refs (they are expensive),
2750 // not _between_ memory refs (that would require us to track the
2751 // flavor of the previous memory refs). Requirements (2) and (3)
2752 // require some barriers before volatile stores and after volatile
2753 // loads. These nearly cover requirement (1) but miss the
2754 // volatile-store-volatile-load case. This final case is placed after
2755 // volatile-stores although it could just as well go before
2756 // volatile-loads.
2757
2758 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2759 // Helper function to insert a is-volatile test and memory barrier
2760 __ membar(order_constraint);
2761 }
2762
2763 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2764 Register cache,
2765 Register index) {
2766 const Register temp = rbx;
2767 assert_different_registers(cache, index, temp);
2768
2769 Label L_clinit_barrier_slow;
2770 Label resolved;
2771
2772 Bytecodes::Code code = bytecode();
2773
2774 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2775
2776 __ load_method_entry(cache, index);
2777 switch(byte_no) {
2778 case f1_byte:
2779 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2780 break;
2781 case f2_byte:
2782 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2783 break;
2784 default:
2785 ShouldNotReachHere();
2786 }
2787 __ cmpl(temp, code); // have we resolved this bytecode?
2788 __ jcc(Assembler::equal, resolved);
2789
2790 // resolve first time through
2791 // Class initialization barrier slow path lands here as well.
2792 __ bind(L_clinit_barrier_slow);
2793 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2794 __ movl(temp, code);
2795 __ call_VM(noreg, entry, temp);
2796 // Update registers with resolved info
2797 __ load_method_entry(cache, index);
2798
2799 __ bind(resolved);
2800
2801 // Class initialization barrier for static methods
2802 if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2803 const Register method = temp;
2804 const Register klass = temp;
2805 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2806 assert(thread != noreg, "x86_32 not supported");
2807
2808 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2809 __ load_method_holder(klass, method);
2810 __ clinit_barrier(klass, thread, nullptr /*L_fast_path*/, &L_clinit_barrier_slow);
2811 }
2812 }
2813
2814 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2815 Register cache,
2816 Register index) {
2817 const Register temp = rbx;
2818 assert_different_registers(cache, index, temp);
2819
2820 Label resolved;
2821
2822 Bytecodes::Code code = bytecode();
2823 switch (code) {
2824 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2825 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2826 default: break;
2827 }
2828
2829 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2830 __ load_field_entry(cache, index);
2831 if (byte_no == f1_byte) {
2832 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2833 } else {
2834 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2835 }
2836 __ cmpl(temp, code); // have we resolved this bytecode?
2837 __ jcc(Assembler::equal, resolved);
2838
2839 // resolve first time through
2840 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2841 __ movl(temp, code);
2842 __ call_VM(noreg, entry, temp);
2843 // Update registers with resolved info
2844 __ load_field_entry(cache, index);
2845
2846 __ bind(resolved);
2847 }
2848
2849 void TemplateTable::load_resolved_field_entry(Register obj,
2850 Register cache,
2851 Register tos_state,
2852 Register offset,
2853 Register flags,
2854 bool is_static = false) {
2855 assert_different_registers(cache, tos_state, flags, offset);
2856
2857 // Field offset
2858 __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2859
2860 // Flags
2861 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2862
2863 // TOS state
2864 __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2865
2866 // Klass overwrite register
2867 if (is_static) {
2868 __ movptr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2869 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2870 __ movptr(obj, Address(obj, mirror_offset));
2871 __ resolve_oop_handle(obj, rscratch2);
2872 }
2873
2874 }
2875
2876 void TemplateTable::load_invokedynamic_entry(Register method) {
2877 // setup registers
2878 const Register appendix = rax;
2879 const Register cache = rcx;
2880 const Register index = rdx;
2881 assert_different_registers(method, appendix, cache, index);
2882
2883 __ save_bcp();
2884
2885 Label resolved;
2886
2887 __ load_resolved_indy_entry(cache, index);
2888 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2889
2890 // Compare the method to zero
2891 __ testptr(method, method);
2892 __ jcc(Assembler::notZero, resolved);
2893
2894 Bytecodes::Code code = bytecode();
2895
2896 // Call to the interpreter runtime to resolve invokedynamic
2897 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2898 __ movl(method, code); // this is essentially Bytecodes::_invokedynamic
2899 __ call_VM(noreg, entry, method);
2900 // Update registers with resolved info
2901 __ load_resolved_indy_entry(cache, index);
2902 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2903
2904 #ifdef ASSERT
2905 __ testptr(method, method);
2906 __ jcc(Assembler::notZero, resolved);
2907 __ stop("Should be resolved by now");
2908 #endif // ASSERT
2909 __ bind(resolved);
2910
2911 Label L_no_push;
2912 // Check if there is an appendix
2913 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2914 __ testl(index, (1 << ResolvedIndyEntry::has_appendix_shift));
2915 __ jcc(Assembler::zero, L_no_push);
2916
2917 // Get appendix
2918 __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2919 // Push the appendix as a trailing parameter
2920 // since the parameter_size includes it.
2921 __ load_resolved_reference_at_index(appendix, index);
2922 __ verify_oop(appendix);
2923 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2924 __ bind(L_no_push);
2925
2926 // compute return type
2927 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2928 // load return address
2929 {
2930 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2931 ExternalAddress table(table_addr);
2932 #ifdef _LP64
2933 __ lea(rscratch1, table);
2934 __ movptr(index, Address(rscratch1, index, Address::times_ptr));
2935 #else
2936 __ movptr(index, ArrayAddress(table, Address(noreg, index, Address::times_ptr)));
2937 #endif // _LP64
2938 }
2939
2940 // push return address
2941 __ push(index);
2942 }
2943
2944 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2945 Register method,
2946 Register flags) {
2947 // setup registers
2948 const Register index = rdx;
2949 assert_different_registers(cache, index);
2950 assert_different_registers(method, cache, flags);
2951
2952 // determine constant pool cache field offsets
2953 resolve_cache_and_index_for_method(f1_byte, cache, index);
2954 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2955 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2956 }
2957
2958 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2959 Register method,
2960 Register ref_index,
2961 Register flags) {
2962 // setup registers
2963 const Register index = rdx;
2964 assert_different_registers(cache, index);
2965 assert_different_registers(cache, method, ref_index, flags);
2966
2967 // determine constant pool cache field offsets
2968 resolve_cache_and_index_for_method(f1_byte, cache, index);
2969 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2970
2971 // Maybe push appendix
2972 Label L_no_push;
2973 __ testl(flags, (1 << ResolvedMethodEntry::has_appendix_shift));
2974 __ jcc(Assembler::zero, L_no_push);
2975 // invokehandle uses an index into the resolved references array
2976 __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2977 // Push the appendix as a trailing parameter.
2978 // This must be done before we get the receiver,
2979 // since the parameter_size includes it.
2980 Register appendix = method;
2981 __ load_resolved_reference_at_index(appendix, ref_index);
2982 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2983 __ bind(L_no_push);
2984
2985 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2986 }
2987
2988 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2989 Register klass,
2990 Register method_or_table_index,
2991 Register flags) {
2992 // setup registers
2993 const Register index = rdx;
2994 assert_different_registers(cache, klass, method_or_table_index, flags);
2995
2996 // determine constant pool cache field offsets
2997 resolve_cache_and_index_for_method(f1_byte, cache, index);
2998 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2999
3000 // Invokeinterface can behave in different ways:
3001 // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
3002 // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
3003 // vtable index is placed in the register.
3004 // Otherwise, the registers will be populated with the klass and method.
3005
3006 Label NotVirtual; Label NotVFinal; Label Done;
3007 __ testl(flags, 1 << ResolvedMethodEntry::is_forced_virtual_shift);
3008 __ jcc(Assembler::zero, NotVirtual);
3009 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
3010 __ jcc(Assembler::zero, NotVFinal);
3011 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3012 __ jmp(Done);
3013
3014 __ bind(NotVFinal);
3015 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
3016 __ jmp(Done);
3017
3018 __ bind(NotVirtual);
3019 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3020 __ movptr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
3021 __ bind(Done);
3022 }
3023
3024 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
3025 Register method_or_table_index,
3026 Register flags) {
3027 // setup registers
3028 const Register index = rdx;
3029 assert_different_registers(index, cache);
3030 assert_different_registers(method_or_table_index, cache, flags);
3031
3032 // determine constant pool cache field offsets
3033 resolve_cache_and_index_for_method(f2_byte, cache, index);
3034 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
3035
3036 // method_or_table_index can either be an itable index or a method depending on the virtual final flag
3037 Label isVFinal; Label Done;
3038 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
3039 __ jcc(Assembler::notZero, isVFinal);
3040 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
3041 __ jmp(Done);
3042 __ bind(isVFinal);
3043 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
3044 __ bind(Done);
3045 }
3046
3047 // The registers cache and index expected to be set before call.
3048 // Correct values of the cache and index registers are preserved.
3049 void TemplateTable::jvmti_post_field_access(Register cache,
3050 Register index,
3051 bool is_static,
3052 bool has_tos) {
3053 if (JvmtiExport::can_post_field_access()) {
3054 // Check to see if a field access watch has been set before we take
3055 // the time to call into the VM.
3056 Label L1;
3057 assert_different_registers(cache, index, rax);
3058 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3059 __ testl(rax,rax);
3060 __ jcc(Assembler::zero, L1);
3061
3062 // cache entry pointer
3063 __ load_field_entry(cache, index);
3064 if (is_static) {
3065 __ xorptr(rax, rax); // null object reference
3066 } else {
3067 __ pop(atos); // Get the object
3068 __ verify_oop(rax);
3069 __ push(atos); // Restore stack state
3070 }
3071 // rax,: object pointer or null
3072 // cache: cache entry pointer
3073 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
3074 rax, cache);
3075
3076 __ load_field_entry(cache, index);
3077 __ bind(L1);
3078 }
3079 }
3080
3081 void TemplateTable::pop_and_check_object(Register r) {
3082 __ pop_ptr(r);
3083 __ null_check(r); // for field access must check obj.
3084 __ verify_oop(r);
3085 }
3086
3087 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3088 transition(vtos, vtos);
3089
3090 const Register obj = LP64_ONLY(r9) NOT_LP64(rcx);
3091 const Register cache = rcx;
3092 const Register index = rdx;
3093 const Register off = rbx;
3094 const Register tos_state = rax;
3095 const Register flags = rdx;
3096 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
3097
3098 resolve_cache_and_index_for_field(byte_no, cache, index);
3099 jvmti_post_field_access(cache, index, is_static, false);
3100 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3101
3102 const Address field(obj, off, Address::times_1, 0*wordSize);
3103
3104 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notInlineType;
3105
3106 // Make sure we don't need to mask edx after the above shift
3107 assert(btos == 0, "change code, btos != 0");
3108 __ testl(tos_state, tos_state);
3109 __ jcc(Assembler::notZero, notByte);
3110
3111 // btos
3112 if (!is_static) pop_and_check_object(obj);
3113 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3114 __ push(btos);
3115 // Rewrite bytecode to be faster
3116 if (!is_static && rc == may_rewrite) {
3117 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3118 }
3119 __ jmp(Done);
3120
3121 __ bind(notByte);
3122 __ cmpl(tos_state, ztos);
3123 __ jcc(Assembler::notEqual, notBool);
3124 if (!is_static) pop_and_check_object(obj);
3125 // ztos (same code as btos)
3126 __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
3127 __ push(ztos);
3128 // Rewrite bytecode to be faster
3129 if (!is_static && rc == may_rewrite) {
3130 // use btos rewriting, no truncating to t/f bit is needed for getfield.
3131 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3132 }
3133 __ jmp(Done);
3134
3135 __ bind(notBool);
3136 __ cmpl(tos_state, atos);
3137 __ jcc(Assembler::notEqual, notObj);
3138 // atos
3139 if (!EnableValhalla) {
3140 if (!is_static) pop_and_check_object(obj);
3141 do_oop_load(_masm, field, rax);
3142 __ push(atos);
3143 if (!is_static && rc == may_rewrite) {
3144 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3145 }
3146 __ jmp(Done);
3147 } else {
3148 if (is_static) {
3149 __ load_heap_oop(rax, field);
3150 Label is_null_free_inline_type, uninitialized;
3151 // Issue below if the static field has not been initialized yet
3152 __ test_field_is_null_free_inline_type(flags, rscratch1, is_null_free_inline_type);
3153 // field is not a null free inline type
3154 __ push(atos);
3155 __ jmp(Done);
3156 // field is a null free inline type, must not return null even if uninitialized
3157 __ bind(is_null_free_inline_type);
3158 __ testptr(rax, rax);
3159 __ jcc(Assembler::zero, uninitialized);
3160 __ push(atos);
3161 __ jmp(Done);
3162 __ bind(uninitialized);
3163 #ifdef _LP64
3164 Label slow_case, finish;
3165 __ movptr(rbx, Address(obj, java_lang_Class::klass_offset()));
3166 __ cmpb(Address(rbx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3167 __ jcc(Assembler::notEqual, slow_case);
3168 __ get_default_value_oop(rbx, rscratch1, rax);
3169 __ jmp(finish);
3170 __ bind(slow_case);
3171 #endif // LP64
3172 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_inline_type_field),
3173 obj, cache);
3174 #ifdef _LP64
3175 __ bind(finish);
3176 #endif // _LP64
3177 __ verify_oop(rax);
3178 __ push(atos);
3179 __ jmp(Done);
3180 } else {
3181 Label is_flat, nonnull, is_inline_type, rewrite_inline, has_null_marker;
3182 __ test_field_is_null_free_inline_type(flags, rscratch1, is_inline_type);
3183 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3184 // field is not a null free inline type
3185 pop_and_check_object(obj);
3186 __ load_heap_oop(rax, field);
3187 __ push(atos);
3188 if (rc == may_rewrite) {
3189 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3190 }
3191 __ jmp(Done);
3192 __ bind(is_inline_type);
3193 __ test_field_is_flat(flags, rscratch1, is_flat);
3194 // field is not flat
3195 pop_and_check_object(obj);
3196 __ load_heap_oop(rax, field);
3197 __ testptr(rax, rax);
3198 __ jcc(Assembler::notZero, nonnull);
3199 __ load_unsigned_short(flags, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
3200 __ movptr(rcx, Address(cache, ResolvedFieldEntry::field_holder_offset()));
3201 __ get_inline_type_field_klass(rcx, flags, rbx);
3202 __ get_default_value_oop(rbx, rcx, rax);
3203 __ bind(nonnull);
3204 __ verify_oop(rax);
3205 __ push(atos);
3206 __ jmp(rewrite_inline);
3207 __ bind(is_flat);
3208 pop_and_check_object(rax);
3209 __ read_flat_field(rcx, rdx, rbx, rax);
3210 __ verify_oop(rax);
3211 __ push(atos);
3212 __ jmp(rewrite_inline);
3213 __ bind(has_null_marker);
3214 pop_and_check_object(rax);
3215 __ load_field_entry(rcx, rbx);
3216 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), rax, rcx);
3217 __ get_vm_result(rax, r15_thread);
3218 __ push(atos);
3219 __ bind(rewrite_inline);
3220 if (rc == may_rewrite) {
3221 patch_bytecode(Bytecodes::_fast_vgetfield, bc, rbx);
3222 }
3223 __ jmp(Done);
3224 }
3225 }
3226
3227 __ bind(notObj);
3228
3229 if (!is_static) pop_and_check_object(obj);
3230
3231 __ cmpl(tos_state, itos);
3232 __ jcc(Assembler::notEqual, notInt);
3233 // itos
3234 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3235 __ push(itos);
3236 // Rewrite bytecode to be faster
3237 if (!is_static && rc == may_rewrite) {
3238 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
3239 }
3240 __ jmp(Done);
3241
3242 __ bind(notInt);
3243 __ cmpl(tos_state, ctos);
3244 __ jcc(Assembler::notEqual, notChar);
3245 // ctos
3246 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3247 __ push(ctos);
3248 // Rewrite bytecode to be faster
3249 if (!is_static && rc == may_rewrite) {
3250 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3251 }
3252 __ jmp(Done);
3253
3254 __ bind(notChar);
3255 __ cmpl(tos_state, stos);
3256 __ jcc(Assembler::notEqual, notShort);
3257 // stos
3258 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3259 __ push(stos);
3260 // Rewrite bytecode to be faster
3261 if (!is_static && rc == may_rewrite) {
3262 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3263 }
3264 __ jmp(Done);
3265
3266 __ bind(notShort);
3267 __ cmpl(tos_state, ltos);
3268 __ jcc(Assembler::notEqual, notLong);
3269 // ltos
3270 // Generate code as if volatile (x86_32). There just aren't enough registers to
3271 // save that information and this code is faster than the test.
3272 __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
3273 __ push(ltos);
3274 // Rewrite bytecode to be faster
3275 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3276 __ jmp(Done);
3277
3278 __ bind(notLong);
3279 __ cmpl(tos_state, ftos);
3280 __ jcc(Assembler::notEqual, notFloat);
3281 // ftos
3282
3283 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3284 __ push(ftos);
3285 // Rewrite bytecode to be faster
3286 if (!is_static && rc == may_rewrite) {
3287 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3288 }
3289 __ jmp(Done);
3290
3291 __ bind(notFloat);
3292 #ifdef ASSERT
3293 Label notDouble;
3294 __ cmpl(tos_state, dtos);
3295 __ jcc(Assembler::notEqual, notDouble);
3296 #endif
3297 // dtos
3298 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3299 __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
3300 __ push(dtos);
3301 // Rewrite bytecode to be faster
3302 if (!is_static && rc == may_rewrite) {
3303 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3304 }
3305 #ifdef ASSERT
3306 __ jmp(Done);
3307
3308 __ bind(notDouble);
3309 __ stop("Bad state");
3310 #endif
3311
3312 __ bind(Done);
3313 // [jk] not needed currently
3314 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3315 // Assembler::LoadStore));
3316 }
3317
3318 void TemplateTable::getfield(int byte_no) {
3319 getfield_or_static(byte_no, false);
3320 }
3321
3322 void TemplateTable::nofast_getfield(int byte_no) {
3323 getfield_or_static(byte_no, false, may_not_rewrite);
3324 }
3325
3326 void TemplateTable::getstatic(int byte_no) {
3327 getfield_or_static(byte_no, true);
3328 }
3329
3330 // The registers cache and index expected to be set before call.
3331 // The function may destroy various registers, just not the cache and index registers.
3332 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3333 // Cache is rcx and index is rdx
3334 const Register entry = LP64_ONLY(c_rarg2) NOT_LP64(rax); // ResolvedFieldEntry
3335 const Register obj = LP64_ONLY(c_rarg1) NOT_LP64(rbx); // Object pointer
3336 const Register value = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // JValue object
3337
3338 if (JvmtiExport::can_post_field_modification()) {
3339 // Check to see if a field modification watch has been set before
3340 // we take the time to call into the VM.
3341 Label L1;
3342 assert_different_registers(cache, obj, rax);
3343 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3344 __ testl(rax, rax);
3345 __ jcc(Assembler::zero, L1);
3346
3347 __ mov(entry, cache);
3348
3349 if (is_static) {
3350 // Life is simple. Null out the object pointer.
3351 __ xorl(obj, obj);
3352
3353 } else {
3354 // Life is harder. The stack holds the value on top, followed by
3355 // the object. We don't know the size of the value, though; it
3356 // could be one or two words depending on its type. As a result,
3357 // we must find the type to determine where the object is.
3358 #ifndef _LP64
3359 Label two_word, valsize_known;
3360 #endif
3361 __ load_unsigned_byte(value, Address(entry, in_bytes(ResolvedFieldEntry::type_offset())));
3362 #ifdef _LP64
3363 __ movptr(obj, at_tos_p1()); // initially assume a one word jvalue
3364 __ cmpl(value, ltos);
3365 __ cmovptr(Assembler::equal,
3366 obj, at_tos_p2()); // ltos (two word jvalue)
3367 __ cmpl(value, dtos);
3368 __ cmovptr(Assembler::equal,
3369 obj, at_tos_p2()); // dtos (two word jvalue)
3370 #else
3371 __ mov(obj, rsp);
3372 __ cmpl(value, ltos);
3373 __ jccb(Assembler::equal, two_word);
3374 __ cmpl(value, dtos);
3375 __ jccb(Assembler::equal, two_word);
3376 __ addptr(obj, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3377 __ jmpb(valsize_known);
3378
3379 __ bind(two_word);
3380 __ addptr(obj, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3381
3382 __ bind(valsize_known);
3383 // setup object pointer
3384 __ movptr(obj, Address(obj, 0));
3385 #endif
3386 }
3387
3388 // object (tos)
3389 __ mov(value, rsp);
3390 // obj: object pointer set up above (null if static)
3391 // cache: field entry pointer
3392 // value: jvalue object on the stack
3393 __ call_VM(noreg,
3394 CAST_FROM_FN_PTR(address,
3395 InterpreterRuntime::post_field_modification),
3396 obj, entry, value);
3397 // Reload field entry
3398 __ load_field_entry(cache, index);
3399 __ bind(L1);
3400 }
3401 }
3402
3403 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3404 transition(vtos, vtos);
3405
3406 const Register obj = rcx;
3407 const Register cache = rcx;
3408 const Register index = rdx;
3409 const Register tos_state = rdx;
3410 const Register off = rbx;
3411 const Register flags = r9;
3412
3413 resolve_cache_and_index_for_field(byte_no, cache, index);
3414 jvmti_post_field_mod(cache, index, is_static);
3415 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3416
3417 // [jk] not needed currently
3418 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3419 // Assembler::StoreStore));
3420
3421 Label notVolatile, Done;
3422
3423 // Check for volatile store
3424 __ movl(rscratch1, flags);
3425 __ andl(rscratch1, (1 << ResolvedFieldEntry::is_volatile_shift));
3426 __ testl(rscratch1, rscratch1);
3427 __ jcc(Assembler::zero, notVolatile);
3428
3429 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state, flags);
3430 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3431 Assembler::StoreStore));
3432 __ jmp(Done);
3433 __ bind(notVolatile);
3434
3435 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state, flags);
3436
3437 __ bind(Done);
3438 }
3439
3440 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3441 Register obj, Register off, Register tos_state, Register flags) {
3442
3443 // field addresses
3444 const Address field(obj, off, Address::times_1, 0*wordSize);
3445 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3446
3447 Label notByte, notBool, notInt, notShort, notChar,
3448 notLong, notFloat, notObj, notInlineType;
3449 Label Done;
3450
3451 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3452
3453 // Test TOS state
3454 __ testl(tos_state, tos_state);
3455 __ jcc(Assembler::notZero, notByte);
3456
3457 // btos
3458 {
3459 __ pop(btos);
3460 if (!is_static) pop_and_check_object(obj);
3461 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3462 if (!is_static && rc == may_rewrite) {
3463 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3464 }
3465 __ jmp(Done);
3466 }
3467
3468 __ bind(notByte);
3469 __ cmpl(tos_state, ztos);
3470 __ jcc(Assembler::notEqual, notBool);
3471
3472 // ztos
3473 {
3474 __ pop(ztos);
3475 if (!is_static) pop_and_check_object(obj);
3476 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3477 if (!is_static && rc == may_rewrite) {
3478 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3479 }
3480 __ jmp(Done);
3481 }
3482
3483 __ bind(notBool);
3484 __ cmpl(tos_state, atos);
3485 __ jcc(Assembler::notEqual, notObj);
3486
3487 // atos
3488 {
3489 if (!EnableValhalla) {
3490 __ pop(atos);
3491 if (!is_static) pop_and_check_object(obj);
3492 // Store into the field
3493 do_oop_store(_masm, field, rax);
3494 if (!is_static && rc == may_rewrite) {
3495 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3496 }
3497 __ jmp(Done);
3498 } else {
3499 __ pop(atos);
3500 if (is_static) {
3501 Label is_inline_type;
3502 __ test_field_is_not_null_free_inline_type(flags, rscratch1, is_inline_type);
3503 __ null_check(rax);
3504 __ bind(is_inline_type);
3505 do_oop_store(_masm, field, rax);
3506 __ jmp(Done);
3507 } else {
3508 Label is_null_free_inline_type, is_flat, has_null_marker,
3509 write_null, rewrite_not_inline, rewrite_inline;
3510 __ test_field_is_null_free_inline_type(flags, rscratch1, is_null_free_inline_type);
3511 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3512 // Not an inline type
3513 pop_and_check_object(obj);
3514 // Store into the field
3515 do_oop_store(_masm, field, rax);
3516 __ bind(rewrite_not_inline);
3517 if (rc == may_rewrite) {
3518 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3519 }
3520 __ jmp(Done);
3521 // Implementation of the inline type semantic
3522 __ bind(is_null_free_inline_type);
3523 __ null_check(rax);
3524 __ test_field_is_flat(flags, rscratch1, is_flat);
3525 // field is not flat
3526 pop_and_check_object(obj);
3527 // Store into the field
3528 do_oop_store(_masm, field, rax);
3529 __ jmp(rewrite_inline);
3530 __ bind(is_flat);
3531 // field is flat
3532 __ load_unsigned_short(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_index_offset())));
3533 __ movptr(r9, Address(rcx, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3534 pop_and_check_object(obj); // obj = rcx
3535 __ load_klass(r8, rax, rscratch1);
3536 __ data_for_oop(rax, rax, r8);
3537 __ addptr(obj, off);
3538 __ inline_layout_info(r9, rdx, rbx);
3539 // because we use InlineLayoutInfo, we need special value access code specialized for fields (arrays will need a different API)
3540 __ flat_field_copy(IN_HEAP, rax, obj, rbx);
3541 __ jmp(rewrite_inline);
3542 __ bind(has_null_marker); // has null marker means the field is flat with a null marker
3543 pop_and_check_object(rbx);
3544 __ load_field_entry(rcx, rdx);
3545 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), rbx, rax, rcx);
3546 __ bind(rewrite_inline);
3547 if (rc == may_rewrite) {
3548 patch_bytecode(Bytecodes::_fast_vputfield, bc, rbx, true, byte_no);
3549 }
3550 __ jmp(Done);
3551 }
3552 }
3553 }
3554
3555 __ bind(notObj);
3556 __ cmpl(tos_state, itos);
3557 __ jcc(Assembler::notEqual, notInt);
3558
3559 // itos
3560 {
3561 __ pop(itos);
3562 if (!is_static) pop_and_check_object(obj);
3563 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3564 if (!is_static && rc == may_rewrite) {
3565 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3566 }
3567 __ jmp(Done);
3568 }
3569
3570 __ bind(notInt);
3571 __ cmpl(tos_state, ctos);
3572 __ jcc(Assembler::notEqual, notChar);
3573
3574 // ctos
3575 {
3576 __ pop(ctos);
3577 if (!is_static) pop_and_check_object(obj);
3578 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3579 if (!is_static && rc == may_rewrite) {
3580 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3581 }
3582 __ jmp(Done);
3583 }
3584
3585 __ bind(notChar);
3586 __ cmpl(tos_state, stos);
3587 __ jcc(Assembler::notEqual, notShort);
3588
3589 // stos
3590 {
3591 __ pop(stos);
3592 if (!is_static) pop_and_check_object(obj);
3593 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3594 if (!is_static && rc == may_rewrite) {
3595 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3596 }
3597 __ jmp(Done);
3598 }
3599
3600 __ bind(notShort);
3601 __ cmpl(tos_state, ltos);
3602 __ jcc(Assembler::notEqual, notLong);
3603
3604 // ltos
3605 {
3606 __ pop(ltos);
3607 if (!is_static) pop_and_check_object(obj);
3608 // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3609 __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg, noreg);
3610 #ifdef _LP64
3611 if (!is_static && rc == may_rewrite) {
3612 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3613 }
3614 #endif // _LP64
3615 __ jmp(Done);
3616 }
3617
3618 __ bind(notLong);
3619 __ cmpl(tos_state, ftos);
3620 __ jcc(Assembler::notEqual, notFloat);
3621
3622 // ftos
3623 {
3624 __ pop(ftos);
3625 if (!is_static) pop_and_check_object(obj);
3626 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3627 if (!is_static && rc == may_rewrite) {
3628 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3629 }
3630 __ jmp(Done);
3631 }
3632
3633 __ bind(notFloat);
3634 #ifdef ASSERT
3635 Label notDouble;
3636 __ cmpl(tos_state, dtos);
3637 __ jcc(Assembler::notEqual, notDouble);
3638 #endif
3639
3640 // dtos
3641 {
3642 __ pop(dtos);
3643 if (!is_static) pop_and_check_object(obj);
3644 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3645 __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg, noreg);
3646 if (!is_static && rc == may_rewrite) {
3647 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3648 }
3649 }
3650
3651 #ifdef ASSERT
3652 __ jmp(Done);
3653
3654 __ bind(notDouble);
3655 __ stop("Bad state");
3656 #endif
3657
3658 __ bind(Done);
3659 }
3660
3661 void TemplateTable::putfield(int byte_no) {
3662 putfield_or_static(byte_no, false);
3663 }
3664
3665 void TemplateTable::nofast_putfield(int byte_no) {
3666 putfield_or_static(byte_no, false, may_not_rewrite);
3667 }
3668
3669 void TemplateTable::putstatic(int byte_no) {
3670 putfield_or_static(byte_no, true);
3671 }
3672
3673 void TemplateTable::jvmti_post_fast_field_mod() {
3674
3675 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3676
3677 if (JvmtiExport::can_post_field_modification()) {
3678 // Check to see if a field modification watch has been set before
3679 // we take the time to call into the VM.
3680 Label L2;
3681 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3682 __ testl(scratch, scratch);
3683 __ jcc(Assembler::zero, L2);
3684 __ pop_ptr(rbx); // copy the object pointer from tos
3685 __ verify_oop(rbx);
3686 __ push_ptr(rbx); // put the object pointer back on tos
3687 // Save tos values before call_VM() clobbers them. Since we have
3688 // to do it for every data type, we use the saved values as the
3689 // jvalue object.
3690 switch (bytecode()) { // load values into the jvalue object
3691 case Bytecodes::_fast_vputfield: //fall through
3692 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3693 case Bytecodes::_fast_bputfield: // fall through
3694 case Bytecodes::_fast_zputfield: // fall through
3695 case Bytecodes::_fast_sputfield: // fall through
3696 case Bytecodes::_fast_cputfield: // fall through
3697 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3698 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3699 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3700 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3701
3702 default:
3703 ShouldNotReachHere();
3704 }
3705 __ mov(scratch, rsp); // points to jvalue on the stack
3706 // access constant pool cache entry
3707 LP64_ONLY(__ load_field_entry(c_rarg2, rax));
3708 NOT_LP64(__ load_field_entry(rax, rdx));
3709 __ verify_oop(rbx);
3710 // rbx: object pointer copied above
3711 // c_rarg2: cache entry pointer
3712 // c_rarg3: jvalue object on the stack
3713 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3714 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3715
3716 switch (bytecode()) { // restore tos values
3717 case Bytecodes::_fast_vputfield: // fall through
3718 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3719 case Bytecodes::_fast_bputfield: // fall through
3720 case Bytecodes::_fast_zputfield: // fall through
3721 case Bytecodes::_fast_sputfield: // fall through
3722 case Bytecodes::_fast_cputfield: // fall through
3723 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3724 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3725 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3726 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3727 default: break;
3728 }
3729 __ bind(L2);
3730 }
3731 }
3732
3733 void TemplateTable::fast_storefield(TosState state) {
3734 transition(state, vtos);
3735
3736 Label notVolatile, Done;
3737
3738 jvmti_post_fast_field_mod();
3739
3740 __ push(rax);
3741 __ load_field_entry(rcx, rax);
3742 load_resolved_field_entry(noreg, rcx, rax, rbx, rdx);
3743 __ pop(rax);
3744 // RBX: field offset, RCX: RAX: TOS, RDX: flags
3745
3746 // Get object from stack
3747 pop_and_check_object(rcx);
3748
3749 // field address
3750 const Address field(rcx, rbx, Address::times_1);
3751
3752 // Check for volatile store
3753 __ movl(rscratch2, rdx); // saving flags for is_flat test
3754 __ andl(rscratch2, (1 << ResolvedFieldEntry::is_volatile_shift));
3755 __ testl(rscratch2, rscratch2);
3756 __ jcc(Assembler::zero, notVolatile);
3757
3758 fast_storefield_helper(field, rax, rdx);
3759 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3760 Assembler::StoreStore));
3761 __ jmp(Done);
3762 __ bind(notVolatile);
3763
3764 fast_storefield_helper(field, rax, rdx);
3765
3766 __ bind(Done);
3767 }
3768
3769 void TemplateTable::fast_storefield_helper(Address field, Register rax, Register flags) {
3770
3771 // DANGER: 'field' argument depends on rcx and rbx
3772
3773 // access field
3774 switch (bytecode()) {
3775 case Bytecodes::_fast_vputfield:
3776 {
3777 Label is_flat, has_null_marker, write_null, done;
3778 __ test_field_has_null_marker(flags, rscratch1, has_null_marker);
3779 // Null free field cases: flat or not flat
3780 __ null_check(rax);
3781 __ test_field_is_flat(flags, rscratch1, is_flat);
3782 // field is not flat
3783 do_oop_store(_masm, field, rax);
3784 __ jmp(done);
3785 __ bind(is_flat);
3786 __ load_field_entry(r8, r9);
3787 __ load_unsigned_short(r9, Address(r8, in_bytes(ResolvedFieldEntry::field_index_offset())));
3788 __ movptr(r8, Address(r8, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3789 __ inline_layout_info(r8, r9, r8);
3790 __ load_klass(rdx, rax, rscratch1);
3791 __ data_for_oop(rax, rax, rdx);
3792 __ lea(rcx, field);
3793 __ flat_field_copy(IN_HEAP, rax, rcx, r8);
3794 __ jmp(done);
3795 __ bind(has_null_marker); // has null marker means the field is flat with a null marker
3796 __ movptr(rbx, rcx);
3797 __ load_field_entry(rcx, rdx);
3798 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), rbx, rax, rcx);
3799 __ bind(done);
3800 }
3801 break;
3802 case Bytecodes::_fast_aputfield:
3803 {
3804 do_oop_store(_masm, field, rax);
3805 }
3806 break;
3807 case Bytecodes::_fast_lputfield:
3808 #ifdef _LP64
3809 __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg, noreg);
3810 #else
3811 __ stop("should not be rewritten");
3812 #endif
3813 break;
3814 case Bytecodes::_fast_iputfield:
3815 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3816 break;
3817 case Bytecodes::_fast_zputfield:
3818 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3819 break;
3820 case Bytecodes::_fast_bputfield:
3821 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3822 break;
3823 case Bytecodes::_fast_sputfield:
3824 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3825 break;
3826 case Bytecodes::_fast_cputfield:
3827 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3828 break;
3829 case Bytecodes::_fast_fputfield:
3830 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg, noreg);
3831 break;
3832 case Bytecodes::_fast_dputfield:
3833 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg, noreg);
3834 break;
3835 default:
3836 ShouldNotReachHere();
3837 }
3838 }
3839
3840 void TemplateTable::fast_accessfield(TosState state) {
3841 transition(atos, state);
3842
3843 // Do the JVMTI work here to avoid disturbing the register state below
3844 if (JvmtiExport::can_post_field_access()) {
3845 // Check to see if a field access watch has been set before we
3846 // take the time to call into the VM.
3847 Label L1;
3848 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3849 __ testl(rcx, rcx);
3850 __ jcc(Assembler::zero, L1);
3851 // access constant pool cache entry
3852 LP64_ONLY(__ load_field_entry(c_rarg2, rcx));
3853 NOT_LP64(__ load_field_entry(rcx, rdx));
3854 __ verify_oop(rax);
3855 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3856 LP64_ONLY(__ mov(c_rarg1, rax));
3857 // c_rarg1: object pointer copied above
3858 // c_rarg2: cache entry pointer
3859 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3860 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3861 __ pop_ptr(rax); // restore object pointer
3862 __ bind(L1);
3863 }
3864
3865 // access constant pool cache
3866 __ load_field_entry(rcx, rbx);
3867 __ load_sized_value(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3868
3869 // rax: object
3870 __ verify_oop(rax);
3871 __ null_check(rax);
3872 Address field(rax, rdx, Address::times_1);
3873
3874 // access field
3875 switch (bytecode()) {
3876 case Bytecodes::_fast_vgetfield:
3877 {
3878 Label is_flat, nonnull, Done, has_null_marker;
3879 __ load_unsigned_byte(rscratch1, Address(rcx, in_bytes(ResolvedFieldEntry::flags_offset())));
3880 __ test_field_has_null_marker(rscratch1, rscratch2, has_null_marker);
3881 __ test_field_is_flat(rscratch1, rscratch2, is_flat);
3882 // field is not flat
3883 __ load_heap_oop(rax, field);
3884 __ testptr(rax, rax);
3885 __ jcc(Assembler::notZero, nonnull);
3886 __ load_unsigned_short(rdx, Address(rcx, in_bytes(ResolvedFieldEntry::field_index_offset())));
3887 __ movptr(rcx, Address(rcx, ResolvedFieldEntry::field_holder_offset()));
3888 __ get_inline_type_field_klass(rcx, rdx, rbx);
3889 __ get_default_value_oop(rbx, rcx, rax);
3890 __ bind(nonnull);
3891 __ verify_oop(rax);
3892 __ jmp(Done);
3893 __ bind(is_flat);
3894 // field is flat
3895 __ read_flat_field(rcx, rdx, rbx, rax);
3896 __ jmp(Done);
3897 __ bind(has_null_marker);
3898 // rax = instance, rcx = resolved entry
3899 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), rax, rcx);
3900 __ get_vm_result(rax, r15_thread);
3901 __ bind(Done);
3902 __ verify_oop(rax);
3903 }
3904 break;
3905 case Bytecodes::_fast_agetfield:
3906 do_oop_load(_masm, field, rax);
3907 __ verify_oop(rax);
3908 break;
3909 case Bytecodes::_fast_lgetfield:
3910 #ifdef _LP64
3911 __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3912 #else
3913 __ stop("should not be rewritten");
3914 #endif
3915 break;
3916 case Bytecodes::_fast_igetfield:
3917 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3918 break;
3919 case Bytecodes::_fast_bgetfield:
3920 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3921 break;
3922 case Bytecodes::_fast_sgetfield:
3923 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3924 break;
3925 case Bytecodes::_fast_cgetfield:
3926 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3927 break;
3928 case Bytecodes::_fast_fgetfield:
3929 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3930 break;
3931 case Bytecodes::_fast_dgetfield:
3932 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3933 break;
3934 default:
3935 ShouldNotReachHere();
3936 }
3937 // [jk] not needed currently
3938 // Label notVolatile;
3939 // __ testl(rdx, rdx);
3940 // __ jcc(Assembler::zero, notVolatile);
3941 // __ membar(Assembler::LoadLoad);
3942 // __ bind(notVolatile);
3943 }
3944
3945 void TemplateTable::fast_xaccess(TosState state) {
3946 transition(vtos, state);
3947
3948 // get receiver
3949 __ movptr(rax, aaddress(0));
3950 // access constant pool cache
3951 __ load_field_entry(rcx, rdx, 2);
3952 __ load_sized_value(rbx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3953
3954 // make sure exception is reported in correct bcp range (getfield is
3955 // next instruction)
3956 __ increment(rbcp);
3957 __ null_check(rax);
3958 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3959 switch (state) {
3960 case itos:
3961 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3962 break;
3963 case atos:
3964 do_oop_load(_masm, field, rax);
3965 __ verify_oop(rax);
3966 break;
3967 case ftos:
3968 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3969 break;
3970 default:
3971 ShouldNotReachHere();
3972 }
3973
3974 // [jk] not needed currently
3975 // Label notVolatile;
3976 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3977 // in_bytes(ConstantPoolCache::base_offset() +
3978 // ConstantPoolCacheEntry::flags_offset())));
3979 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3980 // __ testl(rdx, 0x1);
3981 // __ jcc(Assembler::zero, notVolatile);
3982 // __ membar(Assembler::LoadLoad);
3983 // __ bind(notVolatile);
3984
3985 __ decrement(rbcp);
3986 }
3987
3988 //-----------------------------------------------------------------------------
3989 // Calls
3990
3991 void TemplateTable::prepare_invoke(Register cache, Register recv, Register flags) {
3992 // determine flags
3993 const Bytecodes::Code code = bytecode();
3994 const bool load_receiver = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3995 assert_different_registers(recv, flags);
3996
3997 // save 'interpreter return address'
3998 __ save_bcp();
3999
4000 // Save flags and load TOS
4001 __ movl(rbcp, flags);
4002 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
4003
4004 // load receiver if needed (after appendix is pushed so parameter size is correct)
4005 // Note: no return address pushed yet
4006 if (load_receiver) {
4007 __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
4008 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
4009 const int receiver_is_at_end = -1; // back off one slot to get receiver
4010 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
4011 __ movptr(recv, recv_addr);
4012 __ verify_oop(recv);
4013 }
4014
4015 // load return address
4016 {
4017 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
4018 ExternalAddress table(table_addr);
4019 #ifdef _LP64
4020 __ lea(rscratch1, table);
4021 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
4022 #else
4023 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
4024 #endif // _LP64
4025 }
4026
4027 // push return address
4028 __ push(flags);
4029
4030 // Restore flags value from the constant pool cache entry, and restore rsi
4031 // for later null checks. r13 is the bytecode pointer
4032 __ movl(flags, rbcp);
4033 __ restore_bcp();
4034 }
4035
4036 void TemplateTable::invokevirtual_helper(Register index,
4037 Register recv,
4038 Register flags) {
4039 // Uses temporary registers rax, rdx
4040 assert_different_registers(index, recv, rax, rdx);
4041 assert(index == rbx, "");
4042 assert(recv == rcx, "");
4043
4044 // Test for an invoke of a final method
4045 Label notFinal;
4046 __ movl(rax, flags);
4047 __ andl(rax, (1 << ResolvedMethodEntry::is_vfinal_shift));
4048 __ jcc(Assembler::zero, notFinal);
4049
4050 const Register method = index; // method must be rbx
4051 assert(method == rbx,
4052 "Method* must be rbx for interpreter calling convention");
4053
4054 // do the call - the index is actually the method to call
4055 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
4056
4057 // It's final, need a null check here!
4058 __ null_check(recv);
4059
4060 // profile this call
4061 __ profile_final_call(rax);
4062 __ profile_arguments_type(rax, method, rbcp, true);
4063
4064 __ jump_from_interpreted(method, rax);
4065
4066 __ bind(notFinal);
4067
4068 // get receiver klass
4069 __ load_klass(rax, recv, rscratch1);
4070
4071 // profile this call
4072 __ profile_virtual_call(rax, rlocals, rdx);
4073 // get target Method* & entry point
4074 __ lookup_virtual_method(rax, index, method);
4075
4076 __ profile_arguments_type(rdx, method, rbcp, true);
4077 __ jump_from_interpreted(method, rdx);
4078 }
4079
4080 void TemplateTable::invokevirtual(int byte_no) {
4081 transition(vtos, vtos);
4082 assert(byte_no == f2_byte, "use this argument");
4083
4084 load_resolved_method_entry_virtual(rcx, // ResolvedMethodEntry*
4085 rbx, // Method or itable index
4086 rdx); // Flags
4087 prepare_invoke(rcx, // ResolvedMethodEntry*
4088 rcx, // Receiver
4089 rdx); // flags
4090
4091 // rbx: index
4092 // rcx: receiver
4093 // rdx: flags
4094 invokevirtual_helper(rbx, rcx, rdx);
4095 }
4096
4097 void TemplateTable::invokespecial(int byte_no) {
4098 transition(vtos, vtos);
4099 assert(byte_no == f1_byte, "use this argument");
4100
4101 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
4102 rbx, // Method*
4103 rdx); // flags
4104 prepare_invoke(rcx,
4105 rcx, // get receiver also for null check
4106 rdx); // flags
4107
4108 __ verify_oop(rcx);
4109 __ null_check(rcx);
4110 // do the call
4111 __ profile_call(rax);
4112 __ profile_arguments_type(rax, rbx, rbcp, false);
4113 __ jump_from_interpreted(rbx, rax);
4114 }
4115
4116 void TemplateTable::invokestatic(int byte_no) {
4117 transition(vtos, vtos);
4118 assert(byte_no == f1_byte, "use this argument");
4119
4120 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
4121 rbx, // Method*
4122 rdx // flags
4123 );
4124 prepare_invoke(rcx, rcx, rdx); // cache and flags
4125
4126 // do the call
4127 __ profile_call(rax);
4128 __ profile_arguments_type(rax, rbx, rbcp, false);
4129 __ jump_from_interpreted(rbx, rax);
4130 }
4131
4132
4133 void TemplateTable::fast_invokevfinal(int byte_no) {
4134 transition(vtos, vtos);
4135 assert(byte_no == f2_byte, "use this argument");
4136 __ stop("fast_invokevfinal not used on x86");
4137 }
4138
4139
4140 void TemplateTable::invokeinterface(int byte_no) {
4141 transition(vtos, vtos);
4142 assert(byte_no == f1_byte, "use this argument");
4143
4144 load_resolved_method_entry_interface(rcx, // ResolvedMethodEntry*
4145 rax, // Klass*
4146 rbx, // Method* or itable/vtable index
4147 rdx); // flags
4148 prepare_invoke(rcx, rcx, rdx); // receiver, flags
4149
4150 // First check for Object case, then private interface method,
4151 // then regular interface method.
4152
4153 // Special case of invokeinterface called for virtual method of
4154 // java.lang.Object. See cpCache.cpp for details.
4155 Label notObjectMethod;
4156 __ movl(rlocals, rdx);
4157 __ andl(rlocals, (1 << ResolvedMethodEntry::is_forced_virtual_shift));
4158 __ jcc(Assembler::zero, notObjectMethod);
4159
4160 invokevirtual_helper(rbx, rcx, rdx);
4161 // no return from above
4162 __ bind(notObjectMethod);
4163
4164 Label no_such_interface; // for receiver subtype check
4165 Register recvKlass; // used for exception processing
4166
4167 // Check for private method invocation - indicated by vfinal
4168 Label notVFinal;
4169 __ movl(rlocals, rdx);
4170 __ andl(rlocals, (1 << ResolvedMethodEntry::is_vfinal_shift));
4171 __ jcc(Assembler::zero, notVFinal);
4172
4173 // Get receiver klass into rlocals - also a null check
4174 __ load_klass(rlocals, rcx, rscratch1);
4175
4176 Label subtype;
4177 __ check_klass_subtype(rlocals, rax, rbcp, subtype);
4178 // If we get here the typecheck failed
4179 recvKlass = rdx;
4180 __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
4181 __ jmp(no_such_interface);
4182
4183 __ bind(subtype);
4184
4185 // do the call - rbx is actually the method to call
4186
4187 __ profile_final_call(rdx);
4188 __ profile_arguments_type(rdx, rbx, rbcp, true);
4189
4190 __ jump_from_interpreted(rbx, rdx);
4191 // no return from above
4192 __ bind(notVFinal);
4193
4194 // Get receiver klass into rdx - also a null check
4195 __ restore_locals(); // restore r14
4196 __ load_klass(rdx, rcx, rscratch1);
4197
4198 Label no_such_method;
4199
4200 // Preserve method for throw_AbstractMethodErrorVerbose.
4201 __ mov(rcx, rbx);
4202 // Receiver subtype check against REFC.
4203 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
4204 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4205 rdx, rax, noreg,
4206 // outputs: scan temp. reg, scan temp. reg
4207 rbcp, rlocals,
4208 no_such_interface,
4209 /*return_method=*/false);
4210
4211 // profile this call
4212 __ restore_bcp(); // rbcp was destroyed by receiver type check
4213 __ profile_virtual_call(rdx, rbcp, rlocals);
4214
4215 // Get declaring interface class from method, and itable index
4216 __ load_method_holder(rax, rbx);
4217 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
4218 __ subl(rbx, Method::itable_index_max);
4219 __ negl(rbx);
4220
4221 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
4222 __ mov(rlocals, rdx);
4223 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4224 rlocals, rax, rbx,
4225 // outputs: method, scan temp. reg
4226 rbx, rbcp,
4227 no_such_interface);
4228
4229 // rbx: Method* to call
4230 // rcx: receiver
4231 // Check for abstract method error
4232 // Note: This should be done more efficiently via a throw_abstract_method_error
4233 // interpreter entry point and a conditional jump to it in case of a null
4234 // method.
4235 __ testptr(rbx, rbx);
4236 __ jcc(Assembler::zero, no_such_method);
4237
4238 __ profile_arguments_type(rdx, rbx, rbcp, true);
4239
4240 // do the call
4241 // rcx: receiver
4242 // rbx,: Method*
4243 __ jump_from_interpreted(rbx, rdx);
4244 __ should_not_reach_here();
4245
4246 // exception handling code follows...
4247 // note: must restore interpreter registers to canonical
4248 // state for exception handling to work correctly!
4249
4250 __ bind(no_such_method);
4251 // throw exception
4252 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4253 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4254 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4255 // Pass arguments for generating a verbose error message.
4256 #ifdef _LP64
4257 recvKlass = c_rarg1;
4258 Register method = c_rarg2;
4259 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
4260 if (method != rcx) { __ movq(method, rcx); }
4261 #else
4262 recvKlass = rdx;
4263 Register method = rcx;
4264 #endif
4265 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
4266 recvKlass, method);
4267 // The call_VM checks for exception, so we should never return here.
4268 __ should_not_reach_here();
4269
4270 __ bind(no_such_interface);
4271 // throw exception
4272 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4273 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4274 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4275 // Pass arguments for generating a verbose error message.
4276 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
4277 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
4278 recvKlass, rax);
4279 // the call_VM checks for exception, so we should never return here.
4280 __ should_not_reach_here();
4281 }
4282
4283 void TemplateTable::invokehandle(int byte_no) {
4284 transition(vtos, vtos);
4285 assert(byte_no == f1_byte, "use this argument");
4286 const Register rbx_method = rbx;
4287 const Register rax_mtype = rax;
4288 const Register rcx_recv = rcx;
4289 const Register rdx_flags = rdx;
4290
4291 load_resolved_method_entry_handle(rcx, rbx_method, rax_mtype, rdx_flags);
4292 prepare_invoke(rcx, rcx_recv, rdx_flags);
4293
4294 __ verify_method_ptr(rbx_method);
4295 __ verify_oop(rcx_recv);
4296 __ null_check(rcx_recv);
4297
4298 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4299 // rbx: MH.invokeExact_MT method
4300
4301 // Note: rax_mtype is already pushed (if necessary)
4302
4303 // FIXME: profile the LambdaForm also
4304 __ profile_final_call(rax);
4305 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4306
4307 __ jump_from_interpreted(rbx_method, rdx);
4308 }
4309
4310 void TemplateTable::invokedynamic(int byte_no) {
4311 transition(vtos, vtos);
4312 assert(byte_no == f1_byte, "use this argument");
4313
4314 const Register rbx_method = rbx;
4315 const Register rax_callsite = rax;
4316
4317 load_invokedynamic_entry(rbx_method);
4318 // rax: CallSite object (from cpool->resolved_references[])
4319 // rbx: MH.linkToCallSite method
4320
4321 // Note: rax_callsite is already pushed
4322
4323 // %%% should make a type profile for any invokedynamic that takes a ref argument
4324 // profile this call
4325 __ profile_call(rbcp);
4326 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4327
4328 __ verify_oop(rax_callsite);
4329
4330 __ jump_from_interpreted(rbx_method, rdx);
4331 }
4332
4333 //-----------------------------------------------------------------------------
4334 // Allocation
4335
4336 void TemplateTable::_new() {
4337 transition(vtos, atos);
4338 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4339 Label slow_case;
4340 Label done;
4341
4342 __ get_cpool_and_tags(rcx, rax);
4343
4344 // Make sure the class we're about to instantiate has been resolved.
4345 // This is done before loading InstanceKlass to be consistent with the order
4346 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4347 const int tags_offset = Array<u1>::base_offset_in_bytes();
4348 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4349 __ jcc(Assembler::notEqual, slow_case);
4350
4351 // get InstanceKlass
4352 __ load_resolved_klass_at_index(rcx, rcx, rdx);
4353
4354 // make sure klass is initialized
4355 // init_state needs acquire, but x86 is TSO, and so we are already good.
4356 #ifdef _LP64
4357 assert(VM_Version::supports_fast_class_init_checks(), "must support fast class initialization checks");
4358 __ clinit_barrier(rcx, r15_thread, nullptr /*L_fast_path*/, &slow_case);
4359 #else
4360 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4361 __ jcc(Assembler::notEqual, slow_case);
4362 #endif
4363
4364 __ allocate_instance(rcx, rax, rdx, rbx, true, slow_case);
4365 if (DTraceAllocProbes) {
4366 // Trigger dtrace event for fastpath
4367 __ push(atos);
4368 __ call_VM_leaf(
4369 CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), rax);
4370 __ pop(atos);
4371 }
4372 __ jmp(done);
4373
4374 // slow case
4375 __ bind(slow_case);
4376
4377 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4378 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4379
4380 __ get_constant_pool(rarg1);
4381 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4382 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4383 __ verify_oop(rax);
4384
4385 // continue
4386 __ bind(done);
4387 }
4388
4389 void TemplateTable::newarray() {
4390 transition(itos, atos);
4391 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4392 __ load_unsigned_byte(rarg1, at_bcp(1));
4393 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4394 rarg1, rax);
4395 }
4396
4397 void TemplateTable::anewarray() {
4398 transition(itos, atos);
4399
4400 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4401 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4402
4403 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4404 __ get_constant_pool(rarg1);
4405 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4406 rarg1, rarg2, rax);
4407 }
4408
4409 void TemplateTable::arraylength() {
4410 transition(atos, itos);
4411 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4412 }
4413
4414 void TemplateTable::checkcast() {
4415 transition(atos, atos);
4416 Label done, is_null, ok_is_subtype, quicked, resolved;
4417 __ testptr(rax, rax); // object is in rax
4418 __ jcc(Assembler::zero, is_null);
4419
4420 // Get cpool & tags index
4421 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4422 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4423 // See if bytecode has already been quicked
4424 __ movzbl(rdx, Address(rdx, rbx,
4425 Address::times_1,
4426 Array<u1>::base_offset_in_bytes()));
4427 __ cmpl(rdx, JVM_CONSTANT_Class);
4428 __ jcc(Assembler::equal, quicked);
4429 __ push(atos); // save receiver for result, and for GC
4430 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4431
4432 // vm_result_2 has metadata result
4433 #ifndef _LP64
4434 // borrow rdi from locals
4435 __ get_thread(rdi);
4436 __ get_vm_result_2(rax, rdi);
4437 __ restore_locals();
4438 #else
4439 __ get_vm_result_2(rax, r15_thread);
4440 #endif
4441
4442 __ pop_ptr(rdx); // restore receiver
4443 __ jmpb(resolved);
4444
4445 // Get superklass in rax and subklass in rbx
4446 __ bind(quicked);
4447 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4448 __ load_resolved_klass_at_index(rax, rcx, rbx);
4449
4450 __ bind(resolved);
4451 __ load_klass(rbx, rdx, rscratch1);
4452
4453 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4454 // Superklass in rax. Subklass in rbx.
4455 __ gen_subtype_check(rbx, ok_is_subtype);
4456
4457 // Come here on failure
4458 __ push_ptr(rdx);
4459 // object is at TOS
4460 __ jump(RuntimeAddress(Interpreter::_throw_ClassCastException_entry));
4461
4462 // Come here on success
4463 __ bind(ok_is_subtype);
4464 __ mov(rax, rdx); // Restore object in rdx
4465 __ jmp(done);
4466
4467 __ bind(is_null);
4468
4469 // Collect counts on whether this check-cast sees nulls a lot or not.
4470 if (ProfileInterpreter) {
4471 __ profile_null_seen(rcx);
4472 }
4473
4474 __ bind(done);
4475 }
4476
4477 void TemplateTable::instanceof() {
4478 transition(atos, itos);
4479 Label done, is_null, ok_is_subtype, quicked, resolved;
4480 __ testptr(rax, rax);
4481 __ jcc(Assembler::zero, is_null);
4482
4483 // Get cpool & tags index
4484 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4485 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4486 // See if bytecode has already been quicked
4487 __ movzbl(rdx, Address(rdx, rbx,
4488 Address::times_1,
4489 Array<u1>::base_offset_in_bytes()));
4490 __ cmpl(rdx, JVM_CONSTANT_Class);
4491 __ jcc(Assembler::equal, quicked);
4492
4493 __ push(atos); // save receiver for result, and for GC
4494 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4495 // vm_result_2 has metadata result
4496
4497 #ifndef _LP64
4498 // borrow rdi from locals
4499 __ get_thread(rdi);
4500 __ get_vm_result_2(rax, rdi);
4501 __ restore_locals();
4502 #else
4503 __ get_vm_result_2(rax, r15_thread);
4504 #endif
4505
4506 __ pop_ptr(rdx); // restore receiver
4507 __ verify_oop(rdx);
4508 __ load_klass(rdx, rdx, rscratch1);
4509 __ jmpb(resolved);
4510
4511 // Get superklass in rax and subklass in rdx
4512 __ bind(quicked);
4513 __ load_klass(rdx, rax, rscratch1);
4514 __ load_resolved_klass_at_index(rax, rcx, rbx);
4515
4516 __ bind(resolved);
4517
4518 // Generate subtype check. Blows rcx, rdi
4519 // Superklass in rax. Subklass in rdx.
4520 __ gen_subtype_check(rdx, ok_is_subtype);
4521
4522 // Come here on failure
4523 __ xorl(rax, rax);
4524 __ jmpb(done);
4525 // Come here on success
4526 __ bind(ok_is_subtype);
4527 __ movl(rax, 1);
4528
4529 // Collect counts on whether this test sees nulls a lot or not.
4530 if (ProfileInterpreter) {
4531 __ jmp(done);
4532 __ bind(is_null);
4533 __ profile_null_seen(rcx);
4534 } else {
4535 __ bind(is_null); // same as 'done'
4536 }
4537 __ bind(done);
4538 // rax = 0: obj == nullptr or obj is not an instanceof the specified klass
4539 // rax = 1: obj != nullptr and obj is an instanceof the specified klass
4540 }
4541
4542 //----------------------------------------------------------------------------------------------------
4543 // Breakpoints
4544 void TemplateTable::_breakpoint() {
4545 // Note: We get here even if we are single stepping..
4546 // jbug insists on setting breakpoints at every bytecode
4547 // even if we are in single step mode.
4548
4549 transition(vtos, vtos);
4550
4551 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4552
4553 // get the unpatched byte code
4554 __ get_method(rarg);
4555 __ call_VM(noreg,
4556 CAST_FROM_FN_PTR(address,
4557 InterpreterRuntime::get_original_bytecode_at),
4558 rarg, rbcp);
4559 __ mov(rbx, rax); // why?
4560
4561 // post the breakpoint event
4562 __ get_method(rarg);
4563 __ call_VM(noreg,
4564 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4565 rarg, rbcp);
4566
4567 // complete the execution of original bytecode
4568 __ dispatch_only_normal(vtos);
4569 }
4570
4571 //-----------------------------------------------------------------------------
4572 // Exceptions
4573
4574 void TemplateTable::athrow() {
4575 transition(atos, vtos);
4576 __ null_check(rax);
4577 __ jump(RuntimeAddress(Interpreter::throw_exception_entry()));
4578 }
4579
4580 //-----------------------------------------------------------------------------
4581 // Synchronization
4582 //
4583 // Note: monitorenter & exit are symmetric routines; which is reflected
4584 // in the assembly code structure as well
4585 //
4586 // Stack layout:
4587 //
4588 // [expressions ] <--- rsp = expression stack top
4589 // ..
4590 // [expressions ]
4591 // [monitor entry] <--- monitor block top = expression stack bot
4592 // ..
4593 // [monitor entry]
4594 // [frame data ] <--- monitor block bot
4595 // ...
4596 // [saved rbp ] <--- rbp
4597 void TemplateTable::monitorenter() {
4598 transition(atos, vtos);
4599
4600 // check for null object
4601 __ null_check(rax);
4602
4603 Label is_inline_type;
4604 __ movptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
4605 __ test_markword_is_inline_type(rbx, is_inline_type);
4606
4607 const Address monitor_block_top(
4608 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4609 const Address monitor_block_bot(
4610 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4611 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4612
4613 Label allocated;
4614
4615 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4616 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4617 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4618
4619 // initialize entry pointer
4620 __ xorl(rmon, rmon); // points to free slot or null
4621
4622 // find a free slot in the monitor block (result in rmon)
4623 {
4624 Label entry, loop, exit;
4625 __ movptr(rtop, monitor_block_top); // derelativize pointer
4626 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4627 // rtop points to current entry, starting with top-most entry
4628
4629 __ lea(rbot, monitor_block_bot); // points to word before bottom
4630 // of monitor block
4631 __ jmpb(entry);
4632
4633 __ bind(loop);
4634 // check if current entry is used
4635 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset()), NULL_WORD);
4636 // if not used then remember entry in rmon
4637 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4638 // check if current entry is for same object
4639 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4640 // if same object then stop searching
4641 __ jccb(Assembler::equal, exit);
4642 // otherwise advance to next entry
4643 __ addptr(rtop, entry_size);
4644 __ bind(entry);
4645 // check if bottom reached
4646 __ cmpptr(rtop, rbot);
4647 // if not at bottom then check this entry
4648 __ jcc(Assembler::notEqual, loop);
4649 __ bind(exit);
4650 }
4651
4652 __ testptr(rmon, rmon); // check if a slot has been found
4653 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4654
4655 // allocate one if there's no free slot
4656 {
4657 Label entry, loop;
4658 // 1. compute new pointers // rsp: old expression stack top
4659 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4660 __ lea(rmon, Address(rbp, rmon, Address::times_ptr));
4661 __ subptr(rsp, entry_size); // move expression stack top
4662 __ subptr(rmon, entry_size); // move expression stack bottom
4663 __ mov(rtop, rsp); // set start value for copy loop
4664 __ subptr(monitor_block_bot, entry_size / wordSize); // set new monitor block bottom
4665 __ jmp(entry);
4666 // 2. move expression stack contents
4667 __ bind(loop);
4668 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4669 // word from old location
4670 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4671 __ addptr(rtop, wordSize); // advance to next word
4672 __ bind(entry);
4673 __ cmpptr(rtop, rmon); // check if bottom reached
4674 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4675 // copy next word
4676 }
4677
4678 // call run-time routine
4679 // rmon: points to monitor entry
4680 __ bind(allocated);
4681
4682 // Increment bcp to point to the next bytecode, so exception
4683 // handling for async. exceptions work correctly.
4684 // The object has already been popped from the stack, so the
4685 // expression stack looks correct.
4686 __ increment(rbcp);
4687
4688 // store object
4689 __ movptr(Address(rmon, BasicObjectLock::obj_offset()), rax);
4690 __ lock_object(rmon);
4691
4692 // check to make sure this monitor doesn't cause stack overflow after locking
4693 __ save_bcp(); // in case of exception
4694 __ generate_stack_overflow_check(0);
4695
4696 // The bcp has already been incremented. Just need to dispatch to
4697 // next instruction.
4698 __ dispatch_next(vtos);
4699
4700 __ bind(is_inline_type);
4701 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4702 InterpreterRuntime::throw_identity_exception), rax);
4703 __ should_not_reach_here();
4704 }
4705
4706 void TemplateTable::monitorexit() {
4707 transition(atos, vtos);
4708
4709 // check for null object
4710 __ null_check(rax);
4711
4712 const int is_inline_type_mask = markWord::inline_type_pattern;
4713 Label has_identity;
4714 __ movptr(rbx, Address(rax, oopDesc::mark_offset_in_bytes()));
4715 __ andptr(rbx, is_inline_type_mask);
4716 __ cmpl(rbx, is_inline_type_mask);
4717 __ jcc(Assembler::notEqual, has_identity);
4718 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4719 InterpreterRuntime::throw_illegal_monitor_state_exception));
4720 __ should_not_reach_here();
4721 __ bind(has_identity);
4722
4723 const Address monitor_block_top(
4724 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4725 const Address monitor_block_bot(
4726 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4727 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4728
4729 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4730 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4731
4732 Label found;
4733
4734 // find matching slot
4735 {
4736 Label entry, loop;
4737 __ movptr(rtop, monitor_block_top); // derelativize pointer
4738 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4739 // rtop points to current entry, starting with top-most entry
4740
4741 __ lea(rbot, monitor_block_bot); // points to word before bottom
4742 // of monitor block
4743 __ jmpb(entry);
4744
4745 __ bind(loop);
4746 // check if current entry is for same object
4747 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4748 // if same object then stop searching
4749 __ jcc(Assembler::equal, found);
4750 // otherwise advance to next entry
4751 __ addptr(rtop, entry_size);
4752 __ bind(entry);
4753 // check if bottom reached
4754 __ cmpptr(rtop, rbot);
4755 // if not at bottom then check this entry
4756 __ jcc(Assembler::notEqual, loop);
4757 }
4758
4759 // error handling. Unlocking was not block-structured
4760 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4761 InterpreterRuntime::throw_illegal_monitor_state_exception));
4762 __ should_not_reach_here();
4763
4764 // call run-time routine
4765 __ bind(found);
4766 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4767 __ unlock_object(rtop);
4768 __ pop_ptr(rax); // discard object
4769 }
4770
4771 // Wide instructions
4772 void TemplateTable::wide() {
4773 transition(vtos, vtos);
4774 __ load_unsigned_byte(rbx, at_bcp(1));
4775 ExternalAddress wtable((address)Interpreter::_wentry_point);
4776 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)), rscratch1);
4777 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4778 }
4779
4780 // Multi arrays
4781 void TemplateTable::multianewarray() {
4782 transition(vtos, atos);
4783
4784 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4785 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4786 // last dim is on top of stack; we want address of first one:
4787 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4788 // the latter wordSize to point to the beginning of the array.
4789 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4790 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4791 __ load_unsigned_byte(rbx, at_bcp(3));
4792 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4793 }