1 /*
2 * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "compiler/disassembler.hpp"
27 #include "gc/shared/collectedHeap.hpp"
28 #include "gc/shared/gc_globals.hpp"
29 #include "gc/shared/tlab_globals.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "memory/universe.hpp"
35 #include "oops/methodCounters.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "oops/resolvedFieldEntry.hpp"
40 #include "oops/resolvedIndyEntry.hpp"
41 #include "oops/resolvedMethodEntry.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "prims/methodHandles.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/safepointMechanism.hpp"
46 #include "runtime/sharedRuntime.hpp"
47 #include "runtime/stubRoutines.hpp"
48 #include "runtime/synchronizer.hpp"
49 #include "utilities/macros.hpp"
50
51 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
52
53 // Global Register Names
54 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
55 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
56
57 // Address Computation: local variables
58 static inline Address iaddress(int n) {
59 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
60 }
61
62 static inline Address laddress(int n) {
63 return iaddress(n + 1);
64 }
65
66 #ifndef _LP64
67 static inline Address haddress(int n) {
68 return iaddress(n + 0);
69 }
70 #endif
71
72 static inline Address faddress(int n) {
73 return iaddress(n);
74 }
75
76 static inline Address daddress(int n) {
77 return laddress(n);
78 }
79
80 static inline Address aaddress(int n) {
81 return iaddress(n);
82 }
83
84 static inline Address iaddress(Register r) {
85 return Address(rlocals, r, Address::times_ptr);
86 }
87
88 static inline Address laddress(Register r) {
89 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
90 }
91
92 #ifndef _LP64
93 static inline Address haddress(Register r) {
94 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
95 }
96 #endif
97
98 static inline Address faddress(Register r) {
99 return iaddress(r);
100 }
101
102 static inline Address daddress(Register r) {
103 return laddress(r);
104 }
105
106 static inline Address aaddress(Register r) {
107 return iaddress(r);
108 }
109
110
111 // expression stack
112 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
113 // data beyond the rsp which is potentially unsafe in an MT environment;
114 // an interrupt may overwrite that data.)
115 static inline Address at_rsp () {
116 return Address(rsp, 0);
117 }
118
119 // At top of Java expression stack which may be different than esp(). It
120 // isn't for category 1 objects.
121 static inline Address at_tos () {
122 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
123 }
124
125 static inline Address at_tos_p1() {
126 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
127 }
128
129 static inline Address at_tos_p2() {
130 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
131 }
132
133 // Condition conversion
134 static Assembler::Condition j_not(TemplateTable::Condition cc) {
135 switch (cc) {
136 case TemplateTable::equal : return Assembler::notEqual;
137 case TemplateTable::not_equal : return Assembler::equal;
138 case TemplateTable::less : return Assembler::greaterEqual;
139 case TemplateTable::less_equal : return Assembler::greater;
140 case TemplateTable::greater : return Assembler::lessEqual;
141 case TemplateTable::greater_equal: return Assembler::less;
142 }
143 ShouldNotReachHere();
144 return Assembler::zero;
145 }
146
147
148
149 // Miscellaneous helper routines
150 // Store an oop (or null) at the address described by obj.
151 // If val == noreg this means store a null
152
153
154 static void do_oop_store(InterpreterMacroAssembler* _masm,
155 Address dst,
156 Register val,
157 DecoratorSet decorators = 0) {
158 assert(val == noreg || val == rax, "parameter is just for looks");
159 __ store_heap_oop(dst, val,
160 NOT_LP64(rdx) LP64_ONLY(rscratch2),
161 NOT_LP64(rbx) LP64_ONLY(r9),
162 NOT_LP64(rsi) LP64_ONLY(r8), decorators);
163 }
164
165 static void do_oop_load(InterpreterMacroAssembler* _masm,
166 Address src,
167 Register dst,
168 DecoratorSet decorators = 0) {
169 __ load_heap_oop(dst, src, rdx, rbx, decorators);
170 }
171
172 Address TemplateTable::at_bcp(int offset) {
173 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
174 return Address(rbcp, offset);
175 }
176
177
178 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
179 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
180 int byte_no) {
181 if (!RewriteBytecodes) return;
182 Label L_patch_done;
183
184 switch (bc) {
185 case Bytecodes::_fast_aputfield:
186 case Bytecodes::_fast_bputfield:
187 case Bytecodes::_fast_zputfield:
188 case Bytecodes::_fast_cputfield:
189 case Bytecodes::_fast_dputfield:
190 case Bytecodes::_fast_fputfield:
191 case Bytecodes::_fast_iputfield:
192 case Bytecodes::_fast_lputfield:
193 case Bytecodes::_fast_sputfield:
194 {
195 // We skip bytecode quickening for putfield instructions when
196 // the put_code written to the constant pool cache is zero.
197 // This is required so that every execution of this instruction
198 // calls out to InterpreterRuntime::resolve_get_put to do
199 // additional, required work.
200 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
201 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
202 __ load_field_entry(temp_reg, bc_reg);
203 if (byte_no == f1_byte) {
204 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::get_code_offset())));
205 } else {
206 __ load_unsigned_byte(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::put_code_offset())));
207 }
208
209 __ movl(bc_reg, bc);
210 __ cmpl(temp_reg, (int) 0);
211 __ jcc(Assembler::zero, L_patch_done); // don't patch
212 }
213 break;
214 default:
215 assert(byte_no == -1, "sanity");
216 // the pair bytecodes have already done the load.
217 if (load_bc_into_bc_reg) {
218 __ movl(bc_reg, bc);
219 }
220 }
221
222 if (JvmtiExport::can_post_breakpoint()) {
223 Label L_fast_patch;
224 // if a breakpoint is present we can't rewrite the stream directly
225 __ movzbl(temp_reg, at_bcp(0));
226 __ cmpl(temp_reg, Bytecodes::_breakpoint);
227 __ jcc(Assembler::notEqual, L_fast_patch);
228 __ get_method(temp_reg);
229 // Let breakpoint table handling rewrite to quicker bytecode
230 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
231 #ifndef ASSERT
232 __ jmpb(L_patch_done);
233 #else
234 __ jmp(L_patch_done);
235 #endif
236 __ bind(L_fast_patch);
237 }
238
239 #ifdef ASSERT
240 Label L_okay;
241 __ load_unsigned_byte(temp_reg, at_bcp(0));
242 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
243 __ jcc(Assembler::equal, L_okay);
244 __ cmpl(temp_reg, bc_reg);
245 __ jcc(Assembler::equal, L_okay);
246 __ stop("patching the wrong bytecode");
247 __ bind(L_okay);
248 #endif
249
250 // patch bytecode
251 __ movb(at_bcp(0), bc_reg);
252 __ bind(L_patch_done);
253 }
254 // Individual instructions
255
256
257 void TemplateTable::nop() {
258 transition(vtos, vtos);
259 // nothing to do
260 }
261
262 void TemplateTable::shouldnotreachhere() {
263 transition(vtos, vtos);
264 __ stop("shouldnotreachhere bytecode");
265 }
266
267 void TemplateTable::aconst_null() {
268 transition(vtos, atos);
269 __ xorl(rax, rax);
270 }
271
272 void TemplateTable::iconst(int value) {
273 transition(vtos, itos);
274 if (value == 0) {
275 __ xorl(rax, rax);
276 } else {
277 __ movl(rax, value);
278 }
279 }
280
281 void TemplateTable::lconst(int value) {
282 transition(vtos, ltos);
283 if (value == 0) {
284 __ xorl(rax, rax);
285 } else {
286 __ movl(rax, value);
287 }
288 #ifndef _LP64
289 assert(value >= 0, "check this code");
290 __ xorptr(rdx, rdx);
291 #endif
292 }
293
294
295
296 void TemplateTable::fconst(int value) {
297 transition(vtos, ftos);
298 if (UseSSE >= 1) {
299 static float one = 1.0f, two = 2.0f;
300 switch (value) {
301 case 0:
302 __ xorps(xmm0, xmm0);
303 break;
304 case 1:
305 __ movflt(xmm0, ExternalAddress((address) &one), rscratch1);
306 break;
307 case 2:
308 __ movflt(xmm0, ExternalAddress((address) &two), rscratch1);
309 break;
310 default:
311 ShouldNotReachHere();
312 break;
313 }
314 } else {
315 #ifdef _LP64
316 ShouldNotReachHere();
317 #else
318 if (value == 0) { __ fldz();
319 } else if (value == 1) { __ fld1();
320 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
321 } else { ShouldNotReachHere();
322 }
323 #endif // _LP64
324 }
325 }
326
327 void TemplateTable::dconst(int value) {
328 transition(vtos, dtos);
329 if (UseSSE >= 2) {
330 static double one = 1.0;
331 switch (value) {
332 case 0:
333 __ xorpd(xmm0, xmm0);
334 break;
335 case 1:
336 __ movdbl(xmm0, ExternalAddress((address) &one), rscratch1);
337 break;
338 default:
339 ShouldNotReachHere();
340 break;
341 }
342 } else {
343 #ifdef _LP64
344 ShouldNotReachHere();
345 #else
346 if (value == 0) { __ fldz();
347 } else if (value == 1) { __ fld1();
348 } else { ShouldNotReachHere();
349 }
350 #endif
351 }
352 }
353
354 void TemplateTable::bipush() {
355 transition(vtos, itos);
356 __ load_signed_byte(rax, at_bcp(1));
357 }
358
359 void TemplateTable::sipush() {
360 transition(vtos, itos);
361 __ load_unsigned_short(rax, at_bcp(1));
362 __ bswapl(rax);
363 __ sarl(rax, 16);
364 }
365
366 void TemplateTable::ldc(LdcType type) {
367 transition(vtos, vtos);
368 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
369 Label call_ldc, notFloat, notClass, notInt, Done;
370
371 if (is_ldc_wide(type)) {
372 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
373 } else {
374 __ load_unsigned_byte(rbx, at_bcp(1));
375 }
376
377 __ get_cpool_and_tags(rcx, rax);
378 const int base_offset = ConstantPool::header_size() * wordSize;
379 const int tags_offset = Array<u1>::base_offset_in_bytes();
380
381 // get type
382 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
383
384 // unresolved class - get the resolved class
385 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
386 __ jccb(Assembler::equal, call_ldc);
387
388 // unresolved class in error state - call into runtime to throw the error
389 // from the first resolution attempt
390 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
391 __ jccb(Assembler::equal, call_ldc);
392
393 // resolved class - need to call vm to get java mirror of the class
394 __ cmpl(rdx, JVM_CONSTANT_Class);
395 __ jcc(Assembler::notEqual, notClass);
396
397 __ bind(call_ldc);
398
399 __ movl(rarg, is_ldc_wide(type) ? 1 : 0);
400 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
401
402 __ push(atos);
403 __ jmp(Done);
404
405 __ bind(notClass);
406 __ cmpl(rdx, JVM_CONSTANT_Float);
407 __ jccb(Assembler::notEqual, notFloat);
408
409 // ftos
410 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
411 __ push(ftos);
412 __ jmp(Done);
413
414 __ bind(notFloat);
415 __ cmpl(rdx, JVM_CONSTANT_Integer);
416 __ jccb(Assembler::notEqual, notInt);
417
418 // itos
419 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
420 __ push(itos);
421 __ jmp(Done);
422
423 // assume the tag is for condy; if not, the VM runtime will tell us
424 __ bind(notInt);
425 condy_helper(Done);
426
427 __ bind(Done);
428 }
429
430 // Fast path for caching oop constants.
431 void TemplateTable::fast_aldc(LdcType type) {
432 transition(vtos, atos);
433
434 Register result = rax;
435 Register tmp = rdx;
436 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
437 int index_size = is_ldc_wide(type) ? sizeof(u2) : sizeof(u1);
438
439 Label resolved;
440
441 // We are resolved if the resolved reference cache entry contains a
442 // non-null object (String, MethodType, etc.)
443 assert_different_registers(result, tmp);
444 __ get_cache_index_at_bcp(tmp, 1, index_size);
445 __ load_resolved_reference_at_index(result, tmp);
446 __ testptr(result, result);
447 __ jcc(Assembler::notZero, resolved);
448
449 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
450
451 // first time invocation - must resolve first
452 __ movl(rarg, (int)bytecode());
453 __ call_VM(result, entry, rarg);
454 __ bind(resolved);
455
456 { // Check for the null sentinel.
457 // If we just called the VM, it already did the mapping for us,
458 // but it's harmless to retry.
459 Label notNull;
460 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
461 __ movptr(tmp, null_sentinel);
462 __ resolve_oop_handle(tmp, rscratch2);
463 __ cmpoop(tmp, result);
464 __ jccb(Assembler::notEqual, notNull);
465 __ xorptr(result, result); // null object reference
466 __ bind(notNull);
467 }
468
469 if (VerifyOops) {
470 __ verify_oop(result);
471 }
472 }
473
474 void TemplateTable::ldc2_w() {
475 transition(vtos, vtos);
476 Label notDouble, notLong, Done;
477 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
478
479 __ get_cpool_and_tags(rcx, rax);
480 const int base_offset = ConstantPool::header_size() * wordSize;
481 const int tags_offset = Array<u1>::base_offset_in_bytes();
482
483 // get type
484 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
485 __ cmpl(rdx, JVM_CONSTANT_Double);
486 __ jccb(Assembler::notEqual, notDouble);
487
488 // dtos
489 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
490 __ push(dtos);
491
492 __ jmp(Done);
493 __ bind(notDouble);
494 __ cmpl(rdx, JVM_CONSTANT_Long);
495 __ jccb(Assembler::notEqual, notLong);
496
497 // ltos
498 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
499 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
500 __ push(ltos);
501 __ jmp(Done);
502
503 __ bind(notLong);
504 condy_helper(Done);
505
506 __ bind(Done);
507 }
508
509 void TemplateTable::condy_helper(Label& Done) {
510 const Register obj = rax;
511 const Register off = rbx;
512 const Register flags = rcx;
513 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
514 __ movl(rarg, (int)bytecode());
515 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
516 #ifndef _LP64
517 // borrow rdi from locals
518 __ get_thread(rdi);
519 __ get_vm_result_2(flags, rdi);
520 __ restore_locals();
521 #else
522 __ get_vm_result_2(flags, r15_thread);
523 #endif
524 // VMr = obj = base address to find primitive value to push
525 // VMr2 = flags = (tos, off) using format of CPCE::_flags
526 __ movl(off, flags);
527 __ andl(off, ConstantPoolCache::field_index_mask);
528 const Address field(obj, off, Address::times_1, 0*wordSize);
529
530 // What sort of thing are we loading?
531 __ shrl(flags, ConstantPoolCache::tos_state_shift);
532 __ andl(flags, ConstantPoolCache::tos_state_mask);
533
534 switch (bytecode()) {
535 case Bytecodes::_ldc:
536 case Bytecodes::_ldc_w:
537 {
538 // tos in (itos, ftos, stos, btos, ctos, ztos)
539 Label notInt, notFloat, notShort, notByte, notChar, notBool;
540 __ cmpl(flags, itos);
541 __ jccb(Assembler::notEqual, notInt);
542 // itos
543 __ movl(rax, field);
544 __ push(itos);
545 __ jmp(Done);
546
547 __ bind(notInt);
548 __ cmpl(flags, ftos);
549 __ jccb(Assembler::notEqual, notFloat);
550 // ftos
551 __ load_float(field);
552 __ push(ftos);
553 __ jmp(Done);
554
555 __ bind(notFloat);
556 __ cmpl(flags, stos);
557 __ jccb(Assembler::notEqual, notShort);
558 // stos
559 __ load_signed_short(rax, field);
560 __ push(stos);
561 __ jmp(Done);
562
563 __ bind(notShort);
564 __ cmpl(flags, btos);
565 __ jccb(Assembler::notEqual, notByte);
566 // btos
567 __ load_signed_byte(rax, field);
568 __ push(btos);
569 __ jmp(Done);
570
571 __ bind(notByte);
572 __ cmpl(flags, ctos);
573 __ jccb(Assembler::notEqual, notChar);
574 // ctos
575 __ load_unsigned_short(rax, field);
576 __ push(ctos);
577 __ jmp(Done);
578
579 __ bind(notChar);
580 __ cmpl(flags, ztos);
581 __ jccb(Assembler::notEqual, notBool);
582 // ztos
583 __ load_signed_byte(rax, field);
584 __ push(ztos);
585 __ jmp(Done);
586
587 __ bind(notBool);
588 break;
589 }
590
591 case Bytecodes::_ldc2_w:
592 {
593 Label notLong, notDouble;
594 __ cmpl(flags, ltos);
595 __ jccb(Assembler::notEqual, notLong);
596 // ltos
597 // Loading high word first because movptr clobbers rax
598 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
599 __ movptr(rax, field);
600 __ push(ltos);
601 __ jmp(Done);
602
603 __ bind(notLong);
604 __ cmpl(flags, dtos);
605 __ jccb(Assembler::notEqual, notDouble);
606 // dtos
607 __ load_double(field);
608 __ push(dtos);
609 __ jmp(Done);
610
611 __ bind(notDouble);
612 break;
613 }
614
615 default:
616 ShouldNotReachHere();
617 }
618
619 __ stop("bad ldc/condy");
620 }
621
622 void TemplateTable::locals_index(Register reg, int offset) {
623 __ load_unsigned_byte(reg, at_bcp(offset));
624 __ negptr(reg);
625 }
626
627 void TemplateTable::iload() {
628 iload_internal();
629 }
630
631 void TemplateTable::nofast_iload() {
632 iload_internal(may_not_rewrite);
633 }
634
635 void TemplateTable::iload_internal(RewriteControl rc) {
636 transition(vtos, itos);
637 if (RewriteFrequentPairs && rc == may_rewrite) {
638 Label rewrite, done;
639 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
640 LP64_ONLY(assert(rbx != bc, "register damaged"));
641
642 // get next byte
643 __ load_unsigned_byte(rbx,
644 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
645 // if _iload, wait to rewrite to iload2. We only want to rewrite the
646 // last two iloads in a pair. Comparing against fast_iload means that
647 // the next bytecode is neither an iload or a caload, and therefore
648 // an iload pair.
649 __ cmpl(rbx, Bytecodes::_iload);
650 __ jcc(Assembler::equal, done);
651
652 __ cmpl(rbx, Bytecodes::_fast_iload);
653 __ movl(bc, Bytecodes::_fast_iload2);
654
655 __ jccb(Assembler::equal, rewrite);
656
657 // if _caload, rewrite to fast_icaload
658 __ cmpl(rbx, Bytecodes::_caload);
659 __ movl(bc, Bytecodes::_fast_icaload);
660 __ jccb(Assembler::equal, rewrite);
661
662 // rewrite so iload doesn't check again.
663 __ movl(bc, Bytecodes::_fast_iload);
664
665 // rewrite
666 // bc: fast bytecode
667 __ bind(rewrite);
668 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
669 __ bind(done);
670 }
671
672 // Get the local value into tos
673 locals_index(rbx);
674 __ movl(rax, iaddress(rbx));
675 }
676
677 void TemplateTable::fast_iload2() {
678 transition(vtos, itos);
679 locals_index(rbx);
680 __ movl(rax, iaddress(rbx));
681 __ push(itos);
682 locals_index(rbx, 3);
683 __ movl(rax, iaddress(rbx));
684 }
685
686 void TemplateTable::fast_iload() {
687 transition(vtos, itos);
688 locals_index(rbx);
689 __ movl(rax, iaddress(rbx));
690 }
691
692 void TemplateTable::lload() {
693 transition(vtos, ltos);
694 locals_index(rbx);
695 __ movptr(rax, laddress(rbx));
696 NOT_LP64(__ movl(rdx, haddress(rbx)));
697 }
698
699 void TemplateTable::fload() {
700 transition(vtos, ftos);
701 locals_index(rbx);
702 __ load_float(faddress(rbx));
703 }
704
705 void TemplateTable::dload() {
706 transition(vtos, dtos);
707 locals_index(rbx);
708 __ load_double(daddress(rbx));
709 }
710
711 void TemplateTable::aload() {
712 transition(vtos, atos);
713 locals_index(rbx);
714 __ movptr(rax, aaddress(rbx));
715 }
716
717 void TemplateTable::locals_index_wide(Register reg) {
718 __ load_unsigned_short(reg, at_bcp(2));
719 __ bswapl(reg);
720 __ shrl(reg, 16);
721 __ negptr(reg);
722 }
723
724 void TemplateTable::wide_iload() {
725 transition(vtos, itos);
726 locals_index_wide(rbx);
727 __ movl(rax, iaddress(rbx));
728 }
729
730 void TemplateTable::wide_lload() {
731 transition(vtos, ltos);
732 locals_index_wide(rbx);
733 __ movptr(rax, laddress(rbx));
734 NOT_LP64(__ movl(rdx, haddress(rbx)));
735 }
736
737 void TemplateTable::wide_fload() {
738 transition(vtos, ftos);
739 locals_index_wide(rbx);
740 __ load_float(faddress(rbx));
741 }
742
743 void TemplateTable::wide_dload() {
744 transition(vtos, dtos);
745 locals_index_wide(rbx);
746 __ load_double(daddress(rbx));
747 }
748
749 void TemplateTable::wide_aload() {
750 transition(vtos, atos);
751 locals_index_wide(rbx);
752 __ movptr(rax, aaddress(rbx));
753 }
754
755 void TemplateTable::index_check(Register array, Register index) {
756 // Pop ptr into array
757 __ pop_ptr(array);
758 index_check_without_pop(array, index);
759 }
760
761 void TemplateTable::index_check_without_pop(Register array, Register index) {
762 // destroys rbx
763 // sign extend index for use by indexed load
764 __ movl2ptr(index, index);
765 // check index
766 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
767 if (index != rbx) {
768 // ??? convention: move aberrant index into rbx for exception message
769 assert(rbx != array, "different registers");
770 __ movl(rbx, index);
771 }
772 Label skip;
773 __ jccb(Assembler::below, skip);
774 // Pass array to create more detailed exceptions.
775 __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
776 __ jump(RuntimeAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
777 __ bind(skip);
778 }
779
780 void TemplateTable::iaload() {
781 transition(itos, itos);
782 // rax: index
783 // rdx: array
784 index_check(rdx, rax); // kills rbx
785 __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
786 Address(rdx, rax, Address::times_4,
787 arrayOopDesc::base_offset_in_bytes(T_INT)),
788 noreg, noreg);
789 }
790
791 void TemplateTable::laload() {
792 transition(itos, ltos);
793 // rax: index
794 // rdx: array
795 index_check(rdx, rax); // kills rbx
796 NOT_LP64(__ mov(rbx, rax));
797 // rbx,: index
798 __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
799 Address(rdx, rbx, Address::times_8,
800 arrayOopDesc::base_offset_in_bytes(T_LONG)),
801 noreg, noreg);
802 }
803
804
805
806 void TemplateTable::faload() {
807 transition(itos, ftos);
808 // rax: index
809 // rdx: array
810 index_check(rdx, rax); // kills rbx
811 __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
812 Address(rdx, rax,
813 Address::times_4,
814 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
815 noreg, noreg);
816 }
817
818 void TemplateTable::daload() {
819 transition(itos, dtos);
820 // rax: index
821 // rdx: array
822 index_check(rdx, rax); // kills rbx
823 __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
824 Address(rdx, rax,
825 Address::times_8,
826 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
827 noreg, noreg);
828 }
829
830 void TemplateTable::aaload() {
831 transition(itos, atos);
832 // rax: index
833 // rdx: array
834 index_check(rdx, rax); // kills rbx
835 do_oop_load(_masm,
836 Address(rdx, rax,
837 UseCompressedOops ? Address::times_4 : Address::times_ptr,
838 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
839 rax,
840 IS_ARRAY);
841 }
842
843 void TemplateTable::baload() {
844 transition(itos, itos);
845 // rax: index
846 // rdx: array
847 index_check(rdx, rax); // kills rbx
848 __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
849 Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
850 noreg, noreg);
851 }
852
853 void TemplateTable::caload() {
854 transition(itos, itos);
855 // rax: index
856 // rdx: array
857 index_check(rdx, rax); // kills rbx
858 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
859 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
860 noreg, noreg);
861 }
862
863 // iload followed by caload frequent pair
864 void TemplateTable::fast_icaload() {
865 transition(vtos, itos);
866 // load index out of locals
867 locals_index(rbx);
868 __ movl(rax, iaddress(rbx));
869
870 // rax: index
871 // rdx: array
872 index_check(rdx, rax); // kills rbx
873 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
874 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
875 noreg, noreg);
876 }
877
878
879 void TemplateTable::saload() {
880 transition(itos, itos);
881 // rax: index
882 // rdx: array
883 index_check(rdx, rax); // kills rbx
884 __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
885 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
886 noreg, noreg);
887 }
888
889 void TemplateTable::iload(int n) {
890 transition(vtos, itos);
891 __ movl(rax, iaddress(n));
892 }
893
894 void TemplateTable::lload(int n) {
895 transition(vtos, ltos);
896 __ movptr(rax, laddress(n));
897 NOT_LP64(__ movptr(rdx, haddress(n)));
898 }
899
900 void TemplateTable::fload(int n) {
901 transition(vtos, ftos);
902 __ load_float(faddress(n));
903 }
904
905 void TemplateTable::dload(int n) {
906 transition(vtos, dtos);
907 __ load_double(daddress(n));
908 }
909
910 void TemplateTable::aload(int n) {
911 transition(vtos, atos);
912 __ movptr(rax, aaddress(n));
913 }
914
915 void TemplateTable::aload_0() {
916 aload_0_internal();
917 }
918
919 void TemplateTable::nofast_aload_0() {
920 aload_0_internal(may_not_rewrite);
921 }
922
923 void TemplateTable::aload_0_internal(RewriteControl rc) {
924 transition(vtos, atos);
925 // According to bytecode histograms, the pairs:
926 //
927 // _aload_0, _fast_igetfield
928 // _aload_0, _fast_agetfield
929 // _aload_0, _fast_fgetfield
930 //
931 // occur frequently. If RewriteFrequentPairs is set, the (slow)
932 // _aload_0 bytecode checks if the next bytecode is either
933 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
934 // rewrites the current bytecode into a pair bytecode; otherwise it
935 // rewrites the current bytecode into _fast_aload_0 that doesn't do
936 // the pair check anymore.
937 //
938 // Note: If the next bytecode is _getfield, the rewrite must be
939 // delayed, otherwise we may miss an opportunity for a pair.
940 //
941 // Also rewrite frequent pairs
942 // aload_0, aload_1
943 // aload_0, iload_1
944 // These bytecodes with a small amount of code are most profitable
945 // to rewrite
946 if (RewriteFrequentPairs && rc == may_rewrite) {
947 Label rewrite, done;
948
949 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
950 LP64_ONLY(assert(rbx != bc, "register damaged"));
951
952 // get next byte
953 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
954
955 // if _getfield then wait with rewrite
956 __ cmpl(rbx, Bytecodes::_getfield);
957 __ jcc(Assembler::equal, done);
958
959 // if _igetfield then rewrite to _fast_iaccess_0
960 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
961 __ cmpl(rbx, Bytecodes::_fast_igetfield);
962 __ movl(bc, Bytecodes::_fast_iaccess_0);
963 __ jccb(Assembler::equal, rewrite);
964
965 // if _agetfield then rewrite to _fast_aaccess_0
966 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
967 __ cmpl(rbx, Bytecodes::_fast_agetfield);
968 __ movl(bc, Bytecodes::_fast_aaccess_0);
969 __ jccb(Assembler::equal, rewrite);
970
971 // if _fgetfield then rewrite to _fast_faccess_0
972 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
973 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
974 __ movl(bc, Bytecodes::_fast_faccess_0);
975 __ jccb(Assembler::equal, rewrite);
976
977 // else rewrite to _fast_aload0
978 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
979 __ movl(bc, Bytecodes::_fast_aload_0);
980
981 // rewrite
982 // bc: fast bytecode
983 __ bind(rewrite);
984 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
985
986 __ bind(done);
987 }
988
989 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
990 aload(0);
991 }
992
993 void TemplateTable::istore() {
994 transition(itos, vtos);
995 locals_index(rbx);
996 __ movl(iaddress(rbx), rax);
997 }
998
999
1000 void TemplateTable::lstore() {
1001 transition(ltos, vtos);
1002 locals_index(rbx);
1003 __ movptr(laddress(rbx), rax);
1004 NOT_LP64(__ movptr(haddress(rbx), rdx));
1005 }
1006
1007 void TemplateTable::fstore() {
1008 transition(ftos, vtos);
1009 locals_index(rbx);
1010 __ store_float(faddress(rbx));
1011 }
1012
1013 void TemplateTable::dstore() {
1014 transition(dtos, vtos);
1015 locals_index(rbx);
1016 __ store_double(daddress(rbx));
1017 }
1018
1019 void TemplateTable::astore() {
1020 transition(vtos, vtos);
1021 __ pop_ptr(rax);
1022 locals_index(rbx);
1023 __ movptr(aaddress(rbx), rax);
1024 }
1025
1026 void TemplateTable::wide_istore() {
1027 transition(vtos, vtos);
1028 __ pop_i();
1029 locals_index_wide(rbx);
1030 __ movl(iaddress(rbx), rax);
1031 }
1032
1033 void TemplateTable::wide_lstore() {
1034 transition(vtos, vtos);
1035 NOT_LP64(__ pop_l(rax, rdx));
1036 LP64_ONLY(__ pop_l());
1037 locals_index_wide(rbx);
1038 __ movptr(laddress(rbx), rax);
1039 NOT_LP64(__ movl(haddress(rbx), rdx));
1040 }
1041
1042 void TemplateTable::wide_fstore() {
1043 #ifdef _LP64
1044 transition(vtos, vtos);
1045 __ pop_f(xmm0);
1046 locals_index_wide(rbx);
1047 __ movflt(faddress(rbx), xmm0);
1048 #else
1049 wide_istore();
1050 #endif
1051 }
1052
1053 void TemplateTable::wide_dstore() {
1054 #ifdef _LP64
1055 transition(vtos, vtos);
1056 __ pop_d(xmm0);
1057 locals_index_wide(rbx);
1058 __ movdbl(daddress(rbx), xmm0);
1059 #else
1060 wide_lstore();
1061 #endif
1062 }
1063
1064 void TemplateTable::wide_astore() {
1065 transition(vtos, vtos);
1066 __ pop_ptr(rax);
1067 locals_index_wide(rbx);
1068 __ movptr(aaddress(rbx), rax);
1069 }
1070
1071 void TemplateTable::iastore() {
1072 transition(itos, vtos);
1073 __ pop_i(rbx);
1074 // rax: value
1075 // rbx: index
1076 // rdx: array
1077 index_check(rdx, rbx); // prefer index in rbx
1078 __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1079 Address(rdx, rbx, Address::times_4,
1080 arrayOopDesc::base_offset_in_bytes(T_INT)),
1081 rax, noreg, noreg, noreg);
1082 }
1083
1084 void TemplateTable::lastore() {
1085 transition(ltos, vtos);
1086 __ pop_i(rbx);
1087 // rax,: low(value)
1088 // rcx: array
1089 // rdx: high(value)
1090 index_check(rcx, rbx); // prefer index in rbx,
1091 // rbx,: index
1092 __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1093 Address(rcx, rbx, Address::times_8,
1094 arrayOopDesc::base_offset_in_bytes(T_LONG)),
1095 noreg /* ltos */, noreg, noreg, noreg);
1096 }
1097
1098
1099 void TemplateTable::fastore() {
1100 transition(ftos, vtos);
1101 __ pop_i(rbx);
1102 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1103 // rbx: index
1104 // rdx: array
1105 index_check(rdx, rbx); // prefer index in rbx
1106 __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1107 Address(rdx, rbx, Address::times_4,
1108 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1109 noreg /* ftos */, noreg, noreg, noreg);
1110 }
1111
1112 void TemplateTable::dastore() {
1113 transition(dtos, vtos);
1114 __ pop_i(rbx);
1115 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1116 // rbx: index
1117 // rdx: array
1118 index_check(rdx, rbx); // prefer index in rbx
1119 __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1120 Address(rdx, rbx, Address::times_8,
1121 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1122 noreg /* dtos */, noreg, noreg, noreg);
1123 }
1124
1125 void TemplateTable::aastore() {
1126 Label is_null, ok_is_subtype, done;
1127 transition(vtos, vtos);
1128 // stack: ..., array, index, value
1129 __ movptr(rax, at_tos()); // value
1130 __ movl(rcx, at_tos_p1()); // index
1131 __ movptr(rdx, at_tos_p2()); // array
1132
1133 Address element_address(rdx, rcx,
1134 UseCompressedOops? Address::times_4 : Address::times_ptr,
1135 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1136
1137 index_check_without_pop(rdx, rcx); // kills rbx
1138 __ testptr(rax, rax);
1139 __ jcc(Assembler::zero, is_null);
1140
1141 // Move subklass into rbx
1142 __ load_klass(rbx, rax, rscratch1);
1143 // Move superklass into rax
1144 __ load_klass(rax, rdx, rscratch1);
1145 __ movptr(rax, Address(rax,
1146 ObjArrayKlass::element_klass_offset()));
1147
1148 // Generate subtype check. Blows rcx, rdi
1149 // Superklass in rax. Subklass in rbx.
1150 __ gen_subtype_check(rbx, ok_is_subtype);
1151
1152 // Come here on failure
1153 // object is at TOS
1154 __ jump(RuntimeAddress(Interpreter::_throw_ArrayStoreException_entry));
1155
1156 // Come here on success
1157 __ bind(ok_is_subtype);
1158
1159 // Get the value we will store
1160 __ movptr(rax, at_tos());
1161 __ movl(rcx, at_tos_p1()); // index
1162 // Now store using the appropriate barrier
1163 do_oop_store(_masm, element_address, rax, IS_ARRAY);
1164 __ jmp(done);
1165
1166 // Have a null in rax, rdx=array, ecx=index. Store null at ary[idx]
1167 __ bind(is_null);
1168 __ profile_null_seen(rbx);
1169
1170 // Store a null
1171 do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1172
1173 // Pop stack arguments
1174 __ bind(done);
1175 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1176 }
1177
1178 void TemplateTable::bastore() {
1179 transition(itos, vtos);
1180 __ pop_i(rbx);
1181 // rax: value
1182 // rbx: index
1183 // rdx: array
1184 index_check(rdx, rbx); // prefer index in rbx
1185 // Need to check whether array is boolean or byte
1186 // since both types share the bastore bytecode.
1187 __ load_klass(rcx, rdx, rscratch1);
1188 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1189 int diffbit = Klass::layout_helper_boolean_diffbit();
1190 __ testl(rcx, diffbit);
1191 Label L_skip;
1192 __ jccb(Assembler::zero, L_skip);
1193 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1194 __ bind(L_skip);
1195 __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1196 Address(rdx, rbx,Address::times_1,
1197 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1198 rax, noreg, noreg, noreg);
1199 }
1200
1201 void TemplateTable::castore() {
1202 transition(itos, vtos);
1203 __ pop_i(rbx);
1204 // rax: value
1205 // rbx: index
1206 // rdx: array
1207 index_check(rdx, rbx); // prefer index in rbx
1208 __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1209 Address(rdx, rbx, Address::times_2,
1210 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1211 rax, noreg, noreg, noreg);
1212 }
1213
1214
1215 void TemplateTable::sastore() {
1216 castore();
1217 }
1218
1219 void TemplateTable::istore(int n) {
1220 transition(itos, vtos);
1221 __ movl(iaddress(n), rax);
1222 }
1223
1224 void TemplateTable::lstore(int n) {
1225 transition(ltos, vtos);
1226 __ movptr(laddress(n), rax);
1227 NOT_LP64(__ movptr(haddress(n), rdx));
1228 }
1229
1230 void TemplateTable::fstore(int n) {
1231 transition(ftos, vtos);
1232 __ store_float(faddress(n));
1233 }
1234
1235 void TemplateTable::dstore(int n) {
1236 transition(dtos, vtos);
1237 __ store_double(daddress(n));
1238 }
1239
1240
1241 void TemplateTable::astore(int n) {
1242 transition(vtos, vtos);
1243 __ pop_ptr(rax);
1244 __ movptr(aaddress(n), rax);
1245 }
1246
1247 void TemplateTable::pop() {
1248 transition(vtos, vtos);
1249 __ addptr(rsp, Interpreter::stackElementSize);
1250 }
1251
1252 void TemplateTable::pop2() {
1253 transition(vtos, vtos);
1254 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1255 }
1256
1257
1258 void TemplateTable::dup() {
1259 transition(vtos, vtos);
1260 __ load_ptr(0, rax);
1261 __ push_ptr(rax);
1262 // stack: ..., a, a
1263 }
1264
1265 void TemplateTable::dup_x1() {
1266 transition(vtos, vtos);
1267 // stack: ..., a, b
1268 __ load_ptr( 0, rax); // load b
1269 __ load_ptr( 1, rcx); // load a
1270 __ store_ptr(1, rax); // store b
1271 __ store_ptr(0, rcx); // store a
1272 __ push_ptr(rax); // push b
1273 // stack: ..., b, a, b
1274 }
1275
1276 void TemplateTable::dup_x2() {
1277 transition(vtos, vtos);
1278 // stack: ..., a, b, c
1279 __ load_ptr( 0, rax); // load c
1280 __ load_ptr( 2, rcx); // load a
1281 __ store_ptr(2, rax); // store c in a
1282 __ push_ptr(rax); // push c
1283 // stack: ..., c, b, c, c
1284 __ load_ptr( 2, rax); // load b
1285 __ store_ptr(2, rcx); // store a in b
1286 // stack: ..., c, a, c, c
1287 __ store_ptr(1, rax); // store b in c
1288 // stack: ..., c, a, b, c
1289 }
1290
1291 void TemplateTable::dup2() {
1292 transition(vtos, vtos);
1293 // stack: ..., a, b
1294 __ load_ptr(1, rax); // load a
1295 __ push_ptr(rax); // push a
1296 __ load_ptr(1, rax); // load b
1297 __ push_ptr(rax); // push b
1298 // stack: ..., a, b, a, b
1299 }
1300
1301
1302 void TemplateTable::dup2_x1() {
1303 transition(vtos, vtos);
1304 // stack: ..., a, b, c
1305 __ load_ptr( 0, rcx); // load c
1306 __ load_ptr( 1, rax); // load b
1307 __ push_ptr(rax); // push b
1308 __ push_ptr(rcx); // push c
1309 // stack: ..., a, b, c, b, c
1310 __ store_ptr(3, rcx); // store c in b
1311 // stack: ..., a, c, c, b, c
1312 __ load_ptr( 4, rcx); // load a
1313 __ store_ptr(2, rcx); // store a in 2nd c
1314 // stack: ..., a, c, a, b, c
1315 __ store_ptr(4, rax); // store b in a
1316 // stack: ..., b, c, a, b, c
1317 }
1318
1319 void TemplateTable::dup2_x2() {
1320 transition(vtos, vtos);
1321 // stack: ..., a, b, c, d
1322 __ load_ptr( 0, rcx); // load d
1323 __ load_ptr( 1, rax); // load c
1324 __ push_ptr(rax); // push c
1325 __ push_ptr(rcx); // push d
1326 // stack: ..., a, b, c, d, c, d
1327 __ load_ptr( 4, rax); // load b
1328 __ store_ptr(2, rax); // store b in d
1329 __ store_ptr(4, rcx); // store d in b
1330 // stack: ..., a, d, c, b, c, d
1331 __ load_ptr( 5, rcx); // load a
1332 __ load_ptr( 3, rax); // load c
1333 __ store_ptr(3, rcx); // store a in c
1334 __ store_ptr(5, rax); // store c in a
1335 // stack: ..., c, d, a, b, c, d
1336 }
1337
1338 void TemplateTable::swap() {
1339 transition(vtos, vtos);
1340 // stack: ..., a, b
1341 __ load_ptr( 1, rcx); // load a
1342 __ load_ptr( 0, rax); // load b
1343 __ store_ptr(0, rcx); // store a in b
1344 __ store_ptr(1, rax); // store b in a
1345 // stack: ..., b, a
1346 }
1347
1348 void TemplateTable::iop2(Operation op) {
1349 transition(itos, itos);
1350 switch (op) {
1351 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1352 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1353 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1354 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1355 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1356 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1357 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1358 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1359 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1360 default : ShouldNotReachHere();
1361 }
1362 }
1363
1364 void TemplateTable::lop2(Operation op) {
1365 transition(ltos, ltos);
1366 #ifdef _LP64
1367 switch (op) {
1368 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1369 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1370 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1371 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1372 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1373 default : ShouldNotReachHere();
1374 }
1375 #else
1376 __ pop_l(rbx, rcx);
1377 switch (op) {
1378 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1379 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1380 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1381 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1382 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1383 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1384 default : ShouldNotReachHere();
1385 }
1386 #endif
1387 }
1388
1389 void TemplateTable::idiv() {
1390 transition(itos, itos);
1391 __ movl(rcx, rax);
1392 __ pop_i(rax);
1393 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1394 // they are not equal, one could do a normal division (no correction
1395 // needed), which may speed up this implementation for the common case.
1396 // (see also JVM spec., p.243 & p.271)
1397 __ corrected_idivl(rcx);
1398 }
1399
1400 void TemplateTable::irem() {
1401 transition(itos, itos);
1402 __ movl(rcx, rax);
1403 __ pop_i(rax);
1404 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1405 // they are not equal, one could do a normal division (no correction
1406 // needed), which may speed up this implementation for the common case.
1407 // (see also JVM spec., p.243 & p.271)
1408 __ corrected_idivl(rcx);
1409 __ movl(rax, rdx);
1410 }
1411
1412 void TemplateTable::lmul() {
1413 transition(ltos, ltos);
1414 #ifdef _LP64
1415 __ pop_l(rdx);
1416 __ imulq(rax, rdx);
1417 #else
1418 __ pop_l(rbx, rcx);
1419 __ push(rcx); __ push(rbx);
1420 __ push(rdx); __ push(rax);
1421 __ lmul(2 * wordSize, 0);
1422 __ addptr(rsp, 4 * wordSize); // take off temporaries
1423 #endif
1424 }
1425
1426 void TemplateTable::ldiv() {
1427 transition(ltos, ltos);
1428 #ifdef _LP64
1429 __ mov(rcx, rax);
1430 __ pop_l(rax);
1431 // generate explicit div0 check
1432 __ testq(rcx, rcx);
1433 __ jump_cc(Assembler::zero,
1434 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1435 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1436 // they are not equal, one could do a normal division (no correction
1437 // needed), which may speed up this implementation for the common case.
1438 // (see also JVM spec., p.243 & p.271)
1439 __ corrected_idivq(rcx); // kills rbx
1440 #else
1441 __ pop_l(rbx, rcx);
1442 __ push(rcx); __ push(rbx);
1443 __ push(rdx); __ push(rax);
1444 // check if y = 0
1445 __ orl(rax, rdx);
1446 __ jump_cc(Assembler::zero,
1447 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1448 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1449 __ addptr(rsp, 4 * wordSize); // take off temporaries
1450 #endif
1451 }
1452
1453 void TemplateTable::lrem() {
1454 transition(ltos, ltos);
1455 #ifdef _LP64
1456 __ mov(rcx, rax);
1457 __ pop_l(rax);
1458 __ testq(rcx, rcx);
1459 __ jump_cc(Assembler::zero,
1460 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1461 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1462 // they are not equal, one could do a normal division (no correction
1463 // needed), which may speed up this implementation for the common case.
1464 // (see also JVM spec., p.243 & p.271)
1465 __ corrected_idivq(rcx); // kills rbx
1466 __ mov(rax, rdx);
1467 #else
1468 __ pop_l(rbx, rcx);
1469 __ push(rcx); __ push(rbx);
1470 __ push(rdx); __ push(rax);
1471 // check if y = 0
1472 __ orl(rax, rdx);
1473 __ jump_cc(Assembler::zero,
1474 RuntimeAddress(Interpreter::_throw_ArithmeticException_entry));
1475 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1476 __ addptr(rsp, 4 * wordSize);
1477 #endif
1478 }
1479
1480 void TemplateTable::lshl() {
1481 transition(itos, ltos);
1482 __ movl(rcx, rax); // get shift count
1483 #ifdef _LP64
1484 __ pop_l(rax); // get shift value
1485 __ shlq(rax);
1486 #else
1487 __ pop_l(rax, rdx); // get shift value
1488 __ lshl(rdx, rax);
1489 #endif
1490 }
1491
1492 void TemplateTable::lshr() {
1493 #ifdef _LP64
1494 transition(itos, ltos);
1495 __ movl(rcx, rax); // get shift count
1496 __ pop_l(rax); // get shift value
1497 __ sarq(rax);
1498 #else
1499 transition(itos, ltos);
1500 __ mov(rcx, rax); // get shift count
1501 __ pop_l(rax, rdx); // get shift value
1502 __ lshr(rdx, rax, true);
1503 #endif
1504 }
1505
1506 void TemplateTable::lushr() {
1507 transition(itos, ltos);
1508 #ifdef _LP64
1509 __ movl(rcx, rax); // get shift count
1510 __ pop_l(rax); // get shift value
1511 __ shrq(rax);
1512 #else
1513 __ mov(rcx, rax); // get shift count
1514 __ pop_l(rax, rdx); // get shift value
1515 __ lshr(rdx, rax);
1516 #endif
1517 }
1518
1519 void TemplateTable::fop2(Operation op) {
1520 transition(ftos, ftos);
1521
1522 if (UseSSE >= 1) {
1523 switch (op) {
1524 case add:
1525 __ addss(xmm0, at_rsp());
1526 __ addptr(rsp, Interpreter::stackElementSize);
1527 break;
1528 case sub:
1529 __ movflt(xmm1, xmm0);
1530 __ pop_f(xmm0);
1531 __ subss(xmm0, xmm1);
1532 break;
1533 case mul:
1534 __ mulss(xmm0, at_rsp());
1535 __ addptr(rsp, Interpreter::stackElementSize);
1536 break;
1537 case div:
1538 __ movflt(xmm1, xmm0);
1539 __ pop_f(xmm0);
1540 __ divss(xmm0, xmm1);
1541 break;
1542 case rem:
1543 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1544 // modulo operation. The frem method calls the function
1545 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1546 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1547 // (signalling or quiet) is returned.
1548 //
1549 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1550 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1551 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1552 // The fprem instruction used on x86_32 is functionally equivalent to
1553 // SharedRuntime::frem in that it returns a NaN.
1554 #ifdef _LP64
1555 __ movflt(xmm1, xmm0);
1556 __ pop_f(xmm0);
1557 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1558 #else // !_LP64
1559 __ push_f(xmm0);
1560 __ pop_f();
1561 __ fld_s(at_rsp());
1562 __ fremr(rax);
1563 __ f2ieee();
1564 __ pop(rax); // pop second operand off the stack
1565 __ push_f();
1566 __ pop_f(xmm0);
1567 #endif // _LP64
1568 break;
1569 default:
1570 ShouldNotReachHere();
1571 break;
1572 }
1573 } else {
1574 #ifdef _LP64
1575 ShouldNotReachHere();
1576 #else // !_LP64
1577 switch (op) {
1578 case add: __ fadd_s (at_rsp()); break;
1579 case sub: __ fsubr_s(at_rsp()); break;
1580 case mul: __ fmul_s (at_rsp()); break;
1581 case div: __ fdivr_s(at_rsp()); break;
1582 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1583 default : ShouldNotReachHere();
1584 }
1585 __ f2ieee();
1586 __ pop(rax); // pop second operand off the stack
1587 #endif // _LP64
1588 }
1589 }
1590
1591 void TemplateTable::dop2(Operation op) {
1592 transition(dtos, dtos);
1593 if (UseSSE >= 2) {
1594 switch (op) {
1595 case add:
1596 __ addsd(xmm0, at_rsp());
1597 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1598 break;
1599 case sub:
1600 __ movdbl(xmm1, xmm0);
1601 __ pop_d(xmm0);
1602 __ subsd(xmm0, xmm1);
1603 break;
1604 case mul:
1605 __ mulsd(xmm0, at_rsp());
1606 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1607 break;
1608 case div:
1609 __ movdbl(xmm1, xmm0);
1610 __ pop_d(xmm0);
1611 __ divsd(xmm0, xmm1);
1612 break;
1613 case rem:
1614 // Similar to fop2(), the modulo operation is performed using the
1615 // SharedRuntime::drem method (on x86_64 platforms) or using the
1616 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1617 #ifdef _LP64
1618 __ movdbl(xmm1, xmm0);
1619 __ pop_d(xmm0);
1620 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1621 #else // !_LP64
1622 __ push_d(xmm0);
1623 __ pop_d();
1624 __ fld_d(at_rsp());
1625 __ fremr(rax);
1626 __ d2ieee();
1627 __ pop(rax);
1628 __ pop(rdx);
1629 __ push_d();
1630 __ pop_d(xmm0);
1631 #endif // _LP64
1632 break;
1633 default:
1634 ShouldNotReachHere();
1635 break;
1636 }
1637 } else {
1638 #ifdef _LP64
1639 ShouldNotReachHere();
1640 #else // !_LP64
1641 switch (op) {
1642 case add: __ fadd_d (at_rsp()); break;
1643 case sub: __ fsubr_d(at_rsp()); break;
1644 case mul: {
1645 // strict semantics
1646 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1647 __ fmulp();
1648 __ fmul_d (at_rsp());
1649 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1650 __ fmulp();
1651 break;
1652 }
1653 case div: {
1654 // strict semantics
1655 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1656 __ fmul_d (at_rsp());
1657 __ fdivrp();
1658 __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1659 __ fmulp();
1660 break;
1661 }
1662 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1663 default : ShouldNotReachHere();
1664 }
1665 __ d2ieee();
1666 // Pop double precision number from rsp.
1667 __ pop(rax);
1668 __ pop(rdx);
1669 #endif // _LP64
1670 }
1671 }
1672
1673 void TemplateTable::ineg() {
1674 transition(itos, itos);
1675 __ negl(rax);
1676 }
1677
1678 void TemplateTable::lneg() {
1679 transition(ltos, ltos);
1680 LP64_ONLY(__ negq(rax));
1681 NOT_LP64(__ lneg(rdx, rax));
1682 }
1683
1684 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1685 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1686 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1687 // of 128-bits operands for SSE instructions.
1688 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1689 // Store the value to a 128-bits operand.
1690 operand[0] = lo;
1691 operand[1] = hi;
1692 return operand;
1693 }
1694
1695 // Buffer for 128-bits masks used by SSE instructions.
1696 static jlong float_signflip_pool[2*2];
1697 static jlong double_signflip_pool[2*2];
1698
1699 void TemplateTable::fneg() {
1700 transition(ftos, ftos);
1701 if (UseSSE >= 1) {
1702 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1703 __ xorps(xmm0, ExternalAddress((address) float_signflip), rscratch1);
1704 } else {
1705 LP64_ONLY(ShouldNotReachHere());
1706 NOT_LP64(__ fchs());
1707 }
1708 }
1709
1710 void TemplateTable::dneg() {
1711 transition(dtos, dtos);
1712 if (UseSSE >= 2) {
1713 static jlong *double_signflip =
1714 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1715 __ xorpd(xmm0, ExternalAddress((address) double_signflip), rscratch1);
1716 } else {
1717 #ifdef _LP64
1718 ShouldNotReachHere();
1719 #else
1720 __ fchs();
1721 #endif
1722 }
1723 }
1724
1725 void TemplateTable::iinc() {
1726 transition(vtos, vtos);
1727 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1728 locals_index(rbx);
1729 __ addl(iaddress(rbx), rdx);
1730 }
1731
1732 void TemplateTable::wide_iinc() {
1733 transition(vtos, vtos);
1734 __ movl(rdx, at_bcp(4)); // get constant
1735 locals_index_wide(rbx);
1736 __ bswapl(rdx); // swap bytes & sign-extend constant
1737 __ sarl(rdx, 16);
1738 __ addl(iaddress(rbx), rdx);
1739 // Note: should probably use only one movl to get both
1740 // the index and the constant -> fix this
1741 }
1742
1743 void TemplateTable::convert() {
1744 #ifdef _LP64
1745 // Checking
1746 #ifdef ASSERT
1747 {
1748 TosState tos_in = ilgl;
1749 TosState tos_out = ilgl;
1750 switch (bytecode()) {
1751 case Bytecodes::_i2l: // fall through
1752 case Bytecodes::_i2f: // fall through
1753 case Bytecodes::_i2d: // fall through
1754 case Bytecodes::_i2b: // fall through
1755 case Bytecodes::_i2c: // fall through
1756 case Bytecodes::_i2s: tos_in = itos; break;
1757 case Bytecodes::_l2i: // fall through
1758 case Bytecodes::_l2f: // fall through
1759 case Bytecodes::_l2d: tos_in = ltos; break;
1760 case Bytecodes::_f2i: // fall through
1761 case Bytecodes::_f2l: // fall through
1762 case Bytecodes::_f2d: tos_in = ftos; break;
1763 case Bytecodes::_d2i: // fall through
1764 case Bytecodes::_d2l: // fall through
1765 case Bytecodes::_d2f: tos_in = dtos; break;
1766 default : ShouldNotReachHere();
1767 }
1768 switch (bytecode()) {
1769 case Bytecodes::_l2i: // fall through
1770 case Bytecodes::_f2i: // fall through
1771 case Bytecodes::_d2i: // fall through
1772 case Bytecodes::_i2b: // fall through
1773 case Bytecodes::_i2c: // fall through
1774 case Bytecodes::_i2s: tos_out = itos; break;
1775 case Bytecodes::_i2l: // fall through
1776 case Bytecodes::_f2l: // fall through
1777 case Bytecodes::_d2l: tos_out = ltos; break;
1778 case Bytecodes::_i2f: // fall through
1779 case Bytecodes::_l2f: // fall through
1780 case Bytecodes::_d2f: tos_out = ftos; break;
1781 case Bytecodes::_i2d: // fall through
1782 case Bytecodes::_l2d: // fall through
1783 case Bytecodes::_f2d: tos_out = dtos; break;
1784 default : ShouldNotReachHere();
1785 }
1786 transition(tos_in, tos_out);
1787 }
1788 #endif // ASSERT
1789
1790 static const int64_t is_nan = 0x8000000000000000L;
1791
1792 // Conversion
1793 switch (bytecode()) {
1794 case Bytecodes::_i2l:
1795 __ movslq(rax, rax);
1796 break;
1797 case Bytecodes::_i2f:
1798 __ cvtsi2ssl(xmm0, rax);
1799 break;
1800 case Bytecodes::_i2d:
1801 __ cvtsi2sdl(xmm0, rax);
1802 break;
1803 case Bytecodes::_i2b:
1804 __ movsbl(rax, rax);
1805 break;
1806 case Bytecodes::_i2c:
1807 __ movzwl(rax, rax);
1808 break;
1809 case Bytecodes::_i2s:
1810 __ movswl(rax, rax);
1811 break;
1812 case Bytecodes::_l2i:
1813 __ movl(rax, rax);
1814 break;
1815 case Bytecodes::_l2f:
1816 __ cvtsi2ssq(xmm0, rax);
1817 break;
1818 case Bytecodes::_l2d:
1819 __ cvtsi2sdq(xmm0, rax);
1820 break;
1821 case Bytecodes::_f2i:
1822 {
1823 Label L;
1824 __ cvttss2sil(rax, xmm0);
1825 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1826 __ jcc(Assembler::notEqual, L);
1827 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1828 __ bind(L);
1829 }
1830 break;
1831 case Bytecodes::_f2l:
1832 {
1833 Label L;
1834 __ cvttss2siq(rax, xmm0);
1835 // NaN or overflow/underflow?
1836 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1837 __ jcc(Assembler::notEqual, L);
1838 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1839 __ bind(L);
1840 }
1841 break;
1842 case Bytecodes::_f2d:
1843 __ cvtss2sd(xmm0, xmm0);
1844 break;
1845 case Bytecodes::_d2i:
1846 {
1847 Label L;
1848 __ cvttsd2sil(rax, xmm0);
1849 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1850 __ jcc(Assembler::notEqual, L);
1851 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1852 __ bind(L);
1853 }
1854 break;
1855 case Bytecodes::_d2l:
1856 {
1857 Label L;
1858 __ cvttsd2siq(rax, xmm0);
1859 // NaN or overflow/underflow?
1860 __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1861 __ jcc(Assembler::notEqual, L);
1862 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1863 __ bind(L);
1864 }
1865 break;
1866 case Bytecodes::_d2f:
1867 __ cvtsd2ss(xmm0, xmm0);
1868 break;
1869 default:
1870 ShouldNotReachHere();
1871 }
1872 #else // !_LP64
1873 // Checking
1874 #ifdef ASSERT
1875 { TosState tos_in = ilgl;
1876 TosState tos_out = ilgl;
1877 switch (bytecode()) {
1878 case Bytecodes::_i2l: // fall through
1879 case Bytecodes::_i2f: // fall through
1880 case Bytecodes::_i2d: // fall through
1881 case Bytecodes::_i2b: // fall through
1882 case Bytecodes::_i2c: // fall through
1883 case Bytecodes::_i2s: tos_in = itos; break;
1884 case Bytecodes::_l2i: // fall through
1885 case Bytecodes::_l2f: // fall through
1886 case Bytecodes::_l2d: tos_in = ltos; break;
1887 case Bytecodes::_f2i: // fall through
1888 case Bytecodes::_f2l: // fall through
1889 case Bytecodes::_f2d: tos_in = ftos; break;
1890 case Bytecodes::_d2i: // fall through
1891 case Bytecodes::_d2l: // fall through
1892 case Bytecodes::_d2f: tos_in = dtos; break;
1893 default : ShouldNotReachHere();
1894 }
1895 switch (bytecode()) {
1896 case Bytecodes::_l2i: // fall through
1897 case Bytecodes::_f2i: // fall through
1898 case Bytecodes::_d2i: // fall through
1899 case Bytecodes::_i2b: // fall through
1900 case Bytecodes::_i2c: // fall through
1901 case Bytecodes::_i2s: tos_out = itos; break;
1902 case Bytecodes::_i2l: // fall through
1903 case Bytecodes::_f2l: // fall through
1904 case Bytecodes::_d2l: tos_out = ltos; break;
1905 case Bytecodes::_i2f: // fall through
1906 case Bytecodes::_l2f: // fall through
1907 case Bytecodes::_d2f: tos_out = ftos; break;
1908 case Bytecodes::_i2d: // fall through
1909 case Bytecodes::_l2d: // fall through
1910 case Bytecodes::_f2d: tos_out = dtos; break;
1911 default : ShouldNotReachHere();
1912 }
1913 transition(tos_in, tos_out);
1914 }
1915 #endif // ASSERT
1916
1917 // Conversion
1918 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1919 switch (bytecode()) {
1920 case Bytecodes::_i2l:
1921 __ extend_sign(rdx, rax);
1922 break;
1923 case Bytecodes::_i2f:
1924 if (UseSSE >= 1) {
1925 __ cvtsi2ssl(xmm0, rax);
1926 } else {
1927 __ push(rax); // store int on tos
1928 __ fild_s(at_rsp()); // load int to ST0
1929 __ f2ieee(); // truncate to float size
1930 __ pop(rcx); // adjust rsp
1931 }
1932 break;
1933 case Bytecodes::_i2d:
1934 if (UseSSE >= 2) {
1935 __ cvtsi2sdl(xmm0, rax);
1936 } else {
1937 __ push(rax); // add one slot for d2ieee()
1938 __ push(rax); // store int on tos
1939 __ fild_s(at_rsp()); // load int to ST0
1940 __ d2ieee(); // truncate to double size
1941 __ pop(rcx); // adjust rsp
1942 __ pop(rcx);
1943 }
1944 break;
1945 case Bytecodes::_i2b:
1946 __ shll(rax, 24); // truncate upper 24 bits
1947 __ sarl(rax, 24); // and sign-extend byte
1948 LP64_ONLY(__ movsbl(rax, rax));
1949 break;
1950 case Bytecodes::_i2c:
1951 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1952 LP64_ONLY(__ movzwl(rax, rax));
1953 break;
1954 case Bytecodes::_i2s:
1955 __ shll(rax, 16); // truncate upper 16 bits
1956 __ sarl(rax, 16); // and sign-extend short
1957 LP64_ONLY(__ movswl(rax, rax));
1958 break;
1959 case Bytecodes::_l2i:
1960 /* nothing to do */
1961 break;
1962 case Bytecodes::_l2f:
1963 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1964 // 64-bit long values to floats. On 32-bit platforms it is not possible
1965 // to use that instruction with 64-bit operands, therefore the FPU is
1966 // used to perform the conversion.
1967 __ push(rdx); // store long on tos
1968 __ push(rax);
1969 __ fild_d(at_rsp()); // load long to ST0
1970 __ f2ieee(); // truncate to float size
1971 __ pop(rcx); // adjust rsp
1972 __ pop(rcx);
1973 if (UseSSE >= 1) {
1974 __ push_f();
1975 __ pop_f(xmm0);
1976 }
1977 break;
1978 case Bytecodes::_l2d:
1979 // On 32-bit platforms the FPU is used for conversion because on
1980 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1981 // instruction with 64-bit operands.
1982 __ push(rdx); // store long on tos
1983 __ push(rax);
1984 __ fild_d(at_rsp()); // load long to ST0
1985 __ d2ieee(); // truncate to double size
1986 __ pop(rcx); // adjust rsp
1987 __ pop(rcx);
1988 if (UseSSE >= 2) {
1989 __ push_d();
1990 __ pop_d(xmm0);
1991 }
1992 break;
1993 case Bytecodes::_f2i:
1994 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1995 // as it returns 0 for any NaN.
1996 if (UseSSE >= 1) {
1997 __ push_f(xmm0);
1998 } else {
1999 __ push(rcx); // reserve space for argument
2000 __ fstp_s(at_rsp()); // pass float argument on stack
2001 }
2002 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2003 break;
2004 case Bytecodes::_f2l:
2005 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2006 // as it returns 0 for any NaN.
2007 if (UseSSE >= 1) {
2008 __ push_f(xmm0);
2009 } else {
2010 __ push(rcx); // reserve space for argument
2011 __ fstp_s(at_rsp()); // pass float argument on stack
2012 }
2013 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2014 break;
2015 case Bytecodes::_f2d:
2016 if (UseSSE < 1) {
2017 /* nothing to do */
2018 } else if (UseSSE == 1) {
2019 __ push_f(xmm0);
2020 __ pop_f();
2021 } else { // UseSSE >= 2
2022 __ cvtss2sd(xmm0, xmm0);
2023 }
2024 break;
2025 case Bytecodes::_d2i:
2026 if (UseSSE >= 2) {
2027 __ push_d(xmm0);
2028 } else {
2029 __ push(rcx); // reserve space for argument
2030 __ push(rcx);
2031 __ fstp_d(at_rsp()); // pass double argument on stack
2032 }
2033 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2034 break;
2035 case Bytecodes::_d2l:
2036 if (UseSSE >= 2) {
2037 __ push_d(xmm0);
2038 } else {
2039 __ push(rcx); // reserve space for argument
2040 __ push(rcx);
2041 __ fstp_d(at_rsp()); // pass double argument on stack
2042 }
2043 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2044 break;
2045 case Bytecodes::_d2f:
2046 if (UseSSE <= 1) {
2047 __ push(rcx); // reserve space for f2ieee()
2048 __ f2ieee(); // truncate to float size
2049 __ pop(rcx); // adjust rsp
2050 if (UseSSE == 1) {
2051 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2052 // the conversion is performed using the FPU in this case.
2053 __ push_f();
2054 __ pop_f(xmm0);
2055 }
2056 } else { // UseSSE >= 2
2057 __ cvtsd2ss(xmm0, xmm0);
2058 }
2059 break;
2060 default :
2061 ShouldNotReachHere();
2062 }
2063 #endif // _LP64
2064 }
2065
2066 void TemplateTable::lcmp() {
2067 transition(ltos, itos);
2068 #ifdef _LP64
2069 Label done;
2070 __ pop_l(rdx);
2071 __ cmpq(rdx, rax);
2072 __ movl(rax, -1);
2073 __ jccb(Assembler::less, done);
2074 __ setb(Assembler::notEqual, rax);
2075 __ movzbl(rax, rax);
2076 __ bind(done);
2077 #else
2078
2079 // y = rdx:rax
2080 __ pop_l(rbx, rcx); // get x = rcx:rbx
2081 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2082 __ mov(rax, rcx);
2083 #endif
2084 }
2085
2086 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2087 if ((is_float && UseSSE >= 1) ||
2088 (!is_float && UseSSE >= 2)) {
2089 Label done;
2090 if (is_float) {
2091 // XXX get rid of pop here, use ... reg, mem32
2092 __ pop_f(xmm1);
2093 __ ucomiss(xmm1, xmm0);
2094 } else {
2095 // XXX get rid of pop here, use ... reg, mem64
2096 __ pop_d(xmm1);
2097 __ ucomisd(xmm1, xmm0);
2098 }
2099 if (unordered_result < 0) {
2100 __ movl(rax, -1);
2101 __ jccb(Assembler::parity, done);
2102 __ jccb(Assembler::below, done);
2103 __ setb(Assembler::notEqual, rdx);
2104 __ movzbl(rax, rdx);
2105 } else {
2106 __ movl(rax, 1);
2107 __ jccb(Assembler::parity, done);
2108 __ jccb(Assembler::above, done);
2109 __ movl(rax, 0);
2110 __ jccb(Assembler::equal, done);
2111 __ decrementl(rax);
2112 }
2113 __ bind(done);
2114 } else {
2115 #ifdef _LP64
2116 ShouldNotReachHere();
2117 #else // !_LP64
2118 if (is_float) {
2119 __ fld_s(at_rsp());
2120 } else {
2121 __ fld_d(at_rsp());
2122 __ pop(rdx);
2123 }
2124 __ pop(rcx);
2125 __ fcmp2int(rax, unordered_result < 0);
2126 #endif // _LP64
2127 }
2128 }
2129
2130 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2131 __ get_method(rcx); // rcx holds method
2132 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2133 // holds bumped taken count
2134
2135 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2136 InvocationCounter::counter_offset();
2137 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2138 InvocationCounter::counter_offset();
2139
2140 // Load up edx with the branch displacement
2141 if (is_wide) {
2142 __ movl(rdx, at_bcp(1));
2143 } else {
2144 __ load_signed_short(rdx, at_bcp(1));
2145 }
2146 __ bswapl(rdx);
2147
2148 if (!is_wide) {
2149 __ sarl(rdx, 16);
2150 }
2151 LP64_ONLY(__ movl2ptr(rdx, rdx));
2152
2153 // Handle all the JSR stuff here, then exit.
2154 // It's much shorter and cleaner than intermingling with the non-JSR
2155 // normal-branch stuff occurring below.
2156 if (is_jsr) {
2157 // Pre-load the next target bytecode into rbx
2158 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2159
2160 // compute return address as bci in rax
2161 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2162 in_bytes(ConstMethod::codes_offset())));
2163 __ subptr(rax, Address(rcx, Method::const_offset()));
2164 // Adjust the bcp in r13 by the displacement in rdx
2165 __ addptr(rbcp, rdx);
2166 // jsr returns atos that is not an oop
2167 __ push_i(rax);
2168 __ dispatch_only(vtos, true);
2169 return;
2170 }
2171
2172 // Normal (non-jsr) branch handling
2173
2174 // Adjust the bcp in r13 by the displacement in rdx
2175 __ addptr(rbcp, rdx);
2176
2177 assert(UseLoopCounter || !UseOnStackReplacement,
2178 "on-stack-replacement requires loop counters");
2179 Label backedge_counter_overflow;
2180 Label dispatch;
2181 if (UseLoopCounter) {
2182 // increment backedge counter for backward branches
2183 // rax: MDO
2184 // rbx: MDO bumped taken-count
2185 // rcx: method
2186 // rdx: target offset
2187 // r13: target bcp
2188 // r14: locals pointer
2189 __ testl(rdx, rdx); // check if forward or backward branch
2190 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2191
2192 // check if MethodCounters exists
2193 Label has_counters;
2194 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2195 __ testptr(rax, rax);
2196 __ jcc(Assembler::notZero, has_counters);
2197 __ push(rdx);
2198 __ push(rcx);
2199 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2200 rcx);
2201 __ pop(rcx);
2202 __ pop(rdx);
2203 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2204 __ testptr(rax, rax);
2205 __ jcc(Assembler::zero, dispatch);
2206 __ bind(has_counters);
2207
2208 Label no_mdo;
2209 if (ProfileInterpreter) {
2210 // Are we profiling?
2211 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2212 __ testptr(rbx, rbx);
2213 __ jccb(Assembler::zero, no_mdo);
2214 // Increment the MDO backedge counter
2215 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2216 in_bytes(InvocationCounter::counter_offset()));
2217 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2218 __ increment_mask_and_jump(mdo_backedge_counter, mask, rax,
2219 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2220 __ jmp(dispatch);
2221 }
2222 __ bind(no_mdo);
2223 // Increment backedge counter in MethodCounters*
2224 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2225 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2226 __ increment_mask_and_jump(Address(rcx, be_offset), mask, rax,
2227 UseOnStackReplacement ? &backedge_counter_overflow : nullptr);
2228 __ bind(dispatch);
2229 }
2230
2231 // Pre-load the next target bytecode into rbx
2232 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2233
2234 // continue with the bytecode @ target
2235 // rax: return bci for jsr's, unused otherwise
2236 // rbx: target bytecode
2237 // r13: target bcp
2238 __ dispatch_only(vtos, true);
2239
2240 if (UseLoopCounter) {
2241 if (UseOnStackReplacement) {
2242 Label set_mdp;
2243 // invocation counter overflow
2244 __ bind(backedge_counter_overflow);
2245 __ negptr(rdx);
2246 __ addptr(rdx, rbcp); // branch bcp
2247 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2248 __ call_VM(noreg,
2249 CAST_FROM_FN_PTR(address,
2250 InterpreterRuntime::frequency_counter_overflow),
2251 rdx);
2252
2253 // rax: osr nmethod (osr ok) or null (osr not possible)
2254 // rdx: scratch
2255 // r14: locals pointer
2256 // r13: bcp
2257 __ testptr(rax, rax); // test result
2258 __ jcc(Assembler::zero, dispatch); // no osr if null
2259 // nmethod may have been invalidated (VM may block upon call_VM return)
2260 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2261 __ jcc(Assembler::notEqual, dispatch);
2262
2263 // We have the address of an on stack replacement routine in rax.
2264 // In preparation of invoking it, first we must migrate the locals
2265 // and monitors from off the interpreter frame on the stack.
2266 // Ensure to save the osr nmethod over the migration call,
2267 // it will be preserved in rbx.
2268 __ mov(rbx, rax);
2269
2270 NOT_LP64(__ get_thread(rcx));
2271
2272 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2273
2274 // rax is OSR buffer, move it to expected parameter location
2275 LP64_ONLY(__ mov(j_rarg0, rax));
2276 NOT_LP64(__ mov(rcx, rax));
2277 // We use j_rarg definitions here so that registers don't conflict as parameter
2278 // registers change across platforms as we are in the midst of a calling
2279 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2280
2281 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2282 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2283
2284 // pop the interpreter frame
2285 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2286 __ leave(); // remove frame anchor
2287 __ pop(retaddr); // get return address
2288 __ mov(rsp, sender_sp); // set sp to sender sp
2289 // Ensure compiled code always sees stack at proper alignment
2290 __ andptr(rsp, -(StackAlignmentInBytes));
2291
2292 // unlike x86 we need no specialized return from compiled code
2293 // to the interpreter or the call stub.
2294
2295 // push the return address
2296 __ push(retaddr);
2297
2298 // and begin the OSR nmethod
2299 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2300 }
2301 }
2302 }
2303
2304 void TemplateTable::if_0cmp(Condition cc) {
2305 transition(itos, vtos);
2306 // assume branch is more often taken than not (loops use backward branches)
2307 Label not_taken;
2308 __ testl(rax, rax);
2309 __ jcc(j_not(cc), not_taken);
2310 branch(false, false);
2311 __ bind(not_taken);
2312 __ profile_not_taken_branch(rax);
2313 }
2314
2315 void TemplateTable::if_icmp(Condition cc) {
2316 transition(itos, vtos);
2317 // assume branch is more often taken than not (loops use backward branches)
2318 Label not_taken;
2319 __ pop_i(rdx);
2320 __ cmpl(rdx, rax);
2321 __ jcc(j_not(cc), not_taken);
2322 branch(false, false);
2323 __ bind(not_taken);
2324 __ profile_not_taken_branch(rax);
2325 }
2326
2327 void TemplateTable::if_nullcmp(Condition cc) {
2328 transition(atos, vtos);
2329 // assume branch is more often taken than not (loops use backward branches)
2330 Label not_taken;
2331 __ testptr(rax, rax);
2332 __ jcc(j_not(cc), not_taken);
2333 branch(false, false);
2334 __ bind(not_taken);
2335 __ profile_not_taken_branch(rax);
2336 }
2337
2338 void TemplateTable::if_acmp(Condition cc) {
2339 transition(atos, vtos);
2340 // assume branch is more often taken than not (loops use backward branches)
2341 Label not_taken;
2342 __ pop_ptr(rdx);
2343 __ cmpoop(rdx, rax);
2344 __ jcc(j_not(cc), not_taken);
2345 branch(false, false);
2346 __ bind(not_taken);
2347 __ profile_not_taken_branch(rax);
2348 }
2349
2350 void TemplateTable::ret() {
2351 transition(vtos, vtos);
2352 locals_index(rbx);
2353 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2354 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2355 __ profile_ret(rbx, rcx);
2356 __ get_method(rax);
2357 __ movptr(rbcp, Address(rax, Method::const_offset()));
2358 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2359 ConstMethod::codes_offset()));
2360 __ dispatch_next(vtos, 0, true);
2361 }
2362
2363 void TemplateTable::wide_ret() {
2364 transition(vtos, vtos);
2365 locals_index_wide(rbx);
2366 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2367 __ profile_ret(rbx, rcx);
2368 __ get_method(rax);
2369 __ movptr(rbcp, Address(rax, Method::const_offset()));
2370 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2371 __ dispatch_next(vtos, 0, true);
2372 }
2373
2374 void TemplateTable::tableswitch() {
2375 Label default_case, continue_execution;
2376 transition(itos, vtos);
2377
2378 // align r13/rsi
2379 __ lea(rbx, at_bcp(BytesPerInt));
2380 __ andptr(rbx, -BytesPerInt);
2381 // load lo & hi
2382 __ movl(rcx, Address(rbx, BytesPerInt));
2383 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2384 __ bswapl(rcx);
2385 __ bswapl(rdx);
2386 // check against lo & hi
2387 __ cmpl(rax, rcx);
2388 __ jcc(Assembler::less, default_case);
2389 __ cmpl(rax, rdx);
2390 __ jcc(Assembler::greater, default_case);
2391 // lookup dispatch offset
2392 __ subl(rax, rcx);
2393 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2394 __ profile_switch_case(rax, rbx, rcx);
2395 // continue execution
2396 __ bind(continue_execution);
2397 __ bswapl(rdx);
2398 LP64_ONLY(__ movl2ptr(rdx, rdx));
2399 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2400 __ addptr(rbcp, rdx);
2401 __ dispatch_only(vtos, true);
2402 // handle default
2403 __ bind(default_case);
2404 __ profile_switch_default(rax);
2405 __ movl(rdx, Address(rbx, 0));
2406 __ jmp(continue_execution);
2407 }
2408
2409 void TemplateTable::lookupswitch() {
2410 transition(itos, itos);
2411 __ stop("lookupswitch bytecode should have been rewritten");
2412 }
2413
2414 void TemplateTable::fast_linearswitch() {
2415 transition(itos, vtos);
2416 Label loop_entry, loop, found, continue_execution;
2417 // bswap rax so we can avoid bswapping the table entries
2418 __ bswapl(rax);
2419 // align r13
2420 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2421 // this instruction (change offsets
2422 // below)
2423 __ andptr(rbx, -BytesPerInt);
2424 // set counter
2425 __ movl(rcx, Address(rbx, BytesPerInt));
2426 __ bswapl(rcx);
2427 __ jmpb(loop_entry);
2428 // table search
2429 __ bind(loop);
2430 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2431 __ jcc(Assembler::equal, found);
2432 __ bind(loop_entry);
2433 __ decrementl(rcx);
2434 __ jcc(Assembler::greaterEqual, loop);
2435 // default case
2436 __ profile_switch_default(rax);
2437 __ movl(rdx, Address(rbx, 0));
2438 __ jmp(continue_execution);
2439 // entry found -> get offset
2440 __ bind(found);
2441 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2442 __ profile_switch_case(rcx, rax, rbx);
2443 // continue execution
2444 __ bind(continue_execution);
2445 __ bswapl(rdx);
2446 __ movl2ptr(rdx, rdx);
2447 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2448 __ addptr(rbcp, rdx);
2449 __ dispatch_only(vtos, true);
2450 }
2451
2452 void TemplateTable::fast_binaryswitch() {
2453 transition(itos, vtos);
2454 // Implementation using the following core algorithm:
2455 //
2456 // int binary_search(int key, LookupswitchPair* array, int n) {
2457 // // Binary search according to "Methodik des Programmierens" by
2458 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2459 // int i = 0;
2460 // int j = n;
2461 // while (i+1 < j) {
2462 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2463 // // with Q: for all i: 0 <= i < n: key < a[i]
2464 // // where a stands for the array and assuming that the (inexisting)
2465 // // element a[n] is infinitely big.
2466 // int h = (i + j) >> 1;
2467 // // i < h < j
2468 // if (key < array[h].fast_match()) {
2469 // j = h;
2470 // } else {
2471 // i = h;
2472 // }
2473 // }
2474 // // R: a[i] <= key < a[i+1] or Q
2475 // // (i.e., if key is within array, i is the correct index)
2476 // return i;
2477 // }
2478
2479 // Register allocation
2480 const Register key = rax; // already set (tosca)
2481 const Register array = rbx;
2482 const Register i = rcx;
2483 const Register j = rdx;
2484 const Register h = rdi;
2485 const Register temp = rsi;
2486
2487 // Find array start
2488 NOT_LP64(__ save_bcp());
2489
2490 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2491 // get rid of this
2492 // instruction (change
2493 // offsets below)
2494 __ andptr(array, -BytesPerInt);
2495
2496 // Initialize i & j
2497 __ xorl(i, i); // i = 0;
2498 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2499
2500 // Convert j into native byteordering
2501 __ bswapl(j);
2502
2503 // And start
2504 Label entry;
2505 __ jmp(entry);
2506
2507 // binary search loop
2508 {
2509 Label loop;
2510 __ bind(loop);
2511 // int h = (i + j) >> 1;
2512 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2513 __ sarl(h, 1); // h = (i + j) >> 1;
2514 // if (key < array[h].fast_match()) {
2515 // j = h;
2516 // } else {
2517 // i = h;
2518 // }
2519 // Convert array[h].match to native byte-ordering before compare
2520 __ movl(temp, Address(array, h, Address::times_8));
2521 __ bswapl(temp);
2522 __ cmpl(key, temp);
2523 // j = h if (key < array[h].fast_match())
2524 __ cmov32(Assembler::less, j, h);
2525 // i = h if (key >= array[h].fast_match())
2526 __ cmov32(Assembler::greaterEqual, i, h);
2527 // while (i+1 < j)
2528 __ bind(entry);
2529 __ leal(h, Address(i, 1)); // i+1
2530 __ cmpl(h, j); // i+1 < j
2531 __ jcc(Assembler::less, loop);
2532 }
2533
2534 // end of binary search, result index is i (must check again!)
2535 Label default_case;
2536 // Convert array[i].match to native byte-ordering before compare
2537 __ movl(temp, Address(array, i, Address::times_8));
2538 __ bswapl(temp);
2539 __ cmpl(key, temp);
2540 __ jcc(Assembler::notEqual, default_case);
2541
2542 // entry found -> j = offset
2543 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2544 __ profile_switch_case(i, key, array);
2545 __ bswapl(j);
2546 LP64_ONLY(__ movslq(j, j));
2547
2548 NOT_LP64(__ restore_bcp());
2549 NOT_LP64(__ restore_locals()); // restore rdi
2550
2551 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2552 __ addptr(rbcp, j);
2553 __ dispatch_only(vtos, true);
2554
2555 // default case -> j = default offset
2556 __ bind(default_case);
2557 __ profile_switch_default(i);
2558 __ movl(j, Address(array, -2 * BytesPerInt));
2559 __ bswapl(j);
2560 LP64_ONLY(__ movslq(j, j));
2561
2562 NOT_LP64(__ restore_bcp());
2563 NOT_LP64(__ restore_locals());
2564
2565 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2566 __ addptr(rbcp, j);
2567 __ dispatch_only(vtos, true);
2568 }
2569
2570 void TemplateTable::_return(TosState state) {
2571 transition(state, state);
2572
2573 assert(_desc->calls_vm(),
2574 "inconsistent calls_vm information"); // call in remove_activation
2575
2576 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2577 assert(state == vtos, "only valid state");
2578 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2579 __ movptr(robj, aaddress(0));
2580 __ load_klass(rdi, robj, rscratch1);
2581 __ testb(Address(rdi, Klass::misc_flags_offset()), KlassFlags::_misc_has_finalizer);
2582 Label skip_register_finalizer;
2583 __ jcc(Assembler::zero, skip_register_finalizer);
2584
2585 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2586
2587 __ bind(skip_register_finalizer);
2588 }
2589
2590 if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2591 Label no_safepoint;
2592 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2593 #ifdef _LP64
2594 __ testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2595 #else
2596 const Register thread = rdi;
2597 __ get_thread(thread);
2598 __ testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2599 #endif
2600 __ jcc(Assembler::zero, no_safepoint);
2601 __ push(state);
2602 __ push_cont_fastpath();
2603 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2604 InterpreterRuntime::at_safepoint));
2605 __ pop_cont_fastpath();
2606 __ pop(state);
2607 __ bind(no_safepoint);
2608 }
2609
2610 // Narrow result if state is itos but result type is smaller.
2611 // Need to narrow in the return bytecode rather than in generate_return_entry
2612 // since compiled code callers expect the result to already be narrowed.
2613 if (state == itos) {
2614 __ narrow(rax);
2615 }
2616 __ remove_activation(state, rbcp);
2617
2618 __ jmp(rbcp);
2619 }
2620
2621 // ----------------------------------------------------------------------------
2622 // Volatile variables demand their effects be made known to all CPU's
2623 // in order. Store buffers on most chips allow reads & writes to
2624 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2625 // without some kind of memory barrier (i.e., it's not sufficient that
2626 // the interpreter does not reorder volatile references, the hardware
2627 // also must not reorder them).
2628 //
2629 // According to the new Java Memory Model (JMM):
2630 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2631 // writes act as acquire & release, so:
2632 // (2) A read cannot let unrelated NON-volatile memory refs that
2633 // happen after the read float up to before the read. It's OK for
2634 // non-volatile memory refs that happen before the volatile read to
2635 // float down below it.
2636 // (3) Similar a volatile write cannot let unrelated NON-volatile
2637 // memory refs that happen BEFORE the write float down to after the
2638 // write. It's OK for non-volatile memory refs that happen after the
2639 // volatile write to float up before it.
2640 //
2641 // We only put in barriers around volatile refs (they are expensive),
2642 // not _between_ memory refs (that would require us to track the
2643 // flavor of the previous memory refs). Requirements (2) and (3)
2644 // require some barriers before volatile stores and after volatile
2645 // loads. These nearly cover requirement (1) but miss the
2646 // volatile-store-volatile-load case. This final case is placed after
2647 // volatile-stores although it could just as well go before
2648 // volatile-loads.
2649
2650 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2651 // Helper function to insert a is-volatile test and memory barrier
2652 __ membar(order_constraint);
2653 }
2654
2655 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2656 Register cache,
2657 Register index) {
2658 const Register temp = rbx;
2659 assert_different_registers(cache, index, temp);
2660
2661 Label L_clinit_barrier_slow;
2662 Label resolved;
2663
2664 Bytecodes::Code code = bytecode();
2665
2666 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2667
2668 __ load_method_entry(cache, index);
2669 switch(byte_no) {
2670 case f1_byte:
2671 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2672 break;
2673 case f2_byte:
2674 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2675 break;
2676 default:
2677 ShouldNotReachHere();
2678 }
2679 __ cmpl(temp, code); // have we resolved this bytecode?
2680 __ jcc(Assembler::equal, resolved);
2681
2682 // resolve first time through
2683 // Class initialization barrier slow path lands here as well.
2684 __ bind(L_clinit_barrier_slow);
2685 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2686 __ movl(temp, code);
2687 __ call_VM(noreg, entry, temp);
2688 // Update registers with resolved info
2689 __ load_method_entry(cache, index);
2690
2691 __ bind(resolved);
2692
2693 // Class initialization barrier for static methods
2694 if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2695 const Register method = temp;
2696 const Register klass = temp;
2697 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2698 assert(thread != noreg, "x86_32 not supported");
2699
2700 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2701 __ load_method_holder(klass, method);
2702 __ clinit_barrier(klass, thread, nullptr /*L_fast_path*/, &L_clinit_barrier_slow);
2703 }
2704 }
2705
2706 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2707 Register cache,
2708 Register index) {
2709 const Register temp = rbx;
2710 assert_different_registers(cache, index, temp);
2711
2712 Label L_clinit_barrier_slow;
2713 Label resolved;
2714
2715 Bytecodes::Code code = bytecode();
2716 switch (code) {
2717 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2718 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2719 default: break;
2720 }
2721
2722 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2723 __ load_field_entry(cache, index);
2724 if (byte_no == f1_byte) {
2725 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2726 } else {
2727 __ load_unsigned_byte(temp, Address(cache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2728 }
2729 __ cmpl(temp, code); // have we resolved this bytecode?
2730 __ jcc(Assembler::equal, resolved);
2731
2732 // resolve first time through
2733 __ bind(L_clinit_barrier_slow);
2734 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2735 __ movl(temp, code);
2736 __ call_VM(noreg, entry, temp);
2737 // Update registers with resolved info
2738 __ load_field_entry(cache, index);
2739
2740 __ bind(resolved);
2741
2742 // Class initialization barrier for static fields
2743 if (VM_Version::supports_fast_class_init_checks() &&
2744 (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
2745 const Register field_holder = temp;
2746 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2747 assert(thread != noreg, "x86_32 not supported");
2748
2749 __ movptr(field_holder, Address(cache, in_bytes(ResolvedFieldEntry::field_holder_offset())));
2750 __ clinit_barrier(field_holder, thread, nullptr /*L_fast_path*/, &L_clinit_barrier_slow);
2751 }
2752 }
2753
2754 void TemplateTable::load_resolved_field_entry(Register obj,
2755 Register cache,
2756 Register tos_state,
2757 Register offset,
2758 Register flags,
2759 bool is_static = false) {
2760 assert_different_registers(cache, tos_state, flags, offset);
2761
2762 // Field offset
2763 __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2764
2765 // Flags
2766 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2767
2768 // TOS state
2769 __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2770
2771 // Klass overwrite register
2772 if (is_static) {
2773 __ movptr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2774 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2775 __ movptr(obj, Address(obj, mirror_offset));
2776 __ resolve_oop_handle(obj, rscratch2);
2777 }
2778
2779 }
2780
2781 void TemplateTable::load_invokedynamic_entry(Register method) {
2782 // setup registers
2783 const Register appendix = rax;
2784 const Register cache = rcx;
2785 const Register index = rdx;
2786 assert_different_registers(method, appendix, cache, index);
2787
2788 __ save_bcp();
2789
2790 Label resolved;
2791
2792 __ load_resolved_indy_entry(cache, index);
2793 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2794
2795 // Compare the method to zero
2796 __ testptr(method, method);
2797 __ jcc(Assembler::notZero, resolved);
2798
2799 Bytecodes::Code code = bytecode();
2800
2801 // Call to the interpreter runtime to resolve invokedynamic
2802 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2803 __ movl(method, code); // this is essentially Bytecodes::_invokedynamic
2804 __ call_VM(noreg, entry, method);
2805 // Update registers with resolved info
2806 __ load_resolved_indy_entry(cache, index);
2807 __ movptr(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2808
2809 #ifdef ASSERT
2810 __ testptr(method, method);
2811 __ jcc(Assembler::notZero, resolved);
2812 __ stop("Should be resolved by now");
2813 #endif // ASSERT
2814 __ bind(resolved);
2815
2816 Label L_no_push;
2817 // Check if there is an appendix
2818 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2819 __ testl(index, (1 << ResolvedIndyEntry::has_appendix_shift));
2820 __ jcc(Assembler::zero, L_no_push);
2821
2822 // Get appendix
2823 __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2824 // Push the appendix as a trailing parameter
2825 // since the parameter_size includes it.
2826 __ load_resolved_reference_at_index(appendix, index);
2827 __ verify_oop(appendix);
2828 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2829 __ bind(L_no_push);
2830
2831 // compute return type
2832 __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2833 // load return address
2834 {
2835 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2836 ExternalAddress table(table_addr);
2837 #ifdef _LP64
2838 __ lea(rscratch1, table);
2839 __ movptr(index, Address(rscratch1, index, Address::times_ptr));
2840 #else
2841 __ movptr(index, ArrayAddress(table, Address(noreg, index, Address::times_ptr)));
2842 #endif // _LP64
2843 }
2844
2845 // push return address
2846 __ push(index);
2847 }
2848
2849 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2850 Register method,
2851 Register flags) {
2852 // setup registers
2853 const Register index = rdx;
2854 assert_different_registers(cache, index);
2855 assert_different_registers(method, cache, flags);
2856
2857 // determine constant pool cache field offsets
2858 resolve_cache_and_index_for_method(f1_byte, cache, index);
2859 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2860 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2861 }
2862
2863 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2864 Register method,
2865 Register ref_index,
2866 Register flags) {
2867 // setup registers
2868 const Register index = rdx;
2869 assert_different_registers(cache, index);
2870 assert_different_registers(cache, method, ref_index, flags);
2871
2872 // determine constant pool cache field offsets
2873 resolve_cache_and_index_for_method(f1_byte, cache, index);
2874 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2875
2876 // Maybe push appendix
2877 Label L_no_push;
2878 __ testl(flags, (1 << ResolvedMethodEntry::has_appendix_shift));
2879 __ jcc(Assembler::zero, L_no_push);
2880 // invokehandle uses an index into the resolved references array
2881 __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2882 // Push the appendix as a trailing parameter.
2883 // This must be done before we get the receiver,
2884 // since the parameter_size includes it.
2885 Register appendix = method;
2886 __ load_resolved_reference_at_index(appendix, ref_index);
2887 __ push(appendix); // push appendix (MethodType, CallSite, etc.)
2888 __ bind(L_no_push);
2889
2890 __ movptr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2891 }
2892
2893 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2894 Register klass,
2895 Register method_or_table_index,
2896 Register flags) {
2897 // setup registers
2898 const Register index = rdx;
2899 assert_different_registers(cache, klass, method_or_table_index, flags);
2900
2901 // determine constant pool cache field offsets
2902 resolve_cache_and_index_for_method(f1_byte, cache, index);
2903 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2904
2905 // Invokeinterface can behave in different ways:
2906 // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
2907 // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
2908 // vtable index is placed in the register.
2909 // Otherwise, the registers will be populated with the klass and method.
2910
2911 Label NotVirtual; Label NotVFinal; Label Done;
2912 __ testl(flags, 1 << ResolvedMethodEntry::is_forced_virtual_shift);
2913 __ jcc(Assembler::zero, NotVirtual);
2914 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
2915 __ jcc(Assembler::zero, NotVFinal);
2916 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2917 __ jmp(Done);
2918
2919 __ bind(NotVFinal);
2920 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2921 __ jmp(Done);
2922
2923 __ bind(NotVirtual);
2924 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2925 __ movptr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
2926 __ bind(Done);
2927 }
2928
2929 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
2930 Register method_or_table_index,
2931 Register flags) {
2932 // setup registers
2933 const Register index = rdx;
2934 assert_different_registers(index, cache);
2935 assert_different_registers(method_or_table_index, cache, flags);
2936
2937 // determine constant pool cache field offsets
2938 resolve_cache_and_index_for_method(f2_byte, cache, index);
2939 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2940
2941 // method_or_table_index can either be an itable index or a method depending on the virtual final flag
2942 Label isVFinal; Label Done;
2943 __ testl(flags, (1 << ResolvedMethodEntry::is_vfinal_shift));
2944 __ jcc(Assembler::notZero, isVFinal);
2945 __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2946 __ jmp(Done);
2947 __ bind(isVFinal);
2948 __ movptr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2949 __ bind(Done);
2950 }
2951
2952 // The registers cache and index expected to be set before call.
2953 // Correct values of the cache and index registers are preserved.
2954 void TemplateTable::jvmti_post_field_access(Register cache,
2955 Register index,
2956 bool is_static,
2957 bool has_tos) {
2958 if (JvmtiExport::can_post_field_access()) {
2959 // Check to see if a field access watch has been set before we take
2960 // the time to call into the VM.
2961 Label L1;
2962 assert_different_registers(cache, index, rax);
2963 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2964 __ testl(rax,rax);
2965 __ jcc(Assembler::zero, L1);
2966
2967 // cache entry pointer
2968 __ load_field_entry(cache, index);
2969 if (is_static) {
2970 __ xorptr(rax, rax); // null object reference
2971 } else {
2972 __ pop(atos); // Get the object
2973 __ verify_oop(rax);
2974 __ push(atos); // Restore stack state
2975 }
2976 // rax,: object pointer or null
2977 // cache: cache entry pointer
2978 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2979 rax, cache);
2980
2981 __ load_field_entry(cache, index);
2982 __ bind(L1);
2983 }
2984 }
2985
2986 void TemplateTable::pop_and_check_object(Register r) {
2987 __ pop_ptr(r);
2988 __ null_check(r); // for field access must check obj.
2989 __ verify_oop(r);
2990 }
2991
2992 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2993 transition(vtos, vtos);
2994
2995 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2996 const Register cache = rcx;
2997 const Register index = rdx;
2998 const Register off = rbx;
2999 const Register tos_state = rax;
3000 const Register flags = rdx;
3001 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
3002
3003 resolve_cache_and_index_for_field(byte_no, cache, index);
3004 jvmti_post_field_access(cache, index, is_static, false);
3005 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3006
3007 if (!is_static) pop_and_check_object(obj);
3008
3009 const Address field(obj, off, Address::times_1, 0*wordSize);
3010
3011 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
3012
3013 // Make sure we don't need to mask edx after the above shift
3014 assert(btos == 0, "change code, btos != 0");
3015 __ testl(tos_state, tos_state);
3016 __ jcc(Assembler::notZero, notByte);
3017
3018 // btos
3019 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3020 __ push(btos);
3021 // Rewrite bytecode to be faster
3022 if (!is_static && rc == may_rewrite) {
3023 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3024 }
3025 __ jmp(Done);
3026
3027 __ bind(notByte);
3028 __ cmpl(tos_state, ztos);
3029 __ jcc(Assembler::notEqual, notBool);
3030
3031 // ztos (same code as btos)
3032 __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
3033 __ push(ztos);
3034 // Rewrite bytecode to be faster
3035 if (!is_static && rc == may_rewrite) {
3036 // use btos rewriting, no truncating to t/f bit is needed for getfield.
3037 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3038 }
3039 __ jmp(Done);
3040
3041 __ bind(notBool);
3042 __ cmpl(tos_state, atos);
3043 __ jcc(Assembler::notEqual, notObj);
3044 // atos
3045 do_oop_load(_masm, field, rax);
3046 __ push(atos);
3047 if (!is_static && rc == may_rewrite) {
3048 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3049 }
3050 __ jmp(Done);
3051
3052 __ bind(notObj);
3053 __ cmpl(tos_state, itos);
3054 __ jcc(Assembler::notEqual, notInt);
3055 // itos
3056 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3057 __ push(itos);
3058 // Rewrite bytecode to be faster
3059 if (!is_static && rc == may_rewrite) {
3060 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
3061 }
3062 __ jmp(Done);
3063
3064 __ bind(notInt);
3065 __ cmpl(tos_state, ctos);
3066 __ jcc(Assembler::notEqual, notChar);
3067 // ctos
3068 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3069 __ push(ctos);
3070 // Rewrite bytecode to be faster
3071 if (!is_static && rc == may_rewrite) {
3072 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3073 }
3074 __ jmp(Done);
3075
3076 __ bind(notChar);
3077 __ cmpl(tos_state, stos);
3078 __ jcc(Assembler::notEqual, notShort);
3079 // stos
3080 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3081 __ push(stos);
3082 // Rewrite bytecode to be faster
3083 if (!is_static && rc == may_rewrite) {
3084 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3085 }
3086 __ jmp(Done);
3087
3088 __ bind(notShort);
3089 __ cmpl(tos_state, ltos);
3090 __ jcc(Assembler::notEqual, notLong);
3091 // ltos
3092 // Generate code as if volatile (x86_32). There just aren't enough registers to
3093 // save that information and this code is faster than the test.
3094 __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
3095 __ push(ltos);
3096 // Rewrite bytecode to be faster
3097 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3098 __ jmp(Done);
3099
3100 __ bind(notLong);
3101 __ cmpl(tos_state, ftos);
3102 __ jcc(Assembler::notEqual, notFloat);
3103 // ftos
3104
3105 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3106 __ push(ftos);
3107 // Rewrite bytecode to be faster
3108 if (!is_static && rc == may_rewrite) {
3109 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3110 }
3111 __ jmp(Done);
3112
3113 __ bind(notFloat);
3114 #ifdef ASSERT
3115 Label notDouble;
3116 __ cmpl(tos_state, dtos);
3117 __ jcc(Assembler::notEqual, notDouble);
3118 #endif
3119 // dtos
3120 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3121 __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
3122 __ push(dtos);
3123 // Rewrite bytecode to be faster
3124 if (!is_static && rc == may_rewrite) {
3125 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3126 }
3127 #ifdef ASSERT
3128 __ jmp(Done);
3129
3130 __ bind(notDouble);
3131 __ stop("Bad state");
3132 #endif
3133
3134 __ bind(Done);
3135 // [jk] not needed currently
3136 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3137 // Assembler::LoadStore));
3138 }
3139
3140 void TemplateTable::getfield(int byte_no) {
3141 getfield_or_static(byte_no, false);
3142 }
3143
3144 void TemplateTable::nofast_getfield(int byte_no) {
3145 getfield_or_static(byte_no, false, may_not_rewrite);
3146 }
3147
3148 void TemplateTable::getstatic(int byte_no) {
3149 getfield_or_static(byte_no, true);
3150 }
3151
3152
3153 // The registers cache and index expected to be set before call.
3154 // The function may destroy various registers, just not the cache and index registers.
3155 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3156 // Cache is rcx and index is rdx
3157 const Register entry = LP64_ONLY(c_rarg2) NOT_LP64(rax); // ResolvedFieldEntry
3158 const Register obj = LP64_ONLY(c_rarg1) NOT_LP64(rbx); // Object pointer
3159 const Register value = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // JValue object
3160
3161 if (JvmtiExport::can_post_field_modification()) {
3162 // Check to see if a field modification watch has been set before
3163 // we take the time to call into the VM.
3164 Label L1;
3165 assert_different_registers(cache, obj, rax);
3166 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3167 __ testl(rax, rax);
3168 __ jcc(Assembler::zero, L1);
3169
3170 __ mov(entry, cache);
3171
3172 if (is_static) {
3173 // Life is simple. Null out the object pointer.
3174 __ xorl(obj, obj);
3175
3176 } else {
3177 // Life is harder. The stack holds the value on top, followed by
3178 // the object. We don't know the size of the value, though; it
3179 // could be one or two words depending on its type. As a result,
3180 // we must find the type to determine where the object is.
3181 #ifndef _LP64
3182 Label two_word, valsize_known;
3183 #endif
3184 __ load_unsigned_byte(value, Address(entry, in_bytes(ResolvedFieldEntry::type_offset())));
3185 #ifdef _LP64
3186 __ movptr(obj, at_tos_p1()); // initially assume a one word jvalue
3187 __ cmpl(value, ltos);
3188 __ cmovptr(Assembler::equal,
3189 obj, at_tos_p2()); // ltos (two word jvalue)
3190 __ cmpl(value, dtos);
3191 __ cmovptr(Assembler::equal,
3192 obj, at_tos_p2()); // dtos (two word jvalue)
3193 #else
3194 __ mov(obj, rsp);
3195 __ cmpl(value, ltos);
3196 __ jccb(Assembler::equal, two_word);
3197 __ cmpl(value, dtos);
3198 __ jccb(Assembler::equal, two_word);
3199 __ addptr(obj, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3200 __ jmpb(valsize_known);
3201
3202 __ bind(two_word);
3203 __ addptr(obj, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3204
3205 __ bind(valsize_known);
3206 // setup object pointer
3207 __ movptr(obj, Address(obj, 0));
3208 #endif
3209 }
3210
3211 // object (tos)
3212 __ mov(value, rsp);
3213 // obj: object pointer set up above (null if static)
3214 // cache: field entry pointer
3215 // value: jvalue object on the stack
3216 __ call_VM(noreg,
3217 CAST_FROM_FN_PTR(address,
3218 InterpreterRuntime::post_field_modification),
3219 obj, entry, value);
3220 // Reload field entry
3221 __ load_field_entry(cache, index);
3222 __ bind(L1);
3223 }
3224 }
3225
3226 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3227 transition(vtos, vtos);
3228
3229 const Register obj = rcx;
3230 const Register cache = rcx;
3231 const Register index = rdx;
3232 const Register tos_state = rdx;
3233 const Register off = rbx;
3234 const Register flags = rax;
3235
3236 resolve_cache_and_index_for_field(byte_no, cache, index);
3237 jvmti_post_field_mod(cache, index, is_static);
3238 load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3239
3240 // [jk] not needed currently
3241 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3242 // Assembler::StoreStore));
3243
3244 Label notVolatile, Done;
3245
3246 // Check for volatile store
3247 __ andl(flags, (1 << ResolvedFieldEntry::is_volatile_shift));
3248 __ testl(flags, flags);
3249 __ jcc(Assembler::zero, notVolatile);
3250
3251 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state);
3252 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3253 Assembler::StoreStore));
3254 __ jmp(Done);
3255 __ bind(notVolatile);
3256
3257 putfield_or_static_helper(byte_no, is_static, rc, obj, off, tos_state);
3258
3259 __ bind(Done);
3260 }
3261
3262 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3263 Register obj, Register off, Register tos_state) {
3264
3265 // field addresses
3266 const Address field(obj, off, Address::times_1, 0*wordSize);
3267 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3268
3269 Label notByte, notBool, notInt, notShort, notChar,
3270 notLong, notFloat, notObj;
3271 Label Done;
3272
3273 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3274
3275 // Test TOS state
3276 __ testl(tos_state, tos_state);
3277 __ jcc(Assembler::notZero, notByte);
3278
3279 // btos
3280 {
3281 __ pop(btos);
3282 if (!is_static) pop_and_check_object(obj);
3283 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3284 if (!is_static && rc == may_rewrite) {
3285 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3286 }
3287 __ jmp(Done);
3288 }
3289
3290 __ bind(notByte);
3291 __ cmpl(tos_state, ztos);
3292 __ jcc(Assembler::notEqual, notBool);
3293
3294 // ztos
3295 {
3296 __ pop(ztos);
3297 if (!is_static) pop_and_check_object(obj);
3298 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3299 if (!is_static && rc == may_rewrite) {
3300 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3301 }
3302 __ jmp(Done);
3303 }
3304
3305 __ bind(notBool);
3306 __ cmpl(tos_state, atos);
3307 __ jcc(Assembler::notEqual, notObj);
3308
3309 // atos
3310 {
3311 __ pop(atos);
3312 if (!is_static) pop_and_check_object(obj);
3313 // Store into the field
3314 do_oop_store(_masm, field, rax);
3315 if (!is_static && rc == may_rewrite) {
3316 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3317 }
3318 __ jmp(Done);
3319 }
3320
3321 __ bind(notObj);
3322 __ cmpl(tos_state, itos);
3323 __ jcc(Assembler::notEqual, notInt);
3324
3325 // itos
3326 {
3327 __ pop(itos);
3328 if (!is_static) pop_and_check_object(obj);
3329 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3330 if (!is_static && rc == may_rewrite) {
3331 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3332 }
3333 __ jmp(Done);
3334 }
3335
3336 __ bind(notInt);
3337 __ cmpl(tos_state, ctos);
3338 __ jcc(Assembler::notEqual, notChar);
3339
3340 // ctos
3341 {
3342 __ pop(ctos);
3343 if (!is_static) pop_and_check_object(obj);
3344 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3345 if (!is_static && rc == may_rewrite) {
3346 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3347 }
3348 __ jmp(Done);
3349 }
3350
3351 __ bind(notChar);
3352 __ cmpl(tos_state, stos);
3353 __ jcc(Assembler::notEqual, notShort);
3354
3355 // stos
3356 {
3357 __ pop(stos);
3358 if (!is_static) pop_and_check_object(obj);
3359 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3360 if (!is_static && rc == may_rewrite) {
3361 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3362 }
3363 __ jmp(Done);
3364 }
3365
3366 __ bind(notShort);
3367 __ cmpl(tos_state, ltos);
3368 __ jcc(Assembler::notEqual, notLong);
3369
3370 // ltos
3371 {
3372 __ pop(ltos);
3373 if (!is_static) pop_and_check_object(obj);
3374 // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3375 __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg, noreg);
3376 #ifdef _LP64
3377 if (!is_static && rc == may_rewrite) {
3378 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3379 }
3380 #endif // _LP64
3381 __ jmp(Done);
3382 }
3383
3384 __ bind(notLong);
3385 __ cmpl(tos_state, ftos);
3386 __ jcc(Assembler::notEqual, notFloat);
3387
3388 // ftos
3389 {
3390 __ pop(ftos);
3391 if (!is_static) pop_and_check_object(obj);
3392 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3393 if (!is_static && rc == may_rewrite) {
3394 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3395 }
3396 __ jmp(Done);
3397 }
3398
3399 __ bind(notFloat);
3400 #ifdef ASSERT
3401 Label notDouble;
3402 __ cmpl(tos_state, dtos);
3403 __ jcc(Assembler::notEqual, notDouble);
3404 #endif
3405
3406 // dtos
3407 {
3408 __ pop(dtos);
3409 if (!is_static) pop_and_check_object(obj);
3410 // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3411 __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg, noreg);
3412 if (!is_static && rc == may_rewrite) {
3413 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3414 }
3415 }
3416
3417 #ifdef ASSERT
3418 __ jmp(Done);
3419
3420 __ bind(notDouble);
3421 __ stop("Bad state");
3422 #endif
3423
3424 __ bind(Done);
3425 }
3426
3427 void TemplateTable::putfield(int byte_no) {
3428 putfield_or_static(byte_no, false);
3429 }
3430
3431 void TemplateTable::nofast_putfield(int byte_no) {
3432 putfield_or_static(byte_no, false, may_not_rewrite);
3433 }
3434
3435 void TemplateTable::putstatic(int byte_no) {
3436 putfield_or_static(byte_no, true);
3437 }
3438
3439 void TemplateTable::jvmti_post_fast_field_mod() {
3440
3441 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3442
3443 if (JvmtiExport::can_post_field_modification()) {
3444 // Check to see if a field modification watch has been set before
3445 // we take the time to call into the VM.
3446 Label L2;
3447 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3448 __ testl(scratch, scratch);
3449 __ jcc(Assembler::zero, L2);
3450 __ pop_ptr(rbx); // copy the object pointer from tos
3451 __ verify_oop(rbx);
3452 __ push_ptr(rbx); // put the object pointer back on tos
3453 // Save tos values before call_VM() clobbers them. Since we have
3454 // to do it for every data type, we use the saved values as the
3455 // jvalue object.
3456 switch (bytecode()) { // load values into the jvalue object
3457 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3458 case Bytecodes::_fast_bputfield: // fall through
3459 case Bytecodes::_fast_zputfield: // fall through
3460 case Bytecodes::_fast_sputfield: // fall through
3461 case Bytecodes::_fast_cputfield: // fall through
3462 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3463 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3464 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3465 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3466
3467 default:
3468 ShouldNotReachHere();
3469 }
3470 __ mov(scratch, rsp); // points to jvalue on the stack
3471 // access constant pool cache entry
3472 LP64_ONLY(__ load_field_entry(c_rarg2, rax));
3473 NOT_LP64(__ load_field_entry(rax, rdx));
3474 __ verify_oop(rbx);
3475 // rbx: object pointer copied above
3476 // c_rarg2: cache entry pointer
3477 // c_rarg3: jvalue object on the stack
3478 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3479 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3480
3481 switch (bytecode()) { // restore tos values
3482 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3483 case Bytecodes::_fast_bputfield: // fall through
3484 case Bytecodes::_fast_zputfield: // fall through
3485 case Bytecodes::_fast_sputfield: // fall through
3486 case Bytecodes::_fast_cputfield: // fall through
3487 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3488 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3489 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3490 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3491 default: break;
3492 }
3493 __ bind(L2);
3494 }
3495 }
3496
3497 void TemplateTable::fast_storefield(TosState state) {
3498 transition(state, vtos);
3499
3500 Register cache = rcx;
3501
3502 Label notVolatile, Done;
3503
3504 jvmti_post_fast_field_mod();
3505
3506 __ push(rax);
3507 __ load_field_entry(rcx, rax);
3508 load_resolved_field_entry(noreg, cache, rax, rbx, rdx);
3509 // RBX: field offset, RAX: TOS, RDX: flags
3510 __ andl(rdx, (1 << ResolvedFieldEntry::is_volatile_shift));
3511 __ pop(rax);
3512
3513 // Get object from stack
3514 pop_and_check_object(rcx);
3515
3516 // field address
3517 const Address field(rcx, rbx, Address::times_1);
3518
3519 // Check for volatile store
3520 __ testl(rdx, rdx);
3521 __ jcc(Assembler::zero, notVolatile);
3522
3523 fast_storefield_helper(field, rax);
3524 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3525 Assembler::StoreStore));
3526 __ jmp(Done);
3527 __ bind(notVolatile);
3528
3529 fast_storefield_helper(field, rax);
3530
3531 __ bind(Done);
3532 }
3533
3534 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3535
3536 // access field
3537 switch (bytecode()) {
3538 case Bytecodes::_fast_aputfield:
3539 do_oop_store(_masm, field, rax);
3540 break;
3541 case Bytecodes::_fast_lputfield:
3542 #ifdef _LP64
3543 __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg, noreg);
3544 #else
3545 __ stop("should not be rewritten");
3546 #endif
3547 break;
3548 case Bytecodes::_fast_iputfield:
3549 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3550 break;
3551 case Bytecodes::_fast_zputfield:
3552 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3553 break;
3554 case Bytecodes::_fast_bputfield:
3555 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3556 break;
3557 case Bytecodes::_fast_sputfield:
3558 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3559 break;
3560 case Bytecodes::_fast_cputfield:
3561 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3562 break;
3563 case Bytecodes::_fast_fputfield:
3564 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg, noreg);
3565 break;
3566 case Bytecodes::_fast_dputfield:
3567 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg, noreg);
3568 break;
3569 default:
3570 ShouldNotReachHere();
3571 }
3572 }
3573
3574 void TemplateTable::fast_accessfield(TosState state) {
3575 transition(atos, state);
3576
3577 // Do the JVMTI work here to avoid disturbing the register state below
3578 if (JvmtiExport::can_post_field_access()) {
3579 // Check to see if a field access watch has been set before we
3580 // take the time to call into the VM.
3581 Label L1;
3582 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3583 __ testl(rcx, rcx);
3584 __ jcc(Assembler::zero, L1);
3585 // access constant pool cache entry
3586 LP64_ONLY(__ load_field_entry(c_rarg2, rcx));
3587 NOT_LP64(__ load_field_entry(rcx, rdx));
3588 __ verify_oop(rax);
3589 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3590 LP64_ONLY(__ mov(c_rarg1, rax));
3591 // c_rarg1: object pointer copied above
3592 // c_rarg2: cache entry pointer
3593 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3594 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3595 __ pop_ptr(rax); // restore object pointer
3596 __ bind(L1);
3597 }
3598
3599 // access constant pool cache
3600 __ load_field_entry(rcx, rbx);
3601 __ load_sized_value(rbx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3602
3603 // rax: object
3604 __ verify_oop(rax);
3605 __ null_check(rax);
3606 Address field(rax, rbx, Address::times_1);
3607
3608 // access field
3609 switch (bytecode()) {
3610 case Bytecodes::_fast_agetfield:
3611 do_oop_load(_masm, field, rax);
3612 __ verify_oop(rax);
3613 break;
3614 case Bytecodes::_fast_lgetfield:
3615 #ifdef _LP64
3616 __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3617 #else
3618 __ stop("should not be rewritten");
3619 #endif
3620 break;
3621 case Bytecodes::_fast_igetfield:
3622 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3623 break;
3624 case Bytecodes::_fast_bgetfield:
3625 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3626 break;
3627 case Bytecodes::_fast_sgetfield:
3628 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3629 break;
3630 case Bytecodes::_fast_cgetfield:
3631 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3632 break;
3633 case Bytecodes::_fast_fgetfield:
3634 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3635 break;
3636 case Bytecodes::_fast_dgetfield:
3637 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3638 break;
3639 default:
3640 ShouldNotReachHere();
3641 }
3642 // [jk] not needed currently
3643 // Label notVolatile;
3644 // __ testl(rdx, rdx);
3645 // __ jcc(Assembler::zero, notVolatile);
3646 // __ membar(Assembler::LoadLoad);
3647 // __ bind(notVolatile);
3648 }
3649
3650 void TemplateTable::fast_xaccess(TosState state) {
3651 transition(vtos, state);
3652
3653 // get receiver
3654 __ movptr(rax, aaddress(0));
3655 // access constant pool cache
3656 __ load_field_entry(rcx, rdx, 2);
3657 __ load_sized_value(rbx, Address(rcx, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3658
3659 // make sure exception is reported in correct bcp range (getfield is
3660 // next instruction)
3661 __ increment(rbcp);
3662 __ null_check(rax);
3663 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3664 switch (state) {
3665 case itos:
3666 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3667 break;
3668 case atos:
3669 do_oop_load(_masm, field, rax);
3670 __ verify_oop(rax);
3671 break;
3672 case ftos:
3673 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3674 break;
3675 default:
3676 ShouldNotReachHere();
3677 }
3678
3679 // [jk] not needed currently
3680 // Label notVolatile;
3681 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3682 // in_bytes(ConstantPoolCache::base_offset() +
3683 // ConstantPoolCacheEntry::flags_offset())));
3684 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3685 // __ testl(rdx, 0x1);
3686 // __ jcc(Assembler::zero, notVolatile);
3687 // __ membar(Assembler::LoadLoad);
3688 // __ bind(notVolatile);
3689
3690 __ decrement(rbcp);
3691 }
3692
3693 //-----------------------------------------------------------------------------
3694 // Calls
3695
3696 void TemplateTable::prepare_invoke(Register cache, Register recv, Register flags) {
3697 // determine flags
3698 const Bytecodes::Code code = bytecode();
3699 const bool load_receiver = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3700 assert_different_registers(recv, flags);
3701
3702 // save 'interpreter return address'
3703 __ save_bcp();
3704
3705 // Save flags and load TOS
3706 __ movl(rbcp, flags);
3707 __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
3708
3709 // load receiver if needed (after appendix is pushed so parameter size is correct)
3710 // Note: no return address pushed yet
3711 if (load_receiver) {
3712 __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
3713 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3714 const int receiver_is_at_end = -1; // back off one slot to get receiver
3715 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3716 __ movptr(recv, recv_addr);
3717 __ verify_oop(recv);
3718 }
3719
3720 // load return address
3721 {
3722 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3723 ExternalAddress table(table_addr);
3724 #ifdef _LP64
3725 __ lea(rscratch1, table);
3726 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
3727 #else
3728 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
3729 #endif // _LP64
3730 }
3731
3732 // push return address
3733 __ push(flags);
3734
3735 // Restore flags value from the constant pool cache entry, and restore rsi
3736 // for later null checks. r13 is the bytecode pointer
3737 __ movl(flags, rbcp);
3738 __ restore_bcp();
3739 }
3740
3741 void TemplateTable::invokevirtual_helper(Register index,
3742 Register recv,
3743 Register flags) {
3744 // Uses temporary registers rax, rdx
3745 assert_different_registers(index, recv, rax, rdx);
3746 assert(index == rbx, "");
3747 assert(recv == rcx, "");
3748
3749 // Test for an invoke of a final method
3750 Label notFinal;
3751 __ movl(rax, flags);
3752 __ andl(rax, (1 << ResolvedMethodEntry::is_vfinal_shift));
3753 __ jcc(Assembler::zero, notFinal);
3754
3755 const Register method = index; // method must be rbx
3756 assert(method == rbx,
3757 "Method* must be rbx for interpreter calling convention");
3758
3759 // do the call - the index is actually the method to call
3760 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3761
3762 // It's final, need a null check here!
3763 __ null_check(recv);
3764
3765 // profile this call
3766 __ profile_final_call(rax);
3767 __ profile_arguments_type(rax, method, rbcp, true);
3768
3769 __ jump_from_interpreted(method, rax);
3770
3771 __ bind(notFinal);
3772
3773 // get receiver klass
3774 __ load_klass(rax, recv, rscratch1);
3775
3776 // profile this call
3777 __ profile_virtual_call(rax, rlocals, rdx);
3778 // get target Method* & entry point
3779 __ lookup_virtual_method(rax, index, method);
3780
3781 __ profile_arguments_type(rdx, method, rbcp, true);
3782 __ jump_from_interpreted(method, rdx);
3783 }
3784
3785 void TemplateTable::invokevirtual(int byte_no) {
3786 transition(vtos, vtos);
3787 assert(byte_no == f2_byte, "use this argument");
3788
3789 load_resolved_method_entry_virtual(rcx, // ResolvedMethodEntry*
3790 rbx, // Method or itable index
3791 rdx); // Flags
3792 prepare_invoke(rcx, // ResolvedMethodEntry*
3793 rcx, // Receiver
3794 rdx); // flags
3795
3796 // rbx: index
3797 // rcx: receiver
3798 // rdx: flags
3799 invokevirtual_helper(rbx, rcx, rdx);
3800 }
3801
3802 void TemplateTable::invokespecial(int byte_no) {
3803 transition(vtos, vtos);
3804 assert(byte_no == f1_byte, "use this argument");
3805
3806 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
3807 rbx, // Method*
3808 rdx); // flags
3809 prepare_invoke(rcx,
3810 rcx, // get receiver also for null check
3811 rdx); // flags
3812
3813 __ verify_oop(rcx);
3814 __ null_check(rcx);
3815 // do the call
3816 __ profile_call(rax);
3817 __ profile_arguments_type(rax, rbx, rbcp, false);
3818 __ jump_from_interpreted(rbx, rax);
3819 }
3820
3821 void TemplateTable::invokestatic(int byte_no) {
3822 transition(vtos, vtos);
3823 assert(byte_no == f1_byte, "use this argument");
3824
3825 load_resolved_method_entry_special_or_static(rcx, // ResolvedMethodEntry*
3826 rbx, // Method*
3827 rdx // flags
3828 );
3829 prepare_invoke(rcx, rcx, rdx); // cache and flags
3830
3831 // do the call
3832 __ profile_call(rax);
3833 __ profile_arguments_type(rax, rbx, rbcp, false);
3834 __ jump_from_interpreted(rbx, rax);
3835 }
3836
3837
3838 void TemplateTable::fast_invokevfinal(int byte_no) {
3839 transition(vtos, vtos);
3840 assert(byte_no == f2_byte, "use this argument");
3841 __ stop("fast_invokevfinal not used on x86");
3842 }
3843
3844
3845 void TemplateTable::invokeinterface(int byte_no) {
3846 transition(vtos, vtos);
3847 assert(byte_no == f1_byte, "use this argument");
3848
3849 load_resolved_method_entry_interface(rcx, // ResolvedMethodEntry*
3850 rax, // Klass*
3851 rbx, // Method* or itable/vtable index
3852 rdx); // flags
3853 prepare_invoke(rcx, rcx, rdx); // receiver, flags
3854
3855 // First check for Object case, then private interface method,
3856 // then regular interface method.
3857
3858 // Special case of invokeinterface called for virtual method of
3859 // java.lang.Object. See cpCache.cpp for details.
3860 Label notObjectMethod;
3861 __ movl(rlocals, rdx);
3862 __ andl(rlocals, (1 << ResolvedMethodEntry::is_forced_virtual_shift));
3863 __ jcc(Assembler::zero, notObjectMethod);
3864
3865 invokevirtual_helper(rbx, rcx, rdx);
3866 // no return from above
3867 __ bind(notObjectMethod);
3868
3869 Label no_such_interface; // for receiver subtype check
3870 Register recvKlass; // used for exception processing
3871
3872 // Check for private method invocation - indicated by vfinal
3873 Label notVFinal;
3874 __ movl(rlocals, rdx);
3875 __ andl(rlocals, (1 << ResolvedMethodEntry::is_vfinal_shift));
3876 __ jcc(Assembler::zero, notVFinal);
3877
3878 // Get receiver klass into rlocals - also a null check
3879 __ load_klass(rlocals, rcx, rscratch1);
3880
3881 Label subtype;
3882 __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3883 // If we get here the typecheck failed
3884 recvKlass = rdx;
3885 __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3886 __ jmp(no_such_interface);
3887
3888 __ bind(subtype);
3889
3890 // do the call - rbx is actually the method to call
3891
3892 __ profile_final_call(rdx);
3893 __ profile_arguments_type(rdx, rbx, rbcp, true);
3894
3895 __ jump_from_interpreted(rbx, rdx);
3896 // no return from above
3897 __ bind(notVFinal);
3898
3899 // Get receiver klass into rdx - also a null check
3900 __ restore_locals(); // restore r14
3901 __ load_klass(rdx, rcx, rscratch1);
3902
3903 Label no_such_method;
3904
3905 // Preserve method for throw_AbstractMethodErrorVerbose.
3906 __ mov(rcx, rbx);
3907 // Receiver subtype check against REFC.
3908 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3909 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3910 rdx, rax, noreg,
3911 // outputs: scan temp. reg, scan temp. reg
3912 rbcp, rlocals,
3913 no_such_interface,
3914 /*return_method=*/false);
3915
3916 // profile this call
3917 __ restore_bcp(); // rbcp was destroyed by receiver type check
3918 __ profile_virtual_call(rdx, rbcp, rlocals);
3919
3920 // Get declaring interface class from method, and itable index
3921 __ load_method_holder(rax, rbx);
3922 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3923 __ subl(rbx, Method::itable_index_max);
3924 __ negl(rbx);
3925
3926 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3927 __ mov(rlocals, rdx);
3928 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3929 rlocals, rax, rbx,
3930 // outputs: method, scan temp. reg
3931 rbx, rbcp,
3932 no_such_interface);
3933
3934 // rbx: Method* to call
3935 // rcx: receiver
3936 // Check for abstract method error
3937 // Note: This should be done more efficiently via a throw_abstract_method_error
3938 // interpreter entry point and a conditional jump to it in case of a null
3939 // method.
3940 __ testptr(rbx, rbx);
3941 __ jcc(Assembler::zero, no_such_method);
3942
3943 __ profile_arguments_type(rdx, rbx, rbcp, true);
3944
3945 // do the call
3946 // rcx: receiver
3947 // rbx,: Method*
3948 __ jump_from_interpreted(rbx, rdx);
3949 __ should_not_reach_here();
3950
3951 // exception handling code follows...
3952 // note: must restore interpreter registers to canonical
3953 // state for exception handling to work correctly!
3954
3955 __ bind(no_such_method);
3956 // throw exception
3957 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3958 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3959 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3960 // Pass arguments for generating a verbose error message.
3961 #ifdef _LP64
3962 recvKlass = c_rarg1;
3963 Register method = c_rarg2;
3964 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3965 if (method != rcx) { __ movq(method, rcx); }
3966 #else
3967 recvKlass = rdx;
3968 Register method = rcx;
3969 #endif
3970 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3971 recvKlass, method);
3972 // The call_VM checks for exception, so we should never return here.
3973 __ should_not_reach_here();
3974
3975 __ bind(no_such_interface);
3976 // throw exception
3977 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3978 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3979 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3980 // Pass arguments for generating a verbose error message.
3981 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3982 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3983 recvKlass, rax);
3984 // the call_VM checks for exception, so we should never return here.
3985 __ should_not_reach_here();
3986 }
3987
3988 void TemplateTable::invokehandle(int byte_no) {
3989 transition(vtos, vtos);
3990 assert(byte_no == f1_byte, "use this argument");
3991 const Register rbx_method = rbx;
3992 const Register rax_mtype = rax;
3993 const Register rcx_recv = rcx;
3994 const Register rdx_flags = rdx;
3995
3996 load_resolved_method_entry_handle(rcx, rbx_method, rax_mtype, rdx_flags);
3997 prepare_invoke(rcx, rcx_recv, rdx_flags);
3998
3999 __ verify_method_ptr(rbx_method);
4000 __ verify_oop(rcx_recv);
4001 __ null_check(rcx_recv);
4002
4003 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4004 // rbx: MH.invokeExact_MT method
4005
4006 // Note: rax_mtype is already pushed (if necessary)
4007
4008 // FIXME: profile the LambdaForm also
4009 __ profile_final_call(rax);
4010 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4011
4012 __ jump_from_interpreted(rbx_method, rdx);
4013 }
4014
4015 void TemplateTable::invokedynamic(int byte_no) {
4016 transition(vtos, vtos);
4017 assert(byte_no == f1_byte, "use this argument");
4018
4019 const Register rbx_method = rbx;
4020 const Register rax_callsite = rax;
4021
4022 load_invokedynamic_entry(rbx_method);
4023 // rax: CallSite object (from cpool->resolved_references[])
4024 // rbx: MH.linkToCallSite method
4025
4026 // Note: rax_callsite is already pushed
4027
4028 // %%% should make a type profile for any invokedynamic that takes a ref argument
4029 // profile this call
4030 __ profile_call(rbcp);
4031 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4032
4033 __ verify_oop(rax_callsite);
4034
4035 __ jump_from_interpreted(rbx_method, rdx);
4036 }
4037
4038 //-----------------------------------------------------------------------------
4039 // Allocation
4040
4041 void TemplateTable::_new() {
4042 transition(vtos, atos);
4043 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4044 Label slow_case;
4045 Label slow_case_no_pop;
4046 Label done;
4047 Label initialize_header;
4048
4049 __ get_cpool_and_tags(rcx, rax);
4050
4051 // Make sure the class we're about to instantiate has been resolved.
4052 // This is done before loading InstanceKlass to be consistent with the order
4053 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4054 const int tags_offset = Array<u1>::base_offset_in_bytes();
4055 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4056 __ jcc(Assembler::notEqual, slow_case_no_pop);
4057
4058 // get InstanceKlass
4059 __ load_resolved_klass_at_index(rcx, rcx, rdx);
4060 __ push(rcx); // save the contexts of klass for initializing the header
4061
4062 // make sure klass is initialized
4063 // init_state needs acquire, but x86 is TSO, and so we are already good.
4064 #ifdef _LP64
4065 assert(VM_Version::supports_fast_class_init_checks(), "must support fast class initialization checks");
4066 __ clinit_barrier(rcx, r15_thread, nullptr /*L_fast_path*/, &slow_case);
4067 #else
4068 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4069 __ jcc(Assembler::notEqual, slow_case);
4070 #endif
4071
4072 // get instance_size in InstanceKlass (scaled to a count of bytes)
4073 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4074 // test to see if it is malformed in some way
4075 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4076 __ jcc(Assembler::notZero, slow_case);
4077
4078 // Allocate the instance:
4079 // If TLAB is enabled:
4080 // Try to allocate in the TLAB.
4081 // If fails, go to the slow path.
4082 // Initialize the allocation.
4083 // Exit.
4084 //
4085 // Go to slow path.
4086
4087 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4088
4089 if (UseTLAB) {
4090 NOT_LP64(__ get_thread(thread);)
4091 __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
4092 if (ZeroTLAB) {
4093 // the fields have been already cleared
4094 __ jmp(initialize_header);
4095 }
4096
4097 // The object is initialized before the header. If the object size is
4098 // zero, go directly to the header initialization.
4099 if (UseCompactObjectHeaders) {
4100 assert(is_aligned(oopDesc::base_offset_in_bytes(), BytesPerLong), "oop base offset must be 8-byte-aligned");
4101 __ decrement(rdx, oopDesc::base_offset_in_bytes());
4102 } else {
4103 __ decrement(rdx, sizeof(oopDesc));
4104 }
4105 __ jcc(Assembler::zero, initialize_header);
4106
4107 // Initialize topmost object field, divide rdx by 8, check if odd and
4108 // test if zero.
4109 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4110 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4111
4112 // rdx must have been multiple of 8
4113 #ifdef ASSERT
4114 // make sure rdx was multiple of 8
4115 Label L;
4116 // Ignore partial flag stall after shrl() since it is debug VM
4117 __ jcc(Assembler::carryClear, L);
4118 __ stop("object size is not multiple of 2 - adjust this code");
4119 __ bind(L);
4120 // rdx must be > 0, no extra check needed here
4121 #endif
4122
4123 // initialize remaining object fields: rdx was a multiple of 8
4124 { Label loop;
4125 __ bind(loop);
4126 int header_size_bytes = oopDesc::header_size() * HeapWordSize;
4127 assert(is_aligned(header_size_bytes, BytesPerLong), "oop header size must be 8-byte-aligned");
4128 __ movptr(Address(rax, rdx, Address::times_8, header_size_bytes - 1*oopSize), rcx);
4129 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, header_size_bytes - 2*oopSize), rcx));
4130 __ decrement(rdx);
4131 __ jcc(Assembler::notZero, loop);
4132 }
4133
4134 // initialize object header only.
4135 __ bind(initialize_header);
4136 if (UseCompactObjectHeaders) {
4137 __ pop(rcx); // get saved klass back in the register.
4138 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4139 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rbx);
4140 } else {
4141 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
4142 (intptr_t)markWord::prototype().value()); // header
4143 __ pop(rcx); // get saved klass back in the register.
4144 #ifdef _LP64
4145 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4146 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4147 #endif
4148 __ store_klass(rax, rcx, rscratch1); // klass
4149 }
4150
4151 if (DTraceAllocProbes) {
4152 // Trigger dtrace event for fastpath
4153 __ push(atos);
4154 __ call_VM_leaf(
4155 CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), rax);
4156 __ pop(atos);
4157 }
4158
4159 __ jmp(done);
4160 }
4161
4162 // slow case
4163 __ bind(slow_case);
4164 __ pop(rcx); // restore stack pointer to what it was when we came in.
4165 __ bind(slow_case_no_pop);
4166
4167 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4168 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4169
4170 __ get_constant_pool(rarg1);
4171 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4172 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4173 __ verify_oop(rax);
4174
4175 // continue
4176 __ bind(done);
4177 }
4178
4179 void TemplateTable::newarray() {
4180 transition(itos, atos);
4181 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4182 __ load_unsigned_byte(rarg1, at_bcp(1));
4183 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4184 rarg1, rax);
4185 }
4186
4187 void TemplateTable::anewarray() {
4188 transition(itos, atos);
4189
4190 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4191 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4192
4193 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4194 __ get_constant_pool(rarg1);
4195 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4196 rarg1, rarg2, rax);
4197 }
4198
4199 void TemplateTable::arraylength() {
4200 transition(atos, itos);
4201 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4202 }
4203
4204 void TemplateTable::checkcast() {
4205 transition(atos, atos);
4206 Label done, is_null, ok_is_subtype, quicked, resolved;
4207 __ testptr(rax, rax); // object is in rax
4208 __ jcc(Assembler::zero, is_null);
4209
4210 // Get cpool & tags index
4211 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4212 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4213 // See if bytecode has already been quicked
4214 __ cmpb(Address(rdx, rbx,
4215 Address::times_1,
4216 Array<u1>::base_offset_in_bytes()),
4217 JVM_CONSTANT_Class);
4218 __ jcc(Assembler::equal, quicked);
4219 __ push(atos); // save receiver for result, and for GC
4220 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4221
4222 // vm_result_2 has metadata result
4223 #ifndef _LP64
4224 // borrow rdi from locals
4225 __ get_thread(rdi);
4226 __ get_vm_result_2(rax, rdi);
4227 __ restore_locals();
4228 #else
4229 __ get_vm_result_2(rax, r15_thread);
4230 #endif
4231
4232 __ pop_ptr(rdx); // restore receiver
4233 __ jmpb(resolved);
4234
4235 // Get superklass in rax and subklass in rbx
4236 __ bind(quicked);
4237 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4238 __ load_resolved_klass_at_index(rax, rcx, rbx);
4239
4240 __ bind(resolved);
4241 __ load_klass(rbx, rdx, rscratch1);
4242
4243 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4244 // Superklass in rax. Subklass in rbx.
4245 __ gen_subtype_check(rbx, ok_is_subtype);
4246
4247 // Come here on failure
4248 __ push_ptr(rdx);
4249 // object is at TOS
4250 __ jump(RuntimeAddress(Interpreter::_throw_ClassCastException_entry));
4251
4252 // Come here on success
4253 __ bind(ok_is_subtype);
4254 __ mov(rax, rdx); // Restore object in rdx
4255
4256 // Collect counts on whether this check-cast sees nulls a lot or not.
4257 if (ProfileInterpreter) {
4258 __ jmp(done);
4259 __ bind(is_null);
4260 __ profile_null_seen(rcx);
4261 } else {
4262 __ bind(is_null); // same as 'done'
4263 }
4264 __ bind(done);
4265 }
4266
4267 void TemplateTable::instanceof() {
4268 transition(atos, itos);
4269 Label done, is_null, ok_is_subtype, quicked, resolved;
4270 __ testptr(rax, rax);
4271 __ jcc(Assembler::zero, is_null);
4272
4273 // Get cpool & tags index
4274 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4275 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4276 // See if bytecode has already been quicked
4277 __ cmpb(Address(rdx, rbx,
4278 Address::times_1,
4279 Array<u1>::base_offset_in_bytes()),
4280 JVM_CONSTANT_Class);
4281 __ jcc(Assembler::equal, quicked);
4282
4283 __ push(atos); // save receiver for result, and for GC
4284 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4285 // vm_result_2 has metadata result
4286
4287 #ifndef _LP64
4288 // borrow rdi from locals
4289 __ get_thread(rdi);
4290 __ get_vm_result_2(rax, rdi);
4291 __ restore_locals();
4292 #else
4293 __ get_vm_result_2(rax, r15_thread);
4294 #endif
4295
4296 __ pop_ptr(rdx); // restore receiver
4297 __ verify_oop(rdx);
4298 __ load_klass(rdx, rdx, rscratch1);
4299 __ jmpb(resolved);
4300
4301 // Get superklass in rax and subklass in rdx
4302 __ bind(quicked);
4303 __ load_klass(rdx, rax, rscratch1);
4304 __ load_resolved_klass_at_index(rax, rcx, rbx);
4305
4306 __ bind(resolved);
4307
4308 // Generate subtype check. Blows rcx, rdi
4309 // Superklass in rax. Subklass in rdx.
4310 __ gen_subtype_check(rdx, ok_is_subtype);
4311
4312 // Come here on failure
4313 __ xorl(rax, rax);
4314 __ jmpb(done);
4315 // Come here on success
4316 __ bind(ok_is_subtype);
4317 __ movl(rax, 1);
4318
4319 // Collect counts on whether this test sees nulls a lot or not.
4320 if (ProfileInterpreter) {
4321 __ jmp(done);
4322 __ bind(is_null);
4323 __ profile_null_seen(rcx);
4324 } else {
4325 __ bind(is_null); // same as 'done'
4326 }
4327 __ bind(done);
4328 // rax = 0: obj == nullptr or obj is not an instanceof the specified klass
4329 // rax = 1: obj != nullptr and obj is an instanceof the specified klass
4330 }
4331
4332
4333 //----------------------------------------------------------------------------------------------------
4334 // Breakpoints
4335 void TemplateTable::_breakpoint() {
4336 // Note: We get here even if we are single stepping..
4337 // jbug insists on setting breakpoints at every bytecode
4338 // even if we are in single step mode.
4339
4340 transition(vtos, vtos);
4341
4342 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4343
4344 // get the unpatched byte code
4345 __ get_method(rarg);
4346 __ call_VM(noreg,
4347 CAST_FROM_FN_PTR(address,
4348 InterpreterRuntime::get_original_bytecode_at),
4349 rarg, rbcp);
4350 __ mov(rbx, rax); // why?
4351
4352 // post the breakpoint event
4353 __ get_method(rarg);
4354 __ call_VM(noreg,
4355 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4356 rarg, rbcp);
4357
4358 // complete the execution of original bytecode
4359 __ dispatch_only_normal(vtos);
4360 }
4361
4362 //-----------------------------------------------------------------------------
4363 // Exceptions
4364
4365 void TemplateTable::athrow() {
4366 transition(atos, vtos);
4367 __ null_check(rax);
4368 __ jump(RuntimeAddress(Interpreter::throw_exception_entry()));
4369 }
4370
4371 //-----------------------------------------------------------------------------
4372 // Synchronization
4373 //
4374 // Note: monitorenter & exit are symmetric routines; which is reflected
4375 // in the assembly code structure as well
4376 //
4377 // Stack layout:
4378 //
4379 // [expressions ] <--- rsp = expression stack top
4380 // ..
4381 // [expressions ]
4382 // [monitor entry] <--- monitor block top = expression stack bot
4383 // ..
4384 // [monitor entry]
4385 // [frame data ] <--- monitor block bot
4386 // ...
4387 // [saved rbp ] <--- rbp
4388 void TemplateTable::monitorenter() {
4389 transition(atos, vtos);
4390
4391 // check for null object
4392 __ null_check(rax);
4393
4394 const Address monitor_block_top(
4395 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4396 const Address monitor_block_bot(
4397 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4398 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4399
4400 Label allocated;
4401
4402 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4403 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4404 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4405
4406 // initialize entry pointer
4407 __ xorl(rmon, rmon); // points to free slot or null
4408
4409 // find a free slot in the monitor block (result in rmon)
4410 {
4411 Label entry, loop, exit;
4412 __ movptr(rtop, monitor_block_top); // derelativize pointer
4413 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4414 // rtop points to current entry, starting with top-most entry
4415
4416 __ lea(rbot, monitor_block_bot); // points to word before bottom
4417 // of monitor block
4418 __ jmpb(entry);
4419
4420 __ bind(loop);
4421 // check if current entry is used
4422 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset()), NULL_WORD);
4423 // if not used then remember entry in rmon
4424 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4425 // check if current entry is for same object
4426 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4427 // if same object then stop searching
4428 __ jccb(Assembler::equal, exit);
4429 // otherwise advance to next entry
4430 __ addptr(rtop, entry_size);
4431 __ bind(entry);
4432 // check if bottom reached
4433 __ cmpptr(rtop, rbot);
4434 // if not at bottom then check this entry
4435 __ jcc(Assembler::notEqual, loop);
4436 __ bind(exit);
4437 }
4438
4439 __ testptr(rmon, rmon); // check if a slot has been found
4440 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4441
4442 // allocate one if there's no free slot
4443 {
4444 Label entry, loop;
4445 // 1. compute new pointers // rsp: old expression stack top
4446 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4447 __ lea(rmon, Address(rbp, rmon, Address::times_ptr));
4448 __ subptr(rsp, entry_size); // move expression stack top
4449 __ subptr(rmon, entry_size); // move expression stack bottom
4450 __ mov(rtop, rsp); // set start value for copy loop
4451 __ subptr(monitor_block_bot, entry_size / wordSize); // set new monitor block bottom
4452 __ jmp(entry);
4453 // 2. move expression stack contents
4454 __ bind(loop);
4455 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4456 // word from old location
4457 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4458 __ addptr(rtop, wordSize); // advance to next word
4459 __ bind(entry);
4460 __ cmpptr(rtop, rmon); // check if bottom reached
4461 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4462 // copy next word
4463 }
4464
4465 // call run-time routine
4466 // rmon: points to monitor entry
4467 __ bind(allocated);
4468
4469 // Increment bcp to point to the next bytecode, so exception
4470 // handling for async. exceptions work correctly.
4471 // The object has already been popped from the stack, so the
4472 // expression stack looks correct.
4473 __ increment(rbcp);
4474
4475 // store object
4476 __ movptr(Address(rmon, BasicObjectLock::obj_offset()), rax);
4477 __ lock_object(rmon);
4478
4479 // check to make sure this monitor doesn't cause stack overflow after locking
4480 __ save_bcp(); // in case of exception
4481 __ generate_stack_overflow_check(0);
4482
4483 // The bcp has already been incremented. Just need to dispatch to
4484 // next instruction.
4485 __ dispatch_next(vtos);
4486 }
4487
4488 void TemplateTable::monitorexit() {
4489 transition(atos, vtos);
4490
4491 // check for null object
4492 __ null_check(rax);
4493
4494 const Address monitor_block_top(
4495 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4496 const Address monitor_block_bot(
4497 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4498 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4499
4500 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4501 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4502
4503 Label found;
4504
4505 // find matching slot
4506 {
4507 Label entry, loop;
4508 __ movptr(rtop, monitor_block_top); // derelativize pointer
4509 __ lea(rtop, Address(rbp, rtop, Address::times_ptr));
4510 // rtop points to current entry, starting with top-most entry
4511
4512 __ lea(rbot, monitor_block_bot); // points to word before bottom
4513 // of monitor block
4514 __ jmpb(entry);
4515
4516 __ bind(loop);
4517 // check if current entry is for same object
4518 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset()));
4519 // if same object then stop searching
4520 __ jcc(Assembler::equal, found);
4521 // otherwise advance to next entry
4522 __ addptr(rtop, entry_size);
4523 __ bind(entry);
4524 // check if bottom reached
4525 __ cmpptr(rtop, rbot);
4526 // if not at bottom then check this entry
4527 __ jcc(Assembler::notEqual, loop);
4528 }
4529
4530 // error handling. Unlocking was not block-structured
4531 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4532 InterpreterRuntime::throw_illegal_monitor_state_exception));
4533 __ should_not_reach_here();
4534
4535 // call run-time routine
4536 __ bind(found);
4537 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4538 __ unlock_object(rtop);
4539 __ pop_ptr(rax); // discard object
4540 }
4541
4542 // Wide instructions
4543 void TemplateTable::wide() {
4544 transition(vtos, vtos);
4545 __ load_unsigned_byte(rbx, at_bcp(1));
4546 ExternalAddress wtable((address)Interpreter::_wentry_point);
4547 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)), rscratch1);
4548 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4549 }
4550
4551 // Multi arrays
4552 void TemplateTable::multianewarray() {
4553 transition(vtos, atos);
4554
4555 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4556 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4557 // last dim is on top of stack; we want address of first one:
4558 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4559 // the latter wordSize to point to the beginning of the array.
4560 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4561 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4562 __ load_unsigned_byte(rbx, at_bcp(3));
4563 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4564 }