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