Print this page
| Split |
Close |
| Expand all |
| Collapse all |
--- old/src/cpu/x86/vm/templateTable_x86_32.cpp
+++ new/src/cpu/x86/vm/templateTable_x86_32.cpp
1 1 /*
2 2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
3 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 4 *
5 5 * This code is free software; you can redistribute it and/or modify it
6 6 * under the terms of the GNU General Public License version 2 only, as
7 7 * published by the Free Software Foundation.
8 8 *
9 9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 12 * version 2 for more details (a copy is included in the LICENSE file that
13 13 * accompanied this code).
14 14 *
15 15 * You should have received a copy of the GNU General Public License version
16 16 * 2 along with this work; if not, write to the Free Software Foundation,
17 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 18 *
19 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 20 * or visit www.oracle.com if you need additional information or have any
21 21 * questions.
22 22 *
23 23 */
24 24
25 25 #include "precompiled.hpp"
26 26 #include "asm/assembler.hpp"
27 27 #include "interpreter/interpreter.hpp"
28 28 #include "interpreter/interpreterRuntime.hpp"
29 29 #include "interpreter/templateTable.hpp"
30 30 #include "memory/universe.inline.hpp"
31 31 #include "oops/methodDataOop.hpp"
32 32 #include "oops/objArrayKlass.hpp"
33 33 #include "oops/oop.inline.hpp"
34 34 #include "prims/methodHandles.hpp"
35 35 #include "runtime/sharedRuntime.hpp"
36 36 #include "runtime/stubRoutines.hpp"
37 37 #include "runtime/synchronizer.hpp"
38 38
39 39 #ifndef CC_INTERP
40 40 #define __ _masm->
41 41
42 42 //----------------------------------------------------------------------------------------------------
43 43 // Platform-dependent initialization
44 44
45 45 void TemplateTable::pd_initialize() {
46 46 // No i486 specific initialization
47 47 }
48 48
49 49 //----------------------------------------------------------------------------------------------------
50 50 // Address computation
51 51
52 52 // local variables
53 53 static inline Address iaddress(int n) {
54 54 return Address(rdi, Interpreter::local_offset_in_bytes(n));
55 55 }
56 56
57 57 static inline Address laddress(int n) { return iaddress(n + 1); }
58 58 static inline Address haddress(int n) { return iaddress(n + 0); }
59 59 static inline Address faddress(int n) { return iaddress(n); }
60 60 static inline Address daddress(int n) { return laddress(n); }
61 61 static inline Address aaddress(int n) { return iaddress(n); }
62 62
63 63 static inline Address iaddress(Register r) {
64 64 return Address(rdi, r, Interpreter::stackElementScale());
65 65 }
66 66 static inline Address laddress(Register r) {
67 67 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
68 68 }
69 69 static inline Address haddress(Register r) {
70 70 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
71 71 }
72 72
73 73 static inline Address faddress(Register r) { return iaddress(r); }
74 74 static inline Address daddress(Register r) { return laddress(r); }
75 75 static inline Address aaddress(Register r) { return iaddress(r); }
76 76
77 77 // expression stack
78 78 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
79 79 // data beyond the rsp which is potentially unsafe in an MT environment;
80 80 // an interrupt may overwrite that data.)
81 81 static inline Address at_rsp () {
82 82 return Address(rsp, 0);
83 83 }
84 84
85 85 // At top of Java expression stack which may be different than rsp(). It
86 86 // isn't for category 1 objects.
87 87 static inline Address at_tos () {
88 88 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
89 89 return tos;
90 90 }
91 91
92 92 static inline Address at_tos_p1() {
93 93 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
94 94 }
95 95
96 96 static inline Address at_tos_p2() {
97 97 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
98 98 }
99 99
100 100 // Condition conversion
101 101 static Assembler::Condition j_not(TemplateTable::Condition cc) {
102 102 switch (cc) {
103 103 case TemplateTable::equal : return Assembler::notEqual;
104 104 case TemplateTable::not_equal : return Assembler::equal;
105 105 case TemplateTable::less : return Assembler::greaterEqual;
106 106 case TemplateTable::less_equal : return Assembler::greater;
107 107 case TemplateTable::greater : return Assembler::lessEqual;
108 108 case TemplateTable::greater_equal: return Assembler::less;
109 109 }
110 110 ShouldNotReachHere();
111 111 return Assembler::zero;
112 112 }
113 113
114 114
115 115 //----------------------------------------------------------------------------------------------------
116 116 // Miscelaneous helper routines
117 117
118 118 // Store an oop (or NULL) at the address described by obj.
119 119 // If val == noreg this means store a NULL
120 120
121 121 static void do_oop_store(InterpreterMacroAssembler* _masm,
122 122 Address obj,
123 123 Register val,
124 124 BarrierSet::Name barrier,
125 125 bool precise) {
126 126 assert(val == noreg || val == rax, "parameter is just for looks");
127 127 switch (barrier) {
128 128 #ifndef SERIALGC
129 129 case BarrierSet::G1SATBCT:
130 130 case BarrierSet::G1SATBCTLogging:
131 131 {
132 132 // flatten object address if needed
133 133 // We do it regardless of precise because we need the registers
134 134 if (obj.index() == noreg && obj.disp() == 0) {
135 135 if (obj.base() != rdx) {
136 136 __ movl(rdx, obj.base());
137 137 }
138 138 } else {
139 139 __ leal(rdx, obj);
140 140 }
141 141 __ get_thread(rcx);
142 142 __ save_bcp();
143 143 __ g1_write_barrier_pre(rdx, rcx, rsi, rbx, val != noreg);
144 144
145 145 // Do the actual store
146 146 // noreg means NULL
147 147 if (val == noreg) {
148 148 __ movptr(Address(rdx, 0), NULL_WORD);
149 149 // No post barrier for NULL
150 150 } else {
151 151 __ movl(Address(rdx, 0), val);
152 152 __ g1_write_barrier_post(rdx, rax, rcx, rbx, rsi);
153 153 }
154 154 __ restore_bcp();
155 155
156 156 }
157 157 break;
158 158 #endif // SERIALGC
159 159 case BarrierSet::CardTableModRef:
160 160 case BarrierSet::CardTableExtension:
161 161 {
162 162 if (val == noreg) {
163 163 __ movptr(obj, NULL_WORD);
164 164 } else {
165 165 __ movl(obj, val);
166 166 // flatten object address if needed
167 167 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
168 168 __ store_check(obj.base());
169 169 } else {
170 170 __ leal(rdx, obj);
171 171 __ store_check(rdx);
172 172 }
173 173 }
174 174 }
175 175 break;
176 176 case BarrierSet::ModRef:
177 177 case BarrierSet::Other:
178 178 if (val == noreg) {
179 179 __ movptr(obj, NULL_WORD);
180 180 } else {
181 181 __ movl(obj, val);
182 182 }
183 183 break;
184 184 default :
185 185 ShouldNotReachHere();
186 186
187 187 }
188 188 }
189 189
190 190 Address TemplateTable::at_bcp(int offset) {
191 191 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
192 192 return Address(rsi, offset);
193 193 }
194 194
195 195
196 196 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
197 197 Register scratch,
198 198 bool load_bc_into_scratch/*=true*/) {
199 199
200 200 if (!RewriteBytecodes) return;
201 201 // the pair bytecodes have already done the load.
202 202 if (load_bc_into_scratch) {
203 203 __ movl(bc, bytecode);
204 204 }
205 205 Label patch_done;
206 206 if (JvmtiExport::can_post_breakpoint()) {
207 207 Label fast_patch;
208 208 // if a breakpoint is present we can't rewrite the stream directly
209 209 __ movzbl(scratch, at_bcp(0));
210 210 __ cmpl(scratch, Bytecodes::_breakpoint);
211 211 __ jcc(Assembler::notEqual, fast_patch);
212 212 __ get_method(scratch);
213 213 // Let breakpoint table handling rewrite to quicker bytecode
214 214 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
215 215 #ifndef ASSERT
216 216 __ jmpb(patch_done);
217 217 #else
218 218 __ jmp(patch_done);
219 219 #endif
220 220 __ bind(fast_patch);
221 221 }
222 222 #ifdef ASSERT
223 223 Label okay;
224 224 __ load_unsigned_byte(scratch, at_bcp(0));
225 225 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
226 226 __ jccb(Assembler::equal, okay);
227 227 __ cmpl(scratch, bc);
228 228 __ jcc(Assembler::equal, okay);
229 229 __ stop("patching the wrong bytecode");
230 230 __ bind(okay);
231 231 #endif
232 232 // patch bytecode
233 233 __ movb(at_bcp(0), bc);
234 234 __ bind(patch_done);
235 235 }
236 236
237 237 //----------------------------------------------------------------------------------------------------
238 238 // Individual instructions
239 239
240 240 void TemplateTable::nop() {
241 241 transition(vtos, vtos);
242 242 // nothing to do
243 243 }
244 244
245 245 void TemplateTable::shouldnotreachhere() {
246 246 transition(vtos, vtos);
247 247 __ stop("shouldnotreachhere bytecode");
248 248 }
249 249
250 250
251 251
252 252 void TemplateTable::aconst_null() {
253 253 transition(vtos, atos);
254 254 __ xorptr(rax, rax);
255 255 }
256 256
257 257
258 258 void TemplateTable::iconst(int value) {
259 259 transition(vtos, itos);
260 260 if (value == 0) {
261 261 __ xorptr(rax, rax);
262 262 } else {
263 263 __ movptr(rax, value);
264 264 }
265 265 }
266 266
267 267
268 268 void TemplateTable::lconst(int value) {
269 269 transition(vtos, ltos);
270 270 if (value == 0) {
271 271 __ xorptr(rax, rax);
272 272 } else {
273 273 __ movptr(rax, value);
274 274 }
275 275 assert(value >= 0, "check this code");
276 276 __ xorptr(rdx, rdx);
277 277 }
278 278
279 279
280 280 void TemplateTable::fconst(int value) {
281 281 transition(vtos, ftos);
282 282 if (value == 0) { __ fldz();
283 283 } else if (value == 1) { __ fld1();
284 284 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
285 285 } else { ShouldNotReachHere();
286 286 }
287 287 }
288 288
289 289
290 290 void TemplateTable::dconst(int value) {
291 291 transition(vtos, dtos);
292 292 if (value == 0) { __ fldz();
293 293 } else if (value == 1) { __ fld1();
294 294 } else { ShouldNotReachHere();
295 295 }
296 296 }
297 297
298 298
299 299 void TemplateTable::bipush() {
300 300 transition(vtos, itos);
301 301 __ load_signed_byte(rax, at_bcp(1));
302 302 }
303 303
304 304
305 305 void TemplateTable::sipush() {
306 306 transition(vtos, itos);
307 307 __ load_unsigned_short(rax, at_bcp(1));
308 308 __ bswapl(rax);
309 309 __ sarl(rax, 16);
310 310 }
311 311
312 312 void TemplateTable::ldc(bool wide) {
313 313 transition(vtos, vtos);
314 314 Label call_ldc, notFloat, notClass, Done;
315 315
316 316 if (wide) {
317 317 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
318 318 } else {
319 319 __ load_unsigned_byte(rbx, at_bcp(1));
320 320 }
321 321 __ get_cpool_and_tags(rcx, rax);
322 322 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
323 323 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
324 324
325 325 // get type
326 326 __ xorptr(rdx, rdx);
327 327 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
328 328
329 329 // unresolved string - get the resolved string
330 330 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
331 331 __ jccb(Assembler::equal, call_ldc);
332 332
333 333 // unresolved class - get the resolved class
334 334 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
335 335 __ jccb(Assembler::equal, call_ldc);
336 336
337 337 // unresolved class in error (resolution failed) - call into runtime
338 338 // so that the same error from first resolution attempt is thrown.
339 339 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
340 340 __ jccb(Assembler::equal, call_ldc);
341 341
342 342 // resolved class - need to call vm to get java mirror of the class
343 343 __ cmpl(rdx, JVM_CONSTANT_Class);
344 344 __ jcc(Assembler::notEqual, notClass);
345 345
346 346 __ bind(call_ldc);
347 347 __ movl(rcx, wide);
348 348 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
349 349 __ push(atos);
350 350 __ jmp(Done);
351 351
352 352 __ bind(notClass);
353 353 __ cmpl(rdx, JVM_CONSTANT_Float);
354 354 __ jccb(Assembler::notEqual, notFloat);
355 355 // ftos
356 356 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
357 357 __ push(ftos);
358 358 __ jmp(Done);
359 359
360 360 __ bind(notFloat);
361 361 #ifdef ASSERT
362 362 { Label L;
363 363 __ cmpl(rdx, JVM_CONSTANT_Integer);
364 364 __ jcc(Assembler::equal, L);
365 365 __ cmpl(rdx, JVM_CONSTANT_String);
366 366 __ jcc(Assembler::equal, L);
367 367 __ stop("unexpected tag type in ldc");
368 368 __ bind(L);
369 369 }
370 370 #endif
371 371 Label isOop;
372 372 // atos and itos
373 373 // String is only oop type we will see here
374 374 __ cmpl(rdx, JVM_CONSTANT_String);
375 375 __ jccb(Assembler::equal, isOop);
376 376 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
377 377 __ push(itos);
378 378 __ jmp(Done);
379 379 __ bind(isOop);
380 380 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
381 381 __ push(atos);
382 382
383 383 if (VerifyOops) {
384 384 __ verify_oop(rax);
385 385 }
386 386 __ bind(Done);
387 387 }
388 388
389 389 // Fast path for caching oop constants.
390 390 // %%% We should use this to handle Class and String constants also.
391 391 // %%% It will simplify the ldc/primitive path considerably.
392 392 void TemplateTable::fast_aldc(bool wide) {
393 393 transition(vtos, atos);
394 394
395 395 if (!EnableInvokeDynamic) {
396 396 // We should not encounter this bytecode if !EnableInvokeDynamic.
397 397 // The verifier will stop it. However, if we get past the verifier,
398 398 // this will stop the thread in a reasonable way, without crashing the JVM.
399 399 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
400 400 InterpreterRuntime::throw_IncompatibleClassChangeError));
401 401 // the call_VM checks for exception, so we should never return here.
402 402 __ should_not_reach_here();
403 403 return;
404 404 }
405 405
406 406 const Register cache = rcx;
407 407 const Register index = rdx;
408 408
409 409 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
410 410 if (VerifyOops) {
411 411 __ verify_oop(rax);
412 412 }
413 413
414 414 Label L_done, L_throw_exception;
415 415 const Register con_klass_temp = rcx; // same as Rcache
416 416 __ load_klass(con_klass_temp, rax);
417 417 __ cmpptr(con_klass_temp, ExternalAddress((address)Universe::systemObjArrayKlassObj_addr()));
418 418 __ jcc(Assembler::notEqual, L_done);
419 419 __ cmpl(Address(rax, arrayOopDesc::length_offset_in_bytes()), 0);
420 420 __ jcc(Assembler::notEqual, L_throw_exception);
421 421 __ xorptr(rax, rax);
422 422 __ jmp(L_done);
423 423
424 424 // Load the exception from the system-array which wraps it:
425 425 __ bind(L_throw_exception);
426 426 __ load_heap_oop(rax, Address(rax, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
427 427 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
428 428
429 429 __ bind(L_done);
430 430 }
431 431
432 432 void TemplateTable::ldc2_w() {
433 433 transition(vtos, vtos);
434 434 Label Long, Done;
435 435 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
436 436
437 437 __ get_cpool_and_tags(rcx, rax);
438 438 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
439 439 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
440 440
441 441 // get type
442 442 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
443 443 __ jccb(Assembler::notEqual, Long);
444 444 // dtos
445 445 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
446 446 __ push(dtos);
447 447 __ jmpb(Done);
448 448
449 449 __ bind(Long);
450 450 // ltos
451 451 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
452 452 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
453 453
454 454 __ push(ltos);
455 455
456 456 __ bind(Done);
457 457 }
458 458
459 459
460 460 void TemplateTable::locals_index(Register reg, int offset) {
461 461 __ load_unsigned_byte(reg, at_bcp(offset));
462 462 __ negptr(reg);
463 463 }
464 464
465 465
466 466 void TemplateTable::iload() {
467 467 transition(vtos, itos);
468 468 if (RewriteFrequentPairs) {
469 469 Label rewrite, done;
470 470
471 471 // get next byte
472 472 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
473 473 // if _iload, wait to rewrite to iload2. We only want to rewrite the
474 474 // last two iloads in a pair. Comparing against fast_iload means that
475 475 // the next bytecode is neither an iload or a caload, and therefore
476 476 // an iload pair.
477 477 __ cmpl(rbx, Bytecodes::_iload);
478 478 __ jcc(Assembler::equal, done);
479 479
480 480 __ cmpl(rbx, Bytecodes::_fast_iload);
481 481 __ movl(rcx, Bytecodes::_fast_iload2);
482 482 __ jccb(Assembler::equal, rewrite);
483 483
484 484 // if _caload, rewrite to fast_icaload
485 485 __ cmpl(rbx, Bytecodes::_caload);
486 486 __ movl(rcx, Bytecodes::_fast_icaload);
487 487 __ jccb(Assembler::equal, rewrite);
488 488
489 489 // rewrite so iload doesn't check again.
490 490 __ movl(rcx, Bytecodes::_fast_iload);
491 491
492 492 // rewrite
493 493 // rcx: fast bytecode
494 494 __ bind(rewrite);
495 495 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
496 496 __ bind(done);
497 497 }
498 498
499 499 // Get the local value into tos
500 500 locals_index(rbx);
501 501 __ movl(rax, iaddress(rbx));
502 502 }
503 503
504 504
505 505 void TemplateTable::fast_iload2() {
506 506 transition(vtos, itos);
507 507 locals_index(rbx);
508 508 __ movl(rax, iaddress(rbx));
509 509 __ push(itos);
510 510 locals_index(rbx, 3);
511 511 __ movl(rax, iaddress(rbx));
512 512 }
513 513
514 514 void TemplateTable::fast_iload() {
515 515 transition(vtos, itos);
516 516 locals_index(rbx);
517 517 __ movl(rax, iaddress(rbx));
518 518 }
519 519
520 520
521 521 void TemplateTable::lload() {
522 522 transition(vtos, ltos);
523 523 locals_index(rbx);
524 524 __ movptr(rax, laddress(rbx));
525 525 NOT_LP64(__ movl(rdx, haddress(rbx)));
526 526 }
527 527
528 528
529 529 void TemplateTable::fload() {
530 530 transition(vtos, ftos);
531 531 locals_index(rbx);
532 532 __ fld_s(faddress(rbx));
533 533 }
534 534
535 535
536 536 void TemplateTable::dload() {
537 537 transition(vtos, dtos);
538 538 locals_index(rbx);
539 539 __ fld_d(daddress(rbx));
540 540 }
541 541
542 542
543 543 void TemplateTable::aload() {
544 544 transition(vtos, atos);
545 545 locals_index(rbx);
546 546 __ movptr(rax, aaddress(rbx));
547 547 }
548 548
549 549
550 550 void TemplateTable::locals_index_wide(Register reg) {
551 551 __ movl(reg, at_bcp(2));
552 552 __ bswapl(reg);
553 553 __ shrl(reg, 16);
554 554 __ negptr(reg);
555 555 }
556 556
557 557
558 558 void TemplateTable::wide_iload() {
559 559 transition(vtos, itos);
560 560 locals_index_wide(rbx);
561 561 __ movl(rax, iaddress(rbx));
562 562 }
563 563
564 564
565 565 void TemplateTable::wide_lload() {
566 566 transition(vtos, ltos);
567 567 locals_index_wide(rbx);
568 568 __ movptr(rax, laddress(rbx));
569 569 NOT_LP64(__ movl(rdx, haddress(rbx)));
570 570 }
571 571
572 572
573 573 void TemplateTable::wide_fload() {
574 574 transition(vtos, ftos);
575 575 locals_index_wide(rbx);
576 576 __ fld_s(faddress(rbx));
577 577 }
578 578
579 579
580 580 void TemplateTable::wide_dload() {
581 581 transition(vtos, dtos);
582 582 locals_index_wide(rbx);
583 583 __ fld_d(daddress(rbx));
584 584 }
585 585
586 586
587 587 void TemplateTable::wide_aload() {
588 588 transition(vtos, atos);
589 589 locals_index_wide(rbx);
590 590 __ movptr(rax, aaddress(rbx));
591 591 }
592 592
593 593 void TemplateTable::index_check(Register array, Register index) {
594 594 // Pop ptr into array
595 595 __ pop_ptr(array);
596 596 index_check_without_pop(array, index);
597 597 }
598 598
599 599 void TemplateTable::index_check_without_pop(Register array, Register index) {
600 600 // destroys rbx,
601 601 // check array
602 602 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
603 603 LP64_ONLY(__ movslq(index, index));
604 604 // check index
605 605 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
606 606 if (index != rbx) {
607 607 // ??? convention: move aberrant index into rbx, for exception message
608 608 assert(rbx != array, "different registers");
609 609 __ mov(rbx, index);
610 610 }
611 611 __ jump_cc(Assembler::aboveEqual,
612 612 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
613 613 }
614 614
615 615
616 616 void TemplateTable::iaload() {
617 617 transition(itos, itos);
618 618 // rdx: array
619 619 index_check(rdx, rax); // kills rbx,
620 620 // rax,: index
621 621 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
622 622 }
623 623
624 624
625 625 void TemplateTable::laload() {
626 626 transition(itos, ltos);
627 627 // rax,: index
628 628 // rdx: array
629 629 index_check(rdx, rax);
630 630 __ mov(rbx, rax);
631 631 // rbx,: index
632 632 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
633 633 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
634 634 }
635 635
636 636
637 637 void TemplateTable::faload() {
638 638 transition(itos, ftos);
639 639 // rdx: array
640 640 index_check(rdx, rax); // kills rbx,
641 641 // rax,: index
642 642 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
643 643 }
644 644
645 645
646 646 void TemplateTable::daload() {
647 647 transition(itos, dtos);
648 648 // rdx: array
649 649 index_check(rdx, rax); // kills rbx,
650 650 // rax,: index
651 651 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
652 652 }
653 653
654 654
655 655 void TemplateTable::aaload() {
656 656 transition(itos, atos);
657 657 // rdx: array
658 658 index_check(rdx, rax); // kills rbx,
659 659 // rax,: index
660 660 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
661 661 }
662 662
663 663
664 664 void TemplateTable::baload() {
665 665 transition(itos, itos);
666 666 // rdx: array
667 667 index_check(rdx, rax); // kills rbx,
668 668 // rax,: index
669 669 // can do better code for P5 - fix this at some point
670 670 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
671 671 __ mov(rax, rbx);
672 672 }
673 673
674 674
675 675 void TemplateTable::caload() {
676 676 transition(itos, itos);
677 677 // rdx: array
678 678 index_check(rdx, rax); // kills rbx,
679 679 // rax,: index
680 680 // can do better code for P5 - may want to improve this at some point
681 681 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
682 682 __ mov(rax, rbx);
683 683 }
684 684
685 685 // iload followed by caload frequent pair
686 686 void TemplateTable::fast_icaload() {
687 687 transition(vtos, itos);
688 688 // load index out of locals
689 689 locals_index(rbx);
690 690 __ movl(rax, iaddress(rbx));
691 691
692 692 // rdx: array
693 693 index_check(rdx, rax);
694 694 // rax,: index
695 695 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
696 696 __ mov(rax, rbx);
697 697 }
698 698
699 699 void TemplateTable::saload() {
700 700 transition(itos, itos);
701 701 // rdx: array
702 702 index_check(rdx, rax); // kills rbx,
703 703 // rax,: index
704 704 // can do better code for P5 - may want to improve this at some point
705 705 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
706 706 __ mov(rax, rbx);
707 707 }
708 708
709 709
710 710 void TemplateTable::iload(int n) {
711 711 transition(vtos, itos);
712 712 __ movl(rax, iaddress(n));
713 713 }
714 714
715 715
716 716 void TemplateTable::lload(int n) {
717 717 transition(vtos, ltos);
718 718 __ movptr(rax, laddress(n));
719 719 NOT_LP64(__ movptr(rdx, haddress(n)));
720 720 }
721 721
722 722
723 723 void TemplateTable::fload(int n) {
724 724 transition(vtos, ftos);
725 725 __ fld_s(faddress(n));
726 726 }
727 727
728 728
729 729 void TemplateTable::dload(int n) {
730 730 transition(vtos, dtos);
731 731 __ fld_d(daddress(n));
732 732 }
733 733
734 734
735 735 void TemplateTable::aload(int n) {
736 736 transition(vtos, atos);
737 737 __ movptr(rax, aaddress(n));
738 738 }
739 739
740 740
741 741 void TemplateTable::aload_0() {
742 742 transition(vtos, atos);
743 743 // According to bytecode histograms, the pairs:
744 744 //
745 745 // _aload_0, _fast_igetfield
746 746 // _aload_0, _fast_agetfield
747 747 // _aload_0, _fast_fgetfield
748 748 //
749 749 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
750 750 // bytecode checks if the next bytecode is either _fast_igetfield,
751 751 // _fast_agetfield or _fast_fgetfield and then rewrites the
752 752 // current bytecode into a pair bytecode; otherwise it rewrites the current
753 753 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
754 754 //
755 755 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
756 756 // otherwise we may miss an opportunity for a pair.
757 757 //
758 758 // Also rewrite frequent pairs
759 759 // aload_0, aload_1
760 760 // aload_0, iload_1
761 761 // These bytecodes with a small amount of code are most profitable to rewrite
762 762 if (RewriteFrequentPairs) {
763 763 Label rewrite, done;
764 764 // get next byte
765 765 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
766 766
767 767 // do actual aload_0
768 768 aload(0);
769 769
770 770 // if _getfield then wait with rewrite
771 771 __ cmpl(rbx, Bytecodes::_getfield);
772 772 __ jcc(Assembler::equal, done);
773 773
774 774 // if _igetfield then reqrite to _fast_iaccess_0
775 775 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
776 776 __ cmpl(rbx, Bytecodes::_fast_igetfield);
777 777 __ movl(rcx, Bytecodes::_fast_iaccess_0);
778 778 __ jccb(Assembler::equal, rewrite);
779 779
780 780 // if _agetfield then reqrite to _fast_aaccess_0
781 781 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
782 782 __ cmpl(rbx, Bytecodes::_fast_agetfield);
783 783 __ movl(rcx, Bytecodes::_fast_aaccess_0);
784 784 __ jccb(Assembler::equal, rewrite);
785 785
786 786 // if _fgetfield then reqrite to _fast_faccess_0
787 787 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
788 788 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
789 789 __ movl(rcx, Bytecodes::_fast_faccess_0);
790 790 __ jccb(Assembler::equal, rewrite);
791 791
792 792 // else rewrite to _fast_aload0
793 793 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
794 794 __ movl(rcx, Bytecodes::_fast_aload_0);
795 795
796 796 // rewrite
797 797 // rcx: fast bytecode
798 798 __ bind(rewrite);
799 799 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
800 800
801 801 __ bind(done);
802 802 } else {
803 803 aload(0);
804 804 }
805 805 }
806 806
807 807 void TemplateTable::istore() {
808 808 transition(itos, vtos);
809 809 locals_index(rbx);
810 810 __ movl(iaddress(rbx), rax);
811 811 }
812 812
813 813
814 814 void TemplateTable::lstore() {
815 815 transition(ltos, vtos);
816 816 locals_index(rbx);
817 817 __ movptr(laddress(rbx), rax);
818 818 NOT_LP64(__ movptr(haddress(rbx), rdx));
819 819 }
820 820
821 821
822 822 void TemplateTable::fstore() {
823 823 transition(ftos, vtos);
824 824 locals_index(rbx);
825 825 __ fstp_s(faddress(rbx));
826 826 }
827 827
828 828
829 829 void TemplateTable::dstore() {
830 830 transition(dtos, vtos);
831 831 locals_index(rbx);
832 832 __ fstp_d(daddress(rbx));
833 833 }
834 834
835 835
836 836 void TemplateTable::astore() {
837 837 transition(vtos, vtos);
838 838 __ pop_ptr(rax);
839 839 locals_index(rbx);
840 840 __ movptr(aaddress(rbx), rax);
841 841 }
842 842
843 843
844 844 void TemplateTable::wide_istore() {
845 845 transition(vtos, vtos);
846 846 __ pop_i(rax);
847 847 locals_index_wide(rbx);
848 848 __ movl(iaddress(rbx), rax);
849 849 }
850 850
851 851
852 852 void TemplateTable::wide_lstore() {
853 853 transition(vtos, vtos);
854 854 __ pop_l(rax, rdx);
855 855 locals_index_wide(rbx);
856 856 __ movptr(laddress(rbx), rax);
857 857 NOT_LP64(__ movl(haddress(rbx), rdx));
858 858 }
859 859
860 860
861 861 void TemplateTable::wide_fstore() {
862 862 wide_istore();
863 863 }
864 864
865 865
866 866 void TemplateTable::wide_dstore() {
867 867 wide_lstore();
868 868 }
869 869
870 870
871 871 void TemplateTable::wide_astore() {
872 872 transition(vtos, vtos);
873 873 __ pop_ptr(rax);
874 874 locals_index_wide(rbx);
875 875 __ movptr(aaddress(rbx), rax);
876 876 }
877 877
878 878
879 879 void TemplateTable::iastore() {
880 880 transition(itos, vtos);
881 881 __ pop_i(rbx);
882 882 // rax,: value
883 883 // rdx: array
884 884 index_check(rdx, rbx); // prefer index in rbx,
885 885 // rbx,: index
886 886 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
887 887 }
888 888
889 889
890 890 void TemplateTable::lastore() {
891 891 transition(ltos, vtos);
892 892 __ pop_i(rbx);
893 893 // rax,: low(value)
894 894 // rcx: array
895 895 // rdx: high(value)
896 896 index_check(rcx, rbx); // prefer index in rbx,
897 897 // rbx,: index
898 898 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
899 899 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
900 900 }
901 901
902 902
903 903 void TemplateTable::fastore() {
904 904 transition(ftos, vtos);
905 905 __ pop_i(rbx);
906 906 // rdx: array
907 907 // st0: value
908 908 index_check(rdx, rbx); // prefer index in rbx,
909 909 // rbx,: index
910 910 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
911 911 }
912 912
913 913
914 914 void TemplateTable::dastore() {
915 915 transition(dtos, vtos);
916 916 __ pop_i(rbx);
917 917 // rdx: array
918 918 // st0: value
919 919 index_check(rdx, rbx); // prefer index in rbx,
920 920 // rbx,: index
921 921 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
922 922 }
923 923
924 924
925 925 void TemplateTable::aastore() {
926 926 Label is_null, ok_is_subtype, done;
927 927 transition(vtos, vtos);
928 928 // stack: ..., array, index, value
929 929 __ movptr(rax, at_tos()); // Value
930 930 __ movl(rcx, at_tos_p1()); // Index
931 931 __ movptr(rdx, at_tos_p2()); // Array
932 932
933 933 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
934 934 index_check_without_pop(rdx, rcx); // kills rbx,
935 935 // do array store check - check for NULL value first
936 936 __ testptr(rax, rax);
937 937 __ jcc(Assembler::zero, is_null);
938 938
939 939 // Move subklass into EBX
940 940 __ load_klass(rbx, rax);
941 941 // Move superklass into EAX
942 942 __ load_klass(rax, rdx);
943 943 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
944 944 // Compress array+index*wordSize+12 into a single register. Frees ECX.
945 945 __ lea(rdx, element_address);
946 946
947 947 // Generate subtype check. Blows ECX. Resets EDI to locals.
948 948 // Superklass in EAX. Subklass in EBX.
949 949 __ gen_subtype_check( rbx, ok_is_subtype );
950 950
951 951 // Come here on failure
952 952 // object is at TOS
953 953 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
954 954
955 955 // Come here on success
956 956 __ bind(ok_is_subtype);
957 957
958 958 // Get the value to store
959 959 __ movptr(rax, at_rsp());
960 960 // and store it with appropriate barrier
961 961 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
962 962
963 963 __ jmp(done);
964 964
965 965 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
966 966 __ bind(is_null);
967 967 __ profile_null_seen(rbx);
968 968
969 969 // Store NULL, (noreg means NULL to do_oop_store)
970 970 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
971 971
972 972 // Pop stack arguments
973 973 __ bind(done);
974 974 __ addptr(rsp, 3 * Interpreter::stackElementSize);
975 975 }
976 976
977 977
978 978 void TemplateTable::bastore() {
979 979 transition(itos, vtos);
980 980 __ pop_i(rbx);
981 981 // rax,: value
982 982 // rdx: array
983 983 index_check(rdx, rbx); // prefer index in rbx,
984 984 // rbx,: index
985 985 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
986 986 }
987 987
988 988
989 989 void TemplateTable::castore() {
990 990 transition(itos, vtos);
991 991 __ pop_i(rbx);
992 992 // rax,: value
993 993 // rdx: array
994 994 index_check(rdx, rbx); // prefer index in rbx,
995 995 // rbx,: index
996 996 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
997 997 }
998 998
999 999
1000 1000 void TemplateTable::sastore() {
1001 1001 castore();
1002 1002 }
1003 1003
1004 1004
1005 1005 void TemplateTable::istore(int n) {
1006 1006 transition(itos, vtos);
1007 1007 __ movl(iaddress(n), rax);
1008 1008 }
1009 1009
1010 1010
1011 1011 void TemplateTable::lstore(int n) {
1012 1012 transition(ltos, vtos);
1013 1013 __ movptr(laddress(n), rax);
1014 1014 NOT_LP64(__ movptr(haddress(n), rdx));
1015 1015 }
1016 1016
1017 1017
1018 1018 void TemplateTable::fstore(int n) {
1019 1019 transition(ftos, vtos);
1020 1020 __ fstp_s(faddress(n));
1021 1021 }
1022 1022
1023 1023
1024 1024 void TemplateTable::dstore(int n) {
1025 1025 transition(dtos, vtos);
1026 1026 __ fstp_d(daddress(n));
1027 1027 }
1028 1028
1029 1029
1030 1030 void TemplateTable::astore(int n) {
1031 1031 transition(vtos, vtos);
1032 1032 __ pop_ptr(rax);
1033 1033 __ movptr(aaddress(n), rax);
1034 1034 }
1035 1035
1036 1036
1037 1037 void TemplateTable::pop() {
1038 1038 transition(vtos, vtos);
1039 1039 __ addptr(rsp, Interpreter::stackElementSize);
1040 1040 }
1041 1041
1042 1042
1043 1043 void TemplateTable::pop2() {
1044 1044 transition(vtos, vtos);
1045 1045 __ addptr(rsp, 2*Interpreter::stackElementSize);
1046 1046 }
1047 1047
1048 1048
1049 1049 void TemplateTable::dup() {
1050 1050 transition(vtos, vtos);
1051 1051 // stack: ..., a
1052 1052 __ load_ptr(0, rax);
1053 1053 __ push_ptr(rax);
1054 1054 // stack: ..., a, a
1055 1055 }
1056 1056
1057 1057
1058 1058 void TemplateTable::dup_x1() {
1059 1059 transition(vtos, vtos);
1060 1060 // stack: ..., a, b
1061 1061 __ load_ptr( 0, rax); // load b
1062 1062 __ load_ptr( 1, rcx); // load a
1063 1063 __ store_ptr(1, rax); // store b
1064 1064 __ store_ptr(0, rcx); // store a
1065 1065 __ push_ptr(rax); // push b
1066 1066 // stack: ..., b, a, b
1067 1067 }
1068 1068
1069 1069
1070 1070 void TemplateTable::dup_x2() {
1071 1071 transition(vtos, vtos);
1072 1072 // stack: ..., a, b, c
1073 1073 __ load_ptr( 0, rax); // load c
1074 1074 __ load_ptr( 2, rcx); // load a
1075 1075 __ store_ptr(2, rax); // store c in a
1076 1076 __ push_ptr(rax); // push c
1077 1077 // stack: ..., c, b, c, c
1078 1078 __ load_ptr( 2, rax); // load b
1079 1079 __ store_ptr(2, rcx); // store a in b
1080 1080 // stack: ..., c, a, c, c
1081 1081 __ store_ptr(1, rax); // store b in c
1082 1082 // stack: ..., c, a, b, c
1083 1083 }
1084 1084
1085 1085
1086 1086 void TemplateTable::dup2() {
1087 1087 transition(vtos, vtos);
1088 1088 // stack: ..., a, b
1089 1089 __ load_ptr(1, rax); // load a
1090 1090 __ push_ptr(rax); // push a
1091 1091 __ load_ptr(1, rax); // load b
1092 1092 __ push_ptr(rax); // push b
1093 1093 // stack: ..., a, b, a, b
1094 1094 }
1095 1095
1096 1096
1097 1097 void TemplateTable::dup2_x1() {
1098 1098 transition(vtos, vtos);
1099 1099 // stack: ..., a, b, c
1100 1100 __ load_ptr( 0, rcx); // load c
1101 1101 __ load_ptr( 1, rax); // load b
1102 1102 __ push_ptr(rax); // push b
1103 1103 __ push_ptr(rcx); // push c
1104 1104 // stack: ..., a, b, c, b, c
1105 1105 __ store_ptr(3, rcx); // store c in b
1106 1106 // stack: ..., a, c, c, b, c
1107 1107 __ load_ptr( 4, rcx); // load a
1108 1108 __ store_ptr(2, rcx); // store a in 2nd c
1109 1109 // stack: ..., a, c, a, b, c
1110 1110 __ store_ptr(4, rax); // store b in a
1111 1111 // stack: ..., b, c, a, b, c
1112 1112 // stack: ..., b, c, a, b, c
1113 1113 }
1114 1114
1115 1115
1116 1116 void TemplateTable::dup2_x2() {
1117 1117 transition(vtos, vtos);
1118 1118 // stack: ..., a, b, c, d
1119 1119 __ load_ptr( 0, rcx); // load d
1120 1120 __ load_ptr( 1, rax); // load c
1121 1121 __ push_ptr(rax); // push c
1122 1122 __ push_ptr(rcx); // push d
1123 1123 // stack: ..., a, b, c, d, c, d
1124 1124 __ load_ptr( 4, rax); // load b
1125 1125 __ store_ptr(2, rax); // store b in d
1126 1126 __ store_ptr(4, rcx); // store d in b
1127 1127 // stack: ..., a, d, c, b, c, d
1128 1128 __ load_ptr( 5, rcx); // load a
1129 1129 __ load_ptr( 3, rax); // load c
1130 1130 __ store_ptr(3, rcx); // store a in c
1131 1131 __ store_ptr(5, rax); // store c in a
1132 1132 // stack: ..., c, d, a, b, c, d
1133 1133 // stack: ..., c, d, a, b, c, d
1134 1134 }
1135 1135
1136 1136
1137 1137 void TemplateTable::swap() {
1138 1138 transition(vtos, vtos);
1139 1139 // stack: ..., a, b
1140 1140 __ load_ptr( 1, rcx); // load a
1141 1141 __ load_ptr( 0, rax); // load b
1142 1142 __ store_ptr(0, rcx); // store a in b
1143 1143 __ store_ptr(1, rax); // store b in a
1144 1144 // stack: ..., b, a
1145 1145 }
1146 1146
1147 1147
1148 1148 void TemplateTable::iop2(Operation op) {
1149 1149 transition(itos, itos);
1150 1150 switch (op) {
1151 1151 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1152 1152 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1153 1153 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1154 1154 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1155 1155 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1156 1156 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1157 1157 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1158 1158 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1159 1159 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1160 1160 default : ShouldNotReachHere();
1161 1161 }
1162 1162 }
1163 1163
1164 1164
1165 1165 void TemplateTable::lop2(Operation op) {
1166 1166 transition(ltos, ltos);
1167 1167 __ pop_l(rbx, rcx);
1168 1168 switch (op) {
1169 1169 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1170 1170 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1171 1171 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1172 1172 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1173 1173 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1174 1174 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1175 1175 default : ShouldNotReachHere();
1176 1176 }
1177 1177 }
1178 1178
1179 1179
1180 1180 void TemplateTable::idiv() {
1181 1181 transition(itos, itos);
1182 1182 __ mov(rcx, rax);
1183 1183 __ pop_i(rax);
1184 1184 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1185 1185 // they are not equal, one could do a normal division (no correction
1186 1186 // needed), which may speed up this implementation for the common case.
1187 1187 // (see also JVM spec., p.243 & p.271)
1188 1188 __ corrected_idivl(rcx);
1189 1189 }
1190 1190
1191 1191
1192 1192 void TemplateTable::irem() {
1193 1193 transition(itos, itos);
1194 1194 __ mov(rcx, rax);
1195 1195 __ pop_i(rax);
1196 1196 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1197 1197 // they are not equal, one could do a normal division (no correction
1198 1198 // needed), which may speed up this implementation for the common case.
1199 1199 // (see also JVM spec., p.243 & p.271)
1200 1200 __ corrected_idivl(rcx);
1201 1201 __ mov(rax, rdx);
1202 1202 }
1203 1203
1204 1204
1205 1205 void TemplateTable::lmul() {
1206 1206 transition(ltos, ltos);
1207 1207 __ pop_l(rbx, rcx);
1208 1208 __ push(rcx); __ push(rbx);
1209 1209 __ push(rdx); __ push(rax);
1210 1210 __ lmul(2 * wordSize, 0);
1211 1211 __ addptr(rsp, 4 * wordSize); // take off temporaries
1212 1212 }
1213 1213
1214 1214
1215 1215 void TemplateTable::ldiv() {
1216 1216 transition(ltos, ltos);
1217 1217 __ pop_l(rbx, rcx);
1218 1218 __ push(rcx); __ push(rbx);
1219 1219 __ push(rdx); __ push(rax);
1220 1220 // check if y = 0
1221 1221 __ orl(rax, rdx);
1222 1222 __ jump_cc(Assembler::zero,
1223 1223 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1224 1224 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1225 1225 __ addptr(rsp, 4 * wordSize); // take off temporaries
1226 1226 }
1227 1227
1228 1228
1229 1229 void TemplateTable::lrem() {
1230 1230 transition(ltos, ltos);
1231 1231 __ pop_l(rbx, rcx);
1232 1232 __ push(rcx); __ push(rbx);
1233 1233 __ push(rdx); __ push(rax);
1234 1234 // check if y = 0
1235 1235 __ orl(rax, rdx);
1236 1236 __ jump_cc(Assembler::zero,
1237 1237 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1238 1238 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1239 1239 __ addptr(rsp, 4 * wordSize);
1240 1240 }
1241 1241
1242 1242
1243 1243 void TemplateTable::lshl() {
1244 1244 transition(itos, ltos);
1245 1245 __ movl(rcx, rax); // get shift count
1246 1246 __ pop_l(rax, rdx); // get shift value
1247 1247 __ lshl(rdx, rax);
1248 1248 }
1249 1249
1250 1250
1251 1251 void TemplateTable::lshr() {
1252 1252 transition(itos, ltos);
1253 1253 __ mov(rcx, rax); // get shift count
1254 1254 __ pop_l(rax, rdx); // get shift value
1255 1255 __ lshr(rdx, rax, true);
1256 1256 }
1257 1257
1258 1258
1259 1259 void TemplateTable::lushr() {
1260 1260 transition(itos, ltos);
1261 1261 __ mov(rcx, rax); // get shift count
1262 1262 __ pop_l(rax, rdx); // get shift value
1263 1263 __ lshr(rdx, rax);
1264 1264 }
1265 1265
1266 1266
1267 1267 void TemplateTable::fop2(Operation op) {
1268 1268 transition(ftos, ftos);
1269 1269 switch (op) {
1270 1270 case add: __ fadd_s (at_rsp()); break;
1271 1271 case sub: __ fsubr_s(at_rsp()); break;
1272 1272 case mul: __ fmul_s (at_rsp()); break;
1273 1273 case div: __ fdivr_s(at_rsp()); break;
1274 1274 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1275 1275 default : ShouldNotReachHere();
1276 1276 }
1277 1277 __ f2ieee();
1278 1278 __ pop(rax); // pop float thing off
1279 1279 }
1280 1280
1281 1281
1282 1282 void TemplateTable::dop2(Operation op) {
1283 1283 transition(dtos, dtos);
1284 1284
1285 1285 switch (op) {
1286 1286 case add: __ fadd_d (at_rsp()); break;
1287 1287 case sub: __ fsubr_d(at_rsp()); break;
1288 1288 case mul: {
1289 1289 Label L_strict;
1290 1290 Label L_join;
1291 1291 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1292 1292 __ get_method(rcx);
1293 1293 __ movl(rcx, access_flags);
1294 1294 __ testl(rcx, JVM_ACC_STRICT);
1295 1295 __ jccb(Assembler::notZero, L_strict);
1296 1296 __ fmul_d (at_rsp());
1297 1297 __ jmpb(L_join);
1298 1298 __ bind(L_strict);
1299 1299 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1300 1300 __ fmulp();
1301 1301 __ fmul_d (at_rsp());
1302 1302 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1303 1303 __ fmulp();
1304 1304 __ bind(L_join);
1305 1305 break;
1306 1306 }
1307 1307 case div: {
1308 1308 Label L_strict;
1309 1309 Label L_join;
1310 1310 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1311 1311 __ get_method(rcx);
1312 1312 __ movl(rcx, access_flags);
1313 1313 __ testl(rcx, JVM_ACC_STRICT);
1314 1314 __ jccb(Assembler::notZero, L_strict);
1315 1315 __ fdivr_d(at_rsp());
1316 1316 __ jmp(L_join);
1317 1317 __ bind(L_strict);
1318 1318 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1319 1319 __ fmul_d (at_rsp());
1320 1320 __ fdivrp();
1321 1321 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1322 1322 __ fmulp();
1323 1323 __ bind(L_join);
1324 1324 break;
1325 1325 }
1326 1326 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1327 1327 default : ShouldNotReachHere();
1328 1328 }
1329 1329 __ d2ieee();
1330 1330 // Pop double precision number from rsp.
1331 1331 __ pop(rax);
1332 1332 __ pop(rdx);
1333 1333 }
1334 1334
1335 1335
1336 1336 void TemplateTable::ineg() {
1337 1337 transition(itos, itos);
1338 1338 __ negl(rax);
1339 1339 }
1340 1340
1341 1341
1342 1342 void TemplateTable::lneg() {
1343 1343 transition(ltos, ltos);
1344 1344 __ lneg(rdx, rax);
1345 1345 }
1346 1346
1347 1347
1348 1348 void TemplateTable::fneg() {
1349 1349 transition(ftos, ftos);
1350 1350 __ fchs();
1351 1351 }
1352 1352
1353 1353
1354 1354 void TemplateTable::dneg() {
1355 1355 transition(dtos, dtos);
1356 1356 __ fchs();
1357 1357 }
1358 1358
1359 1359
1360 1360 void TemplateTable::iinc() {
1361 1361 transition(vtos, vtos);
1362 1362 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1363 1363 locals_index(rbx);
1364 1364 __ addl(iaddress(rbx), rdx);
1365 1365 }
1366 1366
1367 1367
1368 1368 void TemplateTable::wide_iinc() {
1369 1369 transition(vtos, vtos);
1370 1370 __ movl(rdx, at_bcp(4)); // get constant
1371 1371 locals_index_wide(rbx);
1372 1372 __ bswapl(rdx); // swap bytes & sign-extend constant
1373 1373 __ sarl(rdx, 16);
1374 1374 __ addl(iaddress(rbx), rdx);
1375 1375 // Note: should probably use only one movl to get both
1376 1376 // the index and the constant -> fix this
1377 1377 }
1378 1378
1379 1379
1380 1380 void TemplateTable::convert() {
1381 1381 // Checking
1382 1382 #ifdef ASSERT
1383 1383 { TosState tos_in = ilgl;
1384 1384 TosState tos_out = ilgl;
1385 1385 switch (bytecode()) {
1386 1386 case Bytecodes::_i2l: // fall through
1387 1387 case Bytecodes::_i2f: // fall through
1388 1388 case Bytecodes::_i2d: // fall through
1389 1389 case Bytecodes::_i2b: // fall through
1390 1390 case Bytecodes::_i2c: // fall through
1391 1391 case Bytecodes::_i2s: tos_in = itos; break;
1392 1392 case Bytecodes::_l2i: // fall through
1393 1393 case Bytecodes::_l2f: // fall through
1394 1394 case Bytecodes::_l2d: tos_in = ltos; break;
1395 1395 case Bytecodes::_f2i: // fall through
1396 1396 case Bytecodes::_f2l: // fall through
1397 1397 case Bytecodes::_f2d: tos_in = ftos; break;
1398 1398 case Bytecodes::_d2i: // fall through
1399 1399 case Bytecodes::_d2l: // fall through
1400 1400 case Bytecodes::_d2f: tos_in = dtos; break;
1401 1401 default : ShouldNotReachHere();
1402 1402 }
1403 1403 switch (bytecode()) {
1404 1404 case Bytecodes::_l2i: // fall through
1405 1405 case Bytecodes::_f2i: // fall through
1406 1406 case Bytecodes::_d2i: // fall through
1407 1407 case Bytecodes::_i2b: // fall through
1408 1408 case Bytecodes::_i2c: // fall through
1409 1409 case Bytecodes::_i2s: tos_out = itos; break;
1410 1410 case Bytecodes::_i2l: // fall through
1411 1411 case Bytecodes::_f2l: // fall through
1412 1412 case Bytecodes::_d2l: tos_out = ltos; break;
1413 1413 case Bytecodes::_i2f: // fall through
1414 1414 case Bytecodes::_l2f: // fall through
1415 1415 case Bytecodes::_d2f: tos_out = ftos; break;
1416 1416 case Bytecodes::_i2d: // fall through
1417 1417 case Bytecodes::_l2d: // fall through
1418 1418 case Bytecodes::_f2d: tos_out = dtos; break;
1419 1419 default : ShouldNotReachHere();
1420 1420 }
1421 1421 transition(tos_in, tos_out);
1422 1422 }
1423 1423 #endif // ASSERT
1424 1424
1425 1425 // Conversion
1426 1426 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1427 1427 switch (bytecode()) {
1428 1428 case Bytecodes::_i2l:
1429 1429 __ extend_sign(rdx, rax);
1430 1430 break;
1431 1431 case Bytecodes::_i2f:
1432 1432 __ push(rax); // store int on tos
1433 1433 __ fild_s(at_rsp()); // load int to ST0
1434 1434 __ f2ieee(); // truncate to float size
1435 1435 __ pop(rcx); // adjust rsp
1436 1436 break;
1437 1437 case Bytecodes::_i2d:
1438 1438 __ push(rax); // add one slot for d2ieee()
1439 1439 __ push(rax); // store int on tos
1440 1440 __ fild_s(at_rsp()); // load int to ST0
1441 1441 __ d2ieee(); // truncate to double size
1442 1442 __ pop(rcx); // adjust rsp
1443 1443 __ pop(rcx);
1444 1444 break;
1445 1445 case Bytecodes::_i2b:
1446 1446 __ shll(rax, 24); // truncate upper 24 bits
1447 1447 __ sarl(rax, 24); // and sign-extend byte
1448 1448 LP64_ONLY(__ movsbl(rax, rax));
1449 1449 break;
1450 1450 case Bytecodes::_i2c:
1451 1451 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1452 1452 LP64_ONLY(__ movzwl(rax, rax));
1453 1453 break;
1454 1454 case Bytecodes::_i2s:
1455 1455 __ shll(rax, 16); // truncate upper 16 bits
1456 1456 __ sarl(rax, 16); // and sign-extend short
1457 1457 LP64_ONLY(__ movswl(rax, rax));
1458 1458 break;
1459 1459 case Bytecodes::_l2i:
1460 1460 /* nothing to do */
1461 1461 break;
1462 1462 case Bytecodes::_l2f:
1463 1463 __ push(rdx); // store long on tos
1464 1464 __ push(rax);
1465 1465 __ fild_d(at_rsp()); // load long to ST0
1466 1466 __ f2ieee(); // truncate to float size
1467 1467 __ pop(rcx); // adjust rsp
1468 1468 __ pop(rcx);
1469 1469 break;
1470 1470 case Bytecodes::_l2d:
1471 1471 __ push(rdx); // store long on tos
1472 1472 __ push(rax);
1473 1473 __ fild_d(at_rsp()); // load long to ST0
1474 1474 __ d2ieee(); // truncate to double size
1475 1475 __ pop(rcx); // adjust rsp
1476 1476 __ pop(rcx);
1477 1477 break;
1478 1478 case Bytecodes::_f2i:
1479 1479 __ push(rcx); // reserve space for argument
1480 1480 __ fstp_s(at_rsp()); // pass float argument on stack
1481 1481 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1482 1482 break;
1483 1483 case Bytecodes::_f2l:
1484 1484 __ push(rcx); // reserve space for argument
1485 1485 __ fstp_s(at_rsp()); // pass float argument on stack
1486 1486 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1487 1487 break;
1488 1488 case Bytecodes::_f2d:
1489 1489 /* nothing to do */
1490 1490 break;
1491 1491 case Bytecodes::_d2i:
1492 1492 __ push(rcx); // reserve space for argument
1493 1493 __ push(rcx);
1494 1494 __ fstp_d(at_rsp()); // pass double argument on stack
1495 1495 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1496 1496 break;
1497 1497 case Bytecodes::_d2l:
1498 1498 __ push(rcx); // reserve space for argument
1499 1499 __ push(rcx);
1500 1500 __ fstp_d(at_rsp()); // pass double argument on stack
1501 1501 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1502 1502 break;
1503 1503 case Bytecodes::_d2f:
1504 1504 __ push(rcx); // reserve space for f2ieee()
1505 1505 __ f2ieee(); // truncate to float size
1506 1506 __ pop(rcx); // adjust rsp
1507 1507 break;
1508 1508 default :
1509 1509 ShouldNotReachHere();
1510 1510 }
1511 1511 }
1512 1512
1513 1513
1514 1514 void TemplateTable::lcmp() {
1515 1515 transition(ltos, itos);
1516 1516 // y = rdx:rax
1517 1517 __ pop_l(rbx, rcx); // get x = rcx:rbx
1518 1518 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1519 1519 __ mov(rax, rcx);
1520 1520 }
1521 1521
1522 1522
1523 1523 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1524 1524 if (is_float) {
1525 1525 __ fld_s(at_rsp());
1526 1526 } else {
1527 1527 __ fld_d(at_rsp());
1528 1528 __ pop(rdx);
1529 1529 }
1530 1530 __ pop(rcx);
1531 1531 __ fcmp2int(rax, unordered_result < 0);
1532 1532 }
1533 1533
1534 1534
1535 1535 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1536 1536 __ get_method(rcx); // ECX holds method
1537 1537 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1538 1538
1539 1539 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1540 1540 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1541 1541 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1542 1542
1543 1543 // Load up EDX with the branch displacement
1544 1544 __ movl(rdx, at_bcp(1));
1545 1545 __ bswapl(rdx);
1546 1546 if (!is_wide) __ sarl(rdx, 16);
1547 1547 LP64_ONLY(__ movslq(rdx, rdx));
1548 1548
1549 1549
1550 1550 // Handle all the JSR stuff here, then exit.
1551 1551 // It's much shorter and cleaner than intermingling with the
1552 1552 // non-JSR normal-branch stuff occurring below.
1553 1553 if (is_jsr) {
1554 1554 // Pre-load the next target bytecode into EBX
1555 1555 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1556 1556
1557 1557 // compute return address as bci in rax,
1558 1558 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1559 1559 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1560 1560 // Adjust the bcp in RSI by the displacement in EDX
1561 1561 __ addptr(rsi, rdx);
1562 1562 // Push return address
1563 1563 __ push_i(rax);
1564 1564 // jsr returns vtos
1565 1565 __ dispatch_only_noverify(vtos);
1566 1566 return;
1567 1567 }
1568 1568
1569 1569 // Normal (non-jsr) branch handling
1570 1570
1571 1571 // Adjust the bcp in RSI by the displacement in EDX
1572 1572 __ addptr(rsi, rdx);
1573 1573
1574 1574 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1575 1575 Label backedge_counter_overflow;
1576 1576 Label profile_method;
1577 1577 Label dispatch;
1578 1578 if (UseLoopCounter) {
1579 1579 // increment backedge counter for backward branches
1580 1580 // rax,: MDO
1581 1581 // rbx,: MDO bumped taken-count
1582 1582 // rcx: method
1583 1583 // rdx: target offset
1584 1584 // rsi: target bcp
1585 1585 // rdi: locals pointer
1586 1586 __ testl(rdx, rdx); // check if forward or backward branch
1587 1587 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1588 1588
1589 1589 if (TieredCompilation) {
1590 1590 Label no_mdo;
1591 1591 int increment = InvocationCounter::count_increment;
1592 1592 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1593 1593 if (ProfileInterpreter) {
1594 1594 // Are we profiling?
1595 1595 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1596 1596 __ testptr(rbx, rbx);
1597 1597 __ jccb(Assembler::zero, no_mdo);
1598 1598 // Increment the MDO backedge counter
1599 1599 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1600 1600 in_bytes(InvocationCounter::counter_offset()));
1601 1601 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1602 1602 rax, false, Assembler::zero, &backedge_counter_overflow);
1603 1603 __ jmp(dispatch);
1604 1604 }
1605 1605 __ bind(no_mdo);
1606 1606 // Increment backedge counter in methodOop
1607 1607 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1608 1608 rax, false, Assembler::zero, &backedge_counter_overflow);
1609 1609 } else {
1610 1610 // increment counter
1611 1611 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1612 1612 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1613 1613 __ movl(Address(rcx, be_offset), rax); // store counter
1614 1614
1615 1615 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1616 1616 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1617 1617 __ addl(rax, Address(rcx, be_offset)); // add both counters
1618 1618
1619 1619 if (ProfileInterpreter) {
1620 1620 // Test to see if we should create a method data oop
1621 1621 __ cmp32(rax,
1622 1622 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1623 1623 __ jcc(Assembler::less, dispatch);
1624 1624
1625 1625 // if no method data exists, go to profile method
1626 1626 __ test_method_data_pointer(rax, profile_method);
1627 1627
1628 1628 if (UseOnStackReplacement) {
1629 1629 // check for overflow against rbx, which is the MDO taken count
1630 1630 __ cmp32(rbx,
1631 1631 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1632 1632 __ jcc(Assembler::below, dispatch);
1633 1633
1634 1634 // When ProfileInterpreter is on, the backedge_count comes from the
1635 1635 // methodDataOop, which value does not get reset on the call to
1636 1636 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1637 1637 // routine while the method is being compiled, add a second test to make
1638 1638 // sure the overflow function is called only once every overflow_frequency.
1639 1639 const int overflow_frequency = 1024;
1640 1640 __ andptr(rbx, overflow_frequency-1);
1641 1641 __ jcc(Assembler::zero, backedge_counter_overflow);
1642 1642 }
1643 1643 } else {
1644 1644 if (UseOnStackReplacement) {
1645 1645 // check for overflow against rax, which is the sum of the counters
1646 1646 __ cmp32(rax,
1647 1647 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1648 1648 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1649 1649
1650 1650 }
1651 1651 }
1652 1652 }
1653 1653 __ bind(dispatch);
1654 1654 }
1655 1655
1656 1656 // Pre-load the next target bytecode into EBX
1657 1657 __ load_unsigned_byte(rbx, Address(rsi, 0));
1658 1658
1659 1659 // continue with the bytecode @ target
1660 1660 // rax,: return bci for jsr's, unused otherwise
1661 1661 // rbx,: target bytecode
1662 1662 // rsi: target bcp
1663 1663 __ dispatch_only(vtos);
1664 1664
1665 1665 if (UseLoopCounter) {
1666 1666 if (ProfileInterpreter) {
1667 1667 // Out-of-line code to allocate method data oop.
1668 1668 __ bind(profile_method);
1669 1669 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1670 1670 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1671 1671 __ set_method_data_pointer_for_bcp();
1672 1672 __ jmp(dispatch);
1673 1673 }
1674 1674
1675 1675 if (UseOnStackReplacement) {
1676 1676
1677 1677 // invocation counter overflow
1678 1678 __ bind(backedge_counter_overflow);
1679 1679 __ negptr(rdx);
1680 1680 __ addptr(rdx, rsi); // branch bcp
1681 1681 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1682 1682 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1683 1683
1684 1684 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1685 1685 // rbx,: target bytecode
1686 1686 // rdx: scratch
1687 1687 // rdi: locals pointer
1688 1688 // rsi: bcp
1689 1689 __ testptr(rax, rax); // test result
1690 1690 __ jcc(Assembler::zero, dispatch); // no osr if null
1691 1691 // nmethod may have been invalidated (VM may block upon call_VM return)
1692 1692 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1693 1693 __ cmpl(rcx, InvalidOSREntryBci);
1694 1694 __ jcc(Assembler::equal, dispatch);
1695 1695
1696 1696 // We have the address of an on stack replacement routine in rax,
1697 1697 // We need to prepare to execute the OSR method. First we must
1698 1698 // migrate the locals and monitors off of the stack.
1699 1699
1700 1700 __ mov(rbx, rax); // save the nmethod
1701 1701
1702 1702 const Register thread = rcx;
1703 1703 __ get_thread(thread);
1704 1704 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1705 1705 // rax, is OSR buffer, move it to expected parameter location
1706 1706 __ mov(rcx, rax);
1707 1707
1708 1708 // pop the interpreter frame
1709 1709 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1710 1710 __ leave(); // remove frame anchor
1711 1711 __ pop(rdi); // get return address
1712 1712 __ mov(rsp, rdx); // set sp to sender sp
1713 1713
1714 1714 // Align stack pointer for compiled code (note that caller is
1715 1715 // responsible for undoing this fixup by remembering the old SP
1716 1716 // in an rbp,-relative location)
1717 1717 __ andptr(rsp, -(StackAlignmentInBytes));
1718 1718
1719 1719 // push the (possibly adjusted) return address
1720 1720 __ push(rdi);
1721 1721
1722 1722 // and begin the OSR nmethod
1723 1723 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1724 1724 }
1725 1725 }
1726 1726 }
1727 1727
1728 1728
1729 1729 void TemplateTable::if_0cmp(Condition cc) {
1730 1730 transition(itos, vtos);
1731 1731 // assume branch is more often taken than not (loops use backward branches)
1732 1732 Label not_taken;
1733 1733 __ testl(rax, rax);
1734 1734 __ jcc(j_not(cc), not_taken);
1735 1735 branch(false, false);
1736 1736 __ bind(not_taken);
1737 1737 __ profile_not_taken_branch(rax);
1738 1738 }
1739 1739
1740 1740
1741 1741 void TemplateTable::if_icmp(Condition cc) {
1742 1742 transition(itos, vtos);
1743 1743 // assume branch is more often taken than not (loops use backward branches)
1744 1744 Label not_taken;
1745 1745 __ pop_i(rdx);
1746 1746 __ cmpl(rdx, rax);
1747 1747 __ jcc(j_not(cc), not_taken);
1748 1748 branch(false, false);
1749 1749 __ bind(not_taken);
1750 1750 __ profile_not_taken_branch(rax);
1751 1751 }
1752 1752
1753 1753
1754 1754 void TemplateTable::if_nullcmp(Condition cc) {
1755 1755 transition(atos, vtos);
1756 1756 // assume branch is more often taken than not (loops use backward branches)
1757 1757 Label not_taken;
1758 1758 __ testptr(rax, rax);
1759 1759 __ jcc(j_not(cc), not_taken);
1760 1760 branch(false, false);
1761 1761 __ bind(not_taken);
1762 1762 __ profile_not_taken_branch(rax);
1763 1763 }
1764 1764
1765 1765
1766 1766 void TemplateTable::if_acmp(Condition cc) {
1767 1767 transition(atos, vtos);
1768 1768 // assume branch is more often taken than not (loops use backward branches)
1769 1769 Label not_taken;
1770 1770 __ pop_ptr(rdx);
1771 1771 __ cmpptr(rdx, rax);
1772 1772 __ jcc(j_not(cc), not_taken);
1773 1773 branch(false, false);
1774 1774 __ bind(not_taken);
1775 1775 __ profile_not_taken_branch(rax);
1776 1776 }
1777 1777
1778 1778
1779 1779 void TemplateTable::ret() {
1780 1780 transition(vtos, vtos);
1781 1781 locals_index(rbx);
1782 1782 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1783 1783 __ profile_ret(rbx, rcx);
1784 1784 __ get_method(rax);
1785 1785 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1786 1786 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1787 1787 constMethodOopDesc::codes_offset()));
1788 1788 __ dispatch_next(vtos);
1789 1789 }
1790 1790
1791 1791
1792 1792 void TemplateTable::wide_ret() {
1793 1793 transition(vtos, vtos);
1794 1794 locals_index_wide(rbx);
1795 1795 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1796 1796 __ profile_ret(rbx, rcx);
1797 1797 __ get_method(rax);
1798 1798 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1799 1799 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1800 1800 __ dispatch_next(vtos);
1801 1801 }
1802 1802
1803 1803
1804 1804 void TemplateTable::tableswitch() {
1805 1805 Label default_case, continue_execution;
1806 1806 transition(itos, vtos);
1807 1807 // align rsi
1808 1808 __ lea(rbx, at_bcp(wordSize));
1809 1809 __ andptr(rbx, -wordSize);
1810 1810 // load lo & hi
1811 1811 __ movl(rcx, Address(rbx, 1 * wordSize));
1812 1812 __ movl(rdx, Address(rbx, 2 * wordSize));
1813 1813 __ bswapl(rcx);
1814 1814 __ bswapl(rdx);
1815 1815 // check against lo & hi
1816 1816 __ cmpl(rax, rcx);
1817 1817 __ jccb(Assembler::less, default_case);
1818 1818 __ cmpl(rax, rdx);
1819 1819 __ jccb(Assembler::greater, default_case);
1820 1820 // lookup dispatch offset
1821 1821 __ subl(rax, rcx);
1822 1822 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1823 1823 __ profile_switch_case(rax, rbx, rcx);
1824 1824 // continue execution
1825 1825 __ bind(continue_execution);
1826 1826 __ bswapl(rdx);
1827 1827 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1828 1828 __ addptr(rsi, rdx);
1829 1829 __ dispatch_only(vtos);
1830 1830 // handle default
1831 1831 __ bind(default_case);
1832 1832 __ profile_switch_default(rax);
1833 1833 __ movl(rdx, Address(rbx, 0));
1834 1834 __ jmp(continue_execution);
1835 1835 }
1836 1836
1837 1837
1838 1838 void TemplateTable::lookupswitch() {
1839 1839 transition(itos, itos);
1840 1840 __ stop("lookupswitch bytecode should have been rewritten");
1841 1841 }
1842 1842
1843 1843
1844 1844 void TemplateTable::fast_linearswitch() {
1845 1845 transition(itos, vtos);
1846 1846 Label loop_entry, loop, found, continue_execution;
1847 1847 // bswapl rax, so we can avoid bswapping the table entries
1848 1848 __ bswapl(rax);
1849 1849 // align rsi
1850 1850 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1851 1851 __ andptr(rbx, -wordSize);
1852 1852 // set counter
1853 1853 __ movl(rcx, Address(rbx, wordSize));
1854 1854 __ bswapl(rcx);
1855 1855 __ jmpb(loop_entry);
1856 1856 // table search
1857 1857 __ bind(loop);
1858 1858 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1859 1859 __ jccb(Assembler::equal, found);
1860 1860 __ bind(loop_entry);
1861 1861 __ decrementl(rcx);
1862 1862 __ jcc(Assembler::greaterEqual, loop);
1863 1863 // default case
1864 1864 __ profile_switch_default(rax);
1865 1865 __ movl(rdx, Address(rbx, 0));
1866 1866 __ jmpb(continue_execution);
1867 1867 // entry found -> get offset
1868 1868 __ bind(found);
1869 1869 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1870 1870 __ profile_switch_case(rcx, rax, rbx);
1871 1871 // continue execution
1872 1872 __ bind(continue_execution);
1873 1873 __ bswapl(rdx);
1874 1874 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1875 1875 __ addptr(rsi, rdx);
1876 1876 __ dispatch_only(vtos);
1877 1877 }
1878 1878
1879 1879
1880 1880 void TemplateTable::fast_binaryswitch() {
1881 1881 transition(itos, vtos);
1882 1882 // Implementation using the following core algorithm:
1883 1883 //
1884 1884 // int binary_search(int key, LookupswitchPair* array, int n) {
1885 1885 // // Binary search according to "Methodik des Programmierens" by
1886 1886 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1887 1887 // int i = 0;
1888 1888 // int j = n;
1889 1889 // while (i+1 < j) {
1890 1890 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1891 1891 // // with Q: for all i: 0 <= i < n: key < a[i]
1892 1892 // // where a stands for the array and assuming that the (inexisting)
1893 1893 // // element a[n] is infinitely big.
1894 1894 // int h = (i + j) >> 1;
1895 1895 // // i < h < j
1896 1896 // if (key < array[h].fast_match()) {
1897 1897 // j = h;
1898 1898 // } else {
1899 1899 // i = h;
1900 1900 // }
1901 1901 // }
1902 1902 // // R: a[i] <= key < a[i+1] or Q
1903 1903 // // (i.e., if key is within array, i is the correct index)
1904 1904 // return i;
1905 1905 // }
1906 1906
1907 1907 // register allocation
1908 1908 const Register key = rax; // already set (tosca)
1909 1909 const Register array = rbx;
1910 1910 const Register i = rcx;
1911 1911 const Register j = rdx;
1912 1912 const Register h = rdi; // needs to be restored
1913 1913 const Register temp = rsi;
1914 1914 // setup array
1915 1915 __ save_bcp();
1916 1916
1917 1917 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1918 1918 __ andptr(array, -wordSize);
1919 1919 // initialize i & j
1920 1920 __ xorl(i, i); // i = 0;
1921 1921 __ movl(j, Address(array, -wordSize)); // j = length(array);
1922 1922 // Convert j into native byteordering
1923 1923 __ bswapl(j);
1924 1924 // and start
1925 1925 Label entry;
1926 1926 __ jmp(entry);
1927 1927
1928 1928 // binary search loop
1929 1929 { Label loop;
1930 1930 __ bind(loop);
1931 1931 // int h = (i + j) >> 1;
1932 1932 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1933 1933 __ sarl(h, 1); // h = (i + j) >> 1;
1934 1934 // if (key < array[h].fast_match()) {
1935 1935 // j = h;
1936 1936 // } else {
1937 1937 // i = h;
1938 1938 // }
1939 1939 // Convert array[h].match to native byte-ordering before compare
1940 1940 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1941 1941 __ bswapl(temp);
1942 1942 __ cmpl(key, temp);
1943 1943 // j = h if (key < array[h].fast_match())
1944 1944 __ cmov32(Assembler::less , j, h);
1945 1945 // i = h if (key >= array[h].fast_match())
1946 1946 __ cmov32(Assembler::greaterEqual, i, h);
1947 1947 // while (i+1 < j)
1948 1948 __ bind(entry);
1949 1949 __ leal(h, Address(i, 1)); // i+1
1950 1950 __ cmpl(h, j); // i+1 < j
1951 1951 __ jcc(Assembler::less, loop);
1952 1952 }
1953 1953
1954 1954 // end of binary search, result index is i (must check again!)
1955 1955 Label default_case;
1956 1956 // Convert array[i].match to native byte-ordering before compare
1957 1957 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1958 1958 __ bswapl(temp);
1959 1959 __ cmpl(key, temp);
1960 1960 __ jcc(Assembler::notEqual, default_case);
1961 1961
1962 1962 // entry found -> j = offset
1963 1963 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1964 1964 __ profile_switch_case(i, key, array);
1965 1965 __ bswapl(j);
1966 1966 LP64_ONLY(__ movslq(j, j));
1967 1967 __ restore_bcp();
1968 1968 __ restore_locals(); // restore rdi
1969 1969 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1970 1970
1971 1971 __ addptr(rsi, j);
1972 1972 __ dispatch_only(vtos);
1973 1973
1974 1974 // default case -> j = default offset
1975 1975 __ bind(default_case);
1976 1976 __ profile_switch_default(i);
1977 1977 __ movl(j, Address(array, -2*wordSize));
1978 1978 __ bswapl(j);
1979 1979 LP64_ONLY(__ movslq(j, j));
1980 1980 __ restore_bcp();
1981 1981 __ restore_locals(); // restore rdi
1982 1982 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1983 1983 __ addptr(rsi, j);
1984 1984 __ dispatch_only(vtos);
1985 1985 }
1986 1986
1987 1987
1988 1988 void TemplateTable::_return(TosState state) {
1989 1989 transition(state, state);
1990 1990 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
1991 1991
1992 1992 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1993 1993 assert(state == vtos, "only valid state");
1994 1994 __ movptr(rax, aaddress(0));
1995 1995 __ load_klass(rdi, rax);
1996 1996 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1997 1997 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
1998 1998 Label skip_register_finalizer;
1999 1999 __ jcc(Assembler::zero, skip_register_finalizer);
2000 2000
2001 2001 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2002 2002
2003 2003 __ bind(skip_register_finalizer);
2004 2004 }
2005 2005
2006 2006 __ remove_activation(state, rsi);
2007 2007 __ jmp(rsi);
2008 2008 }
2009 2009
2010 2010
2011 2011 // ----------------------------------------------------------------------------
2012 2012 // Volatile variables demand their effects be made known to all CPU's in
2013 2013 // order. Store buffers on most chips allow reads & writes to reorder; the
2014 2014 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2015 2015 // memory barrier (i.e., it's not sufficient that the interpreter does not
2016 2016 // reorder volatile references, the hardware also must not reorder them).
2017 2017 //
2018 2018 // According to the new Java Memory Model (JMM):
2019 2019 // (1) All volatiles are serialized wrt to each other.
2020 2020 // ALSO reads & writes act as aquire & release, so:
2021 2021 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2022 2022 // the read float up to before the read. It's OK for non-volatile memory refs
2023 2023 // that happen before the volatile read to float down below it.
2024 2024 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2025 2025 // that happen BEFORE the write float down to after the write. It's OK for
2026 2026 // non-volatile memory refs that happen after the volatile write to float up
2027 2027 // before it.
2028 2028 //
2029 2029 // We only put in barriers around volatile refs (they are expensive), not
2030 2030 // _between_ memory refs (that would require us to track the flavor of the
2031 2031 // previous memory refs). Requirements (2) and (3) require some barriers
2032 2032 // before volatile stores and after volatile loads. These nearly cover
2033 2033 // requirement (1) but miss the volatile-store-volatile-load case. This final
2034 2034 // case is placed after volatile-stores although it could just as well go
2035 2035 // before volatile-loads.
2036 2036 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2037 2037 // Helper function to insert a is-volatile test and memory barrier
2038 2038 if( !os::is_MP() ) return; // Not needed on single CPU
2039 2039 __ membar(order_constraint);
2040 2040 }
2041 2041
2042 2042 void TemplateTable::resolve_cache_and_index(int byte_no,
2043 2043 Register result,
2044 2044 Register Rcache,
2045 2045 Register index,
2046 2046 size_t index_size) {
2047 2047 Register temp = rbx;
2048 2048
2049 2049 assert_different_registers(result, Rcache, index, temp);
2050 2050
2051 2051 Label resolved;
2052 2052 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2053 2053 if (byte_no == f1_oop) {
2054 2054 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2055 2055 // This kind of CP cache entry does not need to match the flags byte, because
2056 2056 // there is a 1-1 relation between bytecode type and CP entry type.
2057 2057 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2058 2058 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2059 2059 __ testptr(result, result);
2060 2060 __ jcc(Assembler::notEqual, resolved);
2061 2061 } else {
2062 2062 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2063 2063 assert(result == noreg, ""); //else change code for setting result
2064 2064 const int shift_count = (1 + byte_no)*BitsPerByte;
2065 2065 __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2066 2066 __ shrl(temp, shift_count);
2067 2067 // have we resolved this bytecode?
2068 2068 __ andl(temp, 0xFF);
2069 2069 __ cmpl(temp, (int)bytecode());
2070 2070 __ jcc(Assembler::equal, resolved);
2071 2071 }
2072 2072
2073 2073 // resolve first time through
2074 2074 address entry;
2075 2075 switch (bytecode()) {
2076 2076 case Bytecodes::_getstatic : // fall through
2077 2077 case Bytecodes::_putstatic : // fall through
2078 2078 case Bytecodes::_getfield : // fall through
2079 2079 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2080 2080 case Bytecodes::_invokevirtual : // fall through
2081 2081 case Bytecodes::_invokespecial : // fall through
2082 2082 case Bytecodes::_invokestatic : // fall through
2083 2083 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2084 2084 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2085 2085 case Bytecodes::_fast_aldc : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2086 2086 case Bytecodes::_fast_aldc_w : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2087 2087 default : ShouldNotReachHere(); break;
2088 2088 }
2089 2089 __ movl(temp, (int)bytecode());
2090 2090 __ call_VM(noreg, entry, temp);
2091 2091 // Update registers with resolved info
2092 2092 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2093 2093 if (result != noreg)
2094 2094 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2095 2095 __ bind(resolved);
2096 2096 }
2097 2097
2098 2098
2099 2099 // The cache and index registers must be set before call
2100 2100 void TemplateTable::load_field_cp_cache_entry(Register obj,
2101 2101 Register cache,
2102 2102 Register index,
2103 2103 Register off,
2104 2104 Register flags,
2105 2105 bool is_static = false) {
2106 2106 assert_different_registers(cache, index, flags, off);
2107 2107
2108 2108 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2109 2109 // Field offset
2110 2110 __ movptr(off, Address(cache, index, Address::times_ptr,
2111 2111 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2112 2112 // Flags
2113 2113 __ movl(flags, Address(cache, index, Address::times_ptr,
2114 2114 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2115 2115
2116 2116 // klass overwrite register
2117 2117 if (is_static) {
2118 2118 __ movptr(obj, Address(cache, index, Address::times_ptr,
2119 2119 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2120 2120 }
2121 2121 }
2122 2122
2123 2123 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2124 2124 Register method,
2125 2125 Register itable_index,
2126 2126 Register flags,
2127 2127 bool is_invokevirtual,
2128 2128 bool is_invokevfinal /*unused*/,
2129 2129 bool is_invokedynamic) {
2130 2130 // setup registers
2131 2131 const Register cache = rcx;
2132 2132 const Register index = rdx;
2133 2133 assert_different_registers(method, flags);
2134 2134 assert_different_registers(method, cache, index);
2135 2135 assert_different_registers(itable_index, flags);
2136 2136 assert_different_registers(itable_index, cache, index);
2137 2137 // determine constant pool cache field offsets
2138 2138 const int method_offset = in_bytes(
2139 2139 constantPoolCacheOopDesc::base_offset() +
2140 2140 (is_invokevirtual
2141 2141 ? ConstantPoolCacheEntry::f2_offset()
2142 2142 : ConstantPoolCacheEntry::f1_offset()
2143 2143 )
2144 2144 );
2145 2145 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2146 2146 ConstantPoolCacheEntry::flags_offset());
2147 2147 // access constant pool cache fields
2148 2148 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2149 2149 ConstantPoolCacheEntry::f2_offset());
2150 2150
2151 2151 if (byte_no == f1_oop) {
2152 2152 // Resolved f1_oop goes directly into 'method' register.
2153 2153 assert(is_invokedynamic, "");
2154 2154 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2155 2155 } else {
2156 2156 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2157 2157 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2158 2158 }
2159 2159 if (itable_index != noreg) {
2160 2160 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2161 2161 }
2162 2162 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2163 2163 }
2164 2164
2165 2165
2166 2166 // The registers cache and index expected to be set before call.
2167 2167 // Correct values of the cache and index registers are preserved.
2168 2168 void TemplateTable::jvmti_post_field_access(Register cache,
2169 2169 Register index,
2170 2170 bool is_static,
2171 2171 bool has_tos) {
2172 2172 if (JvmtiExport::can_post_field_access()) {
2173 2173 // Check to see if a field access watch has been set before we take
2174 2174 // the time to call into the VM.
2175 2175 Label L1;
2176 2176 assert_different_registers(cache, index, rax);
2177 2177 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2178 2178 __ testl(rax,rax);
2179 2179 __ jcc(Assembler::zero, L1);
2180 2180
2181 2181 // cache entry pointer
2182 2182 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2183 2183 __ shll(index, LogBytesPerWord);
2184 2184 __ addptr(cache, index);
2185 2185 if (is_static) {
2186 2186 __ xorptr(rax, rax); // NULL object reference
2187 2187 } else {
2188 2188 __ pop(atos); // Get the object
2189 2189 __ verify_oop(rax);
2190 2190 __ push(atos); // Restore stack state
2191 2191 }
2192 2192 // rax,: object pointer or NULL
2193 2193 // cache: cache entry pointer
2194 2194 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2195 2195 rax, cache);
2196 2196 __ get_cache_and_index_at_bcp(cache, index, 1);
2197 2197 __ bind(L1);
2198 2198 }
2199 2199 }
2200 2200
2201 2201 void TemplateTable::pop_and_check_object(Register r) {
2202 2202 __ pop_ptr(r);
2203 2203 __ null_check(r); // for field access must check obj.
2204 2204 __ verify_oop(r);
2205 2205 }
2206 2206
2207 2207 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2208 2208 transition(vtos, vtos);
2209 2209
2210 2210 const Register cache = rcx;
2211 2211 const Register index = rdx;
2212 2212 const Register obj = rcx;
2213 2213 const Register off = rbx;
2214 2214 const Register flags = rax;
2215 2215
2216 2216 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2217 2217 jvmti_post_field_access(cache, index, is_static, false);
2218 2218 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2219 2219
2220 2220 if (!is_static) pop_and_check_object(obj);
2221 2221
2222 2222 const Address lo(obj, off, Address::times_1, 0*wordSize);
2223 2223 const Address hi(obj, off, Address::times_1, 1*wordSize);
2224 2224
2225 2225 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2226 2226
2227 2227 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2228 2228 assert(btos == 0, "change code, btos != 0");
2229 2229 // btos
2230 2230 __ andptr(flags, 0x0f);
2231 2231 __ jcc(Assembler::notZero, notByte);
2232 2232
2233 2233 __ load_signed_byte(rax, lo );
2234 2234 __ push(btos);
2235 2235 // Rewrite bytecode to be faster
2236 2236 if (!is_static) {
2237 2237 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2238 2238 }
2239 2239 __ jmp(Done);
2240 2240
2241 2241 __ bind(notByte);
2242 2242 // itos
2243 2243 __ cmpl(flags, itos );
2244 2244 __ jcc(Assembler::notEqual, notInt);
2245 2245
2246 2246 __ movl(rax, lo );
2247 2247 __ push(itos);
2248 2248 // Rewrite bytecode to be faster
2249 2249 if (!is_static) {
2250 2250 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2251 2251 }
2252 2252 __ jmp(Done);
2253 2253
2254 2254 __ bind(notInt);
2255 2255 // atos
2256 2256 __ cmpl(flags, atos );
2257 2257 __ jcc(Assembler::notEqual, notObj);
2258 2258
2259 2259 __ movl(rax, lo );
2260 2260 __ push(atos);
2261 2261 if (!is_static) {
2262 2262 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2263 2263 }
2264 2264 __ jmp(Done);
2265 2265
2266 2266 __ bind(notObj);
2267 2267 // ctos
2268 2268 __ cmpl(flags, ctos );
2269 2269 __ jcc(Assembler::notEqual, notChar);
2270 2270
2271 2271 __ load_unsigned_short(rax, lo );
2272 2272 __ push(ctos);
2273 2273 if (!is_static) {
2274 2274 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2275 2275 }
2276 2276 __ jmp(Done);
2277 2277
2278 2278 __ bind(notChar);
2279 2279 // stos
2280 2280 __ cmpl(flags, stos );
2281 2281 __ jcc(Assembler::notEqual, notShort);
2282 2282
2283 2283 __ load_signed_short(rax, lo );
2284 2284 __ push(stos);
2285 2285 if (!is_static) {
2286 2286 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2287 2287 }
2288 2288 __ jmp(Done);
2289 2289
2290 2290 __ bind(notShort);
2291 2291 // ltos
2292 2292 __ cmpl(flags, ltos );
2293 2293 __ jcc(Assembler::notEqual, notLong);
2294 2294
2295 2295 // Generate code as if volatile. There just aren't enough registers to
2296 2296 // save that information and this code is faster than the test.
2297 2297 __ fild_d(lo); // Must load atomically
2298 2298 __ subptr(rsp,2*wordSize); // Make space for store
2299 2299 __ fistp_d(Address(rsp,0));
2300 2300 __ pop(rax);
2301 2301 __ pop(rdx);
2302 2302
2303 2303 __ push(ltos);
2304 2304 // Don't rewrite to _fast_lgetfield for potential volatile case.
2305 2305 __ jmp(Done);
2306 2306
2307 2307 __ bind(notLong);
2308 2308 // ftos
2309 2309 __ cmpl(flags, ftos );
2310 2310 __ jcc(Assembler::notEqual, notFloat);
2311 2311
2312 2312 __ fld_s(lo);
2313 2313 __ push(ftos);
2314 2314 if (!is_static) {
2315 2315 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2316 2316 }
2317 2317 __ jmp(Done);
2318 2318
2319 2319 __ bind(notFloat);
2320 2320 // dtos
2321 2321 __ cmpl(flags, dtos );
2322 2322 __ jcc(Assembler::notEqual, notDouble);
2323 2323
2324 2324 __ fld_d(lo);
2325 2325 __ push(dtos);
2326 2326 if (!is_static) {
2327 2327 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2328 2328 }
2329 2329 __ jmpb(Done);
2330 2330
2331 2331 __ bind(notDouble);
2332 2332
2333 2333 __ stop("Bad state");
2334 2334
2335 2335 __ bind(Done);
2336 2336 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2337 2337 // volatile_barrier( );
2338 2338 }
2339 2339
2340 2340
2341 2341 void TemplateTable::getfield(int byte_no) {
2342 2342 getfield_or_static(byte_no, false);
2343 2343 }
2344 2344
2345 2345
2346 2346 void TemplateTable::getstatic(int byte_no) {
2347 2347 getfield_or_static(byte_no, true);
2348 2348 }
2349 2349
2350 2350 // The registers cache and index expected to be set before call.
2351 2351 // The function may destroy various registers, just not the cache and index registers.
2352 2352 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2353 2353
2354 2354 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2355 2355
2356 2356 if (JvmtiExport::can_post_field_modification()) {
2357 2357 // Check to see if a field modification watch has been set before we take
2358 2358 // the time to call into the VM.
2359 2359 Label L1;
2360 2360 assert_different_registers(cache, index, rax);
2361 2361 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2362 2362 __ testl(rax, rax);
2363 2363 __ jcc(Assembler::zero, L1);
2364 2364
2365 2365 // The cache and index registers have been already set.
2366 2366 // This allows to eliminate this call but the cache and index
2367 2367 // registers have to be correspondingly used after this line.
2368 2368 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2369 2369
2370 2370 if (is_static) {
2371 2371 // Life is simple. Null out the object pointer.
2372 2372 __ xorptr(rbx, rbx);
2373 2373 } else {
2374 2374 // Life is harder. The stack holds the value on top, followed by the object.
2375 2375 // We don't know the size of the value, though; it could be one or two words
2376 2376 // depending on its type. As a result, we must find the type to determine where
2377 2377 // the object is.
2378 2378 Label two_word, valsize_known;
2379 2379 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2380 2380 ConstantPoolCacheEntry::flags_offset())));
2381 2381 __ mov(rbx, rsp);
2382 2382 __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2383 2383 // Make sure we don't need to mask rcx for tosBits after the above shift
2384 2384 ConstantPoolCacheEntry::verify_tosBits();
2385 2385 __ cmpl(rcx, ltos);
2386 2386 __ jccb(Assembler::equal, two_word);
2387 2387 __ cmpl(rcx, dtos);
2388 2388 __ jccb(Assembler::equal, two_word);
2389 2389 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2390 2390 __ jmpb(valsize_known);
2391 2391
2392 2392 __ bind(two_word);
2393 2393 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2394 2394
2395 2395 __ bind(valsize_known);
2396 2396 // setup object pointer
2397 2397 __ movptr(rbx, Address(rbx, 0));
2398 2398 }
2399 2399 // cache entry pointer
2400 2400 __ addptr(rax, in_bytes(cp_base_offset));
2401 2401 __ shll(rdx, LogBytesPerWord);
2402 2402 __ addptr(rax, rdx);
2403 2403 // object (tos)
2404 2404 __ mov(rcx, rsp);
2405 2405 // rbx,: object pointer set up above (NULL if static)
2406 2406 // rax,: cache entry pointer
2407 2407 // rcx: jvalue object on the stack
2408 2408 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2409 2409 rbx, rax, rcx);
2410 2410 __ get_cache_and_index_at_bcp(cache, index, 1);
2411 2411 __ bind(L1);
2412 2412 }
2413 2413 }
2414 2414
2415 2415
2416 2416 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2417 2417 transition(vtos, vtos);
2418 2418
2419 2419 const Register cache = rcx;
2420 2420 const Register index = rdx;
2421 2421 const Register obj = rcx;
2422 2422 const Register off = rbx;
2423 2423 const Register flags = rax;
2424 2424
2425 2425 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2426 2426 jvmti_post_field_mod(cache, index, is_static);
2427 2427 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2428 2428
2429 2429 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2430 2430 // volatile_barrier( );
2431 2431
2432 2432 Label notVolatile, Done;
2433 2433 __ movl(rdx, flags);
2434 2434 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2435 2435 __ andl(rdx, 0x1);
2436 2436
2437 2437 // field addresses
2438 2438 const Address lo(obj, off, Address::times_1, 0*wordSize);
2439 2439 const Address hi(obj, off, Address::times_1, 1*wordSize);
2440 2440
2441 2441 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2442 2442
2443 2443 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2444 2444 assert(btos == 0, "change code, btos != 0");
2445 2445 // btos
2446 2446 __ andl(flags, 0x0f);
2447 2447 __ jcc(Assembler::notZero, notByte);
2448 2448
2449 2449 __ pop(btos);
2450 2450 if (!is_static) pop_and_check_object(obj);
2451 2451 __ movb(lo, rax );
2452 2452 if (!is_static) {
2453 2453 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
2454 2454 }
2455 2455 __ jmp(Done);
2456 2456
2457 2457 __ bind(notByte);
2458 2458 // itos
2459 2459 __ cmpl(flags, itos );
2460 2460 __ jcc(Assembler::notEqual, notInt);
2461 2461
2462 2462 __ pop(itos);
2463 2463 if (!is_static) pop_and_check_object(obj);
2464 2464
2465 2465 __ movl(lo, rax );
2466 2466 if (!is_static) {
2467 2467 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2468 2468 }
2469 2469 __ jmp(Done);
2470 2470
2471 2471 __ bind(notInt);
2472 2472 // atos
2473 2473 __ cmpl(flags, atos );
2474 2474 __ jcc(Assembler::notEqual, notObj);
2475 2475
2476 2476 __ pop(atos);
2477 2477 if (!is_static) pop_and_check_object(obj);
2478 2478
2479 2479 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2480 2480
2481 2481 if (!is_static) {
2482 2482 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2483 2483 }
2484 2484
2485 2485 __ jmp(Done);
2486 2486
2487 2487 __ bind(notObj);
2488 2488 // ctos
2489 2489 __ cmpl(flags, ctos );
2490 2490 __ jcc(Assembler::notEqual, notChar);
2491 2491
2492 2492 __ pop(ctos);
2493 2493 if (!is_static) pop_and_check_object(obj);
2494 2494 __ movw(lo, rax );
2495 2495 if (!is_static) {
2496 2496 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2497 2497 }
2498 2498 __ jmp(Done);
2499 2499
2500 2500 __ bind(notChar);
2501 2501 // stos
2502 2502 __ cmpl(flags, stos );
2503 2503 __ jcc(Assembler::notEqual, notShort);
2504 2504
2505 2505 __ pop(stos);
2506 2506 if (!is_static) pop_and_check_object(obj);
2507 2507 __ movw(lo, rax );
2508 2508 if (!is_static) {
2509 2509 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2510 2510 }
2511 2511 __ jmp(Done);
2512 2512
2513 2513 __ bind(notShort);
2514 2514 // ltos
2515 2515 __ cmpl(flags, ltos );
2516 2516 __ jcc(Assembler::notEqual, notLong);
2517 2517
2518 2518 Label notVolatileLong;
2519 2519 __ testl(rdx, rdx);
2520 2520 __ jcc(Assembler::zero, notVolatileLong);
2521 2521
2522 2522 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2523 2523 if (!is_static) pop_and_check_object(obj);
2524 2524
2525 2525 // Replace with real volatile test
2526 2526 __ push(rdx);
2527 2527 __ push(rax); // Must update atomically with FIST
2528 2528 __ fild_d(Address(rsp,0)); // So load into FPU register
2529 2529 __ fistp_d(lo); // and put into memory atomically
2530 2530 __ addptr(rsp, 2*wordSize);
2531 2531 // volatile_barrier();
2532 2532 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2533 2533 Assembler::StoreStore));
2534 2534 // Don't rewrite volatile version
2535 2535 __ jmp(notVolatile);
2536 2536
2537 2537 __ bind(notVolatileLong);
2538 2538
2539 2539 __ pop(ltos); // overwrites rdx
2540 2540 if (!is_static) pop_and_check_object(obj);
2541 2541 NOT_LP64(__ movptr(hi, rdx));
2542 2542 __ movptr(lo, rax);
2543 2543 if (!is_static) {
2544 2544 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2545 2545 }
2546 2546 __ jmp(notVolatile);
2547 2547
2548 2548 __ bind(notLong);
2549 2549 // ftos
2550 2550 __ cmpl(flags, ftos );
2551 2551 __ jcc(Assembler::notEqual, notFloat);
2552 2552
2553 2553 __ pop(ftos);
2554 2554 if (!is_static) pop_and_check_object(obj);
2555 2555 __ fstp_s(lo);
2556 2556 if (!is_static) {
2557 2557 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2558 2558 }
2559 2559 __ jmp(Done);
2560 2560
2561 2561 __ bind(notFloat);
2562 2562 // dtos
2563 2563 __ cmpl(flags, dtos );
2564 2564 __ jcc(Assembler::notEqual, notDouble);
2565 2565
2566 2566 __ pop(dtos);
2567 2567 if (!is_static) pop_and_check_object(obj);
2568 2568 __ fstp_d(lo);
2569 2569 if (!is_static) {
2570 2570 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2571 2571 }
2572 2572 __ jmp(Done);
2573 2573
2574 2574 __ bind(notDouble);
2575 2575
2576 2576 __ stop("Bad state");
2577 2577
2578 2578 __ bind(Done);
2579 2579
2580 2580 // Check for volatile store
2581 2581 __ testl(rdx, rdx);
2582 2582 __ jcc(Assembler::zero, notVolatile);
2583 2583 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2584 2584 Assembler::StoreStore));
2585 2585 __ bind(notVolatile);
2586 2586 }
2587 2587
2588 2588
2589 2589 void TemplateTable::putfield(int byte_no) {
2590 2590 putfield_or_static(byte_no, false);
2591 2591 }
2592 2592
2593 2593
2594 2594 void TemplateTable::putstatic(int byte_no) {
2595 2595 putfield_or_static(byte_no, true);
2596 2596 }
2597 2597
2598 2598 void TemplateTable::jvmti_post_fast_field_mod() {
2599 2599 if (JvmtiExport::can_post_field_modification()) {
2600 2600 // Check to see if a field modification watch has been set before we take
2601 2601 // the time to call into the VM.
2602 2602 Label L2;
2603 2603 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2604 2604 __ testl(rcx,rcx);
2605 2605 __ jcc(Assembler::zero, L2);
2606 2606 __ pop_ptr(rbx); // copy the object pointer from tos
2607 2607 __ verify_oop(rbx);
2608 2608 __ push_ptr(rbx); // put the object pointer back on tos
2609 2609 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2610 2610 __ mov(rcx, rsp);
2611 2611 __ push_ptr(rbx); // save object pointer so we can steal rbx,
2612 2612 __ xorptr(rbx, rbx);
2613 2613 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2614 2614 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2615 2615 switch (bytecode()) { // load values into the jvalue object
2616 2616 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2617 2617 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2618 2618 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2619 2619 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
2620 2620 case Bytecodes::_fast_lputfield:
2621 2621 NOT_LP64(__ movptr(hi_value, rdx));
2622 2622 __ movptr(lo_value, rax);
2623 2623 break;
2624 2624
2625 2625 // need to call fld_s() after fstp_s() to restore the value for below
2626 2626 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
2627 2627
2628 2628 // need to call fld_d() after fstp_d() to restore the value for below
2629 2629 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
2630 2630
2631 2631 // since rcx is not an object we don't call store_check() here
2632 2632 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break;
2633 2633
2634 2634 default: ShouldNotReachHere();
2635 2635 }
2636 2636 __ pop_ptr(rbx); // restore copy of object pointer
2637 2637
2638 2638 // Save rax, and sometimes rdx because call_VM() will clobber them,
2639 2639 // then use them for JVM/DI purposes
2640 2640 __ push(rax);
2641 2641 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2642 2642 // access constant pool cache entry
2643 2643 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2644 2644 __ verify_oop(rbx);
2645 2645 // rbx,: object pointer copied above
2646 2646 // rax,: cache entry pointer
2647 2647 // rcx: jvalue object on the stack
2648 2648 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2649 2649 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value
2650 2650 __ pop(rax); // restore lower value
2651 2651 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2652 2652 __ bind(L2);
2653 2653 }
2654 2654 }
2655 2655
2656 2656 void TemplateTable::fast_storefield(TosState state) {
2657 2657 transition(state, vtos);
2658 2658
2659 2659 ByteSize base = constantPoolCacheOopDesc::base_offset();
2660 2660
2661 2661 jvmti_post_fast_field_mod();
2662 2662
2663 2663 // access constant pool cache
2664 2664 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2665 2665
2666 2666 // test for volatile with rdx but rdx is tos register for lputfield.
2667 2667 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2668 2668 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2669 2669 ConstantPoolCacheEntry::flags_offset())));
2670 2670
2671 2671 // replace index with field offset from cache entry
2672 2672 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2673 2673
2674 2674 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2675 2675 // volatile_barrier( );
2676 2676
2677 2677 Label notVolatile, Done;
2678 2678 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2679 2679 __ andl(rdx, 0x1);
2680 2680 // Check for volatile store
2681 2681 __ testl(rdx, rdx);
2682 2682 __ jcc(Assembler::zero, notVolatile);
2683 2683
2684 2684 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2685 2685
2686 2686 // Get object from stack
2687 2687 pop_and_check_object(rcx);
2688 2688
2689 2689 // field addresses
2690 2690 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2691 2691 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2692 2692
2693 2693 // access field
2694 2694 switch (bytecode()) {
2695 2695 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2696 2696 case Bytecodes::_fast_sputfield: // fall through
2697 2697 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2698 2698 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2699 2699 case Bytecodes::_fast_lputfield:
2700 2700 NOT_LP64(__ movptr(hi, rdx));
2701 2701 __ movptr(lo, rax);
2702 2702 break;
2703 2703 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2704 2704 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2705 2705 case Bytecodes::_fast_aputfield: {
2706 2706 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2707 2707 break;
2708 2708 }
2709 2709 default:
2710 2710 ShouldNotReachHere();
2711 2711 }
2712 2712
2713 2713 Label done;
2714 2714 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2715 2715 Assembler::StoreStore));
2716 2716 // Barriers are so large that short branch doesn't reach!
2717 2717 __ jmp(done);
2718 2718
2719 2719 // Same code as above, but don't need rdx to test for volatile.
2720 2720 __ bind(notVolatile);
2721 2721
2722 2722 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2723 2723
2724 2724 // Get object from stack
2725 2725 pop_and_check_object(rcx);
2726 2726
2727 2727 // access field
2728 2728 switch (bytecode()) {
2729 2729 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2730 2730 case Bytecodes::_fast_sputfield: // fall through
2731 2731 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2732 2732 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2733 2733 case Bytecodes::_fast_lputfield:
2734 2734 NOT_LP64(__ movptr(hi, rdx));
2735 2735 __ movptr(lo, rax);
2736 2736 break;
2737 2737 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2738 2738 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2739 2739 case Bytecodes::_fast_aputfield: {
2740 2740 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2741 2741 break;
2742 2742 }
2743 2743 default:
2744 2744 ShouldNotReachHere();
2745 2745 }
2746 2746 __ bind(done);
2747 2747 }
2748 2748
2749 2749
2750 2750 void TemplateTable::fast_accessfield(TosState state) {
2751 2751 transition(atos, state);
2752 2752
2753 2753 // do the JVMTI work here to avoid disturbing the register state below
2754 2754 if (JvmtiExport::can_post_field_access()) {
2755 2755 // Check to see if a field access watch has been set before we take
2756 2756 // the time to call into the VM.
2757 2757 Label L1;
2758 2758 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2759 2759 __ testl(rcx,rcx);
2760 2760 __ jcc(Assembler::zero, L1);
2761 2761 // access constant pool cache entry
2762 2762 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2763 2763 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2764 2764 __ verify_oop(rax);
2765 2765 // rax,: object pointer copied above
2766 2766 // rcx: cache entry pointer
2767 2767 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2768 2768 __ pop_ptr(rax); // restore object pointer
2769 2769 __ bind(L1);
2770 2770 }
2771 2771
2772 2772 // access constant pool cache
2773 2773 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2774 2774 // replace index with field offset from cache entry
2775 2775 __ movptr(rbx, Address(rcx,
2776 2776 rbx,
2777 2777 Address::times_ptr,
2778 2778 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2779 2779
2780 2780
2781 2781 // rax,: object
2782 2782 __ verify_oop(rax);
2783 2783 __ null_check(rax);
2784 2784 // field addresses
2785 2785 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2786 2786 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2787 2787
2788 2788 // access field
2789 2789 switch (bytecode()) {
2790 2790 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2791 2791 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break;
2792 2792 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break;
2793 2793 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2794 2794 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2795 2795 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2796 2796 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2797 2797 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2798 2798 default:
2799 2799 ShouldNotReachHere();
2800 2800 }
2801 2801
2802 2802 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2803 2803 // volatile_barrier( );
2804 2804 }
2805 2805
2806 2806 void TemplateTable::fast_xaccess(TosState state) {
2807 2807 transition(vtos, state);
2808 2808 // get receiver
2809 2809 __ movptr(rax, aaddress(0));
2810 2810 // access constant pool cache
2811 2811 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2812 2812 __ movptr(rbx, Address(rcx,
2813 2813 rdx,
2814 2814 Address::times_ptr,
2815 2815 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2816 2816 // make sure exception is reported in correct bcp range (getfield is next instruction)
2817 2817 __ increment(rsi);
2818 2818 __ null_check(rax);
2819 2819 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2820 2820 if (state == itos) {
2821 2821 __ movl(rax, lo);
2822 2822 } else if (state == atos) {
2823 2823 __ movptr(rax, lo);
2824 2824 __ verify_oop(rax);
2825 2825 } else if (state == ftos) {
2826 2826 __ fld_s(lo);
2827 2827 } else {
2828 2828 ShouldNotReachHere();
2829 2829 }
2830 2830 __ decrement(rsi);
2831 2831 }
2832 2832
2833 2833
2834 2834
2835 2835 //----------------------------------------------------------------------------------------------------
2836 2836 // Calls
2837 2837
2838 2838 void TemplateTable::count_calls(Register method, Register temp) {
2839 2839 // implemented elsewhere
2840 2840 ShouldNotReachHere();
2841 2841 }
2842 2842
2843 2843
2844 2844 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2845 2845 // determine flags
2846 2846 Bytecodes::Code code = bytecode();
2847 2847 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2848 2848 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2849 2849 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2850 2850 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2851 2851 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2852 2852 const bool receiver_null_check = is_invokespecial;
2853 2853 const bool save_flags = is_invokeinterface || is_invokevirtual;
2854 2854 // setup registers & access constant pool cache
2855 2855 const Register recv = rcx;
2856 2856 const Register flags = rdx;
2857 2857 assert_different_registers(method, index, recv, flags);
2858 2858
2859 2859 // save 'interpreter return address'
2860 2860 __ save_bcp();
2861 2861
2862 2862 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2863 2863
2864 2864 // load receiver if needed (note: no return address pushed yet)
2865 2865 if (load_receiver) {
2866 2866 assert(!is_invokedynamic, "");
2867 2867 __ movl(recv, flags);
2868 2868 __ andl(recv, 0xFF);
2869 2869 // recv count is 0 based?
2870 2870 Address recv_addr(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1));
2871 2871 __ movptr(recv, recv_addr);
2872 2872 __ verify_oop(recv);
2873 2873 }
2874 2874
2875 2875 // do null check if needed
2876 2876 if (receiver_null_check) {
2877 2877 __ null_check(recv);
2878 2878 }
2879 2879
2880 2880 if (save_flags) {
2881 2881 __ mov(rsi, flags);
2882 2882 }
2883 2883
2884 2884 // compute return type
2885 2885 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2886 2886 // Make sure we don't need to mask flags for tosBits after the above shift
2887 2887 ConstantPoolCacheEntry::verify_tosBits();
2888 2888 // load return address
2889 2889 {
2890 2890 address table_addr;
2891 2891 if (is_invokeinterface || is_invokedynamic)
2892 2892 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2893 2893 else
2894 2894 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2895 2895 ExternalAddress table(table_addr);
2896 2896 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2897 2897 }
2898 2898
2899 2899 // push return address
2900 2900 __ push(flags);
2901 2901
2902 2902 // Restore flag value from the constant pool cache, and restore rsi
2903 2903 // for later null checks. rsi is the bytecode pointer
2904 2904 if (save_flags) {
2905 2905 __ mov(flags, rsi);
2906 2906 __ restore_bcp();
2907 2907 }
2908 2908 }
2909 2909
2910 2910
2911 2911 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2912 2912 Register flags) {
2913 2913
2914 2914 // Uses temporary registers rax, rdx
2915 2915 assert_different_registers(index, recv, rax, rdx);
2916 2916
2917 2917 // Test for an invoke of a final method
2918 2918 Label notFinal;
2919 2919 __ movl(rax, flags);
2920 2920 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2921 2921 __ jcc(Assembler::zero, notFinal);
2922 2922
2923 2923 Register method = index; // method must be rbx,
2924 2924 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2925 2925
2926 2926 // do the call - the index is actually the method to call
2927 2927 __ verify_oop(method);
2928 2928
2929 2929 // It's final, need a null check here!
2930 2930 __ null_check(recv);
2931 2931
2932 2932 // profile this call
2933 2933 __ profile_final_call(rax);
2934 2934
2935 2935 __ jump_from_interpreted(method, rax);
2936 2936
2937 2937 __ bind(notFinal);
2938 2938
2939 2939 // get receiver klass
2940 2940 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2941 2941 // Keep recv in rcx for callee expects it there
2942 2942 __ load_klass(rax, recv);
2943 2943 __ verify_oop(rax);
2944 2944
2945 2945 // profile this call
2946 2946 __ profile_virtual_call(rax, rdi, rdx);
2947 2947
2948 2948 // get target methodOop & entry point
2949 2949 const int base = instanceKlass::vtable_start_offset() * wordSize;
2950 2950 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2951 2951 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2952 2952 __ jump_from_interpreted(method, rdx);
2953 2953 }
2954 2954
2955 2955
2956 2956 void TemplateTable::invokevirtual(int byte_no) {
2957 2957 transition(vtos, vtos);
2958 2958 assert(byte_no == f2_byte, "use this argument");
2959 2959 prepare_invoke(rbx, noreg, byte_no);
2960 2960
2961 2961 // rbx,: index
2962 2962 // rcx: receiver
2963 2963 // rdx: flags
2964 2964
2965 2965 invokevirtual_helper(rbx, rcx, rdx);
2966 2966 }
2967 2967
2968 2968
2969 2969 void TemplateTable::invokespecial(int byte_no) {
2970 2970 transition(vtos, vtos);
2971 2971 assert(byte_no == f1_byte, "use this argument");
2972 2972 prepare_invoke(rbx, noreg, byte_no);
2973 2973 // do the call
2974 2974 __ verify_oop(rbx);
2975 2975 __ profile_call(rax);
2976 2976 __ jump_from_interpreted(rbx, rax);
2977 2977 }
2978 2978
2979 2979
2980 2980 void TemplateTable::invokestatic(int byte_no) {
2981 2981 transition(vtos, vtos);
2982 2982 assert(byte_no == f1_byte, "use this argument");
2983 2983 prepare_invoke(rbx, noreg, byte_no);
2984 2984 // do the call
2985 2985 __ verify_oop(rbx);
2986 2986 __ profile_call(rax);
2987 2987 __ jump_from_interpreted(rbx, rax);
2988 2988 }
2989 2989
2990 2990
2991 2991 void TemplateTable::fast_invokevfinal(int byte_no) {
2992 2992 transition(vtos, vtos);
2993 2993 assert(byte_no == f2_byte, "use this argument");
2994 2994 __ stop("fast_invokevfinal not used on x86");
2995 2995 }
2996 2996
2997 2997
2998 2998 void TemplateTable::invokeinterface(int byte_no) {
2999 2999 transition(vtos, vtos);
3000 3000 assert(byte_no == f1_byte, "use this argument");
3001 3001 prepare_invoke(rax, rbx, byte_no);
3002 3002
3003 3003 // rax,: Interface
3004 3004 // rbx,: index
3005 3005 // rcx: receiver
3006 3006 // rdx: flags
3007 3007
3008 3008 // Special case of invokeinterface called for virtual method of
3009 3009 // java.lang.Object. See cpCacheOop.cpp for details.
3010 3010 // This code isn't produced by javac, but could be produced by
3011 3011 // another compliant java compiler.
3012 3012 Label notMethod;
3013 3013 __ movl(rdi, rdx);
3014 3014 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
3015 3015 __ jcc(Assembler::zero, notMethod);
3016 3016
3017 3017 invokevirtual_helper(rbx, rcx, rdx);
3018 3018 __ bind(notMethod);
3019 3019
3020 3020 // Get receiver klass into rdx - also a null check
3021 3021 __ restore_locals(); // restore rdi
3022 3022 __ load_klass(rdx, rcx);
3023 3023 __ verify_oop(rdx);
3024 3024
3025 3025 // profile this call
3026 3026 __ profile_virtual_call(rdx, rsi, rdi);
3027 3027
3028 3028 Label no_such_interface, no_such_method;
3029 3029
3030 3030 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3031 3031 rdx, rax, rbx,
3032 3032 // outputs: method, scan temp. reg
3033 3033 rbx, rsi,
3034 3034 no_such_interface);
3035 3035
3036 3036 // rbx,: methodOop to call
3037 3037 // rcx: receiver
3038 3038 // Check for abstract method error
3039 3039 // Note: This should be done more efficiently via a throw_abstract_method_error
3040 3040 // interpreter entry point and a conditional jump to it in case of a null
3041 3041 // method.
3042 3042 __ testptr(rbx, rbx);
3043 3043 __ jcc(Assembler::zero, no_such_method);
3044 3044
3045 3045 // do the call
3046 3046 // rcx: receiver
3047 3047 // rbx,: methodOop
3048 3048 __ jump_from_interpreted(rbx, rdx);
3049 3049 __ should_not_reach_here();
3050 3050
3051 3051 // exception handling code follows...
3052 3052 // note: must restore interpreter registers to canonical
3053 3053 // state for exception handling to work correctly!
3054 3054
3055 3055 __ bind(no_such_method);
3056 3056 // throw exception
3057 3057 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3058 3058 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3059 3059 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3060 3060 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3061 3061 // the call_VM checks for exception, so we should never return here.
3062 3062 __ should_not_reach_here();
3063 3063
3064 3064 __ bind(no_such_interface);
3065 3065 // throw exception
3066 3066 __ pop(rbx); // pop return address (pushed by prepare_invoke)
|
↓ open down ↓ |
3066 lines elided |
↑ open up ↑ |
3067 3067 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3068 3068 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3069 3069 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3070 3070 InterpreterRuntime::throw_IncompatibleClassChangeError));
3071 3071 // the call_VM checks for exception, so we should never return here.
3072 3072 __ should_not_reach_here();
3073 3073 }
3074 3074
3075 3075 void TemplateTable::invokedynamic(int byte_no) {
3076 3076 transition(vtos, vtos);
3077 + assert(byte_no == f1_oop, "use this argument");
3077 3078
3078 3079 if (!EnableInvokeDynamic) {
3079 3080 // We should not encounter this bytecode if !EnableInvokeDynamic.
3080 3081 // The verifier will stop it. However, if we get past the verifier,
3081 3082 // this will stop the thread in a reasonable way, without crashing the JVM.
3082 3083 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3083 3084 InterpreterRuntime::throw_IncompatibleClassChangeError));
3084 3085 // the call_VM checks for exception, so we should never return here.
3085 3086 __ should_not_reach_here();
3086 3087 return;
3087 3088 }
3088 3089
3089 - assert(byte_no == f1_oop, "use this argument");
3090 3090 prepare_invoke(rax, rbx, byte_no);
3091 3091
3092 3092 // rax: CallSite object (f1)
3093 3093 // rbx: unused (f2)
3094 3094 // rcx: receiver address
3095 3095 // rdx: flags (unused)
3096 3096
3097 3097 Register rax_callsite = rax;
3098 3098 Register rcx_method_handle = rcx;
3099 3099
3100 - if (ProfileInterpreter) {
3101 - // %%% should make a type profile for any invokedynamic that takes a ref argument
3102 - // profile this call
3103 - __ profile_call(rsi);
3104 - }
3100 + // %%% should make a type profile for any invokedynamic that takes a ref argument
3101 + // profile this call
3102 + __ profile_call(rsi);
3105 3103
3106 - __ load_heap_oop(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, rcx)));
3104 + __ verify_oop(rax_callsite);
3105 + __ load_heap_oop(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, rdx)));
3107 3106 __ null_check(rcx_method_handle);
3107 + __ verify_oop(rcx_method_handle);
3108 3108 __ prepare_to_jump_from_interpreted();
3109 3109 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3110 3110 }
3111 3111
3112 3112 //----------------------------------------------------------------------------------------------------
3113 3113 // Allocation
3114 3114
3115 3115 void TemplateTable::_new() {
3116 3116 transition(vtos, atos);
3117 3117 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3118 3118 Label slow_case;
3119 3119 Label slow_case_no_pop;
3120 3120 Label done;
3121 3121 Label initialize_header;
3122 3122 Label initialize_object; // including clearing the fields
3123 3123 Label allocate_shared;
3124 3124
3125 3125 __ get_cpool_and_tags(rcx, rax);
3126 3126
3127 3127 // Make sure the class we're about to instantiate has been resolved.
3128 3128 // This is done before loading instanceKlass to be consistent with the order
3129 3129 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3130 3130 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3131 3131 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3132 3132 __ jcc(Assembler::notEqual, slow_case_no_pop);
3133 3133
3134 3134 // get instanceKlass
3135 3135 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3136 3136 __ push(rcx); // save the contexts of klass for initializing the header
3137 3137
3138 3138 // make sure klass is initialized & doesn't have finalizer
3139 3139 // make sure klass is fully initialized
3140 3140 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3141 3141 __ jcc(Assembler::notEqual, slow_case);
3142 3142
3143 3143 // get instance_size in instanceKlass (scaled to a count of bytes)
3144 3144 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3145 3145 // test to see if it has a finalizer or is malformed in some way
3146 3146 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3147 3147 __ jcc(Assembler::notZero, slow_case);
3148 3148
3149 3149 //
3150 3150 // Allocate the instance
3151 3151 // 1) Try to allocate in the TLAB
3152 3152 // 2) if fail and the object is large allocate in the shared Eden
3153 3153 // 3) if the above fails (or is not applicable), go to a slow case
3154 3154 // (creates a new TLAB, etc.)
3155 3155
3156 3156 const bool allow_shared_alloc =
3157 3157 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3158 3158
3159 3159 const Register thread = rcx;
3160 3160 if (UseTLAB || allow_shared_alloc) {
3161 3161 __ get_thread(thread);
3162 3162 }
3163 3163
3164 3164 if (UseTLAB) {
3165 3165 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3166 3166 __ lea(rbx, Address(rax, rdx, Address::times_1));
3167 3167 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3168 3168 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3169 3169 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3170 3170 if (ZeroTLAB) {
3171 3171 // the fields have been already cleared
3172 3172 __ jmp(initialize_header);
3173 3173 } else {
3174 3174 // initialize both the header and fields
3175 3175 __ jmp(initialize_object);
3176 3176 }
3177 3177 }
3178 3178
3179 3179 // Allocation in the shared Eden, if allowed.
3180 3180 //
3181 3181 // rdx: instance size in bytes
3182 3182 if (allow_shared_alloc) {
3183 3183 __ bind(allocate_shared);
3184 3184
3185 3185 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3186 3186
3187 3187 Label retry;
3188 3188 __ bind(retry);
3189 3189 __ movptr(rax, heap_top);
3190 3190 __ lea(rbx, Address(rax, rdx, Address::times_1));
3191 3191 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3192 3192 __ jcc(Assembler::above, slow_case);
3193 3193
3194 3194 // Compare rax, with the top addr, and if still equal, store the new
3195 3195 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3196 3196 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3197 3197 //
3198 3198 // rax,: object begin
3199 3199 // rbx,: object end
3200 3200 // rdx: instance size in bytes
3201 3201 __ locked_cmpxchgptr(rbx, heap_top);
3202 3202
3203 3203 // if someone beat us on the allocation, try again, otherwise continue
3204 3204 __ jcc(Assembler::notEqual, retry);
3205 3205
3206 3206 __ incr_allocated_bytes(thread, rdx, 0);
3207 3207 }
3208 3208
3209 3209 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3210 3210 // The object is initialized before the header. If the object size is
3211 3211 // zero, go directly to the header initialization.
3212 3212 __ bind(initialize_object);
3213 3213 __ decrement(rdx, sizeof(oopDesc));
3214 3214 __ jcc(Assembler::zero, initialize_header);
3215 3215
3216 3216 // Initialize topmost object field, divide rdx by 8, check if odd and
3217 3217 // test if zero.
3218 3218 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3219 3219 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3220 3220
3221 3221 // rdx must have been multiple of 8
3222 3222 #ifdef ASSERT
3223 3223 // make sure rdx was multiple of 8
3224 3224 Label L;
3225 3225 // Ignore partial flag stall after shrl() since it is debug VM
3226 3226 __ jccb(Assembler::carryClear, L);
3227 3227 __ stop("object size is not multiple of 2 - adjust this code");
3228 3228 __ bind(L);
3229 3229 // rdx must be > 0, no extra check needed here
3230 3230 #endif
3231 3231
3232 3232 // initialize remaining object fields: rdx was a multiple of 8
3233 3233 { Label loop;
3234 3234 __ bind(loop);
3235 3235 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3236 3236 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3237 3237 __ decrement(rdx);
3238 3238 __ jcc(Assembler::notZero, loop);
3239 3239 }
3240 3240
3241 3241 // initialize object header only.
3242 3242 __ bind(initialize_header);
3243 3243 if (UseBiasedLocking) {
3244 3244 __ pop(rcx); // get saved klass back in the register.
3245 3245 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3246 3246 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3247 3247 } else {
3248 3248 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3249 3249 (int32_t)markOopDesc::prototype()); // header
3250 3250 __ pop(rcx); // get saved klass back in the register.
3251 3251 }
3252 3252 __ store_klass(rax, rcx); // klass
3253 3253
3254 3254 {
3255 3255 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3256 3256 // Trigger dtrace event for fastpath
3257 3257 __ push(atos);
3258 3258 __ call_VM_leaf(
3259 3259 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3260 3260 __ pop(atos);
3261 3261 }
3262 3262
3263 3263 __ jmp(done);
3264 3264 }
3265 3265
3266 3266 // slow case
3267 3267 __ bind(slow_case);
3268 3268 __ pop(rcx); // restore stack pointer to what it was when we came in.
3269 3269 __ bind(slow_case_no_pop);
3270 3270 __ get_constant_pool(rax);
3271 3271 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3272 3272 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3273 3273
3274 3274 // continue
3275 3275 __ bind(done);
3276 3276 }
3277 3277
3278 3278
3279 3279 void TemplateTable::newarray() {
3280 3280 transition(itos, atos);
3281 3281 __ push_i(rax); // make sure everything is on the stack
3282 3282 __ load_unsigned_byte(rdx, at_bcp(1));
3283 3283 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3284 3284 __ pop_i(rdx); // discard size
3285 3285 }
3286 3286
3287 3287
3288 3288 void TemplateTable::anewarray() {
3289 3289 transition(itos, atos);
3290 3290 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3291 3291 __ get_constant_pool(rcx);
3292 3292 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3293 3293 }
3294 3294
3295 3295
3296 3296 void TemplateTable::arraylength() {
3297 3297 transition(atos, itos);
3298 3298 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3299 3299 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3300 3300 }
3301 3301
3302 3302
3303 3303 void TemplateTable::checkcast() {
3304 3304 transition(atos, atos);
3305 3305 Label done, is_null, ok_is_subtype, quicked, resolved;
3306 3306 __ testptr(rax, rax); // Object is in EAX
3307 3307 __ jcc(Assembler::zero, is_null);
3308 3308
3309 3309 // Get cpool & tags index
3310 3310 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3311 3311 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3312 3312 // See if bytecode has already been quicked
3313 3313 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3314 3314 __ jcc(Assembler::equal, quicked);
3315 3315
3316 3316 __ push(atos);
3317 3317 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3318 3318 __ pop_ptr(rdx);
3319 3319 __ jmpb(resolved);
3320 3320
3321 3321 // Get superklass in EAX and subklass in EBX
3322 3322 __ bind(quicked);
3323 3323 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3324 3324 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3325 3325
3326 3326 __ bind(resolved);
3327 3327 __ load_klass(rbx, rdx);
3328 3328
3329 3329 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3330 3330 // Superklass in EAX. Subklass in EBX.
3331 3331 __ gen_subtype_check( rbx, ok_is_subtype );
3332 3332
3333 3333 // Come here on failure
3334 3334 __ push(rdx);
3335 3335 // object is at TOS
3336 3336 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3337 3337
3338 3338 // Come here on success
3339 3339 __ bind(ok_is_subtype);
3340 3340 __ mov(rax,rdx); // Restore object in EDX
3341 3341
3342 3342 // Collect counts on whether this check-cast sees NULLs a lot or not.
3343 3343 if (ProfileInterpreter) {
3344 3344 __ jmp(done);
3345 3345 __ bind(is_null);
3346 3346 __ profile_null_seen(rcx);
3347 3347 } else {
3348 3348 __ bind(is_null); // same as 'done'
3349 3349 }
3350 3350 __ bind(done);
3351 3351 }
3352 3352
3353 3353
3354 3354 void TemplateTable::instanceof() {
3355 3355 transition(atos, itos);
3356 3356 Label done, is_null, ok_is_subtype, quicked, resolved;
3357 3357 __ testptr(rax, rax);
3358 3358 __ jcc(Assembler::zero, is_null);
3359 3359
3360 3360 // Get cpool & tags index
3361 3361 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3362 3362 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3363 3363 // See if bytecode has already been quicked
3364 3364 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3365 3365 __ jcc(Assembler::equal, quicked);
3366 3366
3367 3367 __ push(atos);
3368 3368 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3369 3369 __ pop_ptr(rdx);
3370 3370 __ load_klass(rdx, rdx);
3371 3371 __ jmp(resolved);
3372 3372
3373 3373 // Get superklass in EAX and subklass in EDX
3374 3374 __ bind(quicked);
3375 3375 __ load_klass(rdx, rax);
3376 3376 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3377 3377
3378 3378 __ bind(resolved);
3379 3379
3380 3380 // Generate subtype check. Blows ECX. Resets EDI.
3381 3381 // Superklass in EAX. Subklass in EDX.
3382 3382 __ gen_subtype_check( rdx, ok_is_subtype );
3383 3383
3384 3384 // Come here on failure
3385 3385 __ xorl(rax,rax);
3386 3386 __ jmpb(done);
3387 3387 // Come here on success
3388 3388 __ bind(ok_is_subtype);
3389 3389 __ movl(rax, 1);
3390 3390
3391 3391 // Collect counts on whether this test sees NULLs a lot or not.
3392 3392 if (ProfileInterpreter) {
3393 3393 __ jmp(done);
3394 3394 __ bind(is_null);
3395 3395 __ profile_null_seen(rcx);
3396 3396 } else {
3397 3397 __ bind(is_null); // same as 'done'
3398 3398 }
3399 3399 __ bind(done);
3400 3400 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3401 3401 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3402 3402 }
3403 3403
3404 3404
3405 3405 //----------------------------------------------------------------------------------------------------
3406 3406 // Breakpoints
3407 3407 void TemplateTable::_breakpoint() {
3408 3408
3409 3409 // Note: We get here even if we are single stepping..
3410 3410 // jbug inists on setting breakpoints at every bytecode
3411 3411 // even if we are in single step mode.
3412 3412
3413 3413 transition(vtos, vtos);
3414 3414
3415 3415 // get the unpatched byte code
3416 3416 __ get_method(rcx);
3417 3417 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3418 3418 __ mov(rbx, rax);
3419 3419
3420 3420 // post the breakpoint event
3421 3421 __ get_method(rcx);
3422 3422 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3423 3423
3424 3424 // complete the execution of original bytecode
3425 3425 __ dispatch_only_normal(vtos);
3426 3426 }
3427 3427
3428 3428
3429 3429 //----------------------------------------------------------------------------------------------------
3430 3430 // Exceptions
3431 3431
3432 3432 void TemplateTable::athrow() {
3433 3433 transition(atos, vtos);
3434 3434 __ null_check(rax);
3435 3435 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3436 3436 }
3437 3437
3438 3438
3439 3439 //----------------------------------------------------------------------------------------------------
3440 3440 // Synchronization
3441 3441 //
3442 3442 // Note: monitorenter & exit are symmetric routines; which is reflected
3443 3443 // in the assembly code structure as well
3444 3444 //
3445 3445 // Stack layout:
3446 3446 //
3447 3447 // [expressions ] <--- rsp = expression stack top
3448 3448 // ..
3449 3449 // [expressions ]
3450 3450 // [monitor entry] <--- monitor block top = expression stack bot
3451 3451 // ..
3452 3452 // [monitor entry]
3453 3453 // [frame data ] <--- monitor block bot
3454 3454 // ...
3455 3455 // [saved rbp, ] <--- rbp,
3456 3456
3457 3457
3458 3458 void TemplateTable::monitorenter() {
3459 3459 transition(atos, vtos);
3460 3460
3461 3461 // check for NULL object
3462 3462 __ null_check(rax);
3463 3463
3464 3464 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3465 3465 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3466 3466 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3467 3467 Label allocated;
3468 3468
3469 3469 // initialize entry pointer
3470 3470 __ xorl(rdx, rdx); // points to free slot or NULL
3471 3471
3472 3472 // find a free slot in the monitor block (result in rdx)
3473 3473 { Label entry, loop, exit;
3474 3474 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3475 3475
3476 3476 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3477 3477 __ jmpb(entry);
3478 3478
3479 3479 __ bind(loop);
3480 3480 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3481 3481 __ cmovptr(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3482 3482 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3483 3483 __ jccb(Assembler::equal, exit); // if same object then stop searching
3484 3484 __ addptr(rcx, entry_size); // otherwise advance to next entry
3485 3485 __ bind(entry);
3486 3486 __ cmpptr(rcx, rbx); // check if bottom reached
3487 3487 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3488 3488 __ bind(exit);
3489 3489 }
3490 3490
3491 3491 __ testptr(rdx, rdx); // check if a slot has been found
3492 3492 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3493 3493
3494 3494 // allocate one if there's no free slot
3495 3495 { Label entry, loop;
3496 3496 // 1. compute new pointers // rsp: old expression stack top
3497 3497 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3498 3498 __ subptr(rsp, entry_size); // move expression stack top
3499 3499 __ subptr(rdx, entry_size); // move expression stack bottom
3500 3500 __ mov(rcx, rsp); // set start value for copy loop
3501 3501 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3502 3502 __ jmp(entry);
3503 3503 // 2. move expression stack contents
3504 3504 __ bind(loop);
3505 3505 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3506 3506 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3507 3507 __ addptr(rcx, wordSize); // advance to next word
3508 3508 __ bind(entry);
3509 3509 __ cmpptr(rcx, rdx); // check if bottom reached
3510 3510 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3511 3511 }
3512 3512
3513 3513 // call run-time routine
3514 3514 // rdx: points to monitor entry
3515 3515 __ bind(allocated);
3516 3516
3517 3517 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3518 3518 // The object has already been poped from the stack, so the expression stack looks correct.
3519 3519 __ increment(rsi);
3520 3520
3521 3521 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3522 3522 __ lock_object(rdx);
3523 3523
3524 3524 // check to make sure this monitor doesn't cause stack overflow after locking
3525 3525 __ save_bcp(); // in case of exception
3526 3526 __ generate_stack_overflow_check(0);
3527 3527
3528 3528 // The bcp has already been incremented. Just need to dispatch to next instruction.
3529 3529 __ dispatch_next(vtos);
3530 3530 }
3531 3531
3532 3532
3533 3533 void TemplateTable::monitorexit() {
3534 3534 transition(atos, vtos);
3535 3535
3536 3536 // check for NULL object
3537 3537 __ null_check(rax);
3538 3538
3539 3539 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3540 3540 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3541 3541 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3542 3542 Label found;
3543 3543
3544 3544 // find matching slot
3545 3545 { Label entry, loop;
3546 3546 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3547 3547 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3548 3548 __ jmpb(entry);
3549 3549
3550 3550 __ bind(loop);
3551 3551 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3552 3552 __ jcc(Assembler::equal, found); // if same object then stop searching
3553 3553 __ addptr(rdx, entry_size); // otherwise advance to next entry
3554 3554 __ bind(entry);
3555 3555 __ cmpptr(rdx, rbx); // check if bottom reached
3556 3556 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3557 3557 }
3558 3558
3559 3559 // error handling. Unlocking was not block-structured
3560 3560 Label end;
3561 3561 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3562 3562 __ should_not_reach_here();
3563 3563
3564 3564 // call run-time routine
3565 3565 // rcx: points to monitor entry
3566 3566 __ bind(found);
3567 3567 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3568 3568 __ unlock_object(rdx);
3569 3569 __ pop_ptr(rax); // discard object
3570 3570 __ bind(end);
3571 3571 }
3572 3572
3573 3573
3574 3574 //----------------------------------------------------------------------------------------------------
3575 3575 // Wide instructions
3576 3576
3577 3577 void TemplateTable::wide() {
3578 3578 transition(vtos, vtos);
3579 3579 __ load_unsigned_byte(rbx, at_bcp(1));
3580 3580 ExternalAddress wtable((address)Interpreter::_wentry_point);
3581 3581 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3582 3582 // Note: the rsi increment step is part of the individual wide bytecode implementations
3583 3583 }
3584 3584
3585 3585
3586 3586 //----------------------------------------------------------------------------------------------------
3587 3587 // Multi arrays
3588 3588
3589 3589 void TemplateTable::multianewarray() {
3590 3590 transition(vtos, atos);
3591 3591 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3592 3592 // last dim is on top of stack; we want address of first one:
3593 3593 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3594 3594 // the latter wordSize to point to the beginning of the array.
3595 3595 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3596 3596 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3597 3597 __ load_unsigned_byte(rbx, at_bcp(3));
3598 3598 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3599 3599 }
3600 3600
3601 3601 #endif /* !CC_INTERP */
|
↓ open down ↓ |
484 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX