1 /* 2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "interpreter/interpreter.hpp" 27 #include "interpreter/interpreterRuntime.hpp" 28 #include "interpreter/templateTable.hpp" 29 #include "memory/universe.inline.hpp" 30 #include "oops/methodDataOop.hpp" 31 #include "oops/objArrayKlass.hpp" 32 #include "oops/oop.inline.hpp" 33 #include "prims/methodHandles.hpp" 34 #include "runtime/sharedRuntime.hpp" 35 #include "runtime/stubRoutines.hpp" 36 #include "runtime/synchronizer.hpp" 37 38 #ifndef CC_INTERP 39 #define __ _masm-> 40 41 //---------------------------------------------------------------------------------------------------- 42 // Platform-dependent initialization 43 44 void TemplateTable::pd_initialize() { 45 // No i486 specific initialization 46 } 47 48 //---------------------------------------------------------------------------------------------------- 49 // Address computation 50 51 // local variables 52 static inline Address iaddress(int n) { 53 return Address(rdi, Interpreter::local_offset_in_bytes(n)); 54 } 55 56 static inline Address laddress(int n) { return iaddress(n + 1); } 57 static inline Address haddress(int n) { return iaddress(n + 0); } 58 static inline Address faddress(int n) { return iaddress(n); } 59 static inline Address daddress(int n) { return laddress(n); } 60 static inline Address aaddress(int n) { return iaddress(n); } 61 62 static inline Address iaddress(Register r) { 63 return Address(rdi, r, Interpreter::stackElementScale()); 64 } 65 static inline Address laddress(Register r) { 66 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1)); 67 } 68 static inline Address haddress(Register r) { 69 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0)); 70 } 71 72 static inline Address faddress(Register r) { return iaddress(r); } 73 static inline Address daddress(Register r) { return laddress(r); } 74 static inline Address aaddress(Register r) { return iaddress(r); } 75 76 // expression stack 77 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store 78 // data beyond the rsp which is potentially unsafe in an MT environment; 79 // an interrupt may overwrite that data.) 80 static inline Address at_rsp () { 81 return Address(rsp, 0); 82 } 83 84 // At top of Java expression stack which may be different than rsp(). It 85 // isn't for category 1 objects. 86 static inline Address at_tos () { 87 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0)); 88 return tos; 89 } 90 91 static inline Address at_tos_p1() { 92 return Address(rsp, Interpreter::expr_offset_in_bytes(1)); 93 } 94 95 static inline Address at_tos_p2() { 96 return Address(rsp, Interpreter::expr_offset_in_bytes(2)); 97 } 98 99 // Condition conversion 100 static Assembler::Condition j_not(TemplateTable::Condition cc) { 101 switch (cc) { 102 case TemplateTable::equal : return Assembler::notEqual; 103 case TemplateTable::not_equal : return Assembler::equal; 104 case TemplateTable::less : return Assembler::greaterEqual; 105 case TemplateTable::less_equal : return Assembler::greater; 106 case TemplateTable::greater : return Assembler::lessEqual; 107 case TemplateTable::greater_equal: return Assembler::less; 108 } 109 ShouldNotReachHere(); 110 return Assembler::zero; 111 } 112 113 114 //---------------------------------------------------------------------------------------------------- 115 // Miscelaneous helper routines 116 117 // Store an oop (or NULL) at the address described by obj. 118 // If val == noreg this means store a NULL 119 120 static void do_oop_store(InterpreterMacroAssembler* _masm, 121 Address obj, 122 Register val, 123 BarrierSet::Name barrier, 124 bool precise) { 125 assert(val == noreg || val == rax, "parameter is just for looks"); 126 switch (barrier) { 127 #ifndef SERIALGC 128 case BarrierSet::G1SATBCT: 129 case BarrierSet::G1SATBCTLogging: 130 { 131 // flatten object address if needed 132 // We do it regardless of precise because we need the registers 133 if (obj.index() == noreg && obj.disp() == 0) { 134 if (obj.base() != rdx) { 135 __ movl(rdx, obj.base()); 136 } 137 } else { 138 __ leal(rdx, obj); 139 } 140 __ get_thread(rcx); 141 __ save_bcp(); 142 __ g1_write_barrier_pre(rdx, rcx, rsi, rbx, val != noreg); 143 144 // Do the actual store 145 // noreg means NULL 146 if (val == noreg) { 147 __ movptr(Address(rdx, 0), NULL_WORD); 148 // No post barrier for NULL 149 } else { 150 __ movl(Address(rdx, 0), val); 151 __ g1_write_barrier_post(rdx, rax, rcx, rbx, rsi); 152 } 153 __ restore_bcp(); 154 155 } 156 break; 157 #endif // SERIALGC 158 case BarrierSet::CardTableModRef: 159 case BarrierSet::CardTableExtension: 160 { 161 if (val == noreg) { 162 __ movptr(obj, NULL_WORD); 163 } else { 164 __ movl(obj, val); 165 // flatten object address if needed 166 if (!precise || (obj.index() == noreg && obj.disp() == 0)) { 167 __ store_check(obj.base()); 168 } else { 169 __ leal(rdx, obj); 170 __ store_check(rdx); 171 } 172 } 173 } 174 break; 175 case BarrierSet::ModRef: 176 case BarrierSet::Other: 177 if (val == noreg) { 178 __ movptr(obj, NULL_WORD); 179 } else { 180 __ movl(obj, val); 181 } 182 break; 183 default : 184 ShouldNotReachHere(); 185 186 } 187 } 188 189 Address TemplateTable::at_bcp(int offset) { 190 assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 191 return Address(rsi, offset); 192 } 193 194 195 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc, 196 Register scratch, 197 bool load_bc_into_scratch/*=true*/) { 198 199 if (!RewriteBytecodes) return; 200 // the pair bytecodes have already done the load. 201 if (load_bc_into_scratch) { 202 __ movl(bc, bytecode); 203 } 204 Label patch_done; 205 if (JvmtiExport::can_post_breakpoint()) { 206 Label fast_patch; 207 // if a breakpoint is present we can't rewrite the stream directly 208 __ movzbl(scratch, at_bcp(0)); 209 __ cmpl(scratch, Bytecodes::_breakpoint); 210 __ jcc(Assembler::notEqual, fast_patch); 211 __ get_method(scratch); 212 // Let breakpoint table handling rewrite to quicker bytecode 213 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc); 214 #ifndef ASSERT 215 __ jmpb(patch_done); 216 #else 217 __ jmp(patch_done); 218 #endif 219 __ bind(fast_patch); 220 } 221 #ifdef ASSERT 222 Label okay; 223 __ load_unsigned_byte(scratch, at_bcp(0)); 224 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode)); 225 __ jccb(Assembler::equal, okay); 226 __ cmpl(scratch, bc); 227 __ jcc(Assembler::equal, okay); 228 __ stop("patching the wrong bytecode"); 229 __ bind(okay); 230 #endif 231 // patch bytecode 232 __ movb(at_bcp(0), bc); 233 __ bind(patch_done); 234 } 235 236 //---------------------------------------------------------------------------------------------------- 237 // Individual instructions 238 239 void TemplateTable::nop() { 240 transition(vtos, vtos); 241 // nothing to do 242 } 243 244 void TemplateTable::shouldnotreachhere() { 245 transition(vtos, vtos); 246 __ stop("shouldnotreachhere bytecode"); 247 } 248 249 250 251 void TemplateTable::aconst_null() { 252 transition(vtos, atos); 253 __ xorptr(rax, rax); 254 } 255 256 257 void TemplateTable::iconst(int value) { 258 transition(vtos, itos); 259 if (value == 0) { 260 __ xorptr(rax, rax); 261 } else { 262 __ movptr(rax, value); 263 } 264 } 265 266 267 void TemplateTable::lconst(int value) { 268 transition(vtos, ltos); 269 if (value == 0) { 270 __ xorptr(rax, rax); 271 } else { 272 __ movptr(rax, value); 273 } 274 assert(value >= 0, "check this code"); 275 __ xorptr(rdx, rdx); 276 } 277 278 279 void TemplateTable::fconst(int value) { 280 transition(vtos, ftos); 281 if (value == 0) { __ fldz(); 282 } else if (value == 1) { __ fld1(); 283 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here 284 } else { ShouldNotReachHere(); 285 } 286 } 287 288 289 void TemplateTable::dconst(int value) { 290 transition(vtos, dtos); 291 if (value == 0) { __ fldz(); 292 } else if (value == 1) { __ fld1(); 293 } else { ShouldNotReachHere(); 294 } 295 } 296 297 298 void TemplateTable::bipush() { 299 transition(vtos, itos); 300 __ load_signed_byte(rax, at_bcp(1)); 301 } 302 303 304 void TemplateTable::sipush() { 305 transition(vtos, itos); 306 __ load_unsigned_short(rax, at_bcp(1)); 307 __ bswapl(rax); 308 __ sarl(rax, 16); 309 } 310 311 void TemplateTable::ldc(bool wide) { 312 transition(vtos, vtos); 313 Label call_ldc, notFloat, notClass, Done; 314 315 if (wide) { 316 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 317 } else { 318 __ load_unsigned_byte(rbx, at_bcp(1)); 319 } 320 __ get_cpool_and_tags(rcx, rax); 321 const int base_offset = constantPoolOopDesc::header_size() * wordSize; 322 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize; 323 324 // get type 325 __ xorptr(rdx, rdx); 326 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset)); 327 328 // unresolved string - get the resolved string 329 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString); 330 __ jccb(Assembler::equal, call_ldc); 331 332 // unresolved class - get the resolved class 333 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass); 334 __ jccb(Assembler::equal, call_ldc); 335 336 // unresolved class in error (resolution failed) - call into runtime 337 // so that the same error from first resolution attempt is thrown. 338 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError); 339 __ jccb(Assembler::equal, call_ldc); 340 341 // resolved class - need to call vm to get java mirror of the class 342 __ cmpl(rdx, JVM_CONSTANT_Class); 343 __ jcc(Assembler::notEqual, notClass); 344 345 __ bind(call_ldc); 346 __ movl(rcx, wide); 347 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx); 348 __ push(atos); 349 __ jmp(Done); 350 351 __ bind(notClass); 352 __ cmpl(rdx, JVM_CONSTANT_Float); 353 __ jccb(Assembler::notEqual, notFloat); 354 // ftos 355 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset)); 356 __ push(ftos); 357 __ jmp(Done); 358 359 __ bind(notFloat); 360 #ifdef ASSERT 361 { Label L; 362 __ cmpl(rdx, JVM_CONSTANT_Integer); 363 __ jcc(Assembler::equal, L); 364 __ cmpl(rdx, JVM_CONSTANT_String); 365 __ jcc(Assembler::equal, L); 366 __ stop("unexpected tag type in ldc"); 367 __ bind(L); 368 } 369 #endif 370 Label isOop; 371 // atos and itos 372 // String is only oop type we will see here 373 __ cmpl(rdx, JVM_CONSTANT_String); 374 __ jccb(Assembler::equal, isOop); 375 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 376 __ push(itos); 377 __ jmp(Done); 378 __ bind(isOop); 379 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 380 __ push(atos); 381 382 if (VerifyOops) { 383 __ verify_oop(rax); 384 } 385 __ bind(Done); 386 } 387 388 // Fast path for caching oop constants. 389 // %%% We should use this to handle Class and String constants also. 390 // %%% It will simplify the ldc/primitive path considerably. 391 void TemplateTable::fast_aldc(bool wide) { 392 transition(vtos, atos); 393 394 if (!EnableInvokeDynamic) { 395 // We should not encounter this bytecode if !EnableInvokeDynamic. 396 // The verifier will stop it. However, if we get past the verifier, 397 // this will stop the thread in a reasonable way, without crashing the JVM. 398 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 399 InterpreterRuntime::throw_IncompatibleClassChangeError)); 400 // the call_VM checks for exception, so we should never return here. 401 __ should_not_reach_here(); 402 return; 403 } 404 405 const Register cache = rcx; 406 const Register index = rdx; 407 408 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1)); 409 if (VerifyOops) { 410 __ verify_oop(rax); 411 } 412 413 Label L_done, L_throw_exception; 414 const Register con_klass_temp = rcx; // same as Rcache 415 __ movptr(con_klass_temp, Address(rax, oopDesc::klass_offset_in_bytes())); 416 __ cmpptr(con_klass_temp, ExternalAddress((address)Universe::systemObjArrayKlassObj_addr())); 417 __ jcc(Assembler::notEqual, L_done); 418 __ cmpl(Address(rax, arrayOopDesc::length_offset_in_bytes()), 0); 419 __ jcc(Assembler::notEqual, L_throw_exception); 420 __ xorptr(rax, rax); 421 __ jmp(L_done); 422 423 // Load the exception from the system-array which wraps it: 424 __ bind(L_throw_exception); 425 __ movptr(rax, Address(rax, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 426 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 427 428 __ bind(L_done); 429 } 430 431 void TemplateTable::ldc2_w() { 432 transition(vtos, vtos); 433 Label Long, Done; 434 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 435 436 __ get_cpool_and_tags(rcx, rax); 437 const int base_offset = constantPoolOopDesc::header_size() * wordSize; 438 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize; 439 440 // get type 441 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double); 442 __ jccb(Assembler::notEqual, Long); 443 // dtos 444 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset)); 445 __ push(dtos); 446 __ jmpb(Done); 447 448 __ bind(Long); 449 // ltos 450 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize)); 451 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize))); 452 453 __ push(ltos); 454 455 __ bind(Done); 456 } 457 458 459 void TemplateTable::locals_index(Register reg, int offset) { 460 __ load_unsigned_byte(reg, at_bcp(offset)); 461 __ negptr(reg); 462 } 463 464 465 void TemplateTable::iload() { 466 transition(vtos, itos); 467 if (RewriteFrequentPairs) { 468 Label rewrite, done; 469 470 // get next byte 471 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 472 // if _iload, wait to rewrite to iload2. We only want to rewrite the 473 // last two iloads in a pair. Comparing against fast_iload means that 474 // the next bytecode is neither an iload or a caload, and therefore 475 // an iload pair. 476 __ cmpl(rbx, Bytecodes::_iload); 477 __ jcc(Assembler::equal, done); 478 479 __ cmpl(rbx, Bytecodes::_fast_iload); 480 __ movl(rcx, Bytecodes::_fast_iload2); 481 __ jccb(Assembler::equal, rewrite); 482 483 // if _caload, rewrite to fast_icaload 484 __ cmpl(rbx, Bytecodes::_caload); 485 __ movl(rcx, Bytecodes::_fast_icaload); 486 __ jccb(Assembler::equal, rewrite); 487 488 // rewrite so iload doesn't check again. 489 __ movl(rcx, Bytecodes::_fast_iload); 490 491 // rewrite 492 // rcx: fast bytecode 493 __ bind(rewrite); 494 patch_bytecode(Bytecodes::_iload, rcx, rbx, false); 495 __ bind(done); 496 } 497 498 // Get the local value into tos 499 locals_index(rbx); 500 __ movl(rax, iaddress(rbx)); 501 } 502 503 504 void TemplateTable::fast_iload2() { 505 transition(vtos, itos); 506 locals_index(rbx); 507 __ movl(rax, iaddress(rbx)); 508 __ push(itos); 509 locals_index(rbx, 3); 510 __ movl(rax, iaddress(rbx)); 511 } 512 513 void TemplateTable::fast_iload() { 514 transition(vtos, itos); 515 locals_index(rbx); 516 __ movl(rax, iaddress(rbx)); 517 } 518 519 520 void TemplateTable::lload() { 521 transition(vtos, ltos); 522 locals_index(rbx); 523 __ movptr(rax, laddress(rbx)); 524 NOT_LP64(__ movl(rdx, haddress(rbx))); 525 } 526 527 528 void TemplateTable::fload() { 529 transition(vtos, ftos); 530 locals_index(rbx); 531 __ fld_s(faddress(rbx)); 532 } 533 534 535 void TemplateTable::dload() { 536 transition(vtos, dtos); 537 locals_index(rbx); 538 __ fld_d(daddress(rbx)); 539 } 540 541 542 void TemplateTable::aload() { 543 transition(vtos, atos); 544 locals_index(rbx); 545 __ movptr(rax, aaddress(rbx)); 546 } 547 548 549 void TemplateTable::locals_index_wide(Register reg) { 550 __ movl(reg, at_bcp(2)); 551 __ bswapl(reg); 552 __ shrl(reg, 16); 553 __ negptr(reg); 554 } 555 556 557 void TemplateTable::wide_iload() { 558 transition(vtos, itos); 559 locals_index_wide(rbx); 560 __ movl(rax, iaddress(rbx)); 561 } 562 563 564 void TemplateTable::wide_lload() { 565 transition(vtos, ltos); 566 locals_index_wide(rbx); 567 __ movptr(rax, laddress(rbx)); 568 NOT_LP64(__ movl(rdx, haddress(rbx))); 569 } 570 571 572 void TemplateTable::wide_fload() { 573 transition(vtos, ftos); 574 locals_index_wide(rbx); 575 __ fld_s(faddress(rbx)); 576 } 577 578 579 void TemplateTable::wide_dload() { 580 transition(vtos, dtos); 581 locals_index_wide(rbx); 582 __ fld_d(daddress(rbx)); 583 } 584 585 586 void TemplateTable::wide_aload() { 587 transition(vtos, atos); 588 locals_index_wide(rbx); 589 __ movptr(rax, aaddress(rbx)); 590 } 591 592 void TemplateTable::index_check(Register array, Register index) { 593 // Pop ptr into array 594 __ pop_ptr(array); 595 index_check_without_pop(array, index); 596 } 597 598 void TemplateTable::index_check_without_pop(Register array, Register index) { 599 // destroys rbx, 600 // check array 601 __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 602 LP64_ONLY(__ movslq(index, index)); 603 // check index 604 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes())); 605 if (index != rbx) { 606 // ??? convention: move aberrant index into rbx, for exception message 607 assert(rbx != array, "different registers"); 608 __ mov(rbx, index); 609 } 610 __ jump_cc(Assembler::aboveEqual, 611 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry)); 612 } 613 614 615 void TemplateTable::iaload() { 616 transition(itos, itos); 617 // rdx: array 618 index_check(rdx, rax); // kills rbx, 619 // rax,: index 620 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT))); 621 } 622 623 624 void TemplateTable::laload() { 625 transition(itos, ltos); 626 // rax,: index 627 // rdx: array 628 index_check(rdx, rax); 629 __ mov(rbx, rax); 630 // rbx,: index 631 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize)); 632 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize))); 633 } 634 635 636 void TemplateTable::faload() { 637 transition(itos, ftos); 638 // rdx: array 639 index_check(rdx, rax); // kills rbx, 640 // rax,: index 641 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 642 } 643 644 645 void TemplateTable::daload() { 646 transition(itos, dtos); 647 // rdx: array 648 index_check(rdx, rax); // kills rbx, 649 // rax,: index 650 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 651 } 652 653 654 void TemplateTable::aaload() { 655 transition(itos, atos); 656 // rdx: array 657 index_check(rdx, rax); // kills rbx, 658 // rax,: index 659 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 660 } 661 662 663 void TemplateTable::baload() { 664 transition(itos, itos); 665 // rdx: array 666 index_check(rdx, rax); // kills rbx, 667 // rax,: index 668 // can do better code for P5 - fix this at some point 669 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 670 __ mov(rax, rbx); 671 } 672 673 674 void TemplateTable::caload() { 675 transition(itos, itos); 676 // rdx: array 677 index_check(rdx, rax); // kills rbx, 678 // rax,: index 679 // can do better code for P5 - may want to improve this at some point 680 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 681 __ mov(rax, rbx); 682 } 683 684 // iload followed by caload frequent pair 685 void TemplateTable::fast_icaload() { 686 transition(vtos, itos); 687 // load index out of locals 688 locals_index(rbx); 689 __ movl(rax, iaddress(rbx)); 690 691 // rdx: array 692 index_check(rdx, rax); 693 // rax,: index 694 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 695 __ mov(rax, rbx); 696 } 697 698 void TemplateTable::saload() { 699 transition(itos, itos); 700 // rdx: array 701 index_check(rdx, rax); // kills rbx, 702 // rax,: index 703 // can do better code for P5 - may want to improve this at some point 704 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 705 __ mov(rax, rbx); 706 } 707 708 709 void TemplateTable::iload(int n) { 710 transition(vtos, itos); 711 __ movl(rax, iaddress(n)); 712 } 713 714 715 void TemplateTable::lload(int n) { 716 transition(vtos, ltos); 717 __ movptr(rax, laddress(n)); 718 NOT_LP64(__ movptr(rdx, haddress(n))); 719 } 720 721 722 void TemplateTable::fload(int n) { 723 transition(vtos, ftos); 724 __ fld_s(faddress(n)); 725 } 726 727 728 void TemplateTable::dload(int n) { 729 transition(vtos, dtos); 730 __ fld_d(daddress(n)); 731 } 732 733 734 void TemplateTable::aload(int n) { 735 transition(vtos, atos); 736 __ movptr(rax, aaddress(n)); 737 } 738 739 740 void TemplateTable::aload_0() { 741 transition(vtos, atos); 742 // According to bytecode histograms, the pairs: 743 // 744 // _aload_0, _fast_igetfield 745 // _aload_0, _fast_agetfield 746 // _aload_0, _fast_fgetfield 747 // 748 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0 749 // bytecode checks if the next bytecode is either _fast_igetfield, 750 // _fast_agetfield or _fast_fgetfield and then rewrites the 751 // current bytecode into a pair bytecode; otherwise it rewrites the current 752 // bytecode into _fast_aload_0 that doesn't do the pair check anymore. 753 // 754 // Note: If the next bytecode is _getfield, the rewrite must be delayed, 755 // otherwise we may miss an opportunity for a pair. 756 // 757 // Also rewrite frequent pairs 758 // aload_0, aload_1 759 // aload_0, iload_1 760 // These bytecodes with a small amount of code are most profitable to rewrite 761 if (RewriteFrequentPairs) { 762 Label rewrite, done; 763 // get next byte 764 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 765 766 // do actual aload_0 767 aload(0); 768 769 // if _getfield then wait with rewrite 770 __ cmpl(rbx, Bytecodes::_getfield); 771 __ jcc(Assembler::equal, done); 772 773 // if _igetfield then reqrite to _fast_iaccess_0 774 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 775 __ cmpl(rbx, Bytecodes::_fast_igetfield); 776 __ movl(rcx, Bytecodes::_fast_iaccess_0); 777 __ jccb(Assembler::equal, rewrite); 778 779 // if _agetfield then reqrite to _fast_aaccess_0 780 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 781 __ cmpl(rbx, Bytecodes::_fast_agetfield); 782 __ movl(rcx, Bytecodes::_fast_aaccess_0); 783 __ jccb(Assembler::equal, rewrite); 784 785 // if _fgetfield then reqrite to _fast_faccess_0 786 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 787 __ cmpl(rbx, Bytecodes::_fast_fgetfield); 788 __ movl(rcx, Bytecodes::_fast_faccess_0); 789 __ jccb(Assembler::equal, rewrite); 790 791 // else rewrite to _fast_aload0 792 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 793 __ movl(rcx, Bytecodes::_fast_aload_0); 794 795 // rewrite 796 // rcx: fast bytecode 797 __ bind(rewrite); 798 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false); 799 800 __ bind(done); 801 } else { 802 aload(0); 803 } 804 } 805 806 void TemplateTable::istore() { 807 transition(itos, vtos); 808 locals_index(rbx); 809 __ movl(iaddress(rbx), rax); 810 } 811 812 813 void TemplateTable::lstore() { 814 transition(ltos, vtos); 815 locals_index(rbx); 816 __ movptr(laddress(rbx), rax); 817 NOT_LP64(__ movptr(haddress(rbx), rdx)); 818 } 819 820 821 void TemplateTable::fstore() { 822 transition(ftos, vtos); 823 locals_index(rbx); 824 __ fstp_s(faddress(rbx)); 825 } 826 827 828 void TemplateTable::dstore() { 829 transition(dtos, vtos); 830 locals_index(rbx); 831 __ fstp_d(daddress(rbx)); 832 } 833 834 835 void TemplateTable::astore() { 836 transition(vtos, vtos); 837 __ pop_ptr(rax); 838 locals_index(rbx); 839 __ movptr(aaddress(rbx), rax); 840 } 841 842 843 void TemplateTable::wide_istore() { 844 transition(vtos, vtos); 845 __ pop_i(rax); 846 locals_index_wide(rbx); 847 __ movl(iaddress(rbx), rax); 848 } 849 850 851 void TemplateTable::wide_lstore() { 852 transition(vtos, vtos); 853 __ pop_l(rax, rdx); 854 locals_index_wide(rbx); 855 __ movptr(laddress(rbx), rax); 856 NOT_LP64(__ movl(haddress(rbx), rdx)); 857 } 858 859 860 void TemplateTable::wide_fstore() { 861 wide_istore(); 862 } 863 864 865 void TemplateTable::wide_dstore() { 866 wide_lstore(); 867 } 868 869 870 void TemplateTable::wide_astore() { 871 transition(vtos, vtos); 872 __ pop_ptr(rax); 873 locals_index_wide(rbx); 874 __ movptr(aaddress(rbx), rax); 875 } 876 877 878 void TemplateTable::iastore() { 879 transition(itos, vtos); 880 __ pop_i(rbx); 881 // rax,: value 882 // rdx: array 883 index_check(rdx, rbx); // prefer index in rbx, 884 // rbx,: index 885 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax); 886 } 887 888 889 void TemplateTable::lastore() { 890 transition(ltos, vtos); 891 __ pop_i(rbx); 892 // rax,: low(value) 893 // rcx: array 894 // rdx: high(value) 895 index_check(rcx, rbx); // prefer index in rbx, 896 // rbx,: index 897 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax); 898 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx)); 899 } 900 901 902 void TemplateTable::fastore() { 903 transition(ftos, vtos); 904 __ pop_i(rbx); 905 // rdx: array 906 // st0: value 907 index_check(rdx, rbx); // prefer index in rbx, 908 // rbx,: index 909 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 910 } 911 912 913 void TemplateTable::dastore() { 914 transition(dtos, vtos); 915 __ pop_i(rbx); 916 // rdx: array 917 // st0: value 918 index_check(rdx, rbx); // prefer index in rbx, 919 // rbx,: index 920 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 921 } 922 923 924 void TemplateTable::aastore() { 925 Label is_null, ok_is_subtype, done; 926 transition(vtos, vtos); 927 // stack: ..., array, index, value 928 __ movptr(rax, at_tos()); // Value 929 __ movl(rcx, at_tos_p1()); // Index 930 __ movptr(rdx, at_tos_p2()); // Array 931 932 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 933 index_check_without_pop(rdx, rcx); // kills rbx, 934 // do array store check - check for NULL value first 935 __ testptr(rax, rax); 936 __ jcc(Assembler::zero, is_null); 937 938 // Move subklass into EBX 939 __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes())); 940 // Move superklass into EAX 941 __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes())); 942 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes())); 943 // Compress array+index*wordSize+12 into a single register. Frees ECX. 944 __ lea(rdx, element_address); 945 946 // Generate subtype check. Blows ECX. Resets EDI to locals. 947 // Superklass in EAX. Subklass in EBX. 948 __ gen_subtype_check( rbx, ok_is_subtype ); 949 950 // Come here on failure 951 // object is at TOS 952 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry)); 953 954 // Come here on success 955 __ bind(ok_is_subtype); 956 957 // Get the value to store 958 __ movptr(rax, at_rsp()); 959 // and store it with appropriate barrier 960 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true); 961 962 __ jmp(done); 963 964 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx] 965 __ bind(is_null); 966 __ profile_null_seen(rbx); 967 968 // Store NULL, (noreg means NULL to do_oop_store) 969 do_oop_store(_masm, element_address, noreg, _bs->kind(), true); 970 971 // Pop stack arguments 972 __ bind(done); 973 __ addptr(rsp, 3 * Interpreter::stackElementSize); 974 } 975 976 977 void TemplateTable::bastore() { 978 transition(itos, vtos); 979 __ pop_i(rbx); 980 // rax,: value 981 // rdx: array 982 index_check(rdx, rbx); // prefer index in rbx, 983 // rbx,: index 984 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax); 985 } 986 987 988 void TemplateTable::castore() { 989 transition(itos, vtos); 990 __ pop_i(rbx); 991 // rax,: value 992 // rdx: array 993 index_check(rdx, rbx); // prefer index in rbx, 994 // rbx,: index 995 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax); 996 } 997 998 999 void TemplateTable::sastore() { 1000 castore(); 1001 } 1002 1003 1004 void TemplateTable::istore(int n) { 1005 transition(itos, vtos); 1006 __ movl(iaddress(n), rax); 1007 } 1008 1009 1010 void TemplateTable::lstore(int n) { 1011 transition(ltos, vtos); 1012 __ movptr(laddress(n), rax); 1013 NOT_LP64(__ movptr(haddress(n), rdx)); 1014 } 1015 1016 1017 void TemplateTable::fstore(int n) { 1018 transition(ftos, vtos); 1019 __ fstp_s(faddress(n)); 1020 } 1021 1022 1023 void TemplateTable::dstore(int n) { 1024 transition(dtos, vtos); 1025 __ fstp_d(daddress(n)); 1026 } 1027 1028 1029 void TemplateTable::astore(int n) { 1030 transition(vtos, vtos); 1031 __ pop_ptr(rax); 1032 __ movptr(aaddress(n), rax); 1033 } 1034 1035 1036 void TemplateTable::pop() { 1037 transition(vtos, vtos); 1038 __ addptr(rsp, Interpreter::stackElementSize); 1039 } 1040 1041 1042 void TemplateTable::pop2() { 1043 transition(vtos, vtos); 1044 __ addptr(rsp, 2*Interpreter::stackElementSize); 1045 } 1046 1047 1048 void TemplateTable::dup() { 1049 transition(vtos, vtos); 1050 // stack: ..., a 1051 __ load_ptr(0, rax); 1052 __ push_ptr(rax); 1053 // stack: ..., a, a 1054 } 1055 1056 1057 void TemplateTable::dup_x1() { 1058 transition(vtos, vtos); 1059 // stack: ..., a, b 1060 __ load_ptr( 0, rax); // load b 1061 __ load_ptr( 1, rcx); // load a 1062 __ store_ptr(1, rax); // store b 1063 __ store_ptr(0, rcx); // store a 1064 __ push_ptr(rax); // push b 1065 // stack: ..., b, a, b 1066 } 1067 1068 1069 void TemplateTable::dup_x2() { 1070 transition(vtos, vtos); 1071 // stack: ..., a, b, c 1072 __ load_ptr( 0, rax); // load c 1073 __ load_ptr( 2, rcx); // load a 1074 __ store_ptr(2, rax); // store c in a 1075 __ push_ptr(rax); // push c 1076 // stack: ..., c, b, c, c 1077 __ load_ptr( 2, rax); // load b 1078 __ store_ptr(2, rcx); // store a in b 1079 // stack: ..., c, a, c, c 1080 __ store_ptr(1, rax); // store b in c 1081 // stack: ..., c, a, b, c 1082 } 1083 1084 1085 void TemplateTable::dup2() { 1086 transition(vtos, vtos); 1087 // stack: ..., a, b 1088 __ load_ptr(1, rax); // load a 1089 __ push_ptr(rax); // push a 1090 __ load_ptr(1, rax); // load b 1091 __ push_ptr(rax); // push b 1092 // stack: ..., a, b, a, b 1093 } 1094 1095 1096 void TemplateTable::dup2_x1() { 1097 transition(vtos, vtos); 1098 // stack: ..., a, b, c 1099 __ load_ptr( 0, rcx); // load c 1100 __ load_ptr( 1, rax); // load b 1101 __ push_ptr(rax); // push b 1102 __ push_ptr(rcx); // push c 1103 // stack: ..., a, b, c, b, c 1104 __ store_ptr(3, rcx); // store c in b 1105 // stack: ..., a, c, c, b, c 1106 __ load_ptr( 4, rcx); // load a 1107 __ store_ptr(2, rcx); // store a in 2nd c 1108 // stack: ..., a, c, a, b, c 1109 __ store_ptr(4, rax); // store b in a 1110 // stack: ..., b, c, a, b, c 1111 // stack: ..., b, c, a, b, c 1112 } 1113 1114 1115 void TemplateTable::dup2_x2() { 1116 transition(vtos, vtos); 1117 // stack: ..., a, b, c, d 1118 __ load_ptr( 0, rcx); // load d 1119 __ load_ptr( 1, rax); // load c 1120 __ push_ptr(rax); // push c 1121 __ push_ptr(rcx); // push d 1122 // stack: ..., a, b, c, d, c, d 1123 __ load_ptr( 4, rax); // load b 1124 __ store_ptr(2, rax); // store b in d 1125 __ store_ptr(4, rcx); // store d in b 1126 // stack: ..., a, d, c, b, c, d 1127 __ load_ptr( 5, rcx); // load a 1128 __ load_ptr( 3, rax); // load c 1129 __ store_ptr(3, rcx); // store a in c 1130 __ store_ptr(5, rax); // store c in a 1131 // stack: ..., c, d, a, b, c, d 1132 // stack: ..., c, d, a, b, c, d 1133 } 1134 1135 1136 void TemplateTable::swap() { 1137 transition(vtos, vtos); 1138 // stack: ..., a, b 1139 __ load_ptr( 1, rcx); // load a 1140 __ load_ptr( 0, rax); // load b 1141 __ store_ptr(0, rcx); // store a in b 1142 __ store_ptr(1, rax); // store b in a 1143 // stack: ..., b, a 1144 } 1145 1146 1147 void TemplateTable::iop2(Operation op) { 1148 transition(itos, itos); 1149 switch (op) { 1150 case add : __ pop_i(rdx); __ addl (rax, rdx); break; 1151 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break; 1152 case mul : __ pop_i(rdx); __ imull(rax, rdx); break; 1153 case _and : __ pop_i(rdx); __ andl (rax, rdx); break; 1154 case _or : __ pop_i(rdx); __ orl (rax, rdx); break; 1155 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break; 1156 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1157 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1158 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1159 default : ShouldNotReachHere(); 1160 } 1161 } 1162 1163 1164 void TemplateTable::lop2(Operation op) { 1165 transition(ltos, ltos); 1166 __ pop_l(rbx, rcx); 1167 switch (op) { 1168 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break; 1169 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx); 1170 __ mov (rax, rbx); __ mov (rdx, rcx); break; 1171 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break; 1172 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break; 1173 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break; 1174 default : ShouldNotReachHere(); 1175 } 1176 } 1177 1178 1179 void TemplateTable::idiv() { 1180 transition(itos, itos); 1181 __ mov(rcx, rax); 1182 __ pop_i(rax); 1183 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If 1184 // they are not equal, one could do a normal division (no correction 1185 // needed), which may speed up this implementation for the common case. 1186 // (see also JVM spec., p.243 & p.271) 1187 __ corrected_idivl(rcx); 1188 } 1189 1190 1191 void TemplateTable::irem() { 1192 transition(itos, itos); 1193 __ mov(rcx, rax); 1194 __ pop_i(rax); 1195 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If 1196 // they are not equal, one could do a normal division (no correction 1197 // needed), which may speed up this implementation for the common case. 1198 // (see also JVM spec., p.243 & p.271) 1199 __ corrected_idivl(rcx); 1200 __ mov(rax, rdx); 1201 } 1202 1203 1204 void TemplateTable::lmul() { 1205 transition(ltos, ltos); 1206 __ pop_l(rbx, rcx); 1207 __ push(rcx); __ push(rbx); 1208 __ push(rdx); __ push(rax); 1209 __ lmul(2 * wordSize, 0); 1210 __ addptr(rsp, 4 * wordSize); // take off temporaries 1211 } 1212 1213 1214 void TemplateTable::ldiv() { 1215 transition(ltos, ltos); 1216 __ pop_l(rbx, rcx); 1217 __ push(rcx); __ push(rbx); 1218 __ push(rdx); __ push(rax); 1219 // check if y = 0 1220 __ orl(rax, rdx); 1221 __ jump_cc(Assembler::zero, 1222 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1223 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv)); 1224 __ addptr(rsp, 4 * wordSize); // take off temporaries 1225 } 1226 1227 1228 void TemplateTable::lrem() { 1229 transition(ltos, ltos); 1230 __ pop_l(rbx, rcx); 1231 __ push(rcx); __ push(rbx); 1232 __ push(rdx); __ push(rax); 1233 // check if y = 0 1234 __ orl(rax, rdx); 1235 __ jump_cc(Assembler::zero, 1236 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1237 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem)); 1238 __ addptr(rsp, 4 * wordSize); 1239 } 1240 1241 1242 void TemplateTable::lshl() { 1243 transition(itos, ltos); 1244 __ movl(rcx, rax); // get shift count 1245 __ pop_l(rax, rdx); // get shift value 1246 __ lshl(rdx, rax); 1247 } 1248 1249 1250 void TemplateTable::lshr() { 1251 transition(itos, ltos); 1252 __ mov(rcx, rax); // get shift count 1253 __ pop_l(rax, rdx); // get shift value 1254 __ lshr(rdx, rax, true); 1255 } 1256 1257 1258 void TemplateTable::lushr() { 1259 transition(itos, ltos); 1260 __ mov(rcx, rax); // get shift count 1261 __ pop_l(rax, rdx); // get shift value 1262 __ lshr(rdx, rax); 1263 } 1264 1265 1266 void TemplateTable::fop2(Operation op) { 1267 transition(ftos, ftos); 1268 switch (op) { 1269 case add: __ fadd_s (at_rsp()); break; 1270 case sub: __ fsubr_s(at_rsp()); break; 1271 case mul: __ fmul_s (at_rsp()); break; 1272 case div: __ fdivr_s(at_rsp()); break; 1273 case rem: __ fld_s (at_rsp()); __ fremr(rax); break; 1274 default : ShouldNotReachHere(); 1275 } 1276 __ f2ieee(); 1277 __ pop(rax); // pop float thing off 1278 } 1279 1280 1281 void TemplateTable::dop2(Operation op) { 1282 transition(dtos, dtos); 1283 1284 switch (op) { 1285 case add: __ fadd_d (at_rsp()); break; 1286 case sub: __ fsubr_d(at_rsp()); break; 1287 case mul: { 1288 Label L_strict; 1289 Label L_join; 1290 const Address access_flags (rcx, methodOopDesc::access_flags_offset()); 1291 __ get_method(rcx); 1292 __ movl(rcx, access_flags); 1293 __ testl(rcx, JVM_ACC_STRICT); 1294 __ jccb(Assembler::notZero, L_strict); 1295 __ fmul_d (at_rsp()); 1296 __ jmpb(L_join); 1297 __ bind(L_strict); 1298 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1299 __ fmulp(); 1300 __ fmul_d (at_rsp()); 1301 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1302 __ fmulp(); 1303 __ bind(L_join); 1304 break; 1305 } 1306 case div: { 1307 Label L_strict; 1308 Label L_join; 1309 const Address access_flags (rcx, methodOopDesc::access_flags_offset()); 1310 __ get_method(rcx); 1311 __ movl(rcx, access_flags); 1312 __ testl(rcx, JVM_ACC_STRICT); 1313 __ jccb(Assembler::notZero, L_strict); 1314 __ fdivr_d(at_rsp()); 1315 __ jmp(L_join); 1316 __ bind(L_strict); 1317 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1318 __ fmul_d (at_rsp()); 1319 __ fdivrp(); 1320 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1321 __ fmulp(); 1322 __ bind(L_join); 1323 break; 1324 } 1325 case rem: __ fld_d (at_rsp()); __ fremr(rax); break; 1326 default : ShouldNotReachHere(); 1327 } 1328 __ d2ieee(); 1329 // Pop double precision number from rsp. 1330 __ pop(rax); 1331 __ pop(rdx); 1332 } 1333 1334 1335 void TemplateTable::ineg() { 1336 transition(itos, itos); 1337 __ negl(rax); 1338 } 1339 1340 1341 void TemplateTable::lneg() { 1342 transition(ltos, ltos); 1343 __ lneg(rdx, rax); 1344 } 1345 1346 1347 void TemplateTable::fneg() { 1348 transition(ftos, ftos); 1349 __ fchs(); 1350 } 1351 1352 1353 void TemplateTable::dneg() { 1354 transition(dtos, dtos); 1355 __ fchs(); 1356 } 1357 1358 1359 void TemplateTable::iinc() { 1360 transition(vtos, vtos); 1361 __ load_signed_byte(rdx, at_bcp(2)); // get constant 1362 locals_index(rbx); 1363 __ addl(iaddress(rbx), rdx); 1364 } 1365 1366 1367 void TemplateTable::wide_iinc() { 1368 transition(vtos, vtos); 1369 __ movl(rdx, at_bcp(4)); // get constant 1370 locals_index_wide(rbx); 1371 __ bswapl(rdx); // swap bytes & sign-extend constant 1372 __ sarl(rdx, 16); 1373 __ addl(iaddress(rbx), rdx); 1374 // Note: should probably use only one movl to get both 1375 // the index and the constant -> fix this 1376 } 1377 1378 1379 void TemplateTable::convert() { 1380 // Checking 1381 #ifdef ASSERT 1382 { TosState tos_in = ilgl; 1383 TosState tos_out = ilgl; 1384 switch (bytecode()) { 1385 case Bytecodes::_i2l: // fall through 1386 case Bytecodes::_i2f: // fall through 1387 case Bytecodes::_i2d: // fall through 1388 case Bytecodes::_i2b: // fall through 1389 case Bytecodes::_i2c: // fall through 1390 case Bytecodes::_i2s: tos_in = itos; break; 1391 case Bytecodes::_l2i: // fall through 1392 case Bytecodes::_l2f: // fall through 1393 case Bytecodes::_l2d: tos_in = ltos; break; 1394 case Bytecodes::_f2i: // fall through 1395 case Bytecodes::_f2l: // fall through 1396 case Bytecodes::_f2d: tos_in = ftos; break; 1397 case Bytecodes::_d2i: // fall through 1398 case Bytecodes::_d2l: // fall through 1399 case Bytecodes::_d2f: tos_in = dtos; break; 1400 default : ShouldNotReachHere(); 1401 } 1402 switch (bytecode()) { 1403 case Bytecodes::_l2i: // fall through 1404 case Bytecodes::_f2i: // fall through 1405 case Bytecodes::_d2i: // fall through 1406 case Bytecodes::_i2b: // fall through 1407 case Bytecodes::_i2c: // fall through 1408 case Bytecodes::_i2s: tos_out = itos; break; 1409 case Bytecodes::_i2l: // fall through 1410 case Bytecodes::_f2l: // fall through 1411 case Bytecodes::_d2l: tos_out = ltos; break; 1412 case Bytecodes::_i2f: // fall through 1413 case Bytecodes::_l2f: // fall through 1414 case Bytecodes::_d2f: tos_out = ftos; break; 1415 case Bytecodes::_i2d: // fall through 1416 case Bytecodes::_l2d: // fall through 1417 case Bytecodes::_f2d: tos_out = dtos; break; 1418 default : ShouldNotReachHere(); 1419 } 1420 transition(tos_in, tos_out); 1421 } 1422 #endif // ASSERT 1423 1424 // Conversion 1425 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation) 1426 switch (bytecode()) { 1427 case Bytecodes::_i2l: 1428 __ extend_sign(rdx, rax); 1429 break; 1430 case Bytecodes::_i2f: 1431 __ push(rax); // store int on tos 1432 __ fild_s(at_rsp()); // load int to ST0 1433 __ f2ieee(); // truncate to float size 1434 __ pop(rcx); // adjust rsp 1435 break; 1436 case Bytecodes::_i2d: 1437 __ push(rax); // add one slot for d2ieee() 1438 __ push(rax); // store int on tos 1439 __ fild_s(at_rsp()); // load int to ST0 1440 __ d2ieee(); // truncate to double size 1441 __ pop(rcx); // adjust rsp 1442 __ pop(rcx); 1443 break; 1444 case Bytecodes::_i2b: 1445 __ shll(rax, 24); // truncate upper 24 bits 1446 __ sarl(rax, 24); // and sign-extend byte 1447 LP64_ONLY(__ movsbl(rax, rax)); 1448 break; 1449 case Bytecodes::_i2c: 1450 __ andl(rax, 0xFFFF); // truncate upper 16 bits 1451 LP64_ONLY(__ movzwl(rax, rax)); 1452 break; 1453 case Bytecodes::_i2s: 1454 __ shll(rax, 16); // truncate upper 16 bits 1455 __ sarl(rax, 16); // and sign-extend short 1456 LP64_ONLY(__ movswl(rax, rax)); 1457 break; 1458 case Bytecodes::_l2i: 1459 /* nothing to do */ 1460 break; 1461 case Bytecodes::_l2f: 1462 __ push(rdx); // store long on tos 1463 __ push(rax); 1464 __ fild_d(at_rsp()); // load long to ST0 1465 __ f2ieee(); // truncate to float size 1466 __ pop(rcx); // adjust rsp 1467 __ pop(rcx); 1468 break; 1469 case Bytecodes::_l2d: 1470 __ push(rdx); // store long on tos 1471 __ push(rax); 1472 __ fild_d(at_rsp()); // load long to ST0 1473 __ d2ieee(); // truncate to double size 1474 __ pop(rcx); // adjust rsp 1475 __ pop(rcx); 1476 break; 1477 case Bytecodes::_f2i: 1478 __ push(rcx); // reserve space for argument 1479 __ fstp_s(at_rsp()); // pass float argument on stack 1480 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1481 break; 1482 case Bytecodes::_f2l: 1483 __ push(rcx); // reserve space for argument 1484 __ fstp_s(at_rsp()); // pass float argument on stack 1485 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1486 break; 1487 case Bytecodes::_f2d: 1488 /* nothing to do */ 1489 break; 1490 case Bytecodes::_d2i: 1491 __ push(rcx); // reserve space for argument 1492 __ push(rcx); 1493 __ fstp_d(at_rsp()); // pass double argument on stack 1494 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2); 1495 break; 1496 case Bytecodes::_d2l: 1497 __ push(rcx); // reserve space for argument 1498 __ push(rcx); 1499 __ fstp_d(at_rsp()); // pass double argument on stack 1500 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2); 1501 break; 1502 case Bytecodes::_d2f: 1503 __ push(rcx); // reserve space for f2ieee() 1504 __ f2ieee(); // truncate to float size 1505 __ pop(rcx); // adjust rsp 1506 break; 1507 default : 1508 ShouldNotReachHere(); 1509 } 1510 } 1511 1512 1513 void TemplateTable::lcmp() { 1514 transition(ltos, itos); 1515 // y = rdx:rax 1516 __ pop_l(rbx, rcx); // get x = rcx:rbx 1517 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y) 1518 __ mov(rax, rcx); 1519 } 1520 1521 1522 void TemplateTable::float_cmp(bool is_float, int unordered_result) { 1523 if (is_float) { 1524 __ fld_s(at_rsp()); 1525 } else { 1526 __ fld_d(at_rsp()); 1527 __ pop(rdx); 1528 } 1529 __ pop(rcx); 1530 __ fcmp2int(rax, unordered_result < 0); 1531 } 1532 1533 1534 void TemplateTable::branch(bool is_jsr, bool is_wide) { 1535 __ get_method(rcx); // ECX holds method 1536 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count 1537 1538 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset(); 1539 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset(); 1540 const int method_offset = frame::interpreter_frame_method_offset * wordSize; 1541 1542 // Load up EDX with the branch displacement 1543 __ movl(rdx, at_bcp(1)); 1544 __ bswapl(rdx); 1545 if (!is_wide) __ sarl(rdx, 16); 1546 LP64_ONLY(__ movslq(rdx, rdx)); 1547 1548 1549 // Handle all the JSR stuff here, then exit. 1550 // It's much shorter and cleaner than intermingling with the 1551 // non-JSR normal-branch stuff occurring below. 1552 if (is_jsr) { 1553 // Pre-load the next target bytecode into EBX 1554 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0)); 1555 1556 // compute return address as bci in rax, 1557 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset()))); 1558 __ subptr(rax, Address(rcx, methodOopDesc::const_offset())); 1559 // Adjust the bcp in RSI by the displacement in EDX 1560 __ addptr(rsi, rdx); 1561 // Push return address 1562 __ push_i(rax); 1563 // jsr returns vtos 1564 __ dispatch_only_noverify(vtos); 1565 return; 1566 } 1567 1568 // Normal (non-jsr) branch handling 1569 1570 // Adjust the bcp in RSI by the displacement in EDX 1571 __ addptr(rsi, rdx); 1572 1573 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters"); 1574 Label backedge_counter_overflow; 1575 Label profile_method; 1576 Label dispatch; 1577 if (UseLoopCounter) { 1578 // increment backedge counter for backward branches 1579 // rax,: MDO 1580 // rbx,: MDO bumped taken-count 1581 // rcx: method 1582 // rdx: target offset 1583 // rsi: target bcp 1584 // rdi: locals pointer 1585 __ testl(rdx, rdx); // check if forward or backward branch 1586 __ jcc(Assembler::positive, dispatch); // count only if backward branch 1587 1588 if (TieredCompilation) { 1589 Label no_mdo; 1590 int increment = InvocationCounter::count_increment; 1591 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 1592 if (ProfileInterpreter) { 1593 // Are we profiling? 1594 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset()))); 1595 __ testptr(rbx, rbx); 1596 __ jccb(Assembler::zero, no_mdo); 1597 // Increment the MDO backedge counter 1598 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) + 1599 in_bytes(InvocationCounter::counter_offset())); 1600 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 1601 rax, false, Assembler::zero, &backedge_counter_overflow); 1602 __ jmp(dispatch); 1603 } 1604 __ bind(no_mdo); 1605 // Increment backedge counter in methodOop 1606 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask, 1607 rax, false, Assembler::zero, &backedge_counter_overflow); 1608 } else { 1609 // increment counter 1610 __ movl(rax, Address(rcx, be_offset)); // load backedge counter 1611 __ incrementl(rax, InvocationCounter::count_increment); // increment counter 1612 __ movl(Address(rcx, be_offset), rax); // store counter 1613 1614 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter 1615 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits 1616 __ addl(rax, Address(rcx, be_offset)); // add both counters 1617 1618 if (ProfileInterpreter) { 1619 // Test to see if we should create a method data oop 1620 __ cmp32(rax, 1621 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit)); 1622 __ jcc(Assembler::less, dispatch); 1623 1624 // if no method data exists, go to profile method 1625 __ test_method_data_pointer(rax, profile_method); 1626 1627 if (UseOnStackReplacement) { 1628 // check for overflow against rbx, which is the MDO taken count 1629 __ cmp32(rbx, 1630 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1631 __ jcc(Assembler::below, dispatch); 1632 1633 // When ProfileInterpreter is on, the backedge_count comes from the 1634 // methodDataOop, which value does not get reset on the call to 1635 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1636 // routine while the method is being compiled, add a second test to make 1637 // sure the overflow function is called only once every overflow_frequency. 1638 const int overflow_frequency = 1024; 1639 __ andptr(rbx, overflow_frequency-1); 1640 __ jcc(Assembler::zero, backedge_counter_overflow); 1641 } 1642 } else { 1643 if (UseOnStackReplacement) { 1644 // check for overflow against rax, which is the sum of the counters 1645 __ cmp32(rax, 1646 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1647 __ jcc(Assembler::aboveEqual, backedge_counter_overflow); 1648 1649 } 1650 } 1651 } 1652 __ bind(dispatch); 1653 } 1654 1655 // Pre-load the next target bytecode into EBX 1656 __ load_unsigned_byte(rbx, Address(rsi, 0)); 1657 1658 // continue with the bytecode @ target 1659 // rax,: return bci for jsr's, unused otherwise 1660 // rbx,: target bytecode 1661 // rsi: target bcp 1662 __ dispatch_only(vtos); 1663 1664 if (UseLoopCounter) { 1665 if (ProfileInterpreter) { 1666 // Out-of-line code to allocate method data oop. 1667 __ bind(profile_method); 1668 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1669 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1670 __ set_method_data_pointer_for_bcp(); 1671 __ jmp(dispatch); 1672 } 1673 1674 if (UseOnStackReplacement) { 1675 1676 // invocation counter overflow 1677 __ bind(backedge_counter_overflow); 1678 __ negptr(rdx); 1679 __ addptr(rdx, rsi); // branch bcp 1680 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx); 1681 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1682 1683 // rax,: osr nmethod (osr ok) or NULL (osr not possible) 1684 // rbx,: target bytecode 1685 // rdx: scratch 1686 // rdi: locals pointer 1687 // rsi: bcp 1688 __ testptr(rax, rax); // test result 1689 __ jcc(Assembler::zero, dispatch); // no osr if null 1690 // nmethod may have been invalidated (VM may block upon call_VM return) 1691 __ movl(rcx, Address(rax, nmethod::entry_bci_offset())); 1692 __ cmpl(rcx, InvalidOSREntryBci); 1693 __ jcc(Assembler::equal, dispatch); 1694 1695 // We have the address of an on stack replacement routine in rax, 1696 // We need to prepare to execute the OSR method. First we must 1697 // migrate the locals and monitors off of the stack. 1698 1699 __ mov(rbx, rax); // save the nmethod 1700 1701 const Register thread = rcx; 1702 __ get_thread(thread); 1703 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 1704 // rax, is OSR buffer, move it to expected parameter location 1705 __ mov(rcx, rax); 1706 1707 // pop the interpreter frame 1708 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 1709 __ leave(); // remove frame anchor 1710 __ pop(rdi); // get return address 1711 __ mov(rsp, rdx); // set sp to sender sp 1712 1713 // Align stack pointer for compiled code (note that caller is 1714 // responsible for undoing this fixup by remembering the old SP 1715 // in an rbp,-relative location) 1716 __ andptr(rsp, -(StackAlignmentInBytes)); 1717 1718 // push the (possibly adjusted) return address 1719 __ push(rdi); 1720 1721 // and begin the OSR nmethod 1722 __ jmp(Address(rbx, nmethod::osr_entry_point_offset())); 1723 } 1724 } 1725 } 1726 1727 1728 void TemplateTable::if_0cmp(Condition cc) { 1729 transition(itos, vtos); 1730 // assume branch is more often taken than not (loops use backward branches) 1731 Label not_taken; 1732 __ testl(rax, rax); 1733 __ jcc(j_not(cc), not_taken); 1734 branch(false, false); 1735 __ bind(not_taken); 1736 __ profile_not_taken_branch(rax); 1737 } 1738 1739 1740 void TemplateTable::if_icmp(Condition cc) { 1741 transition(itos, vtos); 1742 // assume branch is more often taken than not (loops use backward branches) 1743 Label not_taken; 1744 __ pop_i(rdx); 1745 __ cmpl(rdx, rax); 1746 __ jcc(j_not(cc), not_taken); 1747 branch(false, false); 1748 __ bind(not_taken); 1749 __ profile_not_taken_branch(rax); 1750 } 1751 1752 1753 void TemplateTable::if_nullcmp(Condition cc) { 1754 transition(atos, vtos); 1755 // assume branch is more often taken than not (loops use backward branches) 1756 Label not_taken; 1757 __ testptr(rax, rax); 1758 __ jcc(j_not(cc), not_taken); 1759 branch(false, false); 1760 __ bind(not_taken); 1761 __ profile_not_taken_branch(rax); 1762 } 1763 1764 1765 void TemplateTable::if_acmp(Condition cc) { 1766 transition(atos, vtos); 1767 // assume branch is more often taken than not (loops use backward branches) 1768 Label not_taken; 1769 __ pop_ptr(rdx); 1770 __ cmpptr(rdx, rax); 1771 __ jcc(j_not(cc), not_taken); 1772 branch(false, false); 1773 __ bind(not_taken); 1774 __ profile_not_taken_branch(rax); 1775 } 1776 1777 1778 void TemplateTable::ret() { 1779 transition(vtos, vtos); 1780 locals_index(rbx); 1781 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1782 __ profile_ret(rbx, rcx); 1783 __ get_method(rax); 1784 __ movptr(rsi, Address(rax, methodOopDesc::const_offset())); 1785 __ lea(rsi, Address(rsi, rbx, Address::times_1, 1786 constMethodOopDesc::codes_offset())); 1787 __ dispatch_next(vtos); 1788 } 1789 1790 1791 void TemplateTable::wide_ret() { 1792 transition(vtos, vtos); 1793 locals_index_wide(rbx); 1794 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1795 __ profile_ret(rbx, rcx); 1796 __ get_method(rax); 1797 __ movptr(rsi, Address(rax, methodOopDesc::const_offset())); 1798 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset())); 1799 __ dispatch_next(vtos); 1800 } 1801 1802 1803 void TemplateTable::tableswitch() { 1804 Label default_case, continue_execution; 1805 transition(itos, vtos); 1806 // align rsi 1807 __ lea(rbx, at_bcp(wordSize)); 1808 __ andptr(rbx, -wordSize); 1809 // load lo & hi 1810 __ movl(rcx, Address(rbx, 1 * wordSize)); 1811 __ movl(rdx, Address(rbx, 2 * wordSize)); 1812 __ bswapl(rcx); 1813 __ bswapl(rdx); 1814 // check against lo & hi 1815 __ cmpl(rax, rcx); 1816 __ jccb(Assembler::less, default_case); 1817 __ cmpl(rax, rdx); 1818 __ jccb(Assembler::greater, default_case); 1819 // lookup dispatch offset 1820 __ subl(rax, rcx); 1821 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt)); 1822 __ profile_switch_case(rax, rbx, rcx); 1823 // continue execution 1824 __ bind(continue_execution); 1825 __ bswapl(rdx); 1826 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1827 __ addptr(rsi, rdx); 1828 __ dispatch_only(vtos); 1829 // handle default 1830 __ bind(default_case); 1831 __ profile_switch_default(rax); 1832 __ movl(rdx, Address(rbx, 0)); 1833 __ jmp(continue_execution); 1834 } 1835 1836 1837 void TemplateTable::lookupswitch() { 1838 transition(itos, itos); 1839 __ stop("lookupswitch bytecode should have been rewritten"); 1840 } 1841 1842 1843 void TemplateTable::fast_linearswitch() { 1844 transition(itos, vtos); 1845 Label loop_entry, loop, found, continue_execution; 1846 // bswapl rax, so we can avoid bswapping the table entries 1847 __ bswapl(rax); 1848 // align rsi 1849 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1850 __ andptr(rbx, -wordSize); 1851 // set counter 1852 __ movl(rcx, Address(rbx, wordSize)); 1853 __ bswapl(rcx); 1854 __ jmpb(loop_entry); 1855 // table search 1856 __ bind(loop); 1857 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize)); 1858 __ jccb(Assembler::equal, found); 1859 __ bind(loop_entry); 1860 __ decrementl(rcx); 1861 __ jcc(Assembler::greaterEqual, loop); 1862 // default case 1863 __ profile_switch_default(rax); 1864 __ movl(rdx, Address(rbx, 0)); 1865 __ jmpb(continue_execution); 1866 // entry found -> get offset 1867 __ bind(found); 1868 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize)); 1869 __ profile_switch_case(rcx, rax, rbx); 1870 // continue execution 1871 __ bind(continue_execution); 1872 __ bswapl(rdx); 1873 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1874 __ addptr(rsi, rdx); 1875 __ dispatch_only(vtos); 1876 } 1877 1878 1879 void TemplateTable::fast_binaryswitch() { 1880 transition(itos, vtos); 1881 // Implementation using the following core algorithm: 1882 // 1883 // int binary_search(int key, LookupswitchPair* array, int n) { 1884 // // Binary search according to "Methodik des Programmierens" by 1885 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 1886 // int i = 0; 1887 // int j = n; 1888 // while (i+1 < j) { 1889 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 1890 // // with Q: for all i: 0 <= i < n: key < a[i] 1891 // // where a stands for the array and assuming that the (inexisting) 1892 // // element a[n] is infinitely big. 1893 // int h = (i + j) >> 1; 1894 // // i < h < j 1895 // if (key < array[h].fast_match()) { 1896 // j = h; 1897 // } else { 1898 // i = h; 1899 // } 1900 // } 1901 // // R: a[i] <= key < a[i+1] or Q 1902 // // (i.e., if key is within array, i is the correct index) 1903 // return i; 1904 // } 1905 1906 // register allocation 1907 const Register key = rax; // already set (tosca) 1908 const Register array = rbx; 1909 const Register i = rcx; 1910 const Register j = rdx; 1911 const Register h = rdi; // needs to be restored 1912 const Register temp = rsi; 1913 // setup array 1914 __ save_bcp(); 1915 1916 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1917 __ andptr(array, -wordSize); 1918 // initialize i & j 1919 __ xorl(i, i); // i = 0; 1920 __ movl(j, Address(array, -wordSize)); // j = length(array); 1921 // Convert j into native byteordering 1922 __ bswapl(j); 1923 // and start 1924 Label entry; 1925 __ jmp(entry); 1926 1927 // binary search loop 1928 { Label loop; 1929 __ bind(loop); 1930 // int h = (i + j) >> 1; 1931 __ leal(h, Address(i, j, Address::times_1)); // h = i + j; 1932 __ sarl(h, 1); // h = (i + j) >> 1; 1933 // if (key < array[h].fast_match()) { 1934 // j = h; 1935 // } else { 1936 // i = h; 1937 // } 1938 // Convert array[h].match to native byte-ordering before compare 1939 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize)); 1940 __ bswapl(temp); 1941 __ cmpl(key, temp); 1942 if (VM_Version::supports_cmov()) { 1943 __ cmovl(Assembler::less , j, h); // j = h if (key < array[h].fast_match()) 1944 __ cmovl(Assembler::greaterEqual, i, h); // i = h if (key >= array[h].fast_match()) 1945 } else { 1946 Label set_i, end_of_if; 1947 __ jccb(Assembler::greaterEqual, set_i); // { 1948 __ mov(j, h); // j = h; 1949 __ jmp(end_of_if); // } 1950 __ bind(set_i); // else { 1951 __ mov(i, h); // i = h; 1952 __ bind(end_of_if); // } 1953 } 1954 // while (i+1 < j) 1955 __ bind(entry); 1956 __ leal(h, Address(i, 1)); // i+1 1957 __ cmpl(h, j); // i+1 < j 1958 __ jcc(Assembler::less, loop); 1959 } 1960 1961 // end of binary search, result index is i (must check again!) 1962 Label default_case; 1963 // Convert array[i].match to native byte-ordering before compare 1964 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize)); 1965 __ bswapl(temp); 1966 __ cmpl(key, temp); 1967 __ jcc(Assembler::notEqual, default_case); 1968 1969 // entry found -> j = offset 1970 __ movl(j , Address(array, i, Address::times_8, 1*wordSize)); 1971 __ profile_switch_case(i, key, array); 1972 __ bswapl(j); 1973 LP64_ONLY(__ movslq(j, j)); 1974 __ restore_bcp(); 1975 __ restore_locals(); // restore rdi 1976 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 1977 1978 __ addptr(rsi, j); 1979 __ dispatch_only(vtos); 1980 1981 // default case -> j = default offset 1982 __ bind(default_case); 1983 __ profile_switch_default(i); 1984 __ movl(j, Address(array, -2*wordSize)); 1985 __ bswapl(j); 1986 LP64_ONLY(__ movslq(j, j)); 1987 __ restore_bcp(); 1988 __ restore_locals(); // restore rdi 1989 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 1990 __ addptr(rsi, j); 1991 __ dispatch_only(vtos); 1992 } 1993 1994 1995 void TemplateTable::_return(TosState state) { 1996 transition(state, state); 1997 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation 1998 1999 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 2000 assert(state == vtos, "only valid state"); 2001 __ movptr(rax, aaddress(0)); 2002 __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes())); 2003 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc))); 2004 __ testl(rdi, JVM_ACC_HAS_FINALIZER); 2005 Label skip_register_finalizer; 2006 __ jcc(Assembler::zero, skip_register_finalizer); 2007 2008 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax); 2009 2010 __ bind(skip_register_finalizer); 2011 } 2012 2013 __ remove_activation(state, rsi); 2014 __ jmp(rsi); 2015 } 2016 2017 2018 // ---------------------------------------------------------------------------- 2019 // Volatile variables demand their effects be made known to all CPU's in 2020 // order. Store buffers on most chips allow reads & writes to reorder; the 2021 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of 2022 // memory barrier (i.e., it's not sufficient that the interpreter does not 2023 // reorder volatile references, the hardware also must not reorder them). 2024 // 2025 // According to the new Java Memory Model (JMM): 2026 // (1) All volatiles are serialized wrt to each other. 2027 // ALSO reads & writes act as aquire & release, so: 2028 // (2) A read cannot let unrelated NON-volatile memory refs that happen after 2029 // the read float up to before the read. It's OK for non-volatile memory refs 2030 // that happen before the volatile read to float down below it. 2031 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs 2032 // that happen BEFORE the write float down to after the write. It's OK for 2033 // non-volatile memory refs that happen after the volatile write to float up 2034 // before it. 2035 // 2036 // We only put in barriers around volatile refs (they are expensive), not 2037 // _between_ memory refs (that would require us to track the flavor of the 2038 // previous memory refs). Requirements (2) and (3) require some barriers 2039 // before volatile stores and after volatile loads. These nearly cover 2040 // requirement (1) but miss the volatile-store-volatile-load case. This final 2041 // case is placed after volatile-stores although it could just as well go 2042 // before volatile-loads. 2043 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) { 2044 // Helper function to insert a is-volatile test and memory barrier 2045 if( !os::is_MP() ) return; // Not needed on single CPU 2046 __ membar(order_constraint); 2047 } 2048 2049 void TemplateTable::resolve_cache_and_index(int byte_no, 2050 Register result, 2051 Register Rcache, 2052 Register index, 2053 size_t index_size) { 2054 Register temp = rbx; 2055 2056 assert_different_registers(result, Rcache, index, temp); 2057 2058 Label resolved; 2059 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2060 if (byte_no == f1_oop) { 2061 // We are resolved if the f1 field contains a non-null object (CallSite, etc.) 2062 // This kind of CP cache entry does not need to match the flags byte, because 2063 // there is a 1-1 relation between bytecode type and CP entry type. 2064 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD) 2065 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset())); 2066 __ testptr(result, result); 2067 __ jcc(Assembler::notEqual, resolved); 2068 } else { 2069 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 2070 assert(result == noreg, ""); //else change code for setting result 2071 const int shift_count = (1 + byte_no)*BitsPerByte; 2072 __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); 2073 __ shrl(temp, shift_count); 2074 // have we resolved this bytecode? 2075 __ andl(temp, 0xFF); 2076 __ cmpl(temp, (int)bytecode()); 2077 __ jcc(Assembler::equal, resolved); 2078 } 2079 2080 // resolve first time through 2081 address entry; 2082 switch (bytecode()) { 2083 case Bytecodes::_getstatic : // fall through 2084 case Bytecodes::_putstatic : // fall through 2085 case Bytecodes::_getfield : // fall through 2086 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break; 2087 case Bytecodes::_invokevirtual : // fall through 2088 case Bytecodes::_invokespecial : // fall through 2089 case Bytecodes::_invokestatic : // fall through 2090 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break; 2091 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break; 2092 case Bytecodes::_fast_aldc : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break; 2093 case Bytecodes::_fast_aldc_w : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break; 2094 default : ShouldNotReachHere(); break; 2095 } 2096 __ movl(temp, (int)bytecode()); 2097 __ call_VM(noreg, entry, temp); 2098 // Update registers with resolved info 2099 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2100 if (result != noreg) 2101 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset())); 2102 __ bind(resolved); 2103 } 2104 2105 2106 // The cache and index registers must be set before call 2107 void TemplateTable::load_field_cp_cache_entry(Register obj, 2108 Register cache, 2109 Register index, 2110 Register off, 2111 Register flags, 2112 bool is_static = false) { 2113 assert_different_registers(cache, index, flags, off); 2114 2115 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset(); 2116 // Field offset 2117 __ movptr(off, Address(cache, index, Address::times_ptr, 2118 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()))); 2119 // Flags 2120 __ movl(flags, Address(cache, index, Address::times_ptr, 2121 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()))); 2122 2123 // klass overwrite register 2124 if (is_static) { 2125 __ movptr(obj, Address(cache, index, Address::times_ptr, 2126 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset()))); 2127 } 2128 } 2129 2130 void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 2131 Register method, 2132 Register itable_index, 2133 Register flags, 2134 bool is_invokevirtual, 2135 bool is_invokevfinal /*unused*/, 2136 bool is_invokedynamic) { 2137 // setup registers 2138 const Register cache = rcx; 2139 const Register index = rdx; 2140 assert_different_registers(method, flags); 2141 assert_different_registers(method, cache, index); 2142 assert_different_registers(itable_index, flags); 2143 assert_different_registers(itable_index, cache, index); 2144 // determine constant pool cache field offsets 2145 const int method_offset = in_bytes( 2146 constantPoolCacheOopDesc::base_offset() + 2147 (is_invokevirtual 2148 ? ConstantPoolCacheEntry::f2_offset() 2149 : ConstantPoolCacheEntry::f1_offset() 2150 ) 2151 ); 2152 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() + 2153 ConstantPoolCacheEntry::flags_offset()); 2154 // access constant pool cache fields 2155 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() + 2156 ConstantPoolCacheEntry::f2_offset()); 2157 2158 if (byte_no == f1_oop) { 2159 // Resolved f1_oop goes directly into 'method' register. 2160 assert(is_invokedynamic, ""); 2161 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4)); 2162 } else { 2163 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2)); 2164 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset)); 2165 } 2166 if (itable_index != noreg) { 2167 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset)); 2168 } 2169 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset)); 2170 } 2171 2172 2173 // The registers cache and index expected to be set before call. 2174 // Correct values of the cache and index registers are preserved. 2175 void TemplateTable::jvmti_post_field_access(Register cache, 2176 Register index, 2177 bool is_static, 2178 bool has_tos) { 2179 if (JvmtiExport::can_post_field_access()) { 2180 // Check to see if a field access watch has been set before we take 2181 // the time to call into the VM. 2182 Label L1; 2183 assert_different_registers(cache, index, rax); 2184 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2185 __ testl(rax,rax); 2186 __ jcc(Assembler::zero, L1); 2187 2188 // cache entry pointer 2189 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset())); 2190 __ shll(index, LogBytesPerWord); 2191 __ addptr(cache, index); 2192 if (is_static) { 2193 __ xorptr(rax, rax); // NULL object reference 2194 } else { 2195 __ pop(atos); // Get the object 2196 __ verify_oop(rax); 2197 __ push(atos); // Restore stack state 2198 } 2199 // rax,: object pointer or NULL 2200 // cache: cache entry pointer 2201 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), 2202 rax, cache); 2203 __ get_cache_and_index_at_bcp(cache, index, 1); 2204 __ bind(L1); 2205 } 2206 } 2207 2208 void TemplateTable::pop_and_check_object(Register r) { 2209 __ pop_ptr(r); 2210 __ null_check(r); // for field access must check obj. 2211 __ verify_oop(r); 2212 } 2213 2214 void TemplateTable::getfield_or_static(int byte_no, bool is_static) { 2215 transition(vtos, vtos); 2216 2217 const Register cache = rcx; 2218 const Register index = rdx; 2219 const Register obj = rcx; 2220 const Register off = rbx; 2221 const Register flags = rax; 2222 2223 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2)); 2224 jvmti_post_field_access(cache, index, is_static, false); 2225 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2226 2227 if (!is_static) pop_and_check_object(obj); 2228 2229 const Address lo(obj, off, Address::times_1, 0*wordSize); 2230 const Address hi(obj, off, Address::times_1, 1*wordSize); 2231 2232 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2233 2234 __ shrl(flags, ConstantPoolCacheEntry::tosBits); 2235 assert(btos == 0, "change code, btos != 0"); 2236 // btos 2237 __ andptr(flags, 0x0f); 2238 __ jcc(Assembler::notZero, notByte); 2239 2240 __ load_signed_byte(rax, lo ); 2241 __ push(btos); 2242 // Rewrite bytecode to be faster 2243 if (!is_static) { 2244 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx); 2245 } 2246 __ jmp(Done); 2247 2248 __ bind(notByte); 2249 // itos 2250 __ cmpl(flags, itos ); 2251 __ jcc(Assembler::notEqual, notInt); 2252 2253 __ movl(rax, lo ); 2254 __ push(itos); 2255 // Rewrite bytecode to be faster 2256 if (!is_static) { 2257 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx); 2258 } 2259 __ jmp(Done); 2260 2261 __ bind(notInt); 2262 // atos 2263 __ cmpl(flags, atos ); 2264 __ jcc(Assembler::notEqual, notObj); 2265 2266 __ movl(rax, lo ); 2267 __ push(atos); 2268 if (!is_static) { 2269 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx); 2270 } 2271 __ jmp(Done); 2272 2273 __ bind(notObj); 2274 // ctos 2275 __ cmpl(flags, ctos ); 2276 __ jcc(Assembler::notEqual, notChar); 2277 2278 __ load_unsigned_short(rax, lo ); 2279 __ push(ctos); 2280 if (!is_static) { 2281 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx); 2282 } 2283 __ jmp(Done); 2284 2285 __ bind(notChar); 2286 // stos 2287 __ cmpl(flags, stos ); 2288 __ jcc(Assembler::notEqual, notShort); 2289 2290 __ load_signed_short(rax, lo ); 2291 __ push(stos); 2292 if (!is_static) { 2293 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx); 2294 } 2295 __ jmp(Done); 2296 2297 __ bind(notShort); 2298 // ltos 2299 __ cmpl(flags, ltos ); 2300 __ jcc(Assembler::notEqual, notLong); 2301 2302 // Generate code as if volatile. There just aren't enough registers to 2303 // save that information and this code is faster than the test. 2304 __ fild_d(lo); // Must load atomically 2305 __ subptr(rsp,2*wordSize); // Make space for store 2306 __ fistp_d(Address(rsp,0)); 2307 __ pop(rax); 2308 __ pop(rdx); 2309 2310 __ push(ltos); 2311 // Don't rewrite to _fast_lgetfield for potential volatile case. 2312 __ jmp(Done); 2313 2314 __ bind(notLong); 2315 // ftos 2316 __ cmpl(flags, ftos ); 2317 __ jcc(Assembler::notEqual, notFloat); 2318 2319 __ fld_s(lo); 2320 __ push(ftos); 2321 if (!is_static) { 2322 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx); 2323 } 2324 __ jmp(Done); 2325 2326 __ bind(notFloat); 2327 // dtos 2328 __ cmpl(flags, dtos ); 2329 __ jcc(Assembler::notEqual, notDouble); 2330 2331 __ fld_d(lo); 2332 __ push(dtos); 2333 if (!is_static) { 2334 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx); 2335 } 2336 __ jmpb(Done); 2337 2338 __ bind(notDouble); 2339 2340 __ stop("Bad state"); 2341 2342 __ bind(Done); 2343 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2344 // volatile_barrier( ); 2345 } 2346 2347 2348 void TemplateTable::getfield(int byte_no) { 2349 getfield_or_static(byte_no, false); 2350 } 2351 2352 2353 void TemplateTable::getstatic(int byte_no) { 2354 getfield_or_static(byte_no, true); 2355 } 2356 2357 // The registers cache and index expected to be set before call. 2358 // The function may destroy various registers, just not the cache and index registers. 2359 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 2360 2361 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset(); 2362 2363 if (JvmtiExport::can_post_field_modification()) { 2364 // Check to see if a field modification watch has been set before we take 2365 // the time to call into the VM. 2366 Label L1; 2367 assert_different_registers(cache, index, rax); 2368 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2369 __ testl(rax, rax); 2370 __ jcc(Assembler::zero, L1); 2371 2372 // The cache and index registers have been already set. 2373 // This allows to eliminate this call but the cache and index 2374 // registers have to be correspondingly used after this line. 2375 __ get_cache_and_index_at_bcp(rax, rdx, 1); 2376 2377 if (is_static) { 2378 // Life is simple. Null out the object pointer. 2379 __ xorptr(rbx, rbx); 2380 } else { 2381 // Life is harder. The stack holds the value on top, followed by the object. 2382 // We don't know the size of the value, though; it could be one or two words 2383 // depending on its type. As a result, we must find the type to determine where 2384 // the object is. 2385 Label two_word, valsize_known; 2386 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset + 2387 ConstantPoolCacheEntry::flags_offset()))); 2388 __ mov(rbx, rsp); 2389 __ shrl(rcx, ConstantPoolCacheEntry::tosBits); 2390 // Make sure we don't need to mask rcx for tosBits after the above shift 2391 ConstantPoolCacheEntry::verify_tosBits(); 2392 __ cmpl(rcx, ltos); 2393 __ jccb(Assembler::equal, two_word); 2394 __ cmpl(rcx, dtos); 2395 __ jccb(Assembler::equal, two_word); 2396 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos) 2397 __ jmpb(valsize_known); 2398 2399 __ bind(two_word); 2400 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue 2401 2402 __ bind(valsize_known); 2403 // setup object pointer 2404 __ movptr(rbx, Address(rbx, 0)); 2405 } 2406 // cache entry pointer 2407 __ addptr(rax, in_bytes(cp_base_offset)); 2408 __ shll(rdx, LogBytesPerWord); 2409 __ addptr(rax, rdx); 2410 // object (tos) 2411 __ mov(rcx, rsp); 2412 // rbx,: object pointer set up above (NULL if static) 2413 // rax,: cache entry pointer 2414 // rcx: jvalue object on the stack 2415 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), 2416 rbx, rax, rcx); 2417 __ get_cache_and_index_at_bcp(cache, index, 1); 2418 __ bind(L1); 2419 } 2420 } 2421 2422 2423 void TemplateTable::putfield_or_static(int byte_no, bool is_static) { 2424 transition(vtos, vtos); 2425 2426 const Register cache = rcx; 2427 const Register index = rdx; 2428 const Register obj = rcx; 2429 const Register off = rbx; 2430 const Register flags = rax; 2431 2432 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2)); 2433 jvmti_post_field_mod(cache, index, is_static); 2434 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2435 2436 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2437 // volatile_barrier( ); 2438 2439 Label notVolatile, Done; 2440 __ movl(rdx, flags); 2441 __ shrl(rdx, ConstantPoolCacheEntry::volatileField); 2442 __ andl(rdx, 0x1); 2443 2444 // field addresses 2445 const Address lo(obj, off, Address::times_1, 0*wordSize); 2446 const Address hi(obj, off, Address::times_1, 1*wordSize); 2447 2448 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2449 2450 __ shrl(flags, ConstantPoolCacheEntry::tosBits); 2451 assert(btos == 0, "change code, btos != 0"); 2452 // btos 2453 __ andl(flags, 0x0f); 2454 __ jcc(Assembler::notZero, notByte); 2455 2456 __ pop(btos); 2457 if (!is_static) pop_and_check_object(obj); 2458 __ movb(lo, rax ); 2459 if (!is_static) { 2460 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx); 2461 } 2462 __ jmp(Done); 2463 2464 __ bind(notByte); 2465 // itos 2466 __ cmpl(flags, itos ); 2467 __ jcc(Assembler::notEqual, notInt); 2468 2469 __ pop(itos); 2470 if (!is_static) pop_and_check_object(obj); 2471 2472 __ movl(lo, rax ); 2473 if (!is_static) { 2474 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx); 2475 } 2476 __ jmp(Done); 2477 2478 __ bind(notInt); 2479 // atos 2480 __ cmpl(flags, atos ); 2481 __ jcc(Assembler::notEqual, notObj); 2482 2483 __ pop(atos); 2484 if (!is_static) pop_and_check_object(obj); 2485 2486 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2487 2488 if (!is_static) { 2489 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx); 2490 } 2491 2492 __ jmp(Done); 2493 2494 __ bind(notObj); 2495 // ctos 2496 __ cmpl(flags, ctos ); 2497 __ jcc(Assembler::notEqual, notChar); 2498 2499 __ pop(ctos); 2500 if (!is_static) pop_and_check_object(obj); 2501 __ movw(lo, rax ); 2502 if (!is_static) { 2503 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx); 2504 } 2505 __ jmp(Done); 2506 2507 __ bind(notChar); 2508 // stos 2509 __ cmpl(flags, stos ); 2510 __ jcc(Assembler::notEqual, notShort); 2511 2512 __ pop(stos); 2513 if (!is_static) pop_and_check_object(obj); 2514 __ movw(lo, rax ); 2515 if (!is_static) { 2516 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx); 2517 } 2518 __ jmp(Done); 2519 2520 __ bind(notShort); 2521 // ltos 2522 __ cmpl(flags, ltos ); 2523 __ jcc(Assembler::notEqual, notLong); 2524 2525 Label notVolatileLong; 2526 __ testl(rdx, rdx); 2527 __ jcc(Assembler::zero, notVolatileLong); 2528 2529 __ pop(ltos); // overwrites rdx, do this after testing volatile. 2530 if (!is_static) pop_and_check_object(obj); 2531 2532 // Replace with real volatile test 2533 __ push(rdx); 2534 __ push(rax); // Must update atomically with FIST 2535 __ fild_d(Address(rsp,0)); // So load into FPU register 2536 __ fistp_d(lo); // and put into memory atomically 2537 __ addptr(rsp, 2*wordSize); 2538 // volatile_barrier(); 2539 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2540 Assembler::StoreStore)); 2541 // Don't rewrite volatile version 2542 __ jmp(notVolatile); 2543 2544 __ bind(notVolatileLong); 2545 2546 __ pop(ltos); // overwrites rdx 2547 if (!is_static) pop_and_check_object(obj); 2548 NOT_LP64(__ movptr(hi, rdx)); 2549 __ movptr(lo, rax); 2550 if (!is_static) { 2551 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx); 2552 } 2553 __ jmp(notVolatile); 2554 2555 __ bind(notLong); 2556 // ftos 2557 __ cmpl(flags, ftos ); 2558 __ jcc(Assembler::notEqual, notFloat); 2559 2560 __ pop(ftos); 2561 if (!is_static) pop_and_check_object(obj); 2562 __ fstp_s(lo); 2563 if (!is_static) { 2564 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx); 2565 } 2566 __ jmp(Done); 2567 2568 __ bind(notFloat); 2569 // dtos 2570 __ cmpl(flags, dtos ); 2571 __ jcc(Assembler::notEqual, notDouble); 2572 2573 __ pop(dtos); 2574 if (!is_static) pop_and_check_object(obj); 2575 __ fstp_d(lo); 2576 if (!is_static) { 2577 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx); 2578 } 2579 __ jmp(Done); 2580 2581 __ bind(notDouble); 2582 2583 __ stop("Bad state"); 2584 2585 __ bind(Done); 2586 2587 // Check for volatile store 2588 __ testl(rdx, rdx); 2589 __ jcc(Assembler::zero, notVolatile); 2590 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2591 Assembler::StoreStore)); 2592 __ bind(notVolatile); 2593 } 2594 2595 2596 void TemplateTable::putfield(int byte_no) { 2597 putfield_or_static(byte_no, false); 2598 } 2599 2600 2601 void TemplateTable::putstatic(int byte_no) { 2602 putfield_or_static(byte_no, true); 2603 } 2604 2605 void TemplateTable::jvmti_post_fast_field_mod() { 2606 if (JvmtiExport::can_post_field_modification()) { 2607 // Check to see if a field modification watch has been set before we take 2608 // the time to call into the VM. 2609 Label L2; 2610 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2611 __ testl(rcx,rcx); 2612 __ jcc(Assembler::zero, L2); 2613 __ pop_ptr(rbx); // copy the object pointer from tos 2614 __ verify_oop(rbx); 2615 __ push_ptr(rbx); // put the object pointer back on tos 2616 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object 2617 __ mov(rcx, rsp); 2618 __ push_ptr(rbx); // save object pointer so we can steal rbx, 2619 __ xorptr(rbx, rbx); 2620 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize); 2621 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize); 2622 switch (bytecode()) { // load values into the jvalue object 2623 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break; 2624 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break; 2625 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break; 2626 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break; 2627 case Bytecodes::_fast_lputfield: 2628 NOT_LP64(__ movptr(hi_value, rdx)); 2629 __ movptr(lo_value, rax); 2630 break; 2631 2632 // need to call fld_s() after fstp_s() to restore the value for below 2633 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break; 2634 2635 // need to call fld_d() after fstp_d() to restore the value for below 2636 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break; 2637 2638 // since rcx is not an object we don't call store_check() here 2639 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break; 2640 2641 default: ShouldNotReachHere(); 2642 } 2643 __ pop_ptr(rbx); // restore copy of object pointer 2644 2645 // Save rax, and sometimes rdx because call_VM() will clobber them, 2646 // then use them for JVM/DI purposes 2647 __ push(rax); 2648 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx); 2649 // access constant pool cache entry 2650 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1); 2651 __ verify_oop(rbx); 2652 // rbx,: object pointer copied above 2653 // rax,: cache entry pointer 2654 // rcx: jvalue object on the stack 2655 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx); 2656 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value 2657 __ pop(rax); // restore lower value 2658 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space 2659 __ bind(L2); 2660 } 2661 } 2662 2663 void TemplateTable::fast_storefield(TosState state) { 2664 transition(state, vtos); 2665 2666 ByteSize base = constantPoolCacheOopDesc::base_offset(); 2667 2668 jvmti_post_fast_field_mod(); 2669 2670 // access constant pool cache 2671 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 2672 2673 // test for volatile with rdx but rdx is tos register for lputfield. 2674 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx); 2675 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + 2676 ConstantPoolCacheEntry::flags_offset()))); 2677 2678 // replace index with field offset from cache entry 2679 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 2680 2681 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2682 // volatile_barrier( ); 2683 2684 Label notVolatile, Done; 2685 __ shrl(rdx, ConstantPoolCacheEntry::volatileField); 2686 __ andl(rdx, 0x1); 2687 // Check for volatile store 2688 __ testl(rdx, rdx); 2689 __ jcc(Assembler::zero, notVolatile); 2690 2691 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 2692 2693 // Get object from stack 2694 pop_and_check_object(rcx); 2695 2696 // field addresses 2697 const Address lo(rcx, rbx, Address::times_1, 0*wordSize); 2698 const Address hi(rcx, rbx, Address::times_1, 1*wordSize); 2699 2700 // access field 2701 switch (bytecode()) { 2702 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 2703 case Bytecodes::_fast_sputfield: // fall through 2704 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 2705 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 2706 case Bytecodes::_fast_lputfield: 2707 NOT_LP64(__ movptr(hi, rdx)); 2708 __ movptr(lo, rax); 2709 break; 2710 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 2711 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 2712 case Bytecodes::_fast_aputfield: { 2713 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2714 break; 2715 } 2716 default: 2717 ShouldNotReachHere(); 2718 } 2719 2720 Label done; 2721 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2722 Assembler::StoreStore)); 2723 // Barriers are so large that short branch doesn't reach! 2724 __ jmp(done); 2725 2726 // Same code as above, but don't need rdx to test for volatile. 2727 __ bind(notVolatile); 2728 2729 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 2730 2731 // Get object from stack 2732 pop_and_check_object(rcx); 2733 2734 // access field 2735 switch (bytecode()) { 2736 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 2737 case Bytecodes::_fast_sputfield: // fall through 2738 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 2739 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 2740 case Bytecodes::_fast_lputfield: 2741 NOT_LP64(__ movptr(hi, rdx)); 2742 __ movptr(lo, rax); 2743 break; 2744 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 2745 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 2746 case Bytecodes::_fast_aputfield: { 2747 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2748 break; 2749 } 2750 default: 2751 ShouldNotReachHere(); 2752 } 2753 __ bind(done); 2754 } 2755 2756 2757 void TemplateTable::fast_accessfield(TosState state) { 2758 transition(atos, state); 2759 2760 // do the JVMTI work here to avoid disturbing the register state below 2761 if (JvmtiExport::can_post_field_access()) { 2762 // Check to see if a field access watch has been set before we take 2763 // the time to call into the VM. 2764 Label L1; 2765 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2766 __ testl(rcx,rcx); 2767 __ jcc(Assembler::zero, L1); 2768 // access constant pool cache entry 2769 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1); 2770 __ push_ptr(rax); // save object pointer before call_VM() clobbers it 2771 __ verify_oop(rax); 2772 // rax,: object pointer copied above 2773 // rcx: cache entry pointer 2774 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx); 2775 __ pop_ptr(rax); // restore object pointer 2776 __ bind(L1); 2777 } 2778 2779 // access constant pool cache 2780 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 2781 // replace index with field offset from cache entry 2782 __ movptr(rbx, Address(rcx, 2783 rbx, 2784 Address::times_ptr, 2785 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 2786 2787 2788 // rax,: object 2789 __ verify_oop(rax); 2790 __ null_check(rax); 2791 // field addresses 2792 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 2793 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize); 2794 2795 // access field 2796 switch (bytecode()) { 2797 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break; 2798 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break; 2799 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break; 2800 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break; 2801 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break; 2802 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break; 2803 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break; 2804 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break; 2805 default: 2806 ShouldNotReachHere(); 2807 } 2808 2809 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO) 2810 // volatile_barrier( ); 2811 } 2812 2813 void TemplateTable::fast_xaccess(TosState state) { 2814 transition(vtos, state); 2815 // get receiver 2816 __ movptr(rax, aaddress(0)); 2817 // access constant pool cache 2818 __ get_cache_and_index_at_bcp(rcx, rdx, 2); 2819 __ movptr(rbx, Address(rcx, 2820 rdx, 2821 Address::times_ptr, 2822 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 2823 // make sure exception is reported in correct bcp range (getfield is next instruction) 2824 __ increment(rsi); 2825 __ null_check(rax); 2826 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 2827 if (state == itos) { 2828 __ movl(rax, lo); 2829 } else if (state == atos) { 2830 __ movptr(rax, lo); 2831 __ verify_oop(rax); 2832 } else if (state == ftos) { 2833 __ fld_s(lo); 2834 } else { 2835 ShouldNotReachHere(); 2836 } 2837 __ decrement(rsi); 2838 } 2839 2840 2841 2842 //---------------------------------------------------------------------------------------------------- 2843 // Calls 2844 2845 void TemplateTable::count_calls(Register method, Register temp) { 2846 // implemented elsewhere 2847 ShouldNotReachHere(); 2848 } 2849 2850 2851 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) { 2852 // determine flags 2853 Bytecodes::Code code = bytecode(); 2854 const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 2855 const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 2856 const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 2857 const bool is_invokespecial = code == Bytecodes::_invokespecial; 2858 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic); 2859 const bool receiver_null_check = is_invokespecial; 2860 const bool save_flags = is_invokeinterface || is_invokevirtual; 2861 // setup registers & access constant pool cache 2862 const Register recv = rcx; 2863 const Register flags = rdx; 2864 assert_different_registers(method, index, recv, flags); 2865 2866 // save 'interpreter return address' 2867 __ save_bcp(); 2868 2869 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 2870 2871 // load receiver if needed (note: no return address pushed yet) 2872 if (load_receiver) { 2873 assert(!is_invokedynamic, ""); 2874 __ movl(recv, flags); 2875 __ andl(recv, 0xFF); 2876 // recv count is 0 based? 2877 Address recv_addr(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1)); 2878 __ movptr(recv, recv_addr); 2879 __ verify_oop(recv); 2880 } 2881 2882 // do null check if needed 2883 if (receiver_null_check) { 2884 __ null_check(recv); 2885 } 2886 2887 if (save_flags) { 2888 __ mov(rsi, flags); 2889 } 2890 2891 // compute return type 2892 __ shrl(flags, ConstantPoolCacheEntry::tosBits); 2893 // Make sure we don't need to mask flags for tosBits after the above shift 2894 ConstantPoolCacheEntry::verify_tosBits(); 2895 // load return address 2896 { 2897 address table_addr; 2898 if (is_invokeinterface || is_invokedynamic) 2899 table_addr = (address)Interpreter::return_5_addrs_by_index_table(); 2900 else 2901 table_addr = (address)Interpreter::return_3_addrs_by_index_table(); 2902 ExternalAddress table(table_addr); 2903 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))); 2904 } 2905 2906 // push return address 2907 __ push(flags); 2908 2909 // Restore flag value from the constant pool cache, and restore rsi 2910 // for later null checks. rsi is the bytecode pointer 2911 if (save_flags) { 2912 __ mov(flags, rsi); 2913 __ restore_bcp(); 2914 } 2915 } 2916 2917 2918 void TemplateTable::invokevirtual_helper(Register index, Register recv, 2919 Register flags) { 2920 2921 // Uses temporary registers rax, rdx 2922 assert_different_registers(index, recv, rax, rdx); 2923 2924 // Test for an invoke of a final method 2925 Label notFinal; 2926 __ movl(rax, flags); 2927 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod)); 2928 __ jcc(Assembler::zero, notFinal); 2929 2930 Register method = index; // method must be rbx, 2931 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention"); 2932 2933 // do the call - the index is actually the method to call 2934 __ verify_oop(method); 2935 2936 // It's final, need a null check here! 2937 __ null_check(recv); 2938 2939 // profile this call 2940 __ profile_final_call(rax); 2941 2942 __ jump_from_interpreted(method, rax); 2943 2944 __ bind(notFinal); 2945 2946 // get receiver klass 2947 __ null_check(recv, oopDesc::klass_offset_in_bytes()); 2948 // Keep recv in rcx for callee expects it there 2949 __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes())); 2950 __ verify_oop(rax); 2951 2952 // profile this call 2953 __ profile_virtual_call(rax, rdi, rdx); 2954 2955 // get target methodOop & entry point 2956 const int base = instanceKlass::vtable_start_offset() * wordSize; 2957 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below"); 2958 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes())); 2959 __ jump_from_interpreted(method, rdx); 2960 } 2961 2962 2963 void TemplateTable::invokevirtual(int byte_no) { 2964 transition(vtos, vtos); 2965 assert(byte_no == f2_byte, "use this argument"); 2966 prepare_invoke(rbx, noreg, byte_no); 2967 2968 // rbx,: index 2969 // rcx: receiver 2970 // rdx: flags 2971 2972 invokevirtual_helper(rbx, rcx, rdx); 2973 } 2974 2975 2976 void TemplateTable::invokespecial(int byte_no) { 2977 transition(vtos, vtos); 2978 assert(byte_no == f1_byte, "use this argument"); 2979 prepare_invoke(rbx, noreg, byte_no); 2980 // do the call 2981 __ verify_oop(rbx); 2982 __ profile_call(rax); 2983 __ jump_from_interpreted(rbx, rax); 2984 } 2985 2986 2987 void TemplateTable::invokestatic(int byte_no) { 2988 transition(vtos, vtos); 2989 assert(byte_no == f1_byte, "use this argument"); 2990 prepare_invoke(rbx, noreg, byte_no); 2991 // do the call 2992 __ verify_oop(rbx); 2993 __ profile_call(rax); 2994 __ jump_from_interpreted(rbx, rax); 2995 } 2996 2997 2998 void TemplateTable::fast_invokevfinal(int byte_no) { 2999 transition(vtos, vtos); 3000 assert(byte_no == f2_byte, "use this argument"); 3001 __ stop("fast_invokevfinal not used on x86"); 3002 } 3003 3004 3005 void TemplateTable::invokeinterface(int byte_no) { 3006 transition(vtos, vtos); 3007 assert(byte_no == f1_byte, "use this argument"); 3008 prepare_invoke(rax, rbx, byte_no); 3009 3010 // rax,: Interface 3011 // rbx,: index 3012 // rcx: receiver 3013 // rdx: flags 3014 3015 // Special case of invokeinterface called for virtual method of 3016 // java.lang.Object. See cpCacheOop.cpp for details. 3017 // This code isn't produced by javac, but could be produced by 3018 // another compliant java compiler. 3019 Label notMethod; 3020 __ movl(rdi, rdx); 3021 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface)); 3022 __ jcc(Assembler::zero, notMethod); 3023 3024 invokevirtual_helper(rbx, rcx, rdx); 3025 __ bind(notMethod); 3026 3027 // Get receiver klass into rdx - also a null check 3028 __ restore_locals(); // restore rdi 3029 __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes())); 3030 __ verify_oop(rdx); 3031 3032 // profile this call 3033 __ profile_virtual_call(rdx, rsi, rdi); 3034 3035 Label no_such_interface, no_such_method; 3036 3037 __ lookup_interface_method(// inputs: rec. class, interface, itable index 3038 rdx, rax, rbx, 3039 // outputs: method, scan temp. reg 3040 rbx, rsi, 3041 no_such_interface); 3042 3043 // rbx,: methodOop to call 3044 // rcx: receiver 3045 // Check for abstract method error 3046 // Note: This should be done more efficiently via a throw_abstract_method_error 3047 // interpreter entry point and a conditional jump to it in case of a null 3048 // method. 3049 __ testptr(rbx, rbx); 3050 __ jcc(Assembler::zero, no_such_method); 3051 3052 // do the call 3053 // rcx: receiver 3054 // rbx,: methodOop 3055 __ jump_from_interpreted(rbx, rdx); 3056 __ should_not_reach_here(); 3057 3058 // exception handling code follows... 3059 // note: must restore interpreter registers to canonical 3060 // state for exception handling to work correctly! 3061 3062 __ bind(no_such_method); 3063 // throw exception 3064 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3065 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 3066 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3067 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 3068 // the call_VM checks for exception, so we should never return here. 3069 __ should_not_reach_here(); 3070 3071 __ bind(no_such_interface); 3072 // throw exception 3073 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3074 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 3075 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3076 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3077 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3078 // the call_VM checks for exception, so we should never return here. 3079 __ should_not_reach_here(); 3080 } 3081 3082 void TemplateTable::invokedynamic(int byte_no) { 3083 transition(vtos, vtos); 3084 3085 if (!EnableInvokeDynamic) { 3086 // We should not encounter this bytecode if !EnableInvokeDynamic. 3087 // The verifier will stop it. However, if we get past the verifier, 3088 // this will stop the thread in a reasonable way, without crashing the JVM. 3089 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3090 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3091 // the call_VM checks for exception, so we should never return here. 3092 __ should_not_reach_here(); 3093 return; 3094 } 3095 3096 assert(byte_no == f1_oop, "use this argument"); 3097 prepare_invoke(rax, rbx, byte_no); 3098 3099 // rax: CallSite object (f1) 3100 // rbx: unused (f2) 3101 // rcx: receiver address 3102 // rdx: flags (unused) 3103 3104 Register rax_callsite = rax; 3105 Register rcx_method_handle = rcx; 3106 3107 if (ProfileInterpreter) { 3108 // %%% should make a type profile for any invokedynamic that takes a ref argument 3109 // profile this call 3110 __ profile_call(rsi); 3111 } 3112 3113 __ movptr(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, rcx))); 3114 __ null_check(rcx_method_handle); 3115 __ prepare_to_jump_from_interpreted(); 3116 __ jump_to_method_handle_entry(rcx_method_handle, rdx); 3117 } 3118 3119 //---------------------------------------------------------------------------------------------------- 3120 // Allocation 3121 3122 void TemplateTable::_new() { 3123 transition(vtos, atos); 3124 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3125 Label slow_case; 3126 Label slow_case_no_pop; 3127 Label done; 3128 Label initialize_header; 3129 Label initialize_object; // including clearing the fields 3130 Label allocate_shared; 3131 3132 __ get_cpool_and_tags(rcx, rax); 3133 3134 // Make sure the class we're about to instantiate has been resolved. 3135 // This is done before loading instanceKlass to be consistent with the order 3136 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put) 3137 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize; 3138 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class); 3139 __ jcc(Assembler::notEqual, slow_case_no_pop); 3140 3141 // get instanceKlass 3142 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc))); 3143 __ push(rcx); // save the contexts of klass for initializing the header 3144 3145 // make sure klass is initialized & doesn't have finalizer 3146 // make sure klass is fully initialized 3147 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized); 3148 __ jcc(Assembler::notEqual, slow_case); 3149 3150 // get instance_size in instanceKlass (scaled to a count of bytes) 3151 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc))); 3152 // test to see if it has a finalizer or is malformed in some way 3153 __ testl(rdx, Klass::_lh_instance_slow_path_bit); 3154 __ jcc(Assembler::notZero, slow_case); 3155 3156 // 3157 // Allocate the instance 3158 // 1) Try to allocate in the TLAB 3159 // 2) if fail and the object is large allocate in the shared Eden 3160 // 3) if the above fails (or is not applicable), go to a slow case 3161 // (creates a new TLAB, etc.) 3162 3163 const bool allow_shared_alloc = 3164 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode; 3165 3166 const Register thread = rcx; 3167 if (UseTLAB || allow_shared_alloc) { 3168 __ get_thread(thread); 3169 } 3170 3171 if (UseTLAB) { 3172 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset()))); 3173 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3174 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset()))); 3175 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case); 3176 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx); 3177 if (ZeroTLAB) { 3178 // the fields have been already cleared 3179 __ jmp(initialize_header); 3180 } else { 3181 // initialize both the header and fields 3182 __ jmp(initialize_object); 3183 } 3184 } 3185 3186 // Allocation in the shared Eden, if allowed. 3187 // 3188 // rdx: instance size in bytes 3189 if (allow_shared_alloc) { 3190 __ bind(allocate_shared); 3191 3192 ExternalAddress heap_top((address)Universe::heap()->top_addr()); 3193 3194 Label retry; 3195 __ bind(retry); 3196 __ movptr(rax, heap_top); 3197 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3198 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr())); 3199 __ jcc(Assembler::above, slow_case); 3200 3201 // Compare rax, with the top addr, and if still equal, store the new 3202 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was 3203 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs. 3204 // 3205 // rax,: object begin 3206 // rbx,: object end 3207 // rdx: instance size in bytes 3208 __ locked_cmpxchgptr(rbx, heap_top); 3209 3210 // if someone beat us on the allocation, try again, otherwise continue 3211 __ jcc(Assembler::notEqual, retry); 3212 3213 __ incr_allocated_bytes(thread, rdx, 0); 3214 } 3215 3216 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 3217 // The object is initialized before the header. If the object size is 3218 // zero, go directly to the header initialization. 3219 __ bind(initialize_object); 3220 __ decrement(rdx, sizeof(oopDesc)); 3221 __ jcc(Assembler::zero, initialize_header); 3222 3223 // Initialize topmost object field, divide rdx by 8, check if odd and 3224 // test if zero. 3225 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) 3226 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd 3227 3228 // rdx must have been multiple of 8 3229 #ifdef ASSERT 3230 // make sure rdx was multiple of 8 3231 Label L; 3232 // Ignore partial flag stall after shrl() since it is debug VM 3233 __ jccb(Assembler::carryClear, L); 3234 __ stop("object size is not multiple of 2 - adjust this code"); 3235 __ bind(L); 3236 // rdx must be > 0, no extra check needed here 3237 #endif 3238 3239 // initialize remaining object fields: rdx was a multiple of 8 3240 { Label loop; 3241 __ bind(loop); 3242 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx); 3243 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx)); 3244 __ decrement(rdx); 3245 __ jcc(Assembler::notZero, loop); 3246 } 3247 3248 // initialize object header only. 3249 __ bind(initialize_header); 3250 if (UseBiasedLocking) { 3251 __ pop(rcx); // get saved klass back in the register. 3252 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 3253 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx); 3254 } else { 3255 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), 3256 (int32_t)markOopDesc::prototype()); // header 3257 __ pop(rcx); // get saved klass back in the register. 3258 } 3259 __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass 3260 3261 { 3262 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0); 3263 // Trigger dtrace event for fastpath 3264 __ push(atos); 3265 __ call_VM_leaf( 3266 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); 3267 __ pop(atos); 3268 } 3269 3270 __ jmp(done); 3271 } 3272 3273 // slow case 3274 __ bind(slow_case); 3275 __ pop(rcx); // restore stack pointer to what it was when we came in. 3276 __ bind(slow_case_no_pop); 3277 __ get_constant_pool(rax); 3278 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3279 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx); 3280 3281 // continue 3282 __ bind(done); 3283 } 3284 3285 3286 void TemplateTable::newarray() { 3287 transition(itos, atos); 3288 __ push_i(rax); // make sure everything is on the stack 3289 __ load_unsigned_byte(rdx, at_bcp(1)); 3290 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax); 3291 __ pop_i(rdx); // discard size 3292 } 3293 3294 3295 void TemplateTable::anewarray() { 3296 transition(itos, atos); 3297 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3298 __ get_constant_pool(rcx); 3299 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax); 3300 } 3301 3302 3303 void TemplateTable::arraylength() { 3304 transition(atos, itos); 3305 __ null_check(rax, arrayOopDesc::length_offset_in_bytes()); 3306 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes())); 3307 } 3308 3309 3310 void TemplateTable::checkcast() { 3311 transition(atos, atos); 3312 Label done, is_null, ok_is_subtype, quicked, resolved; 3313 __ testptr(rax, rax); // Object is in EAX 3314 __ jcc(Assembler::zero, is_null); 3315 3316 // Get cpool & tags index 3317 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 3318 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 3319 // See if bytecode has already been quicked 3320 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class); 3321 __ jcc(Assembler::equal, quicked); 3322 3323 __ push(atos); 3324 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 3325 __ pop_ptr(rdx); 3326 __ jmpb(resolved); 3327 3328 // Get superklass in EAX and subklass in EBX 3329 __ bind(quicked); 3330 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check 3331 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc))); 3332 3333 __ bind(resolved); 3334 __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes())); 3335 3336 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX. 3337 // Superklass in EAX. Subklass in EBX. 3338 __ gen_subtype_check( rbx, ok_is_subtype ); 3339 3340 // Come here on failure 3341 __ push(rdx); 3342 // object is at TOS 3343 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry)); 3344 3345 // Come here on success 3346 __ bind(ok_is_subtype); 3347 __ mov(rax,rdx); // Restore object in EDX 3348 3349 // Collect counts on whether this check-cast sees NULLs a lot or not. 3350 if (ProfileInterpreter) { 3351 __ jmp(done); 3352 __ bind(is_null); 3353 __ profile_null_seen(rcx); 3354 } else { 3355 __ bind(is_null); // same as 'done' 3356 } 3357 __ bind(done); 3358 } 3359 3360 3361 void TemplateTable::instanceof() { 3362 transition(atos, itos); 3363 Label done, is_null, ok_is_subtype, quicked, resolved; 3364 __ testptr(rax, rax); 3365 __ jcc(Assembler::zero, is_null); 3366 3367 // Get cpool & tags index 3368 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 3369 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 3370 // See if bytecode has already been quicked 3371 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class); 3372 __ jcc(Assembler::equal, quicked); 3373 3374 __ push(atos); 3375 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 3376 __ pop_ptr(rdx); 3377 __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes())); 3378 __ jmp(resolved); 3379 3380 // Get superklass in EAX and subklass in EDX 3381 __ bind(quicked); 3382 __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes())); 3383 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc))); 3384 3385 __ bind(resolved); 3386 3387 // Generate subtype check. Blows ECX. Resets EDI. 3388 // Superklass in EAX. Subklass in EDX. 3389 __ gen_subtype_check( rdx, ok_is_subtype ); 3390 3391 // Come here on failure 3392 __ xorl(rax,rax); 3393 __ jmpb(done); 3394 // Come here on success 3395 __ bind(ok_is_subtype); 3396 __ movl(rax, 1); 3397 3398 // Collect counts on whether this test sees NULLs a lot or not. 3399 if (ProfileInterpreter) { 3400 __ jmp(done); 3401 __ bind(is_null); 3402 __ profile_null_seen(rcx); 3403 } else { 3404 __ bind(is_null); // same as 'done' 3405 } 3406 __ bind(done); 3407 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass 3408 // rax, = 1: obj != NULL and obj is an instanceof the specified klass 3409 } 3410 3411 3412 //---------------------------------------------------------------------------------------------------- 3413 // Breakpoints 3414 void TemplateTable::_breakpoint() { 3415 3416 // Note: We get here even if we are single stepping.. 3417 // jbug inists on setting breakpoints at every bytecode 3418 // even if we are in single step mode. 3419 3420 transition(vtos, vtos); 3421 3422 // get the unpatched byte code 3423 __ get_method(rcx); 3424 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi); 3425 __ mov(rbx, rax); 3426 3427 // post the breakpoint event 3428 __ get_method(rcx); 3429 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi); 3430 3431 // complete the execution of original bytecode 3432 __ dispatch_only_normal(vtos); 3433 } 3434 3435 3436 //---------------------------------------------------------------------------------------------------- 3437 // Exceptions 3438 3439 void TemplateTable::athrow() { 3440 transition(atos, vtos); 3441 __ null_check(rax); 3442 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 3443 } 3444 3445 3446 //---------------------------------------------------------------------------------------------------- 3447 // Synchronization 3448 // 3449 // Note: monitorenter & exit are symmetric routines; which is reflected 3450 // in the assembly code structure as well 3451 // 3452 // Stack layout: 3453 // 3454 // [expressions ] <--- rsp = expression stack top 3455 // .. 3456 // [expressions ] 3457 // [monitor entry] <--- monitor block top = expression stack bot 3458 // .. 3459 // [monitor entry] 3460 // [frame data ] <--- monitor block bot 3461 // ... 3462 // [saved rbp, ] <--- rbp, 3463 3464 3465 void TemplateTable::monitorenter() { 3466 transition(atos, vtos); 3467 3468 // check for NULL object 3469 __ null_check(rax); 3470 3471 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3472 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 3473 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 3474 Label allocated; 3475 3476 // initialize entry pointer 3477 __ xorl(rdx, rdx); // points to free slot or NULL 3478 3479 // find a free slot in the monitor block (result in rdx) 3480 { Label entry, loop, exit; 3481 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry 3482 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 3483 __ jmpb(entry); 3484 3485 __ bind(loop); 3486 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used 3487 3488 // TODO - need new func here - kbt 3489 if (VM_Version::supports_cmov()) { 3490 __ cmov(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx 3491 } else { 3492 Label L; 3493 __ jccb(Assembler::notEqual, L); 3494 __ mov(rdx, rcx); // if not used then remember entry in rdx 3495 __ bind(L); 3496 } 3497 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 3498 __ jccb(Assembler::equal, exit); // if same object then stop searching 3499 __ addptr(rcx, entry_size); // otherwise advance to next entry 3500 __ bind(entry); 3501 __ cmpptr(rcx, rbx); // check if bottom reached 3502 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 3503 __ bind(exit); 3504 } 3505 3506 __ testptr(rdx, rdx); // check if a slot has been found 3507 __ jccb(Assembler::notZero, allocated); // if found, continue with that one 3508 3509 // allocate one if there's no free slot 3510 { Label entry, loop; 3511 // 1. compute new pointers // rsp: old expression stack top 3512 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom 3513 __ subptr(rsp, entry_size); // move expression stack top 3514 __ subptr(rdx, entry_size); // move expression stack bottom 3515 __ mov(rcx, rsp); // set start value for copy loop 3516 __ movptr(monitor_block_bot, rdx); // set new monitor block top 3517 __ jmp(entry); 3518 // 2. move expression stack contents 3519 __ bind(loop); 3520 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location 3521 __ movptr(Address(rcx, 0), rbx); // and store it at new location 3522 __ addptr(rcx, wordSize); // advance to next word 3523 __ bind(entry); 3524 __ cmpptr(rcx, rdx); // check if bottom reached 3525 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word 3526 } 3527 3528 // call run-time routine 3529 // rdx: points to monitor entry 3530 __ bind(allocated); 3531 3532 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly. 3533 // The object has already been poped from the stack, so the expression stack looks correct. 3534 __ increment(rsi); 3535 3536 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 3537 __ lock_object(rdx); 3538 3539 // check to make sure this monitor doesn't cause stack overflow after locking 3540 __ save_bcp(); // in case of exception 3541 __ generate_stack_overflow_check(0); 3542 3543 // The bcp has already been incremented. Just need to dispatch to next instruction. 3544 __ dispatch_next(vtos); 3545 } 3546 3547 3548 void TemplateTable::monitorexit() { 3549 transition(atos, vtos); 3550 3551 // check for NULL object 3552 __ null_check(rax); 3553 3554 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3555 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 3556 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 3557 Label found; 3558 3559 // find matching slot 3560 { Label entry, loop; 3561 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry 3562 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 3563 __ jmpb(entry); 3564 3565 __ bind(loop); 3566 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 3567 __ jcc(Assembler::equal, found); // if same object then stop searching 3568 __ addptr(rdx, entry_size); // otherwise advance to next entry 3569 __ bind(entry); 3570 __ cmpptr(rdx, rbx); // check if bottom reached 3571 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 3572 } 3573 3574 // error handling. Unlocking was not block-structured 3575 Label end; 3576 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 3577 __ should_not_reach_here(); 3578 3579 // call run-time routine 3580 // rcx: points to monitor entry 3581 __ bind(found); 3582 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps) 3583 __ unlock_object(rdx); 3584 __ pop_ptr(rax); // discard object 3585 __ bind(end); 3586 } 3587 3588 3589 //---------------------------------------------------------------------------------------------------- 3590 // Wide instructions 3591 3592 void TemplateTable::wide() { 3593 transition(vtos, vtos); 3594 __ load_unsigned_byte(rbx, at_bcp(1)); 3595 ExternalAddress wtable((address)Interpreter::_wentry_point); 3596 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr))); 3597 // Note: the rsi increment step is part of the individual wide bytecode implementations 3598 } 3599 3600 3601 //---------------------------------------------------------------------------------------------------- 3602 // Multi arrays 3603 3604 void TemplateTable::multianewarray() { 3605 transition(vtos, atos); 3606 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions 3607 // last dim is on top of stack; we want address of first one: 3608 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize 3609 // the latter wordSize to point to the beginning of the array. 3610 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize)); 3611 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax, 3612 __ load_unsigned_byte(rbx, at_bcp(3)); 3613 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts 3614 } 3615 3616 #endif /* !CC_INTERP */