1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "interpreter/interpreter.hpp" 28 #include "interpreter/interpreterRuntime.hpp" 29 #include "interpreter/interp_masm.hpp" 30 #include "interpreter/templateTable.hpp" 31 #include "memory/universe.inline.hpp" 32 #include "oops/methodData.hpp" 33 #include "oops/objArrayKlass.hpp" 34 #include "oops/oop.inline.hpp" 35 #include "prims/methodHandles.hpp" 36 #include "runtime/sharedRuntime.hpp" 37 #include "runtime/stubRoutines.hpp" 38 #include "runtime/synchronizer.hpp" 39 #include "utilities/macros.hpp" 40 41 #ifndef CC_INTERP 42 #define __ _masm-> 43 44 //---------------------------------------------------------------------------------------------------- 45 // Platform-dependent initialization 46 47 void TemplateTable::pd_initialize() { 48 // No i486 specific initialization 49 } 50 51 //---------------------------------------------------------------------------------------------------- 52 // Address computation 53 54 // local variables 55 static inline Address iaddress(int n) { 56 return Address(rdi, Interpreter::local_offset_in_bytes(n)); 57 } 58 59 static inline Address laddress(int n) { return iaddress(n + 1); } 60 static inline Address haddress(int n) { return iaddress(n + 0); } 61 static inline Address faddress(int n) { return iaddress(n); } 62 static inline Address daddress(int n) { return laddress(n); } 63 static inline Address aaddress(int n) { return iaddress(n); } 64 65 static inline Address iaddress(Register r) { 66 return Address(rdi, r, Interpreter::stackElementScale()); 67 } 68 static inline Address laddress(Register r) { 69 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1)); 70 } 71 static inline Address haddress(Register r) { 72 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0)); 73 } 74 75 static inline Address faddress(Register r) { return iaddress(r); } 76 static inline Address daddress(Register r) { return laddress(r); } 77 static inline Address aaddress(Register r) { return iaddress(r); } 78 79 // expression stack 80 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store 81 // data beyond the rsp which is potentially unsafe in an MT environment; 82 // an interrupt may overwrite that data.) 83 static inline Address at_rsp () { 84 return Address(rsp, 0); 85 } 86 87 // At top of Java expression stack which may be different than rsp(). It 88 // isn't for category 1 objects. 89 static inline Address at_tos () { 90 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0)); 91 return tos; 92 } 93 94 static inline Address at_tos_p1() { 95 return Address(rsp, Interpreter::expr_offset_in_bytes(1)); 96 } 97 98 static inline Address at_tos_p2() { 99 return Address(rsp, Interpreter::expr_offset_in_bytes(2)); 100 } 101 102 // Condition conversion 103 static Assembler::Condition j_not(TemplateTable::Condition cc) { 104 switch (cc) { 105 case TemplateTable::equal : return Assembler::notEqual; 106 case TemplateTable::not_equal : return Assembler::equal; 107 case TemplateTable::less : return Assembler::greaterEqual; 108 case TemplateTable::less_equal : return Assembler::greater; 109 case TemplateTable::greater : return Assembler::lessEqual; 110 case TemplateTable::greater_equal: return Assembler::less; 111 } 112 ShouldNotReachHere(); 113 return Assembler::zero; 114 } 115 116 117 //---------------------------------------------------------------------------------------------------- 118 // Miscelaneous helper routines 119 120 // Store an oop (or NULL) at the address described by obj. 121 // If val == noreg this means store a NULL 122 123 static void do_oop_store(InterpreterMacroAssembler* _masm, 124 Address obj, 125 Register val, 126 BarrierSet::Name barrier, 127 bool precise) { 128 assert(val == noreg || val == rax, "parameter is just for looks"); 129 switch (barrier) { 130 #if INCLUDE_ALL_GCS 131 case BarrierSet::G1SATBCT: 132 case BarrierSet::G1SATBCTLogging: 133 { 134 // flatten object address if needed 135 // We do it regardless of precise because we need the registers 136 if (obj.index() == noreg && obj.disp() == 0) { 137 if (obj.base() != rdx) { 138 __ movl(rdx, obj.base()); 139 } 140 } else { 141 __ leal(rdx, obj); 142 } 143 __ get_thread(rcx); 144 __ save_bcp(); 145 __ g1_write_barrier_pre(rdx /* obj */, 146 rbx /* pre_val */, 147 rcx /* thread */, 148 rsi /* tmp */, 149 val != noreg /* tosca_live */, 150 false /* expand_call */); 151 152 // Do the actual store 153 // noreg means NULL 154 if (val == noreg) { 155 __ movptr(Address(rdx, 0), NULL_WORD); 156 // No post barrier for NULL 157 } else { 158 __ movl(Address(rdx, 0), val); 159 __ g1_write_barrier_post(rdx /* store_adr */, 160 val /* new_val */, 161 rcx /* thread */, 162 rbx /* tmp */, 163 rsi /* tmp2 */); 164 } 165 __ restore_bcp(); 166 167 } 168 break; 169 #endif // INCLUDE_ALL_GCS 170 case BarrierSet::CardTableModRef: 171 case BarrierSet::CardTableExtension: 172 { 173 if (val == noreg) { 174 __ movptr(obj, NULL_WORD); 175 } else { 176 __ movl(obj, val); 177 // flatten object address if needed 178 if (!precise || (obj.index() == noreg && obj.disp() == 0)) { 179 __ store_check(obj.base()); 180 } else { 181 __ leal(rdx, obj); 182 __ store_check(rdx); 183 } 184 } 185 } 186 break; 187 case BarrierSet::ModRef: 188 case BarrierSet::Other: 189 if (val == noreg) { 190 __ movptr(obj, NULL_WORD); 191 } else { 192 __ movl(obj, val); 193 } 194 break; 195 default : 196 ShouldNotReachHere(); 197 198 } 199 } 200 201 Address TemplateTable::at_bcp(int offset) { 202 assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 203 return Address(rsi, offset); 204 } 205 206 207 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg, 208 Register temp_reg, bool load_bc_into_bc_reg/*=true*/, 209 int byte_no) { 210 if (!RewriteBytecodes) return; 211 Label L_patch_done; 212 213 switch (bc) { 214 case Bytecodes::_fast_aputfield: 215 case Bytecodes::_fast_bputfield: 216 case Bytecodes::_fast_cputfield: 217 case Bytecodes::_fast_dputfield: 218 case Bytecodes::_fast_fputfield: 219 case Bytecodes::_fast_iputfield: 220 case Bytecodes::_fast_lputfield: 221 case Bytecodes::_fast_sputfield: 222 { 223 // We skip bytecode quickening for putfield instructions when 224 // the put_code written to the constant pool cache is zero. 225 // This is required so that every execution of this instruction 226 // calls out to InterpreterRuntime::resolve_get_put to do 227 // additional, required work. 228 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 229 assert(load_bc_into_bc_reg, "we use bc_reg as temp"); 230 __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1); 231 __ movl(bc_reg, bc); 232 __ cmpl(temp_reg, (int) 0); 233 __ jcc(Assembler::zero, L_patch_done); // don't patch 234 } 235 break; 236 default: 237 assert(byte_no == -1, "sanity"); 238 // the pair bytecodes have already done the load. 239 if (load_bc_into_bc_reg) { 240 __ movl(bc_reg, bc); 241 } 242 } 243 244 if (JvmtiExport::can_post_breakpoint()) { 245 Label L_fast_patch; 246 // if a breakpoint is present we can't rewrite the stream directly 247 __ movzbl(temp_reg, at_bcp(0)); 248 __ cmpl(temp_reg, Bytecodes::_breakpoint); 249 __ jcc(Assembler::notEqual, L_fast_patch); 250 __ get_method(temp_reg); 251 // Let breakpoint table handling rewrite to quicker bytecode 252 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rsi, bc_reg); 253 #ifndef ASSERT 254 __ jmpb(L_patch_done); 255 #else 256 __ jmp(L_patch_done); 257 #endif 258 __ bind(L_fast_patch); 259 } 260 261 #ifdef ASSERT 262 Label L_okay; 263 __ load_unsigned_byte(temp_reg, at_bcp(0)); 264 __ cmpl(temp_reg, (int)Bytecodes::java_code(bc)); 265 __ jccb(Assembler::equal, L_okay); 266 __ cmpl(temp_reg, bc_reg); 267 __ jcc(Assembler::equal, L_okay); 268 __ stop("patching the wrong bytecode"); 269 __ bind(L_okay); 270 #endif 271 272 // patch bytecode 273 __ movb(at_bcp(0), bc_reg); 274 __ bind(L_patch_done); 275 } 276 277 //---------------------------------------------------------------------------------------------------- 278 // Individual instructions 279 280 void TemplateTable::nop() { 281 transition(vtos, vtos); 282 // nothing to do 283 } 284 285 void TemplateTable::shouldnotreachhere() { 286 transition(vtos, vtos); 287 __ stop("shouldnotreachhere bytecode"); 288 } 289 290 291 292 void TemplateTable::aconst_null() { 293 transition(vtos, atos); 294 __ xorptr(rax, rax); 295 } 296 297 298 void TemplateTable::iconst(int value) { 299 transition(vtos, itos); 300 if (value == 0) { 301 __ xorptr(rax, rax); 302 } else { 303 __ movptr(rax, value); 304 } 305 } 306 307 308 void TemplateTable::lconst(int value) { 309 transition(vtos, ltos); 310 if (value == 0) { 311 __ xorptr(rax, rax); 312 } else { 313 __ movptr(rax, value); 314 } 315 assert(value >= 0, "check this code"); 316 __ xorptr(rdx, rdx); 317 } 318 319 320 void TemplateTable::fconst(int value) { 321 transition(vtos, ftos); 322 if (value == 0) { __ fldz(); 323 } else if (value == 1) { __ fld1(); 324 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here 325 } else { ShouldNotReachHere(); 326 } 327 } 328 329 330 void TemplateTable::dconst(int value) { 331 transition(vtos, dtos); 332 if (value == 0) { __ fldz(); 333 } else if (value == 1) { __ fld1(); 334 } else { ShouldNotReachHere(); 335 } 336 } 337 338 339 void TemplateTable::bipush() { 340 transition(vtos, itos); 341 __ load_signed_byte(rax, at_bcp(1)); 342 } 343 344 345 void TemplateTable::sipush() { 346 transition(vtos, itos); 347 __ load_unsigned_short(rax, at_bcp(1)); 348 __ bswapl(rax); 349 __ sarl(rax, 16); 350 } 351 352 void TemplateTable::ldc(bool wide) { 353 transition(vtos, vtos); 354 Label call_ldc, notFloat, notClass, Done; 355 356 if (wide) { 357 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 358 } else { 359 __ load_unsigned_byte(rbx, at_bcp(1)); 360 } 361 __ get_cpool_and_tags(rcx, rax); 362 const int base_offset = ConstantPool::header_size() * wordSize; 363 const int tags_offset = Array<u1>::base_offset_in_bytes(); 364 365 // get type 366 __ xorptr(rdx, rdx); 367 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset)); 368 369 // unresolved class - get the resolved class 370 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass); 371 __ jccb(Assembler::equal, call_ldc); 372 373 // unresolved class in error (resolution failed) - call into runtime 374 // so that the same error from first resolution attempt is thrown. 375 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError); 376 __ jccb(Assembler::equal, call_ldc); 377 378 // resolved class - need to call vm to get java mirror of the class 379 __ cmpl(rdx, JVM_CONSTANT_Class); 380 __ jcc(Assembler::notEqual, notClass); 381 382 __ bind(call_ldc); 383 __ movl(rcx, wide); 384 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx); 385 __ push(atos); 386 __ jmp(Done); 387 388 __ bind(notClass); 389 __ cmpl(rdx, JVM_CONSTANT_Float); 390 __ jccb(Assembler::notEqual, notFloat); 391 // ftos 392 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset)); 393 __ push(ftos); 394 __ jmp(Done); 395 396 __ bind(notFloat); 397 #ifdef ASSERT 398 { Label L; 399 __ cmpl(rdx, JVM_CONSTANT_Integer); 400 __ jcc(Assembler::equal, L); 401 // String and Object are rewritten to fast_aldc 402 __ stop("unexpected tag type in ldc"); 403 __ bind(L); 404 } 405 #endif 406 // itos JVM_CONSTANT_Integer only 407 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 408 __ push(itos); 409 __ bind(Done); 410 } 411 412 // Fast path for caching oop constants. 413 void TemplateTable::fast_aldc(bool wide) { 414 transition(vtos, atos); 415 416 Register result = rax; 417 Register tmp = rdx; 418 int index_size = wide ? sizeof(u2) : sizeof(u1); 419 420 Label resolved; 421 422 // We are resolved if the resolved reference cache entry contains a 423 // non-null object (String, MethodType, etc.) 424 assert_different_registers(result, tmp); 425 __ get_cache_index_at_bcp(tmp, 1, index_size); 426 __ load_resolved_reference_at_index(result, tmp); 427 __ testl(result, result); 428 __ jcc(Assembler::notZero, resolved); 429 430 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); 431 432 // first time invocation - must resolve first 433 __ movl(tmp, (int)bytecode()); 434 __ call_VM(result, entry, tmp); 435 436 __ bind(resolved); 437 438 if (VerifyOops) { 439 __ verify_oop(result); 440 } 441 } 442 443 void TemplateTable::ldc2_w() { 444 transition(vtos, vtos); 445 Label Long, Done; 446 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 447 448 __ get_cpool_and_tags(rcx, rax); 449 const int base_offset = ConstantPool::header_size() * wordSize; 450 const int tags_offset = Array<u1>::base_offset_in_bytes(); 451 452 // get type 453 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double); 454 __ jccb(Assembler::notEqual, Long); 455 // dtos 456 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset)); 457 __ push(dtos); 458 __ jmpb(Done); 459 460 __ bind(Long); 461 // ltos 462 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize)); 463 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize))); 464 465 __ push(ltos); 466 467 __ bind(Done); 468 } 469 470 471 void TemplateTable::locals_index(Register reg, int offset) { 472 __ load_unsigned_byte(reg, at_bcp(offset)); 473 __ negptr(reg); 474 } 475 476 477 void TemplateTable::iload() { 478 transition(vtos, itos); 479 if (RewriteFrequentPairs) { 480 Label rewrite, done; 481 482 // get next byte 483 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 484 // if _iload, wait to rewrite to iload2. We only want to rewrite the 485 // last two iloads in a pair. Comparing against fast_iload means that 486 // the next bytecode is neither an iload or a caload, and therefore 487 // an iload pair. 488 __ cmpl(rbx, Bytecodes::_iload); 489 __ jcc(Assembler::equal, done); 490 491 __ cmpl(rbx, Bytecodes::_fast_iload); 492 __ movl(rcx, Bytecodes::_fast_iload2); 493 __ jccb(Assembler::equal, rewrite); 494 495 // if _caload, rewrite to fast_icaload 496 __ cmpl(rbx, Bytecodes::_caload); 497 __ movl(rcx, Bytecodes::_fast_icaload); 498 __ jccb(Assembler::equal, rewrite); 499 500 // rewrite so iload doesn't check again. 501 __ movl(rcx, Bytecodes::_fast_iload); 502 503 // rewrite 504 // rcx: fast bytecode 505 __ bind(rewrite); 506 patch_bytecode(Bytecodes::_iload, rcx, rbx, false); 507 __ bind(done); 508 } 509 510 // Get the local value into tos 511 locals_index(rbx); 512 __ movl(rax, iaddress(rbx)); 513 } 514 515 516 void TemplateTable::fast_iload2() { 517 transition(vtos, itos); 518 locals_index(rbx); 519 __ movl(rax, iaddress(rbx)); 520 __ push(itos); 521 locals_index(rbx, 3); 522 __ movl(rax, iaddress(rbx)); 523 } 524 525 void TemplateTable::fast_iload() { 526 transition(vtos, itos); 527 locals_index(rbx); 528 __ movl(rax, iaddress(rbx)); 529 } 530 531 532 void TemplateTable::lload() { 533 transition(vtos, ltos); 534 locals_index(rbx); 535 __ movptr(rax, laddress(rbx)); 536 NOT_LP64(__ movl(rdx, haddress(rbx))); 537 } 538 539 540 void TemplateTable::fload() { 541 transition(vtos, ftos); 542 locals_index(rbx); 543 __ fld_s(faddress(rbx)); 544 } 545 546 547 void TemplateTable::dload() { 548 transition(vtos, dtos); 549 locals_index(rbx); 550 __ fld_d(daddress(rbx)); 551 } 552 553 554 void TemplateTable::aload() { 555 transition(vtos, atos); 556 locals_index(rbx); 557 __ movptr(rax, aaddress(rbx)); 558 } 559 560 561 void TemplateTable::locals_index_wide(Register reg) { 562 __ load_unsigned_short(reg, at_bcp(2)); 563 __ bswapl(reg); 564 __ shrl(reg, 16); 565 __ negptr(reg); 566 } 567 568 569 void TemplateTable::wide_iload() { 570 transition(vtos, itos); 571 locals_index_wide(rbx); 572 __ movl(rax, iaddress(rbx)); 573 } 574 575 576 void TemplateTable::wide_lload() { 577 transition(vtos, ltos); 578 locals_index_wide(rbx); 579 __ movptr(rax, laddress(rbx)); 580 NOT_LP64(__ movl(rdx, haddress(rbx))); 581 } 582 583 584 void TemplateTable::wide_fload() { 585 transition(vtos, ftos); 586 locals_index_wide(rbx); 587 __ fld_s(faddress(rbx)); 588 } 589 590 591 void TemplateTable::wide_dload() { 592 transition(vtos, dtos); 593 locals_index_wide(rbx); 594 __ fld_d(daddress(rbx)); 595 } 596 597 598 void TemplateTable::wide_aload() { 599 transition(vtos, atos); 600 locals_index_wide(rbx); 601 __ movptr(rax, aaddress(rbx)); 602 } 603 604 void TemplateTable::index_check(Register array, Register index) { 605 // Pop ptr into array 606 __ pop_ptr(array); 607 index_check_without_pop(array, index); 608 } 609 610 void TemplateTable::index_check_without_pop(Register array, Register index) { 611 // destroys rbx, 612 // check array 613 __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 614 LP64_ONLY(__ movslq(index, index)); 615 // check index 616 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes())); 617 if (index != rbx) { 618 // ??? convention: move aberrant index into rbx, for exception message 619 assert(rbx != array, "different registers"); 620 __ mov(rbx, index); 621 } 622 __ jump_cc(Assembler::aboveEqual, 623 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry)); 624 } 625 626 627 void TemplateTable::iaload() { 628 transition(itos, itos); 629 // rdx: array 630 index_check(rdx, rax); // kills rbx, 631 // rax,: index 632 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT))); 633 } 634 635 636 void TemplateTable::laload() { 637 transition(itos, ltos); 638 // rax,: index 639 // rdx: array 640 index_check(rdx, rax); 641 __ mov(rbx, rax); 642 // rbx,: index 643 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize)); 644 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize))); 645 } 646 647 648 void TemplateTable::faload() { 649 transition(itos, ftos); 650 // rdx: array 651 index_check(rdx, rax); // kills rbx, 652 // rax,: index 653 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 654 } 655 656 657 void TemplateTable::daload() { 658 transition(itos, dtos); 659 // rdx: array 660 index_check(rdx, rax); // kills rbx, 661 // rax,: index 662 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 663 } 664 665 666 void TemplateTable::aaload() { 667 transition(itos, atos); 668 // rdx: array 669 index_check(rdx, rax); // kills rbx, 670 // rax,: index 671 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 672 } 673 674 675 void TemplateTable::baload() { 676 transition(itos, itos); 677 // rdx: array 678 index_check(rdx, rax); // kills rbx, 679 // rax,: index 680 // can do better code for P5 - fix this at some point 681 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 682 __ mov(rax, rbx); 683 } 684 685 686 void TemplateTable::caload() { 687 transition(itos, itos); 688 // rdx: array 689 index_check(rdx, rax); // kills rbx, 690 // rax,: index 691 // can do better code for P5 - may want to improve this at some point 692 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 693 __ mov(rax, rbx); 694 } 695 696 // iload followed by caload frequent pair 697 void TemplateTable::fast_icaload() { 698 transition(vtos, itos); 699 // load index out of locals 700 locals_index(rbx); 701 __ movl(rax, iaddress(rbx)); 702 703 // rdx: array 704 index_check(rdx, rax); 705 // rax,: index 706 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 707 __ mov(rax, rbx); 708 } 709 710 void TemplateTable::saload() { 711 transition(itos, itos); 712 // rdx: array 713 index_check(rdx, rax); // kills rbx, 714 // rax,: index 715 // can do better code for P5 - may want to improve this at some point 716 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 717 __ mov(rax, rbx); 718 } 719 720 721 void TemplateTable::iload(int n) { 722 transition(vtos, itos); 723 __ movl(rax, iaddress(n)); 724 } 725 726 727 void TemplateTable::lload(int n) { 728 transition(vtos, ltos); 729 __ movptr(rax, laddress(n)); 730 NOT_LP64(__ movptr(rdx, haddress(n))); 731 } 732 733 734 void TemplateTable::fload(int n) { 735 transition(vtos, ftos); 736 __ fld_s(faddress(n)); 737 } 738 739 740 void TemplateTable::dload(int n) { 741 transition(vtos, dtos); 742 __ fld_d(daddress(n)); 743 } 744 745 746 void TemplateTable::aload(int n) { 747 transition(vtos, atos); 748 __ movptr(rax, aaddress(n)); 749 } 750 751 752 void TemplateTable::aload_0() { 753 transition(vtos, atos); 754 // According to bytecode histograms, the pairs: 755 // 756 // _aload_0, _fast_igetfield 757 // _aload_0, _fast_agetfield 758 // _aload_0, _fast_fgetfield 759 // 760 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0 761 // bytecode checks if the next bytecode is either _fast_igetfield, 762 // _fast_agetfield or _fast_fgetfield and then rewrites the 763 // current bytecode into a pair bytecode; otherwise it rewrites the current 764 // bytecode into _fast_aload_0 that doesn't do the pair check anymore. 765 // 766 // Note: If the next bytecode is _getfield, the rewrite must be delayed, 767 // otherwise we may miss an opportunity for a pair. 768 // 769 // Also rewrite frequent pairs 770 // aload_0, aload_1 771 // aload_0, iload_1 772 // These bytecodes with a small amount of code are most profitable to rewrite 773 if (RewriteFrequentPairs) { 774 Label rewrite, done; 775 // get next byte 776 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 777 778 // do actual aload_0 779 aload(0); 780 781 // if _getfield then wait with rewrite 782 __ cmpl(rbx, Bytecodes::_getfield); 783 __ jcc(Assembler::equal, done); 784 785 // if _igetfield then reqrite to _fast_iaccess_0 786 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 787 __ cmpl(rbx, Bytecodes::_fast_igetfield); 788 __ movl(rcx, Bytecodes::_fast_iaccess_0); 789 __ jccb(Assembler::equal, rewrite); 790 791 // if _agetfield then reqrite to _fast_aaccess_0 792 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 793 __ cmpl(rbx, Bytecodes::_fast_agetfield); 794 __ movl(rcx, Bytecodes::_fast_aaccess_0); 795 __ jccb(Assembler::equal, rewrite); 796 797 // if _fgetfield then reqrite to _fast_faccess_0 798 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 799 __ cmpl(rbx, Bytecodes::_fast_fgetfield); 800 __ movl(rcx, Bytecodes::_fast_faccess_0); 801 __ jccb(Assembler::equal, rewrite); 802 803 // else rewrite to _fast_aload0 804 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 805 __ movl(rcx, Bytecodes::_fast_aload_0); 806 807 // rewrite 808 // rcx: fast bytecode 809 __ bind(rewrite); 810 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false); 811 812 __ bind(done); 813 } else { 814 aload(0); 815 } 816 } 817 818 void TemplateTable::istore() { 819 transition(itos, vtos); 820 locals_index(rbx); 821 __ movl(iaddress(rbx), rax); 822 } 823 824 825 void TemplateTable::lstore() { 826 transition(ltos, vtos); 827 locals_index(rbx); 828 __ movptr(laddress(rbx), rax); 829 NOT_LP64(__ movptr(haddress(rbx), rdx)); 830 } 831 832 833 void TemplateTable::fstore() { 834 transition(ftos, vtos); 835 locals_index(rbx); 836 __ fstp_s(faddress(rbx)); 837 } 838 839 840 void TemplateTable::dstore() { 841 transition(dtos, vtos); 842 locals_index(rbx); 843 __ fstp_d(daddress(rbx)); 844 } 845 846 847 void TemplateTable::astore() { 848 transition(vtos, vtos); 849 __ pop_ptr(rax); 850 locals_index(rbx); 851 __ movptr(aaddress(rbx), rax); 852 } 853 854 855 void TemplateTable::wide_istore() { 856 transition(vtos, vtos); 857 __ pop_i(rax); 858 locals_index_wide(rbx); 859 __ movl(iaddress(rbx), rax); 860 } 861 862 863 void TemplateTable::wide_lstore() { 864 transition(vtos, vtos); 865 __ pop_l(rax, rdx); 866 locals_index_wide(rbx); 867 __ movptr(laddress(rbx), rax); 868 NOT_LP64(__ movl(haddress(rbx), rdx)); 869 } 870 871 872 void TemplateTable::wide_fstore() { 873 wide_istore(); 874 } 875 876 877 void TemplateTable::wide_dstore() { 878 wide_lstore(); 879 } 880 881 882 void TemplateTable::wide_astore() { 883 transition(vtos, vtos); 884 __ pop_ptr(rax); 885 locals_index_wide(rbx); 886 __ movptr(aaddress(rbx), rax); 887 } 888 889 890 void TemplateTable::iastore() { 891 transition(itos, vtos); 892 __ pop_i(rbx); 893 // rax,: value 894 // rdx: array 895 index_check(rdx, rbx); // prefer index in rbx, 896 // rbx,: index 897 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax); 898 } 899 900 901 void TemplateTable::lastore() { 902 transition(ltos, vtos); 903 __ pop_i(rbx); 904 // rax,: low(value) 905 // rcx: array 906 // rdx: high(value) 907 index_check(rcx, rbx); // prefer index in rbx, 908 // rbx,: index 909 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax); 910 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx)); 911 } 912 913 914 void TemplateTable::fastore() { 915 transition(ftos, vtos); 916 __ pop_i(rbx); 917 // rdx: array 918 // st0: value 919 index_check(rdx, rbx); // prefer index in rbx, 920 // rbx,: index 921 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 922 } 923 924 925 void TemplateTable::dastore() { 926 transition(dtos, vtos); 927 __ pop_i(rbx); 928 // rdx: array 929 // st0: value 930 index_check(rdx, rbx); // prefer index in rbx, 931 // rbx,: index 932 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 933 } 934 935 936 void TemplateTable::aastore() { 937 Label is_null, ok_is_subtype, done; 938 transition(vtos, vtos); 939 // stack: ..., array, index, value 940 __ movptr(rax, at_tos()); // Value 941 __ movl(rcx, at_tos_p1()); // Index 942 __ movptr(rdx, at_tos_p2()); // Array 943 944 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 945 index_check_without_pop(rdx, rcx); // kills rbx, 946 // do array store check - check for NULL value first 947 __ testptr(rax, rax); 948 __ jcc(Assembler::zero, is_null); 949 950 // Move subklass into EBX 951 __ load_klass(rbx, rax); 952 // Move superklass into EAX 953 __ load_klass(rax, rdx); 954 __ movptr(rax, Address(rax, ObjArrayKlass::element_klass_offset())); 955 // Compress array+index*wordSize+12 into a single register. Frees ECX. 956 __ lea(rdx, element_address); 957 958 // Generate subtype check. Blows ECX. Resets EDI to locals. 959 // Superklass in EAX. Subklass in EBX. 960 __ gen_subtype_check( rbx, ok_is_subtype ); 961 962 // Come here on failure 963 // object is at TOS 964 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry)); 965 966 // Come here on success 967 __ bind(ok_is_subtype); 968 969 // Get the value to store 970 __ movptr(rax, at_rsp()); 971 // and store it with appropriate barrier 972 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true); 973 974 __ jmp(done); 975 976 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx] 977 __ bind(is_null); 978 __ profile_null_seen(rbx); 979 980 // Store NULL, (noreg means NULL to do_oop_store) 981 do_oop_store(_masm, element_address, noreg, _bs->kind(), true); 982 983 // Pop stack arguments 984 __ bind(done); 985 __ addptr(rsp, 3 * Interpreter::stackElementSize); 986 } 987 988 989 void TemplateTable::bastore() { 990 transition(itos, vtos); 991 __ pop_i(rbx); 992 // rax,: value 993 // rdx: array 994 index_check(rdx, rbx); // prefer index in rbx, 995 // rbx,: index 996 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax); 997 } 998 999 1000 void TemplateTable::castore() { 1001 transition(itos, vtos); 1002 __ pop_i(rbx); 1003 // rax,: value 1004 // rdx: array 1005 index_check(rdx, rbx); // prefer index in rbx, 1006 // rbx,: index 1007 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax); 1008 } 1009 1010 1011 void TemplateTable::sastore() { 1012 castore(); 1013 } 1014 1015 1016 void TemplateTable::istore(int n) { 1017 transition(itos, vtos); 1018 __ movl(iaddress(n), rax); 1019 } 1020 1021 1022 void TemplateTable::lstore(int n) { 1023 transition(ltos, vtos); 1024 __ movptr(laddress(n), rax); 1025 NOT_LP64(__ movptr(haddress(n), rdx)); 1026 } 1027 1028 1029 void TemplateTable::fstore(int n) { 1030 transition(ftos, vtos); 1031 __ fstp_s(faddress(n)); 1032 } 1033 1034 1035 void TemplateTable::dstore(int n) { 1036 transition(dtos, vtos); 1037 __ fstp_d(daddress(n)); 1038 } 1039 1040 1041 void TemplateTable::astore(int n) { 1042 transition(vtos, vtos); 1043 __ pop_ptr(rax); 1044 __ movptr(aaddress(n), rax); 1045 } 1046 1047 1048 void TemplateTable::pop() { 1049 transition(vtos, vtos); 1050 __ addptr(rsp, Interpreter::stackElementSize); 1051 } 1052 1053 1054 void TemplateTable::pop2() { 1055 transition(vtos, vtos); 1056 __ addptr(rsp, 2*Interpreter::stackElementSize); 1057 } 1058 1059 1060 void TemplateTable::dup() { 1061 transition(vtos, vtos); 1062 // stack: ..., a 1063 __ load_ptr(0, rax); 1064 __ push_ptr(rax); 1065 // stack: ..., a, a 1066 } 1067 1068 1069 void TemplateTable::dup_x1() { 1070 transition(vtos, vtos); 1071 // stack: ..., a, b 1072 __ load_ptr( 0, rax); // load b 1073 __ load_ptr( 1, rcx); // load a 1074 __ store_ptr(1, rax); // store b 1075 __ store_ptr(0, rcx); // store a 1076 __ push_ptr(rax); // push b 1077 // stack: ..., b, a, b 1078 } 1079 1080 1081 void TemplateTable::dup_x2() { 1082 transition(vtos, vtos); 1083 // stack: ..., a, b, c 1084 __ load_ptr( 0, rax); // load c 1085 __ load_ptr( 2, rcx); // load a 1086 __ store_ptr(2, rax); // store c in a 1087 __ push_ptr(rax); // push c 1088 // stack: ..., c, b, c, c 1089 __ load_ptr( 2, rax); // load b 1090 __ store_ptr(2, rcx); // store a in b 1091 // stack: ..., c, a, c, c 1092 __ store_ptr(1, rax); // store b in c 1093 // stack: ..., c, a, b, c 1094 } 1095 1096 1097 void TemplateTable::dup2() { 1098 transition(vtos, vtos); 1099 // stack: ..., a, b 1100 __ load_ptr(1, rax); // load a 1101 __ push_ptr(rax); // push a 1102 __ load_ptr(1, rax); // load b 1103 __ push_ptr(rax); // push b 1104 // stack: ..., a, b, a, b 1105 } 1106 1107 1108 void TemplateTable::dup2_x1() { 1109 transition(vtos, vtos); 1110 // stack: ..., a, b, c 1111 __ load_ptr( 0, rcx); // load c 1112 __ load_ptr( 1, rax); // load b 1113 __ push_ptr(rax); // push b 1114 __ push_ptr(rcx); // push c 1115 // stack: ..., a, b, c, b, c 1116 __ store_ptr(3, rcx); // store c in b 1117 // stack: ..., a, c, c, b, c 1118 __ load_ptr( 4, rcx); // load a 1119 __ store_ptr(2, rcx); // store a in 2nd c 1120 // stack: ..., a, c, a, b, c 1121 __ store_ptr(4, rax); // store b in a 1122 // stack: ..., b, c, a, b, c 1123 // stack: ..., b, c, a, b, c 1124 } 1125 1126 1127 void TemplateTable::dup2_x2() { 1128 transition(vtos, vtos); 1129 // stack: ..., a, b, c, d 1130 __ load_ptr( 0, rcx); // load d 1131 __ load_ptr( 1, rax); // load c 1132 __ push_ptr(rax); // push c 1133 __ push_ptr(rcx); // push d 1134 // stack: ..., a, b, c, d, c, d 1135 __ load_ptr( 4, rax); // load b 1136 __ store_ptr(2, rax); // store b in d 1137 __ store_ptr(4, rcx); // store d in b 1138 // stack: ..., a, d, c, b, c, d 1139 __ load_ptr( 5, rcx); // load a 1140 __ load_ptr( 3, rax); // load c 1141 __ store_ptr(3, rcx); // store a in c 1142 __ store_ptr(5, rax); // store c in a 1143 // stack: ..., c, d, a, b, c, d 1144 // stack: ..., c, d, a, b, c, d 1145 } 1146 1147 1148 void TemplateTable::swap() { 1149 transition(vtos, vtos); 1150 // stack: ..., a, b 1151 __ load_ptr( 1, rcx); // load a 1152 __ load_ptr( 0, rax); // load b 1153 __ store_ptr(0, rcx); // store a in b 1154 __ store_ptr(1, rax); // store b in a 1155 // stack: ..., b, a 1156 } 1157 1158 1159 void TemplateTable::iop2(Operation op) { 1160 transition(itos, itos); 1161 switch (op) { 1162 case add : __ pop_i(rdx); __ addl (rax, rdx); break; 1163 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break; 1164 case mul : __ pop_i(rdx); __ imull(rax, rdx); break; 1165 case _and : __ pop_i(rdx); __ andl (rax, rdx); break; 1166 case _or : __ pop_i(rdx); __ orl (rax, rdx); break; 1167 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break; 1168 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1169 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1170 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1171 default : ShouldNotReachHere(); 1172 } 1173 } 1174 1175 1176 void TemplateTable::lop2(Operation op) { 1177 transition(ltos, ltos); 1178 __ pop_l(rbx, rcx); 1179 switch (op) { 1180 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break; 1181 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx); 1182 __ mov (rax, rbx); __ mov (rdx, rcx); break; 1183 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break; 1184 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break; 1185 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break; 1186 default : ShouldNotReachHere(); 1187 } 1188 } 1189 1190 1191 void TemplateTable::idiv() { 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 } 1201 1202 1203 void TemplateTable::irem() { 1204 transition(itos, itos); 1205 __ mov(rcx, rax); 1206 __ pop_i(rax); 1207 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If 1208 // they are not equal, one could do a normal division (no correction 1209 // needed), which may speed up this implementation for the common case. 1210 // (see also JVM spec., p.243 & p.271) 1211 __ corrected_idivl(rcx); 1212 __ mov(rax, rdx); 1213 } 1214 1215 1216 void TemplateTable::lmul() { 1217 transition(ltos, ltos); 1218 __ pop_l(rbx, rcx); 1219 __ push(rcx); __ push(rbx); 1220 __ push(rdx); __ push(rax); 1221 __ lmul(2 * wordSize, 0); 1222 __ addptr(rsp, 4 * wordSize); // take off temporaries 1223 } 1224 1225 1226 void TemplateTable::ldiv() { 1227 transition(ltos, ltos); 1228 __ pop_l(rbx, rcx); 1229 __ push(rcx); __ push(rbx); 1230 __ push(rdx); __ push(rax); 1231 // check if y = 0 1232 __ orl(rax, rdx); 1233 __ jump_cc(Assembler::zero, 1234 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1235 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv)); 1236 __ addptr(rsp, 4 * wordSize); // take off temporaries 1237 } 1238 1239 1240 void TemplateTable::lrem() { 1241 transition(ltos, ltos); 1242 __ pop_l(rbx, rcx); 1243 __ push(rcx); __ push(rbx); 1244 __ push(rdx); __ push(rax); 1245 // check if y = 0 1246 __ orl(rax, rdx); 1247 __ jump_cc(Assembler::zero, 1248 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1249 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem)); 1250 __ addptr(rsp, 4 * wordSize); 1251 } 1252 1253 1254 void TemplateTable::lshl() { 1255 transition(itos, ltos); 1256 __ movl(rcx, rax); // get shift count 1257 __ pop_l(rax, rdx); // get shift value 1258 __ lshl(rdx, rax); 1259 } 1260 1261 1262 void TemplateTable::lshr() { 1263 transition(itos, ltos); 1264 __ mov(rcx, rax); // get shift count 1265 __ pop_l(rax, rdx); // get shift value 1266 __ lshr(rdx, rax, true); 1267 } 1268 1269 1270 void TemplateTable::lushr() { 1271 transition(itos, ltos); 1272 __ mov(rcx, rax); // get shift count 1273 __ pop_l(rax, rdx); // get shift value 1274 __ lshr(rdx, rax); 1275 } 1276 1277 1278 void TemplateTable::fop2(Operation op) { 1279 transition(ftos, ftos); 1280 switch (op) { 1281 case add: __ fadd_s (at_rsp()); break; 1282 case sub: __ fsubr_s(at_rsp()); break; 1283 case mul: __ fmul_s (at_rsp()); break; 1284 case div: __ fdivr_s(at_rsp()); break; 1285 case rem: __ fld_s (at_rsp()); __ fremr(rax); break; 1286 default : ShouldNotReachHere(); 1287 } 1288 __ f2ieee(); 1289 __ pop(rax); // pop float thing off 1290 } 1291 1292 1293 void TemplateTable::dop2(Operation op) { 1294 transition(dtos, dtos); 1295 1296 switch (op) { 1297 case add: __ fadd_d (at_rsp()); break; 1298 case sub: __ fsubr_d(at_rsp()); break; 1299 case mul: { 1300 Label L_strict; 1301 Label L_join; 1302 const Address access_flags (rcx, Method::access_flags_offset()); 1303 __ get_method(rcx); 1304 __ movl(rcx, access_flags); 1305 __ testl(rcx, JVM_ACC_STRICT); 1306 __ jccb(Assembler::notZero, L_strict); 1307 __ fmul_d (at_rsp()); 1308 __ jmpb(L_join); 1309 __ bind(L_strict); 1310 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1311 __ fmulp(); 1312 __ fmul_d (at_rsp()); 1313 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1314 __ fmulp(); 1315 __ bind(L_join); 1316 break; 1317 } 1318 case div: { 1319 Label L_strict; 1320 Label L_join; 1321 const Address access_flags (rcx, Method::access_flags_offset()); 1322 __ get_method(rcx); 1323 __ movl(rcx, access_flags); 1324 __ testl(rcx, JVM_ACC_STRICT); 1325 __ jccb(Assembler::notZero, L_strict); 1326 __ fdivr_d(at_rsp()); 1327 __ jmp(L_join); 1328 __ bind(L_strict); 1329 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1330 __ fmul_d (at_rsp()); 1331 __ fdivrp(); 1332 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1333 __ fmulp(); 1334 __ bind(L_join); 1335 break; 1336 } 1337 case rem: __ fld_d (at_rsp()); __ fremr(rax); break; 1338 default : ShouldNotReachHere(); 1339 } 1340 __ d2ieee(); 1341 // Pop double precision number from rsp. 1342 __ pop(rax); 1343 __ pop(rdx); 1344 } 1345 1346 1347 void TemplateTable::ineg() { 1348 transition(itos, itos); 1349 __ negl(rax); 1350 } 1351 1352 1353 void TemplateTable::lneg() { 1354 transition(ltos, ltos); 1355 __ lneg(rdx, rax); 1356 } 1357 1358 1359 void TemplateTable::fneg() { 1360 transition(ftos, ftos); 1361 __ fchs(); 1362 } 1363 1364 1365 void TemplateTable::dneg() { 1366 transition(dtos, dtos); 1367 __ fchs(); 1368 } 1369 1370 1371 void TemplateTable::iinc() { 1372 transition(vtos, vtos); 1373 __ load_signed_byte(rdx, at_bcp(2)); // get constant 1374 locals_index(rbx); 1375 __ addl(iaddress(rbx), rdx); 1376 } 1377 1378 1379 void TemplateTable::wide_iinc() { 1380 transition(vtos, vtos); 1381 __ movl(rdx, at_bcp(4)); // get constant 1382 locals_index_wide(rbx); 1383 __ bswapl(rdx); // swap bytes & sign-extend constant 1384 __ sarl(rdx, 16); 1385 __ addl(iaddress(rbx), rdx); 1386 // Note: should probably use only one movl to get both 1387 // the index and the constant -> fix this 1388 } 1389 1390 1391 void TemplateTable::convert() { 1392 // Checking 1393 #ifdef ASSERT 1394 { TosState tos_in = ilgl; 1395 TosState tos_out = ilgl; 1396 switch (bytecode()) { 1397 case Bytecodes::_i2l: // fall through 1398 case Bytecodes::_i2f: // fall through 1399 case Bytecodes::_i2d: // fall through 1400 case Bytecodes::_i2b: // fall through 1401 case Bytecodes::_i2c: // fall through 1402 case Bytecodes::_i2s: tos_in = itos; break; 1403 case Bytecodes::_l2i: // fall through 1404 case Bytecodes::_l2f: // fall through 1405 case Bytecodes::_l2d: tos_in = ltos; break; 1406 case Bytecodes::_f2i: // fall through 1407 case Bytecodes::_f2l: // fall through 1408 case Bytecodes::_f2d: tos_in = ftos; break; 1409 case Bytecodes::_d2i: // fall through 1410 case Bytecodes::_d2l: // fall through 1411 case Bytecodes::_d2f: tos_in = dtos; break; 1412 default : ShouldNotReachHere(); 1413 } 1414 switch (bytecode()) { 1415 case Bytecodes::_l2i: // fall through 1416 case Bytecodes::_f2i: // fall through 1417 case Bytecodes::_d2i: // fall through 1418 case Bytecodes::_i2b: // fall through 1419 case Bytecodes::_i2c: // fall through 1420 case Bytecodes::_i2s: tos_out = itos; break; 1421 case Bytecodes::_i2l: // fall through 1422 case Bytecodes::_f2l: // fall through 1423 case Bytecodes::_d2l: tos_out = ltos; break; 1424 case Bytecodes::_i2f: // fall through 1425 case Bytecodes::_l2f: // fall through 1426 case Bytecodes::_d2f: tos_out = ftos; break; 1427 case Bytecodes::_i2d: // fall through 1428 case Bytecodes::_l2d: // fall through 1429 case Bytecodes::_f2d: tos_out = dtos; break; 1430 default : ShouldNotReachHere(); 1431 } 1432 transition(tos_in, tos_out); 1433 } 1434 #endif // ASSERT 1435 1436 // Conversion 1437 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation) 1438 switch (bytecode()) { 1439 case Bytecodes::_i2l: 1440 __ extend_sign(rdx, rax); 1441 break; 1442 case Bytecodes::_i2f: 1443 __ push(rax); // store int on tos 1444 __ fild_s(at_rsp()); // load int to ST0 1445 __ f2ieee(); // truncate to float size 1446 __ pop(rcx); // adjust rsp 1447 break; 1448 case Bytecodes::_i2d: 1449 __ push(rax); // add one slot for d2ieee() 1450 __ push(rax); // store int on tos 1451 __ fild_s(at_rsp()); // load int to ST0 1452 __ d2ieee(); // truncate to double size 1453 __ pop(rcx); // adjust rsp 1454 __ pop(rcx); 1455 break; 1456 case Bytecodes::_i2b: 1457 __ shll(rax, 24); // truncate upper 24 bits 1458 __ sarl(rax, 24); // and sign-extend byte 1459 LP64_ONLY(__ movsbl(rax, rax)); 1460 break; 1461 case Bytecodes::_i2c: 1462 __ andl(rax, 0xFFFF); // truncate upper 16 bits 1463 LP64_ONLY(__ movzwl(rax, rax)); 1464 break; 1465 case Bytecodes::_i2s: 1466 __ shll(rax, 16); // truncate upper 16 bits 1467 __ sarl(rax, 16); // and sign-extend short 1468 LP64_ONLY(__ movswl(rax, rax)); 1469 break; 1470 case Bytecodes::_l2i: 1471 /* nothing to do */ 1472 break; 1473 case Bytecodes::_l2f: 1474 __ push(rdx); // store long on tos 1475 __ push(rax); 1476 __ fild_d(at_rsp()); // load long to ST0 1477 __ f2ieee(); // truncate to float size 1478 __ pop(rcx); // adjust rsp 1479 __ pop(rcx); 1480 break; 1481 case Bytecodes::_l2d: 1482 __ push(rdx); // store long on tos 1483 __ push(rax); 1484 __ fild_d(at_rsp()); // load long to ST0 1485 __ d2ieee(); // truncate to double size 1486 __ pop(rcx); // adjust rsp 1487 __ pop(rcx); 1488 break; 1489 case Bytecodes::_f2i: 1490 __ push(rcx); // reserve space for argument 1491 __ fstp_s(at_rsp()); // pass float argument on stack 1492 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1493 break; 1494 case Bytecodes::_f2l: 1495 __ push(rcx); // reserve space for argument 1496 __ fstp_s(at_rsp()); // pass float argument on stack 1497 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1498 break; 1499 case Bytecodes::_f2d: 1500 /* nothing to do */ 1501 break; 1502 case Bytecodes::_d2i: 1503 __ push(rcx); // reserve space for argument 1504 __ push(rcx); 1505 __ fstp_d(at_rsp()); // pass double argument on stack 1506 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2); 1507 break; 1508 case Bytecodes::_d2l: 1509 __ push(rcx); // reserve space for argument 1510 __ push(rcx); 1511 __ fstp_d(at_rsp()); // pass double argument on stack 1512 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2); 1513 break; 1514 case Bytecodes::_d2f: 1515 __ push(rcx); // reserve space for f2ieee() 1516 __ f2ieee(); // truncate to float size 1517 __ pop(rcx); // adjust rsp 1518 break; 1519 default : 1520 ShouldNotReachHere(); 1521 } 1522 } 1523 1524 1525 void TemplateTable::lcmp() { 1526 transition(ltos, itos); 1527 // y = rdx:rax 1528 __ pop_l(rbx, rcx); // get x = rcx:rbx 1529 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y) 1530 __ mov(rax, rcx); 1531 } 1532 1533 1534 void TemplateTable::float_cmp(bool is_float, int unordered_result) { 1535 if (is_float) { 1536 __ fld_s(at_rsp()); 1537 } else { 1538 __ fld_d(at_rsp()); 1539 __ pop(rdx); 1540 } 1541 __ pop(rcx); 1542 __ fcmp2int(rax, unordered_result < 0); 1543 } 1544 1545 1546 void TemplateTable::branch(bool is_jsr, bool is_wide) { 1547 __ get_method(rcx); // ECX holds method 1548 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count 1549 1550 const ByteSize be_offset = MethodCounters::backedge_counter_offset() + 1551 InvocationCounter::counter_offset(); 1552 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() + 1553 InvocationCounter::counter_offset(); 1554 1555 // Load up EDX with the branch displacement 1556 if (is_wide) { 1557 __ movl(rdx, at_bcp(1)); 1558 } else { 1559 __ load_signed_short(rdx, at_bcp(1)); 1560 } 1561 __ bswapl(rdx); 1562 if (!is_wide) __ sarl(rdx, 16); 1563 LP64_ONLY(__ movslq(rdx, rdx)); 1564 1565 1566 // Handle all the JSR stuff here, then exit. 1567 // It's much shorter and cleaner than intermingling with the 1568 // non-JSR normal-branch stuff occurring below. 1569 if (is_jsr) { 1570 // Pre-load the next target bytecode into EBX 1571 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0)); 1572 1573 // compute return address as bci in rax, 1574 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(ConstMethod::codes_offset()))); 1575 __ subptr(rax, Address(rcx, Method::const_offset())); 1576 // Adjust the bcp in RSI by the displacement in EDX 1577 __ addptr(rsi, rdx); 1578 // Push return address 1579 __ push_i(rax); 1580 // jsr returns vtos 1581 __ dispatch_only_noverify(vtos); 1582 return; 1583 } 1584 1585 // Normal (non-jsr) branch handling 1586 1587 // Adjust the bcp in RSI by the displacement in EDX 1588 __ addptr(rsi, rdx); 1589 1590 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters"); 1591 Label backedge_counter_overflow; 1592 Label profile_method; 1593 Label dispatch; 1594 if (UseLoopCounter) { 1595 // increment backedge counter for backward branches 1596 // rax,: MDO 1597 // rbx,: MDO bumped taken-count 1598 // rcx: method 1599 // rdx: target offset 1600 // rsi: target bcp 1601 // rdi: locals pointer 1602 __ testl(rdx, rdx); // check if forward or backward branch 1603 __ jcc(Assembler::positive, dispatch); // count only if backward branch 1604 1605 // check if MethodCounters exists 1606 Label has_counters; 1607 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 1608 __ testptr(rax, rax); 1609 __ jcc(Assembler::notZero, has_counters); 1610 __ push(rdx); 1611 __ push(rcx); 1612 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), 1613 rcx); 1614 __ pop(rcx); 1615 __ pop(rdx); 1616 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 1617 __ testptr(rax, rax); 1618 __ jcc(Assembler::zero, dispatch); 1619 __ bind(has_counters); 1620 1621 if (TieredCompilation) { 1622 Label no_mdo; 1623 int increment = InvocationCounter::count_increment; 1624 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 1625 if (ProfileInterpreter) { 1626 // Are we profiling? 1627 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset()))); 1628 __ testptr(rbx, rbx); 1629 __ jccb(Assembler::zero, no_mdo); 1630 // Increment the MDO backedge counter 1631 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) + 1632 in_bytes(InvocationCounter::counter_offset())); 1633 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 1634 rax, false, Assembler::zero, &backedge_counter_overflow); 1635 __ jmp(dispatch); 1636 } 1637 __ bind(no_mdo); 1638 // Increment backedge counter in MethodCounters* 1639 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 1640 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask, 1641 rax, false, Assembler::zero, &backedge_counter_overflow); 1642 } else { 1643 // increment counter 1644 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 1645 __ movl(rax, Address(rcx, be_offset)); // load backedge counter 1646 __ incrementl(rax, InvocationCounter::count_increment); // increment counter 1647 __ movl(Address(rcx, be_offset), rax); // store counter 1648 1649 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter 1650 1651 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits 1652 __ addl(rax, Address(rcx, be_offset)); // add both counters 1653 1654 if (ProfileInterpreter) { 1655 // Test to see if we should create a method data oop 1656 __ cmp32(rax, 1657 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit)); 1658 __ jcc(Assembler::less, dispatch); 1659 1660 // if no method data exists, go to profile method 1661 __ test_method_data_pointer(rax, profile_method); 1662 1663 if (UseOnStackReplacement) { 1664 // check for overflow against rbx, which is the MDO taken count 1665 __ cmp32(rbx, 1666 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1667 __ jcc(Assembler::below, dispatch); 1668 1669 // When ProfileInterpreter is on, the backedge_count comes from the 1670 // MethodData*, which value does not get reset on the call to 1671 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1672 // routine while the method is being compiled, add a second test to make 1673 // sure the overflow function is called only once every overflow_frequency. 1674 const int overflow_frequency = 1024; 1675 __ andptr(rbx, overflow_frequency-1); 1676 __ jcc(Assembler::zero, backedge_counter_overflow); 1677 } 1678 } else { 1679 if (UseOnStackReplacement) { 1680 // check for overflow against rax, which is the sum of the counters 1681 __ cmp32(rax, 1682 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1683 __ jcc(Assembler::aboveEqual, backedge_counter_overflow); 1684 1685 } 1686 } 1687 } 1688 __ bind(dispatch); 1689 } 1690 1691 // Pre-load the next target bytecode into EBX 1692 __ load_unsigned_byte(rbx, Address(rsi, 0)); 1693 1694 // continue with the bytecode @ target 1695 // rax,: return bci for jsr's, unused otherwise 1696 // rbx,: target bytecode 1697 // rsi: target bcp 1698 __ dispatch_only(vtos); 1699 1700 if (UseLoopCounter) { 1701 if (ProfileInterpreter) { 1702 // Out-of-line code to allocate method data oop. 1703 __ bind(profile_method); 1704 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1705 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1706 __ set_method_data_pointer_for_bcp(); 1707 __ jmp(dispatch); 1708 } 1709 1710 if (UseOnStackReplacement) { 1711 1712 // invocation counter overflow 1713 __ bind(backedge_counter_overflow); 1714 __ negptr(rdx); 1715 __ addptr(rdx, rsi); // branch bcp 1716 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx); 1717 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1718 1719 // rax,: osr nmethod (osr ok) or NULL (osr not possible) 1720 // rbx,: target bytecode 1721 // rdx: scratch 1722 // rdi: locals pointer 1723 // rsi: bcp 1724 __ testptr(rax, rax); // test result 1725 __ jcc(Assembler::zero, dispatch); // no osr if null 1726 // nmethod may have been invalidated (VM may block upon call_VM return) 1727 __ movl(rcx, Address(rax, nmethod::entry_bci_offset())); 1728 __ cmpl(rcx, InvalidOSREntryBci); 1729 __ jcc(Assembler::equal, dispatch); 1730 1731 // We have the address of an on stack replacement routine in rax, 1732 // We need to prepare to execute the OSR method. First we must 1733 // migrate the locals and monitors off of the stack. 1734 1735 __ mov(rbx, rax); // save the nmethod 1736 1737 const Register thread = rcx; 1738 __ get_thread(thread); 1739 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 1740 // rax, is OSR buffer, move it to expected parameter location 1741 __ mov(rcx, rax); 1742 1743 // pop the interpreter frame 1744 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 1745 __ leave(); // remove frame anchor 1746 __ pop(rdi); // get return address 1747 __ mov(rsp, rdx); // set sp to sender sp 1748 1749 // Align stack pointer for compiled code (note that caller is 1750 // responsible for undoing this fixup by remembering the old SP 1751 // in an rbp,-relative location) 1752 __ andptr(rsp, -(StackAlignmentInBytes)); 1753 1754 // push the (possibly adjusted) return address 1755 __ push(rdi); 1756 1757 // and begin the OSR nmethod 1758 __ jmp(Address(rbx, nmethod::osr_entry_point_offset())); 1759 } 1760 } 1761 } 1762 1763 1764 void TemplateTable::if_0cmp(Condition cc) { 1765 transition(itos, vtos); 1766 // assume branch is more often taken than not (loops use backward branches) 1767 Label not_taken; 1768 __ testl(rax, rax); 1769 __ jcc(j_not(cc), not_taken); 1770 branch(false, false); 1771 __ bind(not_taken); 1772 __ profile_not_taken_branch(rax); 1773 } 1774 1775 1776 void TemplateTable::if_icmp(Condition cc) { 1777 transition(itos, vtos); 1778 // assume branch is more often taken than not (loops use backward branches) 1779 Label not_taken; 1780 __ pop_i(rdx); 1781 __ cmpl(rdx, rax); 1782 __ jcc(j_not(cc), not_taken); 1783 branch(false, false); 1784 __ bind(not_taken); 1785 __ profile_not_taken_branch(rax); 1786 } 1787 1788 1789 void TemplateTable::if_nullcmp(Condition cc) { 1790 transition(atos, vtos); 1791 // assume branch is more often taken than not (loops use backward branches) 1792 Label not_taken; 1793 __ testptr(rax, rax); 1794 __ jcc(j_not(cc), not_taken); 1795 branch(false, false); 1796 __ bind(not_taken); 1797 __ profile_not_taken_branch(rax); 1798 } 1799 1800 1801 void TemplateTable::if_acmp(Condition cc) { 1802 transition(atos, vtos); 1803 // assume branch is more often taken than not (loops use backward branches) 1804 Label not_taken; 1805 __ pop_ptr(rdx); 1806 __ cmpptr(rdx, rax); 1807 __ jcc(j_not(cc), not_taken); 1808 branch(false, false); 1809 __ bind(not_taken); 1810 __ profile_not_taken_branch(rax); 1811 } 1812 1813 1814 void TemplateTable::ret() { 1815 transition(vtos, vtos); 1816 locals_index(rbx); 1817 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1818 __ profile_ret(rbx, rcx); 1819 __ get_method(rax); 1820 __ movptr(rsi, Address(rax, Method::const_offset())); 1821 __ lea(rsi, Address(rsi, rbx, Address::times_1, 1822 ConstMethod::codes_offset())); 1823 __ dispatch_next(vtos); 1824 } 1825 1826 1827 void TemplateTable::wide_ret() { 1828 transition(vtos, vtos); 1829 locals_index_wide(rbx); 1830 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1831 __ profile_ret(rbx, rcx); 1832 __ get_method(rax); 1833 __ movptr(rsi, Address(rax, Method::const_offset())); 1834 __ lea(rsi, Address(rsi, rbx, Address::times_1, ConstMethod::codes_offset())); 1835 __ dispatch_next(vtos); 1836 } 1837 1838 1839 void TemplateTable::tableswitch() { 1840 Label default_case, continue_execution; 1841 transition(itos, vtos); 1842 // align rsi 1843 __ lea(rbx, at_bcp(wordSize)); 1844 __ andptr(rbx, -wordSize); 1845 // load lo & hi 1846 __ movl(rcx, Address(rbx, 1 * wordSize)); 1847 __ movl(rdx, Address(rbx, 2 * wordSize)); 1848 __ bswapl(rcx); 1849 __ bswapl(rdx); 1850 // check against lo & hi 1851 __ cmpl(rax, rcx); 1852 __ jccb(Assembler::less, default_case); 1853 __ cmpl(rax, rdx); 1854 __ jccb(Assembler::greater, default_case); 1855 // lookup dispatch offset 1856 __ subl(rax, rcx); 1857 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt)); 1858 __ profile_switch_case(rax, rbx, rcx); 1859 // continue execution 1860 __ bind(continue_execution); 1861 __ bswapl(rdx); 1862 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1863 __ addptr(rsi, rdx); 1864 __ dispatch_only(vtos); 1865 // handle default 1866 __ bind(default_case); 1867 __ profile_switch_default(rax); 1868 __ movl(rdx, Address(rbx, 0)); 1869 __ jmp(continue_execution); 1870 } 1871 1872 1873 void TemplateTable::lookupswitch() { 1874 transition(itos, itos); 1875 __ stop("lookupswitch bytecode should have been rewritten"); 1876 } 1877 1878 1879 void TemplateTable::fast_linearswitch() { 1880 transition(itos, vtos); 1881 Label loop_entry, loop, found, continue_execution; 1882 // bswapl rax, so we can avoid bswapping the table entries 1883 __ bswapl(rax); 1884 // align rsi 1885 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1886 __ andptr(rbx, -wordSize); 1887 // set counter 1888 __ movl(rcx, Address(rbx, wordSize)); 1889 __ bswapl(rcx); 1890 __ jmpb(loop_entry); 1891 // table search 1892 __ bind(loop); 1893 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize)); 1894 __ jccb(Assembler::equal, found); 1895 __ bind(loop_entry); 1896 __ decrementl(rcx); 1897 __ jcc(Assembler::greaterEqual, loop); 1898 // default case 1899 __ profile_switch_default(rax); 1900 __ movl(rdx, Address(rbx, 0)); 1901 __ jmpb(continue_execution); 1902 // entry found -> get offset 1903 __ bind(found); 1904 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize)); 1905 __ profile_switch_case(rcx, rax, rbx); 1906 // continue execution 1907 __ bind(continue_execution); 1908 __ bswapl(rdx); 1909 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1910 __ addptr(rsi, rdx); 1911 __ dispatch_only(vtos); 1912 } 1913 1914 1915 void TemplateTable::fast_binaryswitch() { 1916 transition(itos, vtos); 1917 // Implementation using the following core algorithm: 1918 // 1919 // int binary_search(int key, LookupswitchPair* array, int n) { 1920 // // Binary search according to "Methodik des Programmierens" by 1921 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 1922 // int i = 0; 1923 // int j = n; 1924 // while (i+1 < j) { 1925 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 1926 // // with Q: for all i: 0 <= i < n: key < a[i] 1927 // // where a stands for the array and assuming that the (inexisting) 1928 // // element a[n] is infinitely big. 1929 // int h = (i + j) >> 1; 1930 // // i < h < j 1931 // if (key < array[h].fast_match()) { 1932 // j = h; 1933 // } else { 1934 // i = h; 1935 // } 1936 // } 1937 // // R: a[i] <= key < a[i+1] or Q 1938 // // (i.e., if key is within array, i is the correct index) 1939 // return i; 1940 // } 1941 1942 // register allocation 1943 const Register key = rax; // already set (tosca) 1944 const Register array = rbx; 1945 const Register i = rcx; 1946 const Register j = rdx; 1947 const Register h = rdi; // needs to be restored 1948 const Register temp = rsi; 1949 // setup array 1950 __ save_bcp(); 1951 1952 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1953 __ andptr(array, -wordSize); 1954 // initialize i & j 1955 __ xorl(i, i); // i = 0; 1956 __ movl(j, Address(array, -wordSize)); // j = length(array); 1957 // Convert j into native byteordering 1958 __ bswapl(j); 1959 // and start 1960 Label entry; 1961 __ jmp(entry); 1962 1963 // binary search loop 1964 { Label loop; 1965 __ bind(loop); 1966 // int h = (i + j) >> 1; 1967 __ leal(h, Address(i, j, Address::times_1)); // h = i + j; 1968 __ sarl(h, 1); // h = (i + j) >> 1; 1969 // if (key < array[h].fast_match()) { 1970 // j = h; 1971 // } else { 1972 // i = h; 1973 // } 1974 // Convert array[h].match to native byte-ordering before compare 1975 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize)); 1976 __ bswapl(temp); 1977 __ cmpl(key, temp); 1978 // j = h if (key < array[h].fast_match()) 1979 __ cmov32(Assembler::less , j, h); 1980 // i = h if (key >= array[h].fast_match()) 1981 __ cmov32(Assembler::greaterEqual, i, h); 1982 // while (i+1 < j) 1983 __ bind(entry); 1984 __ leal(h, Address(i, 1)); // i+1 1985 __ cmpl(h, j); // i+1 < j 1986 __ jcc(Assembler::less, loop); 1987 } 1988 1989 // end of binary search, result index is i (must check again!) 1990 Label default_case; 1991 // Convert array[i].match to native byte-ordering before compare 1992 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize)); 1993 __ bswapl(temp); 1994 __ cmpl(key, temp); 1995 __ jcc(Assembler::notEqual, default_case); 1996 1997 // entry found -> j = offset 1998 __ movl(j , Address(array, i, Address::times_8, 1*wordSize)); 1999 __ profile_switch_case(i, key, array); 2000 __ bswapl(j); 2001 LP64_ONLY(__ movslq(j, j)); 2002 __ restore_bcp(); 2003 __ restore_locals(); // restore rdi 2004 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 2005 2006 __ addptr(rsi, j); 2007 __ dispatch_only(vtos); 2008 2009 // default case -> j = default offset 2010 __ bind(default_case); 2011 __ profile_switch_default(i); 2012 __ movl(j, Address(array, -2*wordSize)); 2013 __ bswapl(j); 2014 LP64_ONLY(__ movslq(j, j)); 2015 __ restore_bcp(); 2016 __ restore_locals(); // restore rdi 2017 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 2018 __ addptr(rsi, j); 2019 __ dispatch_only(vtos); 2020 } 2021 2022 2023 void TemplateTable::_return(TosState state) { 2024 transition(state, state); 2025 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation 2026 2027 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 2028 assert(state == vtos, "only valid state"); 2029 __ movptr(rax, aaddress(0)); 2030 __ load_klass(rdi, rax); 2031 __ movl(rdi, Address(rdi, Klass::access_flags_offset())); 2032 __ testl(rdi, JVM_ACC_HAS_FINALIZER); 2033 Label skip_register_finalizer; 2034 __ jcc(Assembler::zero, skip_register_finalizer); 2035 2036 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax); 2037 2038 __ bind(skip_register_finalizer); 2039 } 2040 2041 __ remove_activation(state, rsi); 2042 __ jmp(rsi); 2043 } 2044 2045 2046 // ---------------------------------------------------------------------------- 2047 // Volatile variables demand their effects be made known to all CPU's in 2048 // order. Store buffers on most chips allow reads & writes to reorder; the 2049 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of 2050 // memory barrier (i.e., it's not sufficient that the interpreter does not 2051 // reorder volatile references, the hardware also must not reorder them). 2052 // 2053 // According to the new Java Memory Model (JMM): 2054 // (1) All volatiles are serialized wrt to each other. 2055 // ALSO reads & writes act as aquire & release, so: 2056 // (2) A read cannot let unrelated NON-volatile memory refs that happen after 2057 // the read float up to before the read. It's OK for non-volatile memory refs 2058 // that happen before the volatile read to float down below it. 2059 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs 2060 // that happen BEFORE the write float down to after the write. It's OK for 2061 // non-volatile memory refs that happen after the volatile write to float up 2062 // before it. 2063 // 2064 // We only put in barriers around volatile refs (they are expensive), not 2065 // _between_ memory refs (that would require us to track the flavor of the 2066 // previous memory refs). Requirements (2) and (3) require some barriers 2067 // before volatile stores and after volatile loads. These nearly cover 2068 // requirement (1) but miss the volatile-store-volatile-load case. This final 2069 // case is placed after volatile-stores although it could just as well go 2070 // before volatile-loads. 2071 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) { 2072 // Helper function to insert a is-volatile test and memory barrier 2073 if( !os::is_MP() ) return; // Not needed on single CPU 2074 __ membar(order_constraint); 2075 } 2076 2077 void TemplateTable::resolve_cache_and_index(int byte_no, 2078 Register Rcache, 2079 Register index, 2080 size_t index_size) { 2081 const Register temp = rbx; 2082 assert_different_registers(Rcache, index, temp); 2083 2084 Label resolved; 2085 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 2086 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size); 2087 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode? 2088 __ jcc(Assembler::equal, resolved); 2089 2090 // resolve first time through 2091 address entry; 2092 switch (bytecode()) { 2093 case Bytecodes::_getstatic : // fall through 2094 case Bytecodes::_putstatic : // fall through 2095 case Bytecodes::_getfield : // fall through 2096 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break; 2097 case Bytecodes::_invokevirtual : // fall through 2098 case Bytecodes::_invokespecial : // fall through 2099 case Bytecodes::_invokestatic : // fall through 2100 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break; 2101 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break; 2102 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break; 2103 default: 2104 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode()))); 2105 break; 2106 } 2107 __ movl(temp, (int)bytecode()); 2108 __ call_VM(noreg, entry, temp); 2109 // Update registers with resolved info 2110 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2111 __ bind(resolved); 2112 } 2113 2114 2115 // The cache and index registers must be set before call 2116 void TemplateTable::load_field_cp_cache_entry(Register obj, 2117 Register cache, 2118 Register index, 2119 Register off, 2120 Register flags, 2121 bool is_static = false) { 2122 assert_different_registers(cache, index, flags, off); 2123 2124 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2125 // Field offset 2126 __ movptr(off, Address(cache, index, Address::times_ptr, 2127 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()))); 2128 // Flags 2129 __ movl(flags, Address(cache, index, Address::times_ptr, 2130 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()))); 2131 2132 // klass overwrite register 2133 if (is_static) { 2134 __ movptr(obj, Address(cache, index, Address::times_ptr, 2135 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset()))); 2136 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 2137 __ movptr(obj, Address(obj, mirror_offset)); 2138 } 2139 } 2140 2141 void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 2142 Register method, 2143 Register itable_index, 2144 Register flags, 2145 bool is_invokevirtual, 2146 bool is_invokevfinal, /*unused*/ 2147 bool is_invokedynamic) { 2148 // setup registers 2149 const Register cache = rcx; 2150 const Register index = rdx; 2151 assert_different_registers(method, flags); 2152 assert_different_registers(method, cache, index); 2153 assert_different_registers(itable_index, flags); 2154 assert_different_registers(itable_index, cache, index); 2155 // determine constant pool cache field offsets 2156 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant"); 2157 const int method_offset = in_bytes( 2158 ConstantPoolCache::base_offset() + 2159 ((byte_no == f2_byte) 2160 ? ConstantPoolCacheEntry::f2_offset() 2161 : ConstantPoolCacheEntry::f1_offset())); 2162 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() + 2163 ConstantPoolCacheEntry::flags_offset()); 2164 // access constant pool cache fields 2165 const int index_offset = in_bytes(ConstantPoolCache::base_offset() + 2166 ConstantPoolCacheEntry::f2_offset()); 2167 2168 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2)); 2169 resolve_cache_and_index(byte_no, cache, index, index_size); 2170 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset)); 2171 2172 if (itable_index != noreg) { 2173 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset)); 2174 } 2175 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset)); 2176 } 2177 2178 2179 // The registers cache and index expected to be set before call. 2180 // Correct values of the cache and index registers are preserved. 2181 void TemplateTable::jvmti_post_field_access(Register cache, 2182 Register index, 2183 bool is_static, 2184 bool has_tos) { 2185 if (JvmtiExport::can_post_field_access()) { 2186 // Check to see if a field access watch has been set before we take 2187 // the time to call into the VM. 2188 Label L1; 2189 assert_different_registers(cache, index, rax); 2190 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2191 __ testl(rax,rax); 2192 __ jcc(Assembler::zero, L1); 2193 2194 // cache entry pointer 2195 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 2196 __ shll(index, LogBytesPerWord); 2197 __ addptr(cache, index); 2198 if (is_static) { 2199 __ xorptr(rax, rax); // NULL object reference 2200 } else { 2201 __ pop(atos); // Get the object 2202 __ verify_oop(rax); 2203 __ push(atos); // Restore stack state 2204 } 2205 // rax,: object pointer or NULL 2206 // cache: cache entry pointer 2207 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), 2208 rax, cache); 2209 __ get_cache_and_index_at_bcp(cache, index, 1); 2210 __ bind(L1); 2211 } 2212 } 2213 2214 void TemplateTable::pop_and_check_object(Register r) { 2215 __ pop_ptr(r); 2216 __ null_check(r); // for field access must check obj. 2217 __ verify_oop(r); 2218 } 2219 2220 void TemplateTable::getfield_or_static(int byte_no, bool is_static) { 2221 transition(vtos, vtos); 2222 2223 const Register cache = rcx; 2224 const Register index = rdx; 2225 const Register obj = rcx; 2226 const Register off = rbx; 2227 const Register flags = rax; 2228 2229 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2230 jvmti_post_field_access(cache, index, is_static, false); 2231 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2232 2233 if (!is_static) pop_and_check_object(obj); 2234 2235 const Address lo(obj, off, Address::times_1, 0*wordSize); 2236 const Address hi(obj, off, Address::times_1, 1*wordSize); 2237 2238 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2239 2240 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2241 assert(btos == 0, "change code, btos != 0"); 2242 // btos 2243 __ andptr(flags, ConstantPoolCacheEntry::tos_state_mask); 2244 __ jcc(Assembler::notZero, notByte); 2245 2246 __ load_signed_byte(rax, lo ); 2247 __ push(btos); 2248 // Rewrite bytecode to be faster 2249 if (!is_static) { 2250 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx); 2251 } 2252 __ jmp(Done); 2253 2254 __ bind(notByte); 2255 // itos 2256 __ cmpl(flags, itos ); 2257 __ jcc(Assembler::notEqual, notInt); 2258 2259 __ movl(rax, lo ); 2260 __ push(itos); 2261 // Rewrite bytecode to be faster 2262 if (!is_static) { 2263 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx); 2264 } 2265 __ jmp(Done); 2266 2267 __ bind(notInt); 2268 // atos 2269 __ cmpl(flags, atos ); 2270 __ jcc(Assembler::notEqual, notObj); 2271 2272 __ movl(rax, lo ); 2273 __ push(atos); 2274 if (!is_static) { 2275 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx); 2276 } 2277 __ jmp(Done); 2278 2279 __ bind(notObj); 2280 // ctos 2281 __ cmpl(flags, ctos ); 2282 __ jcc(Assembler::notEqual, notChar); 2283 2284 __ load_unsigned_short(rax, lo ); 2285 __ push(ctos); 2286 if (!is_static) { 2287 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx); 2288 } 2289 __ jmp(Done); 2290 2291 __ bind(notChar); 2292 // stos 2293 __ cmpl(flags, stos ); 2294 __ jcc(Assembler::notEqual, notShort); 2295 2296 __ load_signed_short(rax, lo ); 2297 __ push(stos); 2298 if (!is_static) { 2299 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx); 2300 } 2301 __ jmp(Done); 2302 2303 __ bind(notShort); 2304 // ltos 2305 __ cmpl(flags, ltos ); 2306 __ jcc(Assembler::notEqual, notLong); 2307 2308 // Generate code as if volatile. There just aren't enough registers to 2309 // save that information and this code is faster than the test. 2310 __ fild_d(lo); // Must load atomically 2311 __ subptr(rsp,2*wordSize); // Make space for store 2312 __ fistp_d(Address(rsp,0)); 2313 __ pop(rax); 2314 __ pop(rdx); 2315 2316 __ push(ltos); 2317 // Don't rewrite to _fast_lgetfield for potential volatile case. 2318 __ jmp(Done); 2319 2320 __ bind(notLong); 2321 // ftos 2322 __ cmpl(flags, ftos ); 2323 __ jcc(Assembler::notEqual, notFloat); 2324 2325 __ fld_s(lo); 2326 __ push(ftos); 2327 if (!is_static) { 2328 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx); 2329 } 2330 __ jmp(Done); 2331 2332 __ bind(notFloat); 2333 // dtos 2334 __ cmpl(flags, dtos ); 2335 __ jcc(Assembler::notEqual, notDouble); 2336 2337 __ fld_d(lo); 2338 __ push(dtos); 2339 if (!is_static) { 2340 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx); 2341 } 2342 __ jmpb(Done); 2343 2344 __ bind(notDouble); 2345 2346 __ stop("Bad state"); 2347 2348 __ bind(Done); 2349 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2350 // volatile_barrier( ); 2351 } 2352 2353 2354 void TemplateTable::getfield(int byte_no) { 2355 getfield_or_static(byte_no, false); 2356 } 2357 2358 2359 void TemplateTable::getstatic(int byte_no) { 2360 getfield_or_static(byte_no, true); 2361 } 2362 2363 // The registers cache and index expected to be set before call. 2364 // The function may destroy various registers, just not the cache and index registers. 2365 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 2366 2367 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2368 2369 if (JvmtiExport::can_post_field_modification()) { 2370 // Check to see if a field modification watch has been set before we take 2371 // the time to call into the VM. 2372 Label L1; 2373 assert_different_registers(cache, index, rax); 2374 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2375 __ testl(rax, rax); 2376 __ jcc(Assembler::zero, L1); 2377 2378 // The cache and index registers have been already set. 2379 // This allows to eliminate this call but the cache and index 2380 // registers have to be correspondingly used after this line. 2381 __ get_cache_and_index_at_bcp(rax, rdx, 1); 2382 2383 if (is_static) { 2384 // Life is simple. Null out the object pointer. 2385 __ xorptr(rbx, rbx); 2386 } else { 2387 // Life is harder. The stack holds the value on top, followed by the object. 2388 // We don't know the size of the value, though; it could be one or two words 2389 // depending on its type. As a result, we must find the type to determine where 2390 // the object is. 2391 Label two_word, valsize_known; 2392 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset + 2393 ConstantPoolCacheEntry::flags_offset()))); 2394 __ mov(rbx, rsp); 2395 __ shrl(rcx, ConstantPoolCacheEntry::tos_state_shift); 2396 // Make sure we don't need to mask rcx after the above shift 2397 ConstantPoolCacheEntry::verify_tos_state_shift(); 2398 __ cmpl(rcx, ltos); 2399 __ jccb(Assembler::equal, two_word); 2400 __ cmpl(rcx, dtos); 2401 __ jccb(Assembler::equal, two_word); 2402 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos) 2403 __ jmpb(valsize_known); 2404 2405 __ bind(two_word); 2406 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue 2407 2408 __ bind(valsize_known); 2409 // setup object pointer 2410 __ movptr(rbx, Address(rbx, 0)); 2411 } 2412 // cache entry pointer 2413 __ addptr(rax, in_bytes(cp_base_offset)); 2414 __ shll(rdx, LogBytesPerWord); 2415 __ addptr(rax, rdx); 2416 // object (tos) 2417 __ mov(rcx, rsp); 2418 // rbx,: object pointer set up above (NULL if static) 2419 // rax,: cache entry pointer 2420 // rcx: jvalue object on the stack 2421 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), 2422 rbx, rax, rcx); 2423 __ get_cache_and_index_at_bcp(cache, index, 1); 2424 __ bind(L1); 2425 } 2426 } 2427 2428 2429 void TemplateTable::putfield_or_static(int byte_no, bool is_static) { 2430 transition(vtos, vtos); 2431 2432 const Register cache = rcx; 2433 const Register index = rdx; 2434 const Register obj = rcx; 2435 const Register off = rbx; 2436 const Register flags = rax; 2437 2438 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2439 jvmti_post_field_mod(cache, index, is_static); 2440 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2441 2442 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2443 // volatile_barrier( ); 2444 2445 Label notVolatile, Done; 2446 __ movl(rdx, flags); 2447 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 2448 __ andl(rdx, 0x1); 2449 2450 // field addresses 2451 const Address lo(obj, off, Address::times_1, 0*wordSize); 2452 const Address hi(obj, off, Address::times_1, 1*wordSize); 2453 2454 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2455 2456 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2457 assert(btos == 0, "change code, btos != 0"); 2458 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 2459 __ jcc(Assembler::notZero, notByte); 2460 2461 // btos 2462 { 2463 __ pop(btos); 2464 if (!is_static) pop_and_check_object(obj); 2465 __ movb(lo, rax); 2466 if (!is_static) { 2467 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx, true, byte_no); 2468 } 2469 __ jmp(Done); 2470 } 2471 2472 __ bind(notByte); 2473 __ cmpl(flags, itos); 2474 __ jcc(Assembler::notEqual, notInt); 2475 2476 // itos 2477 { 2478 __ pop(itos); 2479 if (!is_static) pop_and_check_object(obj); 2480 __ movl(lo, rax); 2481 if (!is_static) { 2482 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx, true, byte_no); 2483 } 2484 __ jmp(Done); 2485 } 2486 2487 __ bind(notInt); 2488 __ cmpl(flags, atos); 2489 __ jcc(Assembler::notEqual, notObj); 2490 2491 // atos 2492 { 2493 __ pop(atos); 2494 if (!is_static) pop_and_check_object(obj); 2495 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2496 if (!is_static) { 2497 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx, true, byte_no); 2498 } 2499 __ jmp(Done); 2500 } 2501 2502 __ bind(notObj); 2503 __ cmpl(flags, ctos); 2504 __ jcc(Assembler::notEqual, notChar); 2505 2506 // ctos 2507 { 2508 __ pop(ctos); 2509 if (!is_static) pop_and_check_object(obj); 2510 __ movw(lo, rax); 2511 if (!is_static) { 2512 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx, true, byte_no); 2513 } 2514 __ jmp(Done); 2515 } 2516 2517 __ bind(notChar); 2518 __ cmpl(flags, stos); 2519 __ jcc(Assembler::notEqual, notShort); 2520 2521 // stos 2522 { 2523 __ pop(stos); 2524 if (!is_static) pop_and_check_object(obj); 2525 __ movw(lo, rax); 2526 if (!is_static) { 2527 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx, true, byte_no); 2528 } 2529 __ jmp(Done); 2530 } 2531 2532 __ bind(notShort); 2533 __ cmpl(flags, ltos); 2534 __ jcc(Assembler::notEqual, notLong); 2535 2536 // ltos 2537 { 2538 Label notVolatileLong; 2539 __ testl(rdx, rdx); 2540 __ jcc(Assembler::zero, notVolatileLong); 2541 2542 __ pop(ltos); // overwrites rdx, do this after testing volatile. 2543 if (!is_static) pop_and_check_object(obj); 2544 2545 // Replace with real volatile test 2546 __ push(rdx); 2547 __ push(rax); // Must update atomically with FIST 2548 __ fild_d(Address(rsp,0)); // So load into FPU register 2549 __ fistp_d(lo); // and put into memory atomically 2550 __ addptr(rsp, 2*wordSize); 2551 // volatile_barrier(); 2552 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2553 Assembler::StoreStore)); 2554 // Don't rewrite volatile version 2555 __ jmp(notVolatile); 2556 2557 __ bind(notVolatileLong); 2558 2559 __ pop(ltos); // overwrites rdx 2560 if (!is_static) pop_and_check_object(obj); 2561 NOT_LP64(__ movptr(hi, rdx)); 2562 __ movptr(lo, rax); 2563 if (!is_static) { 2564 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx, true, byte_no); 2565 } 2566 __ jmp(notVolatile); 2567 } 2568 2569 __ bind(notLong); 2570 __ cmpl(flags, ftos); 2571 __ jcc(Assembler::notEqual, notFloat); 2572 2573 // ftos 2574 { 2575 __ pop(ftos); 2576 if (!is_static) pop_and_check_object(obj); 2577 __ fstp_s(lo); 2578 if (!is_static) { 2579 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx, true, byte_no); 2580 } 2581 __ jmp(Done); 2582 } 2583 2584 __ bind(notFloat); 2585 #ifdef ASSERT 2586 __ cmpl(flags, dtos); 2587 __ jcc(Assembler::notEqual, notDouble); 2588 #endif 2589 2590 // dtos 2591 { 2592 __ pop(dtos); 2593 if (!is_static) pop_and_check_object(obj); 2594 __ fstp_d(lo); 2595 if (!is_static) { 2596 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx, true, byte_no); 2597 } 2598 __ jmp(Done); 2599 } 2600 2601 #ifdef ASSERT 2602 __ bind(notDouble); 2603 __ stop("Bad state"); 2604 #endif 2605 2606 __ bind(Done); 2607 2608 // Check for volatile store 2609 __ testl(rdx, rdx); 2610 __ jcc(Assembler::zero, notVolatile); 2611 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2612 Assembler::StoreStore)); 2613 __ bind(notVolatile); 2614 } 2615 2616 2617 void TemplateTable::putfield(int byte_no) { 2618 putfield_or_static(byte_no, false); 2619 } 2620 2621 2622 void TemplateTable::putstatic(int byte_no) { 2623 putfield_or_static(byte_no, true); 2624 } 2625 2626 void TemplateTable::jvmti_post_fast_field_mod() { 2627 if (JvmtiExport::can_post_field_modification()) { 2628 // Check to see if a field modification watch has been set before we take 2629 // the time to call into the VM. 2630 Label L2; 2631 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2632 __ testl(rcx,rcx); 2633 __ jcc(Assembler::zero, L2); 2634 __ pop_ptr(rbx); // copy the object pointer from tos 2635 __ verify_oop(rbx); 2636 __ push_ptr(rbx); // put the object pointer back on tos 2637 2638 // Save tos values before call_VM() clobbers them. Since we have 2639 // to do it for every data type, we use the saved values as the 2640 // jvalue object. 2641 switch (bytecode()) { // load values into the jvalue object 2642 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break; 2643 case Bytecodes::_fast_bputfield: // fall through 2644 case Bytecodes::_fast_sputfield: // fall through 2645 case Bytecodes::_fast_cputfield: // fall through 2646 case Bytecodes::_fast_iputfield: __ push_i(rax); break; 2647 case Bytecodes::_fast_dputfield: __ push_d(); break; 2648 case Bytecodes::_fast_fputfield: __ push_f(); break; 2649 case Bytecodes::_fast_lputfield: __ push_l(rax); break; 2650 2651 default: 2652 ShouldNotReachHere(); 2653 } 2654 __ mov(rcx, rsp); // points to jvalue on the stack 2655 // access constant pool cache entry 2656 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1); 2657 __ verify_oop(rbx); 2658 // rbx,: object pointer copied above 2659 // rax,: cache entry pointer 2660 // rcx: jvalue object on the stack 2661 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx); 2662 2663 switch (bytecode()) { // restore tos values 2664 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break; 2665 case Bytecodes::_fast_bputfield: // fall through 2666 case Bytecodes::_fast_sputfield: // fall through 2667 case Bytecodes::_fast_cputfield: // fall through 2668 case Bytecodes::_fast_iputfield: __ pop_i(rax); break; 2669 case Bytecodes::_fast_dputfield: __ pop_d(); break; 2670 case Bytecodes::_fast_fputfield: __ pop_f(); break; 2671 case Bytecodes::_fast_lputfield: __ pop_l(rax); break; 2672 } 2673 __ bind(L2); 2674 } 2675 } 2676 2677 void TemplateTable::fast_storefield(TosState state) { 2678 transition(state, vtos); 2679 2680 ByteSize base = ConstantPoolCache::base_offset(); 2681 2682 jvmti_post_fast_field_mod(); 2683 2684 // access constant pool cache 2685 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 2686 2687 // test for volatile with rdx but rdx is tos register for lputfield. 2688 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx); 2689 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + 2690 ConstantPoolCacheEntry::flags_offset()))); 2691 2692 // replace index with field offset from cache entry 2693 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 2694 2695 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 2696 // volatile_barrier( ); 2697 2698 Label notVolatile, Done; 2699 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 2700 __ andl(rdx, 0x1); 2701 // Check for volatile store 2702 __ testl(rdx, rdx); 2703 __ jcc(Assembler::zero, notVolatile); 2704 2705 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 2706 2707 // Get object from stack 2708 pop_and_check_object(rcx); 2709 2710 // field addresses 2711 const Address lo(rcx, rbx, Address::times_1, 0*wordSize); 2712 const Address hi(rcx, rbx, Address::times_1, 1*wordSize); 2713 2714 // access field 2715 switch (bytecode()) { 2716 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 2717 case Bytecodes::_fast_sputfield: // fall through 2718 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 2719 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 2720 case Bytecodes::_fast_lputfield: 2721 NOT_LP64(__ movptr(hi, rdx)); 2722 __ movptr(lo, rax); 2723 break; 2724 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 2725 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 2726 case Bytecodes::_fast_aputfield: { 2727 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2728 break; 2729 } 2730 default: 2731 ShouldNotReachHere(); 2732 } 2733 2734 Label done; 2735 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 2736 Assembler::StoreStore)); 2737 // Barriers are so large that short branch doesn't reach! 2738 __ jmp(done); 2739 2740 // Same code as above, but don't need rdx to test for volatile. 2741 __ bind(notVolatile); 2742 2743 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 2744 2745 // Get object from stack 2746 pop_and_check_object(rcx); 2747 2748 // access field 2749 switch (bytecode()) { 2750 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 2751 case Bytecodes::_fast_sputfield: // fall through 2752 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 2753 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 2754 case Bytecodes::_fast_lputfield: 2755 NOT_LP64(__ movptr(hi, rdx)); 2756 __ movptr(lo, rax); 2757 break; 2758 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 2759 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 2760 case Bytecodes::_fast_aputfield: { 2761 do_oop_store(_masm, lo, rax, _bs->kind(), false); 2762 break; 2763 } 2764 default: 2765 ShouldNotReachHere(); 2766 } 2767 __ bind(done); 2768 } 2769 2770 2771 void TemplateTable::fast_accessfield(TosState state) { 2772 transition(atos, state); 2773 2774 // do the JVMTI work here to avoid disturbing the register state below 2775 if (JvmtiExport::can_post_field_access()) { 2776 // Check to see if a field access watch has been set before we take 2777 // the time to call into the VM. 2778 Label L1; 2779 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2780 __ testl(rcx,rcx); 2781 __ jcc(Assembler::zero, L1); 2782 // access constant pool cache entry 2783 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1); 2784 __ push_ptr(rax); // save object pointer before call_VM() clobbers it 2785 __ verify_oop(rax); 2786 // rax,: object pointer copied above 2787 // rcx: cache entry pointer 2788 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx); 2789 __ pop_ptr(rax); // restore object pointer 2790 __ bind(L1); 2791 } 2792 2793 // access constant pool cache 2794 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 2795 // replace index with field offset from cache entry 2796 __ movptr(rbx, Address(rcx, 2797 rbx, 2798 Address::times_ptr, 2799 in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 2800 2801 2802 // rax,: object 2803 __ verify_oop(rax); 2804 __ null_check(rax); 2805 // field addresses 2806 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 2807 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize); 2808 2809 // access field 2810 switch (bytecode()) { 2811 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break; 2812 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break; 2813 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break; 2814 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break; 2815 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break; 2816 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break; 2817 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break; 2818 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break; 2819 default: 2820 ShouldNotReachHere(); 2821 } 2822 2823 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO) 2824 // volatile_barrier( ); 2825 } 2826 2827 void TemplateTable::fast_xaccess(TosState state) { 2828 transition(vtos, state); 2829 // get receiver 2830 __ movptr(rax, aaddress(0)); 2831 // access constant pool cache 2832 __ get_cache_and_index_at_bcp(rcx, rdx, 2); 2833 __ movptr(rbx, Address(rcx, 2834 rdx, 2835 Address::times_ptr, 2836 in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 2837 // make sure exception is reported in correct bcp range (getfield is next instruction) 2838 __ increment(rsi); 2839 __ null_check(rax); 2840 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 2841 if (state == itos) { 2842 __ movl(rax, lo); 2843 } else if (state == atos) { 2844 __ movptr(rax, lo); 2845 __ verify_oop(rax); 2846 } else if (state == ftos) { 2847 __ fld_s(lo); 2848 } else { 2849 ShouldNotReachHere(); 2850 } 2851 __ decrement(rsi); 2852 } 2853 2854 2855 2856 //---------------------------------------------------------------------------------------------------- 2857 // Calls 2858 2859 void TemplateTable::count_calls(Register method, Register temp) { 2860 // implemented elsewhere 2861 ShouldNotReachHere(); 2862 } 2863 2864 2865 void TemplateTable::prepare_invoke(int byte_no, 2866 Register method, // linked method (or i-klass) 2867 Register index, // itable index, MethodType, etc. 2868 Register recv, // if caller wants to see it 2869 Register flags // if caller wants to test it 2870 ) { 2871 // determine flags 2872 const Bytecodes::Code code = bytecode(); 2873 const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 2874 const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 2875 const bool is_invokehandle = code == Bytecodes::_invokehandle; 2876 const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 2877 const bool is_invokespecial = code == Bytecodes::_invokespecial; 2878 const bool load_receiver = (recv != noreg); 2879 const bool save_flags = (flags != noreg); 2880 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), ""); 2881 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal"); 2882 assert(flags == noreg || flags == rdx, ""); 2883 assert(recv == noreg || recv == rcx, ""); 2884 2885 // setup registers & access constant pool cache 2886 if (recv == noreg) recv = rcx; 2887 if (flags == noreg) flags = rdx; 2888 assert_different_registers(method, index, recv, flags); 2889 2890 // save 'interpreter return address' 2891 __ save_bcp(); 2892 2893 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 2894 2895 // maybe push appendix to arguments (just before return address) 2896 if (is_invokedynamic || is_invokehandle) { 2897 Label L_no_push; 2898 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift)); 2899 __ jccb(Assembler::zero, L_no_push); 2900 // Push the appendix as a trailing parameter. 2901 // This must be done before we get the receiver, 2902 // since the parameter_size includes it. 2903 __ push(rbx); 2904 __ mov(rbx, index); 2905 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0"); 2906 __ load_resolved_reference_at_index(index, rbx); 2907 __ pop(rbx); 2908 __ push(index); // push appendix (MethodType, CallSite, etc.) 2909 __ bind(L_no_push); 2910 } 2911 2912 // load receiver if needed (note: no return address pushed yet) 2913 if (load_receiver) { 2914 __ movl(recv, flags); 2915 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask); 2916 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address 2917 const int receiver_is_at_end = -1; // back off one slot to get receiver 2918 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end); 2919 __ movptr(recv, recv_addr); 2920 __ verify_oop(recv); 2921 } 2922 2923 if (save_flags) { 2924 __ mov(rsi, flags); 2925 } 2926 2927 // compute return type 2928 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2929 // Make sure we don't need to mask flags after the above shift 2930 ConstantPoolCacheEntry::verify_tos_state_shift(); 2931 // load return address 2932 { 2933 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code); 2934 ExternalAddress table(table_addr); 2935 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))); 2936 } 2937 2938 // push return address 2939 __ push(flags); 2940 2941 // Restore flags value from the constant pool cache, and restore rsi 2942 // for later null checks. rsi is the bytecode pointer 2943 if (save_flags) { 2944 __ mov(flags, rsi); 2945 __ restore_bcp(); 2946 } 2947 } 2948 2949 2950 void TemplateTable::invokevirtual_helper(Register index, 2951 Register recv, 2952 Register flags) { 2953 // Uses temporary registers rax, rdx 2954 assert_different_registers(index, recv, rax, rdx); 2955 assert(index == rbx, ""); 2956 assert(recv == rcx, ""); 2957 2958 // Test for an invoke of a final method 2959 Label notFinal; 2960 __ movl(rax, flags); 2961 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift)); 2962 __ jcc(Assembler::zero, notFinal); 2963 2964 const Register method = index; // method must be rbx 2965 assert(method == rbx, 2966 "Method* must be rbx for interpreter calling convention"); 2967 2968 // do the call - the index is actually the method to call 2969 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method* 2970 2971 // It's final, need a null check here! 2972 __ null_check(recv); 2973 2974 // profile this call 2975 __ profile_final_call(rax); 2976 __ profile_arguments_type(rax, method, rsi, true); 2977 2978 __ jump_from_interpreted(method, rax); 2979 2980 __ bind(notFinal); 2981 2982 // get receiver klass 2983 __ null_check(recv, oopDesc::klass_offset_in_bytes()); 2984 __ load_klass(rax, recv); 2985 2986 // profile this call 2987 __ profile_virtual_call(rax, rdi, rdx); 2988 2989 // get target Method* & entry point 2990 __ lookup_virtual_method(rax, index, method); 2991 __ profile_arguments_type(rdx, method, rsi, true); 2992 __ jump_from_interpreted(method, rdx); 2993 } 2994 2995 2996 void TemplateTable::invokevirtual(int byte_no) { 2997 transition(vtos, vtos); 2998 assert(byte_no == f2_byte, "use this argument"); 2999 prepare_invoke(byte_no, 3000 rbx, // method or vtable index 3001 noreg, // unused itable index 3002 rcx, rdx); // recv, flags 3003 3004 // rbx: index 3005 // rcx: receiver 3006 // rdx: flags 3007 3008 invokevirtual_helper(rbx, rcx, rdx); 3009 } 3010 3011 3012 void TemplateTable::invokespecial(int byte_no) { 3013 transition(vtos, vtos); 3014 assert(byte_no == f1_byte, "use this argument"); 3015 prepare_invoke(byte_no, rbx, noreg, // get f1 Method* 3016 rcx); // get receiver also for null check 3017 __ verify_oop(rcx); 3018 __ null_check(rcx); 3019 // do the call 3020 __ profile_call(rax); 3021 __ profile_arguments_type(rax, rbx, rsi, false); 3022 __ jump_from_interpreted(rbx, rax); 3023 } 3024 3025 3026 void TemplateTable::invokestatic(int byte_no) { 3027 transition(vtos, vtos); 3028 assert(byte_no == f1_byte, "use this argument"); 3029 prepare_invoke(byte_no, rbx); // get f1 Method* 3030 // do the call 3031 __ profile_call(rax); 3032 __ profile_arguments_type(rax, rbx, rsi, false); 3033 __ jump_from_interpreted(rbx, rax); 3034 } 3035 3036 3037 void TemplateTable::fast_invokevfinal(int byte_no) { 3038 transition(vtos, vtos); 3039 assert(byte_no == f2_byte, "use this argument"); 3040 __ stop("fast_invokevfinal not used on x86"); 3041 } 3042 3043 3044 void TemplateTable::invokeinterface(int byte_no) { 3045 transition(vtos, vtos); 3046 assert(byte_no == f1_byte, "use this argument"); 3047 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index 3048 rcx, rdx); // recv, flags 3049 3050 // rax: interface klass (from f1) 3051 // rbx: itable index (from f2) 3052 // rcx: receiver 3053 // rdx: flags 3054 3055 // Special case of invokeinterface called for virtual method of 3056 // java.lang.Object. See cpCacheOop.cpp for details. 3057 // This code isn't produced by javac, but could be produced by 3058 // another compliant java compiler. 3059 Label notMethod; 3060 __ movl(rdi, rdx); 3061 __ andl(rdi, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift)); 3062 __ jcc(Assembler::zero, notMethod); 3063 3064 invokevirtual_helper(rbx, rcx, rdx); 3065 __ bind(notMethod); 3066 3067 // Get receiver klass into rdx - also a null check 3068 __ restore_locals(); // restore rdi 3069 __ null_check(rcx, oopDesc::klass_offset_in_bytes()); 3070 __ load_klass(rdx, rcx); 3071 3072 // profile this call 3073 __ profile_virtual_call(rdx, rsi, rdi); 3074 3075 Label no_such_interface, no_such_method; 3076 3077 __ lookup_interface_method(// inputs: rec. class, interface, itable index 3078 rdx, rax, rbx, 3079 // outputs: method, scan temp. reg 3080 rbx, rsi, 3081 no_such_interface); 3082 3083 // rbx: Method* to call 3084 // rcx: receiver 3085 // Check for abstract method error 3086 // Note: This should be done more efficiently via a throw_abstract_method_error 3087 // interpreter entry point and a conditional jump to it in case of a null 3088 // method. 3089 __ testptr(rbx, rbx); 3090 __ jcc(Assembler::zero, no_such_method); 3091 3092 __ profile_arguments_type(rdx, rbx, rsi, true); 3093 3094 // do the call 3095 // rcx: receiver 3096 // rbx,: Method* 3097 __ jump_from_interpreted(rbx, rdx); 3098 __ should_not_reach_here(); 3099 3100 // exception handling code follows... 3101 // note: must restore interpreter registers to canonical 3102 // state for exception handling to work correctly! 3103 3104 __ bind(no_such_method); 3105 // throw exception 3106 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3107 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 3108 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3109 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 3110 // the call_VM checks for exception, so we should never return here. 3111 __ should_not_reach_here(); 3112 3113 __ bind(no_such_interface); 3114 // throw exception 3115 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3116 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 3117 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3118 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3119 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3120 // the call_VM checks for exception, so we should never return here. 3121 __ should_not_reach_here(); 3122 } 3123 3124 void TemplateTable::invokehandle(int byte_no) { 3125 transition(vtos, vtos); 3126 assert(byte_no == f1_byte, "use this argument"); 3127 const Register rbx_method = rbx; 3128 const Register rax_mtype = rax; 3129 const Register rcx_recv = rcx; 3130 const Register rdx_flags = rdx; 3131 3132 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv); 3133 __ verify_method_ptr(rbx_method); 3134 __ verify_oop(rcx_recv); 3135 __ null_check(rcx_recv); 3136 3137 // rax: MethodType object (from cpool->resolved_references[f1], if necessary) 3138 // rbx: MH.invokeExact_MT method (from f2) 3139 3140 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke 3141 3142 // FIXME: profile the LambdaForm also 3143 __ profile_final_call(rax); 3144 __ profile_arguments_type(rdx, rbx_method, rsi, true); 3145 3146 __ jump_from_interpreted(rbx_method, rdx); 3147 } 3148 3149 3150 void TemplateTable::invokedynamic(int byte_no) { 3151 transition(vtos, vtos); 3152 assert(byte_no == f1_byte, "use this argument"); 3153 3154 const Register rbx_method = rbx; 3155 const Register rax_callsite = rax; 3156 3157 prepare_invoke(byte_no, rbx_method, rax_callsite); 3158 3159 // rax: CallSite object (from cpool->resolved_references[f1]) 3160 // rbx: MH.linkToCallSite method (from f2) 3161 3162 // Note: rax_callsite is already pushed by prepare_invoke 3163 3164 // %%% should make a type profile for any invokedynamic that takes a ref argument 3165 // profile this call 3166 __ profile_call(rsi); 3167 __ profile_arguments_type(rdx, rbx, rsi, false); 3168 3169 __ verify_oop(rax_callsite); 3170 3171 __ jump_from_interpreted(rbx_method, rdx); 3172 } 3173 3174 //---------------------------------------------------------------------------------------------------- 3175 // Allocation 3176 3177 void TemplateTable::_new() { 3178 transition(vtos, atos); 3179 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3180 Label slow_case; 3181 Label slow_case_no_pop; 3182 Label done; 3183 Label initialize_header; 3184 Label initialize_object; // including clearing the fields 3185 Label allocate_shared; 3186 3187 __ get_cpool_and_tags(rcx, rax); 3188 3189 // Make sure the class we're about to instantiate has been resolved. 3190 // This is done before loading InstanceKlass to be consistent with the order 3191 // how Constant Pool is updated (see ConstantPool::klass_at_put) 3192 const int tags_offset = Array<u1>::base_offset_in_bytes(); 3193 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class); 3194 __ jcc(Assembler::notEqual, slow_case_no_pop); 3195 3196 // get InstanceKlass 3197 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool))); 3198 __ push(rcx); // save the contexts of klass for initializing the header 3199 3200 // make sure klass is initialized & doesn't have finalizer 3201 // make sure klass is fully initialized 3202 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized); 3203 __ jcc(Assembler::notEqual, slow_case); 3204 3205 // get instance_size in InstanceKlass (scaled to a count of bytes) 3206 __ movl(rdx, Address(rcx, Klass::layout_helper_offset())); 3207 // test to see if it has a finalizer or is malformed in some way 3208 __ testl(rdx, Klass::_lh_instance_slow_path_bit); 3209 __ jcc(Assembler::notZero, slow_case); 3210 3211 // 3212 // Allocate the instance 3213 // 1) Try to allocate in the TLAB 3214 // 2) if fail and the object is large allocate in the shared Eden 3215 // 3) if the above fails (or is not applicable), go to a slow case 3216 // (creates a new TLAB, etc.) 3217 3218 const bool allow_shared_alloc = 3219 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode; 3220 3221 const Register thread = rcx; 3222 if (UseTLAB || allow_shared_alloc) { 3223 __ get_thread(thread); 3224 } 3225 3226 if (UseTLAB) { 3227 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset()))); 3228 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3229 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset()))); 3230 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case); 3231 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx); 3232 if (ZeroTLAB) { 3233 // the fields have been already cleared 3234 __ jmp(initialize_header); 3235 } else { 3236 // initialize both the header and fields 3237 __ jmp(initialize_object); 3238 } 3239 } 3240 3241 // Allocation in the shared Eden, if allowed. 3242 // 3243 // rdx: instance size in bytes 3244 if (allow_shared_alloc) { 3245 __ bind(allocate_shared); 3246 3247 ExternalAddress heap_top((address)Universe::heap()->top_addr()); 3248 3249 Label retry; 3250 __ bind(retry); 3251 __ movptr(rax, heap_top); 3252 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3253 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr())); 3254 __ jcc(Assembler::above, slow_case); 3255 3256 // Compare rax, with the top addr, and if still equal, store the new 3257 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was 3258 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs. 3259 // 3260 // rax,: object begin 3261 // rbx,: object end 3262 // rdx: instance size in bytes 3263 __ locked_cmpxchgptr(rbx, heap_top); 3264 3265 // if someone beat us on the allocation, try again, otherwise continue 3266 __ jcc(Assembler::notEqual, retry); 3267 3268 __ incr_allocated_bytes(thread, rdx, 0); 3269 } 3270 3271 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 3272 // The object is initialized before the header. If the object size is 3273 // zero, go directly to the header initialization. 3274 __ bind(initialize_object); 3275 __ decrement(rdx, sizeof(oopDesc)); 3276 __ jcc(Assembler::zero, initialize_header); 3277 3278 // Initialize topmost object field, divide rdx by 8, check if odd and 3279 // test if zero. 3280 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) 3281 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd 3282 3283 // rdx must have been multiple of 8 3284 #ifdef ASSERT 3285 // make sure rdx was multiple of 8 3286 Label L; 3287 // Ignore partial flag stall after shrl() since it is debug VM 3288 __ jccb(Assembler::carryClear, L); 3289 __ stop("object size is not multiple of 2 - adjust this code"); 3290 __ bind(L); 3291 // rdx must be > 0, no extra check needed here 3292 #endif 3293 3294 // initialize remaining object fields: rdx was a multiple of 8 3295 { Label loop; 3296 __ bind(loop); 3297 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx); 3298 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx)); 3299 __ decrement(rdx); 3300 __ jcc(Assembler::notZero, loop); 3301 } 3302 3303 // initialize object header only. 3304 __ bind(initialize_header); 3305 if (UseBiasedLocking) { 3306 __ pop(rcx); // get saved klass back in the register. 3307 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset())); 3308 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx); 3309 } else { 3310 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), 3311 (int32_t)markOopDesc::prototype()); // header 3312 __ pop(rcx); // get saved klass back in the register. 3313 } 3314 __ store_klass(rax, rcx); // klass 3315 3316 { 3317 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0); 3318 // Trigger dtrace event for fastpath 3319 __ push(atos); 3320 __ call_VM_leaf( 3321 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); 3322 __ pop(atos); 3323 } 3324 3325 __ jmp(done); 3326 } 3327 3328 // slow case 3329 __ bind(slow_case); 3330 __ pop(rcx); // restore stack pointer to what it was when we came in. 3331 __ bind(slow_case_no_pop); 3332 __ get_constant_pool(rax); 3333 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3334 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx); 3335 3336 // continue 3337 __ bind(done); 3338 } 3339 3340 3341 void TemplateTable::newarray() { 3342 transition(itos, atos); 3343 __ push_i(rax); // make sure everything is on the stack 3344 __ load_unsigned_byte(rdx, at_bcp(1)); 3345 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax); 3346 __ pop_i(rdx); // discard size 3347 } 3348 3349 3350 void TemplateTable::anewarray() { 3351 transition(itos, atos); 3352 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3353 __ get_constant_pool(rcx); 3354 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax); 3355 } 3356 3357 3358 void TemplateTable::arraylength() { 3359 transition(atos, itos); 3360 __ null_check(rax, arrayOopDesc::length_offset_in_bytes()); 3361 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes())); 3362 } 3363 3364 3365 void TemplateTable::checkcast() { 3366 transition(atos, atos); 3367 Label done, is_null, ok_is_subtype, quicked, resolved; 3368 __ testptr(rax, rax); // Object is in EAX 3369 __ jcc(Assembler::zero, is_null); 3370 3371 // Get cpool & tags index 3372 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 3373 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 3374 // See if bytecode has already been quicked 3375 __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class); 3376 __ jcc(Assembler::equal, quicked); 3377 3378 __ push(atos); 3379 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 3380 // vm_result_2 has metadata result 3381 // borrow rdi from locals 3382 __ get_thread(rdi); 3383 __ get_vm_result_2(rax, rdi); 3384 __ restore_locals(); 3385 __ pop_ptr(rdx); 3386 __ jmpb(resolved); 3387 3388 // Get superklass in EAX and subklass in EBX 3389 __ bind(quicked); 3390 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check 3391 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool))); 3392 3393 __ bind(resolved); 3394 __ load_klass(rbx, rdx); 3395 3396 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX. 3397 // Superklass in EAX. Subklass in EBX. 3398 __ gen_subtype_check( rbx, ok_is_subtype ); 3399 3400 // Come here on failure 3401 __ push(rdx); 3402 // object is at TOS 3403 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry)); 3404 3405 // Come here on success 3406 __ bind(ok_is_subtype); 3407 __ mov(rax,rdx); // Restore object in EDX 3408 3409 // Collect counts on whether this check-cast sees NULLs a lot or not. 3410 if (ProfileInterpreter) { 3411 __ jmp(done); 3412 __ bind(is_null); 3413 __ profile_null_seen(rcx); 3414 } else { 3415 __ bind(is_null); // same as 'done' 3416 } 3417 __ bind(done); 3418 } 3419 3420 3421 void TemplateTable::instanceof() { 3422 transition(atos, itos); 3423 Label done, is_null, ok_is_subtype, quicked, resolved; 3424 __ testptr(rax, rax); 3425 __ jcc(Assembler::zero, is_null); 3426 3427 // Get cpool & tags index 3428 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 3429 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 3430 // See if bytecode has already been quicked 3431 __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class); 3432 __ jcc(Assembler::equal, quicked); 3433 3434 __ push(atos); 3435 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 3436 // vm_result_2 has metadata result 3437 // borrow rdi from locals 3438 __ get_thread(rdi); 3439 __ get_vm_result_2(rax, rdi); 3440 __ restore_locals(); 3441 __ pop_ptr(rdx); 3442 __ load_klass(rdx, rdx); 3443 __ jmp(resolved); 3444 3445 // Get superklass in EAX and subklass in EDX 3446 __ bind(quicked); 3447 __ load_klass(rdx, rax); 3448 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool))); 3449 3450 __ bind(resolved); 3451 3452 // Generate subtype check. Blows ECX. Resets EDI. 3453 // Superklass in EAX. Subklass in EDX. 3454 __ gen_subtype_check( rdx, ok_is_subtype ); 3455 3456 // Come here on failure 3457 __ xorl(rax,rax); 3458 __ jmpb(done); 3459 // Come here on success 3460 __ bind(ok_is_subtype); 3461 __ movl(rax, 1); 3462 3463 // Collect counts on whether this test sees NULLs a lot or not. 3464 if (ProfileInterpreter) { 3465 __ jmp(done); 3466 __ bind(is_null); 3467 __ profile_null_seen(rcx); 3468 } else { 3469 __ bind(is_null); // same as 'done' 3470 } 3471 __ bind(done); 3472 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass 3473 // rax, = 1: obj != NULL and obj is an instanceof the specified klass 3474 } 3475 3476 3477 //---------------------------------------------------------------------------------------------------- 3478 // Breakpoints 3479 void TemplateTable::_breakpoint() { 3480 3481 // Note: We get here even if we are single stepping.. 3482 // jbug inists on setting breakpoints at every bytecode 3483 // even if we are in single step mode. 3484 3485 transition(vtos, vtos); 3486 3487 // get the unpatched byte code 3488 __ get_method(rcx); 3489 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi); 3490 __ mov(rbx, rax); 3491 3492 // post the breakpoint event 3493 __ get_method(rcx); 3494 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi); 3495 3496 // complete the execution of original bytecode 3497 __ dispatch_only_normal(vtos); 3498 } 3499 3500 3501 //---------------------------------------------------------------------------------------------------- 3502 // Exceptions 3503 3504 void TemplateTable::athrow() { 3505 transition(atos, vtos); 3506 __ null_check(rax); 3507 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 3508 } 3509 3510 3511 //---------------------------------------------------------------------------------------------------- 3512 // Synchronization 3513 // 3514 // Note: monitorenter & exit are symmetric routines; which is reflected 3515 // in the assembly code structure as well 3516 // 3517 // Stack layout: 3518 // 3519 // [expressions ] <--- rsp = expression stack top 3520 // .. 3521 // [expressions ] 3522 // [monitor entry] <--- monitor block top = expression stack bot 3523 // .. 3524 // [monitor entry] 3525 // [frame data ] <--- monitor block bot 3526 // ... 3527 // [saved rbp, ] <--- rbp, 3528 3529 3530 void TemplateTable::monitorenter() { 3531 transition(atos, vtos); 3532 3533 // check for NULL object 3534 __ null_check(rax); 3535 3536 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3537 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 3538 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 3539 Label allocated; 3540 3541 // initialize entry pointer 3542 __ xorl(rdx, rdx); // points to free slot or NULL 3543 3544 // find a free slot in the monitor block (result in rdx) 3545 { Label entry, loop, exit; 3546 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry 3547 3548 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 3549 __ jmpb(entry); 3550 3551 __ bind(loop); 3552 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used 3553 __ cmovptr(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx 3554 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 3555 __ jccb(Assembler::equal, exit); // if same object then stop searching 3556 __ addptr(rcx, entry_size); // otherwise advance to next entry 3557 __ bind(entry); 3558 __ cmpptr(rcx, rbx); // check if bottom reached 3559 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 3560 __ bind(exit); 3561 } 3562 3563 __ testptr(rdx, rdx); // check if a slot has been found 3564 __ jccb(Assembler::notZero, allocated); // if found, continue with that one 3565 3566 // allocate one if there's no free slot 3567 { Label entry, loop; 3568 // 1. compute new pointers // rsp: old expression stack top 3569 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom 3570 __ subptr(rsp, entry_size); // move expression stack top 3571 __ subptr(rdx, entry_size); // move expression stack bottom 3572 __ mov(rcx, rsp); // set start value for copy loop 3573 __ movptr(monitor_block_bot, rdx); // set new monitor block top 3574 __ jmp(entry); 3575 // 2. move expression stack contents 3576 __ bind(loop); 3577 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location 3578 __ movptr(Address(rcx, 0), rbx); // and store it at new location 3579 __ addptr(rcx, wordSize); // advance to next word 3580 __ bind(entry); 3581 __ cmpptr(rcx, rdx); // check if bottom reached 3582 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word 3583 } 3584 3585 // call run-time routine 3586 // rdx: points to monitor entry 3587 __ bind(allocated); 3588 3589 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly. 3590 // The object has already been poped from the stack, so the expression stack looks correct. 3591 __ increment(rsi); 3592 3593 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 3594 __ lock_object(rdx); 3595 3596 // check to make sure this monitor doesn't cause stack overflow after locking 3597 __ save_bcp(); // in case of exception 3598 __ generate_stack_overflow_check(0); 3599 3600 // The bcp has already been incremented. Just need to dispatch to next instruction. 3601 __ dispatch_next(vtos); 3602 } 3603 3604 3605 void TemplateTable::monitorexit() { 3606 transition(atos, vtos); 3607 3608 // check for NULL object 3609 __ null_check(rax); 3610 3611 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3612 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 3613 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 3614 Label found; 3615 3616 // find matching slot 3617 { Label entry, loop; 3618 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry 3619 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 3620 __ jmpb(entry); 3621 3622 __ bind(loop); 3623 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 3624 __ jcc(Assembler::equal, found); // if same object then stop searching 3625 __ addptr(rdx, entry_size); // otherwise advance to next entry 3626 __ bind(entry); 3627 __ cmpptr(rdx, rbx); // check if bottom reached 3628 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 3629 } 3630 3631 // error handling. Unlocking was not block-structured 3632 Label end; 3633 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 3634 __ should_not_reach_here(); 3635 3636 // call run-time routine 3637 // rcx: points to monitor entry 3638 __ bind(found); 3639 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps) 3640 __ unlock_object(rdx); 3641 __ pop_ptr(rax); // discard object 3642 __ bind(end); 3643 } 3644 3645 3646 //---------------------------------------------------------------------------------------------------- 3647 // Wide instructions 3648 3649 void TemplateTable::wide() { 3650 transition(vtos, vtos); 3651 __ load_unsigned_byte(rbx, at_bcp(1)); 3652 ExternalAddress wtable((address)Interpreter::_wentry_point); 3653 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr))); 3654 // Note: the rsi increment step is part of the individual wide bytecode implementations 3655 } 3656 3657 3658 //---------------------------------------------------------------------------------------------------- 3659 // Multi arrays 3660 3661 void TemplateTable::multianewarray() { 3662 transition(vtos, atos); 3663 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions 3664 // last dim is on top of stack; we want address of first one: 3665 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize 3666 // the latter wordSize to point to the beginning of the array. 3667 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize)); 3668 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax, 3669 __ load_unsigned_byte(rbx, at_bcp(3)); 3670 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts 3671 } 3672 3673 #endif /* !CC_INTERP */