1 /* 2 * Copyright (c) 1997, 2015, 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 43 #define __ _masm-> 44 45 // Global Register Names 46 Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi); 47 Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi); 48 49 // Platform-dependent initialization 50 void TemplateTable::pd_initialize() { 51 // No x86 specific initialization 52 } 53 54 // Address Computation: local variables 55 static inline Address iaddress(int n) { 56 return Address(rlocals, Interpreter::local_offset_in_bytes(n)); 57 } 58 59 static inline Address laddress(int n) { 60 return iaddress(n + 1); 61 } 62 63 #ifndef _LP64 64 static inline Address haddress(int n) { 65 return iaddress(n + 0); 66 } 67 #endif 68 69 static inline Address faddress(int n) { 70 return iaddress(n); 71 } 72 73 static inline Address daddress(int n) { 74 return laddress(n); 75 } 76 77 static inline Address aaddress(int n) { 78 return iaddress(n); 79 } 80 81 static inline Address iaddress(Register r) { 82 return Address(rlocals, r, Address::times_ptr); 83 } 84 85 static inline Address laddress(Register r) { 86 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1)); 87 } 88 89 #ifndef _LP64 90 static inline Address haddress(Register r) { 91 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0)); 92 } 93 #endif 94 95 static inline Address faddress(Register r) { 96 return iaddress(r); 97 } 98 99 static inline Address daddress(Register r) { 100 return laddress(r); 101 } 102 103 static inline Address aaddress(Register r) { 104 return iaddress(r); 105 } 106 107 108 // expression stack 109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store 110 // data beyond the rsp which is potentially unsafe in an MT environment; 111 // an interrupt may overwrite that data.) 112 static inline Address at_rsp () { 113 return Address(rsp, 0); 114 } 115 116 // At top of Java expression stack which may be different than esp(). It 117 // isn't for category 1 objects. 118 static inline Address at_tos () { 119 return Address(rsp, Interpreter::expr_offset_in_bytes(0)); 120 } 121 122 static inline Address at_tos_p1() { 123 return Address(rsp, Interpreter::expr_offset_in_bytes(1)); 124 } 125 126 static inline Address at_tos_p2() { 127 return Address(rsp, Interpreter::expr_offset_in_bytes(2)); 128 } 129 130 // Condition conversion 131 static Assembler::Condition j_not(TemplateTable::Condition cc) { 132 switch (cc) { 133 case TemplateTable::equal : return Assembler::notEqual; 134 case TemplateTable::not_equal : return Assembler::equal; 135 case TemplateTable::less : return Assembler::greaterEqual; 136 case TemplateTable::less_equal : return Assembler::greater; 137 case TemplateTable::greater : return Assembler::lessEqual; 138 case TemplateTable::greater_equal: return Assembler::less; 139 } 140 ShouldNotReachHere(); 141 return Assembler::zero; 142 } 143 144 145 146 // Miscelaneous helper routines 147 // Store an oop (or NULL) at the address described by obj. 148 // If val == noreg this means store a NULL 149 150 151 static void do_oop_store(InterpreterMacroAssembler* _masm, 152 Address obj, 153 Register val, 154 BarrierSet::Name barrier, 155 bool precise) { 156 assert(val == noreg || val == rax, "parameter is just for looks"); 157 switch (barrier) { 158 #if INCLUDE_ALL_GCS 159 case BarrierSet::G1SATBCTLogging: 160 { 161 // flatten object address if needed 162 // We do it regardless of precise because we need the registers 163 if (obj.index() == noreg && obj.disp() == 0) { 164 if (obj.base() != rdx) { 165 __ movptr(rdx, obj.base()); 166 } 167 } else { 168 __ lea(rdx, obj); 169 } 170 171 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi); 172 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 173 174 NOT_LP64(__ get_thread(rcx)); 175 NOT_LP64(__ save_bcp()); 176 177 __ g1_write_barrier_pre(rdx /* obj */, 178 rbx /* pre_val */, 179 rthread /* thread */, 180 rtmp /* tmp */, 181 val != noreg /* tosca_live */, 182 false /* expand_call */); 183 if (val == noreg) { 184 __ store_heap_oop_null(Address(rdx, 0)); 185 } else { 186 // G1 barrier needs uncompressed oop for region cross check. 187 Register new_val = val; 188 if (UseCompressedOops) { 189 new_val = rbx; 190 __ movptr(new_val, val); 191 } 192 __ store_heap_oop(Address(rdx, 0), val); 193 __ g1_write_barrier_post(rdx /* store_adr */, 194 new_val /* new_val */, 195 rthread /* thread */, 196 rtmp /* tmp */, 197 rbx /* tmp2 */); 198 } 199 NOT_LP64( __ restore_bcp()); 200 } 201 break; 202 #endif // INCLUDE_ALL_GCS 203 case BarrierSet::CardTableModRef: 204 case BarrierSet::CardTableExtension: 205 { 206 if (val == noreg) { 207 __ store_heap_oop_null(obj); 208 } else { 209 __ store_heap_oop(obj, val); 210 // flatten object address if needed 211 if (!precise || (obj.index() == noreg && obj.disp() == 0)) { 212 __ store_check(obj.base()); 213 } else { 214 __ lea(rdx, obj); 215 __ store_check(rdx); 216 } 217 } 218 } 219 break; 220 case BarrierSet::ModRef: 221 if (val == noreg) { 222 __ store_heap_oop_null(obj); 223 } else { 224 __ store_heap_oop(obj, val); 225 } 226 break; 227 default : 228 ShouldNotReachHere(); 229 230 } 231 } 232 233 Address TemplateTable::at_bcp(int offset) { 234 assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 235 return Address(rbcp, offset); 236 } 237 238 239 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg, 240 Register temp_reg, bool load_bc_into_bc_reg/*=true*/, 241 int byte_no) { 242 if (!RewriteBytecodes) return; 243 Label L_patch_done; 244 245 switch (bc) { 246 case Bytecodes::_fast_aputfield: 247 case Bytecodes::_fast_bputfield: 248 case Bytecodes::_fast_cputfield: 249 case Bytecodes::_fast_dputfield: 250 case Bytecodes::_fast_fputfield: 251 case Bytecodes::_fast_iputfield: 252 case Bytecodes::_fast_lputfield: 253 case Bytecodes::_fast_sputfield: 254 { 255 // We skip bytecode quickening for putfield instructions when 256 // the put_code written to the constant pool cache is zero. 257 // This is required so that every execution of this instruction 258 // calls out to InterpreterRuntime::resolve_get_put to do 259 // additional, required work. 260 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 261 assert(load_bc_into_bc_reg, "we use bc_reg as temp"); 262 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1); 263 __ movl(bc_reg, bc); 264 __ cmpl(temp_reg, (int) 0); 265 __ jcc(Assembler::zero, L_patch_done); // don't patch 266 } 267 break; 268 default: 269 assert(byte_no == -1, "sanity"); 270 // the pair bytecodes have already done the load. 271 if (load_bc_into_bc_reg) { 272 __ movl(bc_reg, bc); 273 } 274 } 275 276 if (JvmtiExport::can_post_breakpoint()) { 277 Label L_fast_patch; 278 // if a breakpoint is present we can't rewrite the stream directly 279 __ movzbl(temp_reg, at_bcp(0)); 280 __ cmpl(temp_reg, Bytecodes::_breakpoint); 281 __ jcc(Assembler::notEqual, L_fast_patch); 282 __ get_method(temp_reg); 283 // Let breakpoint table handling rewrite to quicker bytecode 284 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg); 285 #ifndef ASSERT 286 __ jmpb(L_patch_done); 287 #else 288 __ jmp(L_patch_done); 289 #endif 290 __ bind(L_fast_patch); 291 } 292 293 #ifdef ASSERT 294 Label L_okay; 295 __ load_unsigned_byte(temp_reg, at_bcp(0)); 296 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc)); 297 __ jcc(Assembler::equal, L_okay); 298 __ cmpl(temp_reg, bc_reg); 299 __ jcc(Assembler::equal, L_okay); 300 __ stop("patching the wrong bytecode"); 301 __ bind(L_okay); 302 #endif 303 304 // patch bytecode 305 __ movb(at_bcp(0), bc_reg); 306 __ bind(L_patch_done); 307 } 308 // Individual instructions 309 310 311 void TemplateTable::nop() { 312 transition(vtos, vtos); 313 // nothing to do 314 } 315 316 void TemplateTable::shouldnotreachhere() { 317 transition(vtos, vtos); 318 __ stop("shouldnotreachhere bytecode"); 319 } 320 321 void TemplateTable::aconst_null() { 322 transition(vtos, atos); 323 __ xorl(rax, rax); 324 } 325 326 void TemplateTable::iconst(int value) { 327 transition(vtos, itos); 328 if (value == 0) { 329 __ xorl(rax, rax); 330 } else { 331 __ movl(rax, value); 332 } 333 } 334 335 void TemplateTable::lconst(int value) { 336 transition(vtos, ltos); 337 if (value == 0) { 338 __ xorl(rax, rax); 339 } else { 340 __ movl(rax, value); 341 } 342 #ifndef _LP64 343 assert(value >= 0, "check this code"); 344 __ xorptr(rdx, rdx); 345 #endif 346 } 347 348 349 350 void TemplateTable::fconst(int value) { 351 transition(vtos, ftos); 352 if (UseSSE >= 1) { 353 static float one = 1.0f, two = 2.0f; 354 switch (value) { 355 case 0: 356 __ xorps(xmm0, xmm0); 357 break; 358 case 1: 359 __ movflt(xmm0, ExternalAddress((address) &one)); 360 break; 361 case 2: 362 __ movflt(xmm0, ExternalAddress((address) &two)); 363 break; 364 default: 365 ShouldNotReachHere(); 366 break; 367 } 368 } else { 369 #ifdef _LP64 370 ShouldNotReachHere(); 371 #else 372 if (value == 0) { __ fldz(); 373 } else if (value == 1) { __ fld1(); 374 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here 375 } else { ShouldNotReachHere(); 376 } 377 #endif // _LP64 378 } 379 } 380 381 void TemplateTable::dconst(int value) { 382 transition(vtos, dtos); 383 if (UseSSE >= 2) { 384 static double one = 1.0; 385 switch (value) { 386 case 0: 387 __ xorpd(xmm0, xmm0); 388 break; 389 case 1: 390 __ movdbl(xmm0, ExternalAddress((address) &one)); 391 break; 392 default: 393 ShouldNotReachHere(); 394 break; 395 } 396 } else { 397 #ifdef _LP64 398 ShouldNotReachHere(); 399 #else 400 if (value == 0) { __ fldz(); 401 } else if (value == 1) { __ fld1(); 402 } else { ShouldNotReachHere(); 403 } 404 #endif 405 } 406 } 407 408 void TemplateTable::bipush() { 409 transition(vtos, itos); 410 __ load_signed_byte(rax, at_bcp(1)); 411 } 412 413 void TemplateTable::sipush() { 414 transition(vtos, itos); 415 __ load_unsigned_short(rax, at_bcp(1)); 416 __ bswapl(rax); 417 __ sarl(rax, 16); 418 } 419 420 void TemplateTable::ldc(bool wide) { 421 transition(vtos, vtos); 422 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1); 423 Label call_ldc, notFloat, notClass, Done; 424 425 if (wide) { 426 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 427 } else { 428 __ load_unsigned_byte(rbx, at_bcp(1)); 429 } 430 431 __ get_cpool_and_tags(rcx, rax); 432 const int base_offset = ConstantPool::header_size() * wordSize; 433 const int tags_offset = Array<u1>::base_offset_in_bytes(); 434 435 // get type 436 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset)); 437 438 // unresolved class - get the resolved class 439 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass); 440 __ jccb(Assembler::equal, call_ldc); 441 442 // unresolved class in error state - call into runtime to throw the error 443 // from the first resolution attempt 444 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError); 445 __ jccb(Assembler::equal, call_ldc); 446 447 // resolved class - need to call vm to get java mirror of the class 448 __ cmpl(rdx, JVM_CONSTANT_Class); 449 __ jcc(Assembler::notEqual, notClass); 450 451 __ bind(call_ldc); 452 453 __ movl(rarg, wide); 454 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg); 455 456 __ push(atos); 457 __ jmp(Done); 458 459 __ bind(notClass); 460 __ cmpl(rdx, JVM_CONSTANT_Float); 461 __ jccb(Assembler::notEqual, notFloat); 462 463 // ftos 464 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset)); 465 __ push(ftos); 466 __ jmp(Done); 467 468 __ bind(notFloat); 469 #ifdef ASSERT 470 { 471 Label L; 472 __ cmpl(rdx, JVM_CONSTANT_Integer); 473 __ jcc(Assembler::equal, L); 474 // String and Object are rewritten to fast_aldc 475 __ stop("unexpected tag type in ldc"); 476 __ bind(L); 477 } 478 #endif 479 // itos JVM_CONSTANT_Integer only 480 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 481 __ push(itos); 482 __ bind(Done); 483 } 484 485 // Fast path for caching oop constants. 486 void TemplateTable::fast_aldc(bool wide) { 487 transition(vtos, atos); 488 489 Register result = rax; 490 Register tmp = rdx; 491 int index_size = wide ? sizeof(u2) : sizeof(u1); 492 493 Label resolved; 494 495 // We are resolved if the resolved reference cache entry contains a 496 // non-null object (String, MethodType, etc.) 497 assert_different_registers(result, tmp); 498 __ get_cache_index_at_bcp(tmp, 1, index_size); 499 __ load_resolved_reference_at_index(result, tmp); 500 __ testl(result, result); 501 __ jcc(Assembler::notZero, resolved); 502 503 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); 504 505 // first time invocation - must resolve first 506 __ movl(tmp, (int)bytecode()); 507 __ call_VM(result, entry, tmp); 508 509 __ bind(resolved); 510 511 if (VerifyOops) { 512 __ verify_oop(result); 513 } 514 } 515 516 void TemplateTable::ldc2_w() { 517 transition(vtos, vtos); 518 Label Long, Done; 519 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 520 521 __ get_cpool_and_tags(rcx, rax); 522 const int base_offset = ConstantPool::header_size() * wordSize; 523 const int tags_offset = Array<u1>::base_offset_in_bytes(); 524 525 // get type 526 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), 527 JVM_CONSTANT_Double); 528 __ jccb(Assembler::notEqual, Long); 529 530 // dtos 531 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset)); 532 __ push(dtos); 533 534 __ jmpb(Done); 535 __ bind(Long); 536 537 // ltos 538 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize)); 539 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize))); 540 __ push(ltos); 541 542 __ bind(Done); 543 } 544 545 void TemplateTable::locals_index(Register reg, int offset) { 546 __ load_unsigned_byte(reg, at_bcp(offset)); 547 __ negptr(reg); 548 } 549 550 void TemplateTable::iload() { 551 iload_internal(); 552 } 553 554 void TemplateTable::nofast_iload() { 555 iload_internal(may_not_rewrite); 556 } 557 558 void TemplateTable::iload_internal(RewriteControl rc) { 559 transition(vtos, itos); 560 if (RewriteFrequentPairs && rc == may_rewrite) { 561 Label rewrite, done; 562 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 563 LP64_ONLY(assert(rbx != bc, "register damaged")); 564 565 // get next byte 566 __ load_unsigned_byte(rbx, 567 at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 568 // if _iload, wait to rewrite to iload2. We only want to rewrite the 569 // last two iloads in a pair. Comparing against fast_iload means that 570 // the next bytecode is neither an iload or a caload, and therefore 571 // an iload pair. 572 __ cmpl(rbx, Bytecodes::_iload); 573 __ jcc(Assembler::equal, done); 574 575 __ cmpl(rbx, Bytecodes::_fast_iload); 576 __ movl(bc, Bytecodes::_fast_iload2); 577 578 __ jccb(Assembler::equal, rewrite); 579 580 // if _caload, rewrite to fast_icaload 581 __ cmpl(rbx, Bytecodes::_caload); 582 __ movl(bc, Bytecodes::_fast_icaload); 583 __ jccb(Assembler::equal, rewrite); 584 585 // rewrite so iload doesn't check again. 586 __ movl(bc, Bytecodes::_fast_iload); 587 588 // rewrite 589 // bc: fast bytecode 590 __ bind(rewrite); 591 patch_bytecode(Bytecodes::_iload, bc, rbx, false); 592 __ bind(done); 593 } 594 595 // Get the local value into tos 596 locals_index(rbx); 597 __ movl(rax, iaddress(rbx)); 598 } 599 600 void TemplateTable::fast_iload2() { 601 transition(vtos, itos); 602 locals_index(rbx); 603 __ movl(rax, iaddress(rbx)); 604 __ push(itos); 605 locals_index(rbx, 3); 606 __ movl(rax, iaddress(rbx)); 607 } 608 609 void TemplateTable::fast_iload() { 610 transition(vtos, itos); 611 locals_index(rbx); 612 __ movl(rax, iaddress(rbx)); 613 } 614 615 void TemplateTable::lload() { 616 transition(vtos, ltos); 617 locals_index(rbx); 618 __ movptr(rax, laddress(rbx)); 619 NOT_LP64(__ movl(rdx, haddress(rbx))); 620 } 621 622 void TemplateTable::fload() { 623 transition(vtos, ftos); 624 locals_index(rbx); 625 __ load_float(faddress(rbx)); 626 } 627 628 void TemplateTable::dload() { 629 transition(vtos, dtos); 630 locals_index(rbx); 631 __ load_double(daddress(rbx)); 632 } 633 634 void TemplateTable::aload() { 635 transition(vtos, atos); 636 locals_index(rbx); 637 __ movptr(rax, aaddress(rbx)); 638 } 639 640 void TemplateTable::locals_index_wide(Register reg) { 641 __ load_unsigned_short(reg, at_bcp(2)); 642 __ bswapl(reg); 643 __ shrl(reg, 16); 644 __ negptr(reg); 645 } 646 647 void TemplateTable::wide_iload() { 648 transition(vtos, itos); 649 locals_index_wide(rbx); 650 __ movl(rax, iaddress(rbx)); 651 } 652 653 void TemplateTable::wide_lload() { 654 transition(vtos, ltos); 655 locals_index_wide(rbx); 656 __ movptr(rax, laddress(rbx)); 657 NOT_LP64(__ movl(rdx, haddress(rbx))); 658 } 659 660 void TemplateTable::wide_fload() { 661 transition(vtos, ftos); 662 locals_index_wide(rbx); 663 __ load_float(faddress(rbx)); 664 } 665 666 void TemplateTable::wide_dload() { 667 transition(vtos, dtos); 668 locals_index_wide(rbx); 669 __ load_double(daddress(rbx)); 670 } 671 672 void TemplateTable::wide_aload() { 673 transition(vtos, atos); 674 locals_index_wide(rbx); 675 __ movptr(rax, aaddress(rbx)); 676 } 677 678 void TemplateTable::index_check(Register array, Register index) { 679 // Pop ptr into array 680 __ pop_ptr(array); 681 index_check_without_pop(array, index); 682 } 683 684 void TemplateTable::index_check_without_pop(Register array, Register index) { 685 // destroys rbx 686 // check array 687 __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 688 // sign extend index for use by indexed load 689 __ movl2ptr(index, index); 690 // check index 691 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes())); 692 if (index != rbx) { 693 // ??? convention: move aberrant index into rbx for exception message 694 assert(rbx != array, "different registers"); 695 __ movl(rbx, index); 696 } 697 __ jump_cc(Assembler::aboveEqual, 698 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry)); 699 } 700 701 702 void TemplateTable::iaload() { 703 transition(itos, itos); 704 // rax: index 705 // rdx: array 706 index_check(rdx, rax); // kills rbx 707 __ movl(rax, Address(rdx, rax, 708 Address::times_4, 709 arrayOopDesc::base_offset_in_bytes(T_INT))); 710 } 711 712 void TemplateTable::laload() { 713 transition(itos, ltos); 714 // rax: index 715 // rdx: array 716 index_check(rdx, rax); // kills rbx 717 NOT_LP64(__ mov(rbx, rax)); 718 // rbx,: index 719 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize)); 720 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize))); 721 } 722 723 724 725 void TemplateTable::faload() { 726 transition(itos, ftos); 727 // rax: index 728 // rdx: array 729 index_check(rdx, rax); // kills rbx 730 __ load_float(Address(rdx, rax, 731 Address::times_4, 732 arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 733 } 734 735 void TemplateTable::daload() { 736 transition(itos, dtos); 737 // rax: index 738 // rdx: array 739 index_check(rdx, rax); // kills rbx 740 __ load_double(Address(rdx, rax, 741 Address::times_8, 742 arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 743 } 744 745 void TemplateTable::aaload() { 746 transition(itos, atos); 747 // rax: index 748 // rdx: array 749 index_check(rdx, rax); // kills rbx 750 __ load_heap_oop(rax, Address(rdx, rax, 751 UseCompressedOops ? Address::times_4 : Address::times_ptr, 752 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 753 } 754 755 void TemplateTable::baload() { 756 transition(itos, itos); 757 // rax: index 758 // rdx: array 759 index_check(rdx, rax); // kills rbx 760 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 761 } 762 763 void TemplateTable::caload() { 764 transition(itos, itos); 765 // rax: index 766 // rdx: array 767 index_check(rdx, rax); // kills rbx 768 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 769 } 770 771 // iload followed by caload frequent pair 772 void TemplateTable::fast_icaload() { 773 transition(vtos, itos); 774 // load index out of locals 775 locals_index(rbx); 776 __ movl(rax, iaddress(rbx)); 777 778 // rax: index 779 // rdx: array 780 index_check(rdx, rax); // kills rbx 781 __ load_unsigned_short(rax, 782 Address(rdx, rax, 783 Address::times_2, 784 arrayOopDesc::base_offset_in_bytes(T_CHAR))); 785 } 786 787 788 void TemplateTable::saload() { 789 transition(itos, itos); 790 // rax: index 791 // rdx: array 792 index_check(rdx, rax); // kills rbx 793 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 794 } 795 796 void TemplateTable::iload(int n) { 797 transition(vtos, itos); 798 __ movl(rax, iaddress(n)); 799 } 800 801 void TemplateTable::lload(int n) { 802 transition(vtos, ltos); 803 __ movptr(rax, laddress(n)); 804 NOT_LP64(__ movptr(rdx, haddress(n))); 805 } 806 807 void TemplateTable::fload(int n) { 808 transition(vtos, ftos); 809 __ load_float(faddress(n)); 810 } 811 812 void TemplateTable::dload(int n) { 813 transition(vtos, dtos); 814 __ load_double(daddress(n)); 815 } 816 817 void TemplateTable::aload(int n) { 818 transition(vtos, atos); 819 __ movptr(rax, aaddress(n)); 820 } 821 822 void TemplateTable::aload_0() { 823 aload_0_internal(); 824 } 825 826 void TemplateTable::nofast_aload_0() { 827 aload_0_internal(may_not_rewrite); 828 } 829 830 void TemplateTable::aload_0_internal(RewriteControl rc) { 831 transition(vtos, atos); 832 // According to bytecode histograms, the pairs: 833 // 834 // _aload_0, _fast_igetfield 835 // _aload_0, _fast_agetfield 836 // _aload_0, _fast_fgetfield 837 // 838 // occur frequently. If RewriteFrequentPairs is set, the (slow) 839 // _aload_0 bytecode checks if the next bytecode is either 840 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then 841 // rewrites the current bytecode into a pair bytecode; otherwise it 842 // rewrites the current bytecode into _fast_aload_0 that doesn't do 843 // the pair check anymore. 844 // 845 // Note: If the next bytecode is _getfield, the rewrite must be 846 // delayed, otherwise we may miss an opportunity for a pair. 847 // 848 // Also rewrite frequent pairs 849 // aload_0, aload_1 850 // aload_0, iload_1 851 // These bytecodes with a small amount of code are most profitable 852 // to rewrite 853 if (RewriteFrequentPairs && rc == may_rewrite) { 854 Label rewrite, done; 855 856 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 857 LP64_ONLY(assert(rbx != bc, "register damaged")); 858 859 // get next byte 860 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 861 862 // do actual aload_0 863 aload(0); 864 865 // if _getfield then wait with rewrite 866 __ cmpl(rbx, Bytecodes::_getfield); 867 __ jcc(Assembler::equal, done); 868 869 // if _igetfield then reqrite to _fast_iaccess_0 870 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 871 __ cmpl(rbx, Bytecodes::_fast_igetfield); 872 __ movl(bc, Bytecodes::_fast_iaccess_0); 873 __ jccb(Assembler::equal, rewrite); 874 875 // if _agetfield then reqrite to _fast_aaccess_0 876 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 877 __ cmpl(rbx, Bytecodes::_fast_agetfield); 878 __ movl(bc, Bytecodes::_fast_aaccess_0); 879 __ jccb(Assembler::equal, rewrite); 880 881 // if _fgetfield then reqrite to _fast_faccess_0 882 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 883 __ cmpl(rbx, Bytecodes::_fast_fgetfield); 884 __ movl(bc, Bytecodes::_fast_faccess_0); 885 __ jccb(Assembler::equal, rewrite); 886 887 // else rewrite to _fast_aload0 888 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 889 __ movl(bc, Bytecodes::_fast_aload_0); 890 891 // rewrite 892 // bc: fast bytecode 893 __ bind(rewrite); 894 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false); 895 896 __ bind(done); 897 } else { 898 aload(0); 899 } 900 } 901 902 void TemplateTable::istore() { 903 transition(itos, vtos); 904 locals_index(rbx); 905 __ movl(iaddress(rbx), rax); 906 } 907 908 909 void TemplateTable::lstore() { 910 transition(ltos, vtos); 911 locals_index(rbx); 912 __ movptr(laddress(rbx), rax); 913 NOT_LP64(__ movptr(haddress(rbx), rdx)); 914 } 915 916 void TemplateTable::fstore() { 917 transition(ftos, vtos); 918 locals_index(rbx); 919 __ store_float(faddress(rbx)); 920 } 921 922 void TemplateTable::dstore() { 923 transition(dtos, vtos); 924 locals_index(rbx); 925 __ store_double(daddress(rbx)); 926 } 927 928 void TemplateTable::astore() { 929 transition(vtos, vtos); 930 __ pop_ptr(rax); 931 locals_index(rbx); 932 __ movptr(aaddress(rbx), rax); 933 } 934 935 void TemplateTable::wide_istore() { 936 transition(vtos, vtos); 937 __ pop_i(); 938 locals_index_wide(rbx); 939 __ movl(iaddress(rbx), rax); 940 } 941 942 void TemplateTable::wide_lstore() { 943 transition(vtos, vtos); 944 NOT_LP64(__ pop_l(rax, rdx)); 945 LP64_ONLY(__ pop_l()); 946 locals_index_wide(rbx); 947 __ movptr(laddress(rbx), rax); 948 NOT_LP64(__ movl(haddress(rbx), rdx)); 949 } 950 951 void TemplateTable::wide_fstore() { 952 #ifdef _LP64 953 transition(vtos, vtos); 954 __ pop_f(xmm0); 955 locals_index_wide(rbx); 956 __ movflt(faddress(rbx), xmm0); 957 #else 958 wide_istore(); 959 #endif 960 } 961 962 void TemplateTable::wide_dstore() { 963 #ifdef _LP64 964 transition(vtos, vtos); 965 __ pop_d(xmm0); 966 locals_index_wide(rbx); 967 __ movdbl(daddress(rbx), xmm0); 968 #else 969 wide_lstore(); 970 #endif 971 } 972 973 void TemplateTable::wide_astore() { 974 transition(vtos, vtos); 975 __ pop_ptr(rax); 976 locals_index_wide(rbx); 977 __ movptr(aaddress(rbx), rax); 978 } 979 980 void TemplateTable::iastore() { 981 transition(itos, vtos); 982 __ pop_i(rbx); 983 // rax: value 984 // rbx: index 985 // rdx: array 986 index_check(rdx, rbx); // prefer index in rbx 987 __ movl(Address(rdx, rbx, 988 Address::times_4, 989 arrayOopDesc::base_offset_in_bytes(T_INT)), 990 rax); 991 } 992 993 void TemplateTable::lastore() { 994 transition(ltos, vtos); 995 __ pop_i(rbx); 996 // rax,: low(value) 997 // rcx: array 998 // rdx: high(value) 999 index_check(rcx, rbx); // prefer index in rbx, 1000 // rbx,: index 1001 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax); 1002 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx)); 1003 } 1004 1005 1006 void TemplateTable::fastore() { 1007 transition(ftos, vtos); 1008 __ pop_i(rbx); 1009 // value is in UseSSE >= 1 ? xmm0 : ST(0) 1010 // rbx: index 1011 // rdx: array 1012 index_check(rdx, rbx); // prefer index in rbx 1013 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 1014 } 1015 1016 void TemplateTable::dastore() { 1017 transition(dtos, vtos); 1018 __ pop_i(rbx); 1019 // value is in UseSSE >= 2 ? xmm0 : ST(0) 1020 // rbx: index 1021 // rdx: array 1022 index_check(rdx, rbx); // prefer index in rbx 1023 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 1024 } 1025 1026 void TemplateTable::aastore() { 1027 Label is_null, ok_is_subtype, done; 1028 transition(vtos, vtos); 1029 // stack: ..., array, index, value 1030 __ movptr(rax, at_tos()); // value 1031 __ movl(rcx, at_tos_p1()); // index 1032 __ movptr(rdx, at_tos_p2()); // array 1033 1034 Address element_address(rdx, rcx, 1035 UseCompressedOops? Address::times_4 : Address::times_ptr, 1036 arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 1037 1038 index_check_without_pop(rdx, rcx); // kills rbx 1039 __ testptr(rax, rax); 1040 __ jcc(Assembler::zero, is_null); 1041 1042 // Move subklass into rbx 1043 __ load_klass(rbx, rax); 1044 // Move superklass into rax 1045 __ load_klass(rax, rdx); 1046 __ movptr(rax, Address(rax, 1047 ObjArrayKlass::element_klass_offset())); 1048 // Compress array + index*oopSize + 12 into a single register. Frees rcx. 1049 __ lea(rdx, element_address); 1050 1051 // Generate subtype check. Blows rcx, rdi 1052 // Superklass in rax. Subklass in rbx. 1053 __ gen_subtype_check(rbx, ok_is_subtype); 1054 1055 // Come here on failure 1056 // object is at TOS 1057 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry)); 1058 1059 // Come here on success 1060 __ bind(ok_is_subtype); 1061 1062 // Get the value we will store 1063 __ movptr(rax, at_tos()); 1064 // Now store using the appropriate barrier 1065 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true); 1066 __ jmp(done); 1067 1068 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx] 1069 __ bind(is_null); 1070 __ profile_null_seen(rbx); 1071 1072 // Store a NULL 1073 do_oop_store(_masm, element_address, noreg, _bs->kind(), true); 1074 1075 // Pop stack arguments 1076 __ bind(done); 1077 __ addptr(rsp, 3 * Interpreter::stackElementSize); 1078 } 1079 1080 void TemplateTable::bastore() { 1081 transition(itos, vtos); 1082 __ pop_i(rbx); 1083 // rax: value 1084 // rbx: index 1085 // rdx: array 1086 index_check(rdx, rbx); // prefer index in rbx 1087 __ movb(Address(rdx, rbx, 1088 Address::times_1, 1089 arrayOopDesc::base_offset_in_bytes(T_BYTE)), 1090 rax); 1091 } 1092 1093 void TemplateTable::castore() { 1094 transition(itos, vtos); 1095 __ pop_i(rbx); 1096 // rax: value 1097 // rbx: index 1098 // rdx: array 1099 index_check(rdx, rbx); // prefer index in rbx 1100 __ movw(Address(rdx, rbx, 1101 Address::times_2, 1102 arrayOopDesc::base_offset_in_bytes(T_CHAR)), 1103 rax); 1104 } 1105 1106 1107 void TemplateTable::sastore() { 1108 castore(); 1109 } 1110 1111 void TemplateTable::istore(int n) { 1112 transition(itos, vtos); 1113 __ movl(iaddress(n), rax); 1114 } 1115 1116 void TemplateTable::lstore(int n) { 1117 transition(ltos, vtos); 1118 __ movptr(laddress(n), rax); 1119 NOT_LP64(__ movptr(haddress(n), rdx)); 1120 } 1121 1122 void TemplateTable::fstore(int n) { 1123 transition(ftos, vtos); 1124 __ store_float(faddress(n)); 1125 } 1126 1127 void TemplateTable::dstore(int n) { 1128 transition(dtos, vtos); 1129 __ store_double(daddress(n)); 1130 } 1131 1132 1133 void TemplateTable::astore(int n) { 1134 transition(vtos, vtos); 1135 __ pop_ptr(rax); 1136 __ movptr(aaddress(n), rax); 1137 } 1138 1139 void TemplateTable::pop() { 1140 transition(vtos, vtos); 1141 __ addptr(rsp, Interpreter::stackElementSize); 1142 } 1143 1144 void TemplateTable::pop2() { 1145 transition(vtos, vtos); 1146 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1147 } 1148 1149 1150 void TemplateTable::dup() { 1151 transition(vtos, vtos); 1152 __ load_ptr(0, rax); 1153 __ push_ptr(rax); 1154 // stack: ..., a, a 1155 } 1156 1157 void TemplateTable::dup_x1() { 1158 transition(vtos, vtos); 1159 // stack: ..., a, b 1160 __ load_ptr( 0, rax); // load b 1161 __ load_ptr( 1, rcx); // load a 1162 __ store_ptr(1, rax); // store b 1163 __ store_ptr(0, rcx); // store a 1164 __ push_ptr(rax); // push b 1165 // stack: ..., b, a, b 1166 } 1167 1168 void TemplateTable::dup_x2() { 1169 transition(vtos, vtos); 1170 // stack: ..., a, b, c 1171 __ load_ptr( 0, rax); // load c 1172 __ load_ptr( 2, rcx); // load a 1173 __ store_ptr(2, rax); // store c in a 1174 __ push_ptr(rax); // push c 1175 // stack: ..., c, b, c, c 1176 __ load_ptr( 2, rax); // load b 1177 __ store_ptr(2, rcx); // store a in b 1178 // stack: ..., c, a, c, c 1179 __ store_ptr(1, rax); // store b in c 1180 // stack: ..., c, a, b, c 1181 } 1182 1183 void TemplateTable::dup2() { 1184 transition(vtos, vtos); 1185 // stack: ..., a, b 1186 __ load_ptr(1, rax); // load a 1187 __ push_ptr(rax); // push a 1188 __ load_ptr(1, rax); // load b 1189 __ push_ptr(rax); // push b 1190 // stack: ..., a, b, a, b 1191 } 1192 1193 1194 void TemplateTable::dup2_x1() { 1195 transition(vtos, vtos); 1196 // stack: ..., a, b, c 1197 __ load_ptr( 0, rcx); // load c 1198 __ load_ptr( 1, rax); // load b 1199 __ push_ptr(rax); // push b 1200 __ push_ptr(rcx); // push c 1201 // stack: ..., a, b, c, b, c 1202 __ store_ptr(3, rcx); // store c in b 1203 // stack: ..., a, c, c, b, c 1204 __ load_ptr( 4, rcx); // load a 1205 __ store_ptr(2, rcx); // store a in 2nd c 1206 // stack: ..., a, c, a, b, c 1207 __ store_ptr(4, rax); // store b in a 1208 // stack: ..., b, c, a, b, c 1209 } 1210 1211 void TemplateTable::dup2_x2() { 1212 transition(vtos, vtos); 1213 // stack: ..., a, b, c, d 1214 __ load_ptr( 0, rcx); // load d 1215 __ load_ptr( 1, rax); // load c 1216 __ push_ptr(rax); // push c 1217 __ push_ptr(rcx); // push d 1218 // stack: ..., a, b, c, d, c, d 1219 __ load_ptr( 4, rax); // load b 1220 __ store_ptr(2, rax); // store b in d 1221 __ store_ptr(4, rcx); // store d in b 1222 // stack: ..., a, d, c, b, c, d 1223 __ load_ptr( 5, rcx); // load a 1224 __ load_ptr( 3, rax); // load c 1225 __ store_ptr(3, rcx); // store a in c 1226 __ store_ptr(5, rax); // store c in a 1227 // stack: ..., c, d, a, b, c, d 1228 } 1229 1230 void TemplateTable::swap() { 1231 transition(vtos, vtos); 1232 // stack: ..., a, b 1233 __ load_ptr( 1, rcx); // load a 1234 __ load_ptr( 0, rax); // load b 1235 __ store_ptr(0, rcx); // store a in b 1236 __ store_ptr(1, rax); // store b in a 1237 // stack: ..., b, a 1238 } 1239 1240 void TemplateTable::iop2(Operation op) { 1241 transition(itos, itos); 1242 switch (op) { 1243 case add : __ pop_i(rdx); __ addl (rax, rdx); break; 1244 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break; 1245 case mul : __ pop_i(rdx); __ imull(rax, rdx); break; 1246 case _and : __ pop_i(rdx); __ andl (rax, rdx); break; 1247 case _or : __ pop_i(rdx); __ orl (rax, rdx); break; 1248 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break; 1249 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break; 1250 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break; 1251 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break; 1252 default : ShouldNotReachHere(); 1253 } 1254 } 1255 1256 void TemplateTable::lop2(Operation op) { 1257 transition(ltos, ltos); 1258 #ifdef _LP64 1259 switch (op) { 1260 case add : __ pop_l(rdx); __ addptr(rax, rdx); break; 1261 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break; 1262 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break; 1263 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break; 1264 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break; 1265 default : ShouldNotReachHere(); 1266 } 1267 #else 1268 __ pop_l(rbx, rcx); 1269 switch (op) { 1270 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break; 1271 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx); 1272 __ mov (rax, rbx); __ mov (rdx, rcx); break; 1273 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break; 1274 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break; 1275 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break; 1276 default : ShouldNotReachHere(); 1277 } 1278 #endif 1279 } 1280 1281 void TemplateTable::idiv() { 1282 transition(itos, itos); 1283 __ movl(rcx, rax); 1284 __ pop_i(rax); 1285 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If 1286 // they are not equal, one could do a normal division (no correction 1287 // needed), which may speed up this implementation for the common case. 1288 // (see also JVM spec., p.243 & p.271) 1289 __ corrected_idivl(rcx); 1290 } 1291 1292 void TemplateTable::irem() { 1293 transition(itos, itos); 1294 __ movl(rcx, rax); 1295 __ pop_i(rax); 1296 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If 1297 // they are not equal, one could do a normal division (no correction 1298 // needed), which may speed up this implementation for the common case. 1299 // (see also JVM spec., p.243 & p.271) 1300 __ corrected_idivl(rcx); 1301 __ movl(rax, rdx); 1302 } 1303 1304 void TemplateTable::lmul() { 1305 transition(ltos, ltos); 1306 #ifdef _LP64 1307 __ pop_l(rdx); 1308 __ imulq(rax, rdx); 1309 #else 1310 __ pop_l(rbx, rcx); 1311 __ push(rcx); __ push(rbx); 1312 __ push(rdx); __ push(rax); 1313 __ lmul(2 * wordSize, 0); 1314 __ addptr(rsp, 4 * wordSize); // take off temporaries 1315 #endif 1316 } 1317 1318 void TemplateTable::ldiv() { 1319 transition(ltos, ltos); 1320 #ifdef _LP64 1321 __ mov(rcx, rax); 1322 __ pop_l(rax); 1323 // generate explicit div0 check 1324 __ testq(rcx, rcx); 1325 __ jump_cc(Assembler::zero, 1326 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1327 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If 1328 // they are not equal, one could do a normal division (no correction 1329 // needed), which may speed up this implementation for the common case. 1330 // (see also JVM spec., p.243 & p.271) 1331 __ corrected_idivq(rcx); // kills rbx 1332 #else 1333 __ pop_l(rbx, rcx); 1334 __ push(rcx); __ push(rbx); 1335 __ push(rdx); __ push(rax); 1336 // check if y = 0 1337 __ orl(rax, rdx); 1338 __ jump_cc(Assembler::zero, 1339 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1340 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv)); 1341 __ addptr(rsp, 4 * wordSize); // take off temporaries 1342 #endif 1343 } 1344 1345 void TemplateTable::lrem() { 1346 transition(ltos, ltos); 1347 #ifdef _LP64 1348 __ mov(rcx, rax); 1349 __ pop_l(rax); 1350 __ testq(rcx, rcx); 1351 __ jump_cc(Assembler::zero, 1352 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1353 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If 1354 // they are not equal, one could do a normal division (no correction 1355 // needed), which may speed up this implementation for the common case. 1356 // (see also JVM spec., p.243 & p.271) 1357 __ corrected_idivq(rcx); // kills rbx 1358 __ mov(rax, rdx); 1359 #else 1360 __ pop_l(rbx, rcx); 1361 __ push(rcx); __ push(rbx); 1362 __ push(rdx); __ push(rax); 1363 // check if y = 0 1364 __ orl(rax, rdx); 1365 __ jump_cc(Assembler::zero, 1366 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1367 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem)); 1368 __ addptr(rsp, 4 * wordSize); 1369 #endif 1370 } 1371 1372 void TemplateTable::lshl() { 1373 transition(itos, ltos); 1374 __ movl(rcx, rax); // get shift count 1375 #ifdef _LP64 1376 __ pop_l(rax); // get shift value 1377 __ shlq(rax); 1378 #else 1379 __ pop_l(rax, rdx); // get shift value 1380 __ lshl(rdx, rax); 1381 #endif 1382 } 1383 1384 void TemplateTable::lshr() { 1385 #ifdef _LP64 1386 transition(itos, ltos); 1387 __ movl(rcx, rax); // get shift count 1388 __ pop_l(rax); // get shift value 1389 __ sarq(rax); 1390 #else 1391 transition(itos, ltos); 1392 __ mov(rcx, rax); // get shift count 1393 __ pop_l(rax, rdx); // get shift value 1394 __ lshr(rdx, rax, true); 1395 #endif 1396 } 1397 1398 void TemplateTable::lushr() { 1399 transition(itos, ltos); 1400 #ifdef _LP64 1401 __ movl(rcx, rax); // get shift count 1402 __ pop_l(rax); // get shift value 1403 __ shrq(rax); 1404 #else 1405 __ mov(rcx, rax); // get shift count 1406 __ pop_l(rax, rdx); // get shift value 1407 __ lshr(rdx, rax); 1408 #endif 1409 } 1410 1411 void TemplateTable::fop2(Operation op) { 1412 transition(ftos, ftos); 1413 1414 if (UseSSE >= 1) { 1415 switch (op) { 1416 case add: 1417 __ addss(xmm0, at_rsp()); 1418 __ addptr(rsp, Interpreter::stackElementSize); 1419 break; 1420 case sub: 1421 __ movflt(xmm1, xmm0); 1422 __ pop_f(xmm0); 1423 __ subss(xmm0, xmm1); 1424 break; 1425 case mul: 1426 __ mulss(xmm0, at_rsp()); 1427 __ addptr(rsp, Interpreter::stackElementSize); 1428 break; 1429 case div: 1430 __ movflt(xmm1, xmm0); 1431 __ pop_f(xmm0); 1432 __ divss(xmm0, xmm1); 1433 break; 1434 case rem: 1435 // On x86_64 platforms the SharedRuntime::frem method is called to perform the 1436 // modulo operation. The frem method calls the function 1437 // double fmod(double x, double y) in math.h. The documentation of fmod states: 1438 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN 1439 // (signalling or quiet) is returned. 1440 // 1441 // On x86_32 platforms the FPU is used to perform the modulo operation. The 1442 // reason is that on 32-bit Windows the sign of modulo operations diverges from 1443 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f). 1444 // The fprem instruction used on x86_32 is functionally equivalent to 1445 // SharedRuntime::frem in that it returns a NaN. 1446 #ifdef _LP64 1447 __ movflt(xmm1, xmm0); 1448 __ pop_f(xmm0); 1449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2); 1450 #else 1451 __ push_f(xmm0); 1452 __ pop_f(); 1453 __ fld_s(at_rsp()); 1454 __ fremr(rax); 1455 __ f2ieee(); 1456 __ pop(rax); // pop second operand off the stack 1457 __ push_f(); 1458 __ pop_f(xmm0); 1459 #endif 1460 break; 1461 default: 1462 ShouldNotReachHere(); 1463 break; 1464 } 1465 } else { 1466 #ifdef _LP64 1467 ShouldNotReachHere(); 1468 #else 1469 switch (op) { 1470 case add: __ fadd_s (at_rsp()); break; 1471 case sub: __ fsubr_s(at_rsp()); break; 1472 case mul: __ fmul_s (at_rsp()); break; 1473 case div: __ fdivr_s(at_rsp()); break; 1474 case rem: __ fld_s (at_rsp()); __ fremr(rax); break; 1475 default : ShouldNotReachHere(); 1476 } 1477 __ f2ieee(); 1478 __ pop(rax); // pop second operand off the stack 1479 #endif // _LP64 1480 } 1481 } 1482 1483 void TemplateTable::dop2(Operation op) { 1484 transition(dtos, dtos); 1485 if (UseSSE >= 2) { 1486 switch (op) { 1487 case add: 1488 __ addsd(xmm0, at_rsp()); 1489 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1490 break; 1491 case sub: 1492 __ movdbl(xmm1, xmm0); 1493 __ pop_d(xmm0); 1494 __ subsd(xmm0, xmm1); 1495 break; 1496 case mul: 1497 __ mulsd(xmm0, at_rsp()); 1498 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1499 break; 1500 case div: 1501 __ movdbl(xmm1, xmm0); 1502 __ pop_d(xmm0); 1503 __ divsd(xmm0, xmm1); 1504 break; 1505 case rem: 1506 // Similar to fop2(), the modulo operation is performed using the 1507 // SharedRuntime::drem method (on x86_64 platforms) or using the 1508 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2(). 1509 #ifdef _LP64 1510 __ movdbl(xmm1, xmm0); 1511 __ pop_d(xmm0); 1512 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2); 1513 #else 1514 __ push_d(xmm0); 1515 __ pop_d(); 1516 __ fld_d(at_rsp()); 1517 __ fremr(rax); 1518 __ d2ieee(); 1519 __ pop(rax); 1520 __ pop(rdx); 1521 __ push_d(); 1522 __ pop_d(xmm0); 1523 #endif 1524 break; 1525 default: 1526 ShouldNotReachHere(); 1527 break; 1528 } 1529 } else { 1530 #ifdef _LP64 1531 ShouldNotReachHere(); 1532 #else 1533 switch (op) { 1534 case add: __ fadd_d (at_rsp()); break; 1535 case sub: __ fsubr_d(at_rsp()); break; 1536 case mul: { 1537 Label L_strict; 1538 Label L_join; 1539 const Address access_flags (rcx, Method::access_flags_offset()); 1540 __ get_method(rcx); 1541 __ movl(rcx, access_flags); 1542 __ testl(rcx, JVM_ACC_STRICT); 1543 __ jccb(Assembler::notZero, L_strict); 1544 __ fmul_d (at_rsp()); 1545 __ jmpb(L_join); 1546 __ bind(L_strict); 1547 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1548 __ fmulp(); 1549 __ fmul_d (at_rsp()); 1550 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1551 __ fmulp(); 1552 __ bind(L_join); 1553 break; 1554 } 1555 case div: { 1556 Label L_strict; 1557 Label L_join; 1558 const Address access_flags (rcx, Method::access_flags_offset()); 1559 __ get_method(rcx); 1560 __ movl(rcx, access_flags); 1561 __ testl(rcx, JVM_ACC_STRICT); 1562 __ jccb(Assembler::notZero, L_strict); 1563 __ fdivr_d(at_rsp()); 1564 __ jmp(L_join); 1565 __ bind(L_strict); 1566 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1567 __ fmul_d (at_rsp()); 1568 __ fdivrp(); 1569 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1570 __ fmulp(); 1571 __ bind(L_join); 1572 break; 1573 } 1574 case rem: __ fld_d (at_rsp()); __ fremr(rax); break; 1575 default : ShouldNotReachHere(); 1576 } 1577 __ d2ieee(); 1578 // Pop double precision number from rsp. 1579 __ pop(rax); 1580 __ pop(rdx); 1581 #endif 1582 } 1583 } 1584 1585 void TemplateTable::ineg() { 1586 transition(itos, itos); 1587 __ negl(rax); 1588 } 1589 1590 void TemplateTable::lneg() { 1591 transition(ltos, ltos); 1592 LP64_ONLY(__ negq(rax)); 1593 NOT_LP64(__ lneg(rdx, rax)); 1594 } 1595 1596 // Note: 'double' and 'long long' have 32-bits alignment on x86. 1597 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) { 1598 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address 1599 // of 128-bits operands for SSE instructions. 1600 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF))); 1601 // Store the value to a 128-bits operand. 1602 operand[0] = lo; 1603 operand[1] = hi; 1604 return operand; 1605 } 1606 1607 // Buffer for 128-bits masks used by SSE instructions. 1608 static jlong float_signflip_pool[2*2]; 1609 static jlong double_signflip_pool[2*2]; 1610 1611 void TemplateTable::fneg() { 1612 transition(ftos, ftos); 1613 if (UseSSE >= 1) { 1614 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000); 1615 __ xorps(xmm0, ExternalAddress((address) float_signflip)); 1616 } else { 1617 LP64_ONLY(ShouldNotReachHere()); 1618 NOT_LP64(__ fchs()); 1619 } 1620 } 1621 1622 void TemplateTable::dneg() { 1623 transition(dtos, dtos); 1624 if (UseSSE >= 2) { 1625 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000); 1626 __ xorpd(xmm0, ExternalAddress((address) double_signflip)); 1627 } else { 1628 #ifdef _LP64 1629 ShouldNotReachHere(); 1630 #else 1631 __ fchs(); 1632 #endif 1633 } 1634 } 1635 1636 void TemplateTable::iinc() { 1637 transition(vtos, vtos); 1638 __ load_signed_byte(rdx, at_bcp(2)); // get constant 1639 locals_index(rbx); 1640 __ addl(iaddress(rbx), rdx); 1641 } 1642 1643 void TemplateTable::wide_iinc() { 1644 transition(vtos, vtos); 1645 __ movl(rdx, at_bcp(4)); // get constant 1646 locals_index_wide(rbx); 1647 __ bswapl(rdx); // swap bytes & sign-extend constant 1648 __ sarl(rdx, 16); 1649 __ addl(iaddress(rbx), rdx); 1650 // Note: should probably use only one movl to get both 1651 // the index and the constant -> fix this 1652 } 1653 1654 void TemplateTable::convert() { 1655 #ifdef _LP64 1656 // Checking 1657 #ifdef ASSERT 1658 { 1659 TosState tos_in = ilgl; 1660 TosState tos_out = ilgl; 1661 switch (bytecode()) { 1662 case Bytecodes::_i2l: // fall through 1663 case Bytecodes::_i2f: // fall through 1664 case Bytecodes::_i2d: // fall through 1665 case Bytecodes::_i2b: // fall through 1666 case Bytecodes::_i2c: // fall through 1667 case Bytecodes::_i2s: tos_in = itos; break; 1668 case Bytecodes::_l2i: // fall through 1669 case Bytecodes::_l2f: // fall through 1670 case Bytecodes::_l2d: tos_in = ltos; break; 1671 case Bytecodes::_f2i: // fall through 1672 case Bytecodes::_f2l: // fall through 1673 case Bytecodes::_f2d: tos_in = ftos; break; 1674 case Bytecodes::_d2i: // fall through 1675 case Bytecodes::_d2l: // fall through 1676 case Bytecodes::_d2f: tos_in = dtos; break; 1677 default : ShouldNotReachHere(); 1678 } 1679 switch (bytecode()) { 1680 case Bytecodes::_l2i: // fall through 1681 case Bytecodes::_f2i: // fall through 1682 case Bytecodes::_d2i: // fall through 1683 case Bytecodes::_i2b: // fall through 1684 case Bytecodes::_i2c: // fall through 1685 case Bytecodes::_i2s: tos_out = itos; break; 1686 case Bytecodes::_i2l: // fall through 1687 case Bytecodes::_f2l: // fall through 1688 case Bytecodes::_d2l: tos_out = ltos; break; 1689 case Bytecodes::_i2f: // fall through 1690 case Bytecodes::_l2f: // fall through 1691 case Bytecodes::_d2f: tos_out = ftos; break; 1692 case Bytecodes::_i2d: // fall through 1693 case Bytecodes::_l2d: // fall through 1694 case Bytecodes::_f2d: tos_out = dtos; break; 1695 default : ShouldNotReachHere(); 1696 } 1697 transition(tos_in, tos_out); 1698 } 1699 #endif // ASSERT 1700 1701 static const int64_t is_nan = 0x8000000000000000L; 1702 1703 // Conversion 1704 switch (bytecode()) { 1705 case Bytecodes::_i2l: 1706 __ movslq(rax, rax); 1707 break; 1708 case Bytecodes::_i2f: 1709 __ cvtsi2ssl(xmm0, rax); 1710 break; 1711 case Bytecodes::_i2d: 1712 __ cvtsi2sdl(xmm0, rax); 1713 break; 1714 case Bytecodes::_i2b: 1715 __ movsbl(rax, rax); 1716 break; 1717 case Bytecodes::_i2c: 1718 __ movzwl(rax, rax); 1719 break; 1720 case Bytecodes::_i2s: 1721 __ movswl(rax, rax); 1722 break; 1723 case Bytecodes::_l2i: 1724 __ movl(rax, rax); 1725 break; 1726 case Bytecodes::_l2f: 1727 __ cvtsi2ssq(xmm0, rax); 1728 break; 1729 case Bytecodes::_l2d: 1730 __ cvtsi2sdq(xmm0, rax); 1731 break; 1732 case Bytecodes::_f2i: 1733 { 1734 Label L; 1735 __ cvttss2sil(rax, xmm0); 1736 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow? 1737 __ jcc(Assembler::notEqual, L); 1738 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1739 __ bind(L); 1740 } 1741 break; 1742 case Bytecodes::_f2l: 1743 { 1744 Label L; 1745 __ cvttss2siq(rax, xmm0); 1746 // NaN or overflow/underflow? 1747 __ cmp64(rax, ExternalAddress((address) &is_nan)); 1748 __ jcc(Assembler::notEqual, L); 1749 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1750 __ bind(L); 1751 } 1752 break; 1753 case Bytecodes::_f2d: 1754 __ cvtss2sd(xmm0, xmm0); 1755 break; 1756 case Bytecodes::_d2i: 1757 { 1758 Label L; 1759 __ cvttsd2sil(rax, xmm0); 1760 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow? 1761 __ jcc(Assembler::notEqual, L); 1762 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1); 1763 __ bind(L); 1764 } 1765 break; 1766 case Bytecodes::_d2l: 1767 { 1768 Label L; 1769 __ cvttsd2siq(rax, xmm0); 1770 // NaN or overflow/underflow? 1771 __ cmp64(rax, ExternalAddress((address) &is_nan)); 1772 __ jcc(Assembler::notEqual, L); 1773 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1); 1774 __ bind(L); 1775 } 1776 break; 1777 case Bytecodes::_d2f: 1778 __ cvtsd2ss(xmm0, xmm0); 1779 break; 1780 default: 1781 ShouldNotReachHere(); 1782 } 1783 #else 1784 // Checking 1785 #ifdef ASSERT 1786 { TosState tos_in = ilgl; 1787 TosState tos_out = ilgl; 1788 switch (bytecode()) { 1789 case Bytecodes::_i2l: // fall through 1790 case Bytecodes::_i2f: // fall through 1791 case Bytecodes::_i2d: // fall through 1792 case Bytecodes::_i2b: // fall through 1793 case Bytecodes::_i2c: // fall through 1794 case Bytecodes::_i2s: tos_in = itos; break; 1795 case Bytecodes::_l2i: // fall through 1796 case Bytecodes::_l2f: // fall through 1797 case Bytecodes::_l2d: tos_in = ltos; break; 1798 case Bytecodes::_f2i: // fall through 1799 case Bytecodes::_f2l: // fall through 1800 case Bytecodes::_f2d: tos_in = ftos; break; 1801 case Bytecodes::_d2i: // fall through 1802 case Bytecodes::_d2l: // fall through 1803 case Bytecodes::_d2f: tos_in = dtos; break; 1804 default : ShouldNotReachHere(); 1805 } 1806 switch (bytecode()) { 1807 case Bytecodes::_l2i: // fall through 1808 case Bytecodes::_f2i: // fall through 1809 case Bytecodes::_d2i: // fall through 1810 case Bytecodes::_i2b: // fall through 1811 case Bytecodes::_i2c: // fall through 1812 case Bytecodes::_i2s: tos_out = itos; break; 1813 case Bytecodes::_i2l: // fall through 1814 case Bytecodes::_f2l: // fall through 1815 case Bytecodes::_d2l: tos_out = ltos; break; 1816 case Bytecodes::_i2f: // fall through 1817 case Bytecodes::_l2f: // fall through 1818 case Bytecodes::_d2f: tos_out = ftos; break; 1819 case Bytecodes::_i2d: // fall through 1820 case Bytecodes::_l2d: // fall through 1821 case Bytecodes::_f2d: tos_out = dtos; break; 1822 default : ShouldNotReachHere(); 1823 } 1824 transition(tos_in, tos_out); 1825 } 1826 #endif // ASSERT 1827 1828 // Conversion 1829 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation) 1830 switch (bytecode()) { 1831 case Bytecodes::_i2l: 1832 __ extend_sign(rdx, rax); 1833 break; 1834 case Bytecodes::_i2f: 1835 if (UseSSE >= 1) { 1836 __ cvtsi2ssl(xmm0, rax); 1837 } else { 1838 __ push(rax); // store int on tos 1839 __ fild_s(at_rsp()); // load int to ST0 1840 __ f2ieee(); // truncate to float size 1841 __ pop(rcx); // adjust rsp 1842 } 1843 break; 1844 case Bytecodes::_i2d: 1845 if (UseSSE >= 2) { 1846 __ cvtsi2sdl(xmm0, rax); 1847 } else { 1848 __ push(rax); // add one slot for d2ieee() 1849 __ push(rax); // store int on tos 1850 __ fild_s(at_rsp()); // load int to ST0 1851 __ d2ieee(); // truncate to double size 1852 __ pop(rcx); // adjust rsp 1853 __ pop(rcx); 1854 } 1855 break; 1856 case Bytecodes::_i2b: 1857 __ shll(rax, 24); // truncate upper 24 bits 1858 __ sarl(rax, 24); // and sign-extend byte 1859 LP64_ONLY(__ movsbl(rax, rax)); 1860 break; 1861 case Bytecodes::_i2c: 1862 __ andl(rax, 0xFFFF); // truncate upper 16 bits 1863 LP64_ONLY(__ movzwl(rax, rax)); 1864 break; 1865 case Bytecodes::_i2s: 1866 __ shll(rax, 16); // truncate upper 16 bits 1867 __ sarl(rax, 16); // and sign-extend short 1868 LP64_ONLY(__ movswl(rax, rax)); 1869 break; 1870 case Bytecodes::_l2i: 1871 /* nothing to do */ 1872 break; 1873 case Bytecodes::_l2f: 1874 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert 1875 // 64-bit long values to floats. On 32-bit platforms it is not possible 1876 // to use that instruction with 64-bit operands, therefore the FPU is 1877 // used to perform the conversion. 1878 __ push(rdx); // store long on tos 1879 __ push(rax); 1880 __ fild_d(at_rsp()); // load long to ST0 1881 __ f2ieee(); // truncate to float size 1882 __ pop(rcx); // adjust rsp 1883 __ pop(rcx); 1884 if (UseSSE >= 1) { 1885 __ push_f(); 1886 __ pop_f(xmm0); 1887 } 1888 break; 1889 case Bytecodes::_l2d: 1890 // On 32-bit platforms the FPU is used for conversion because on 1891 // 32-bit platforms it is not not possible to use the cvtsi2sdq 1892 // instruction with 64-bit operands. 1893 __ push(rdx); // store long on tos 1894 __ push(rax); 1895 __ fild_d(at_rsp()); // load long to ST0 1896 __ d2ieee(); // truncate to double size 1897 __ pop(rcx); // adjust rsp 1898 __ pop(rcx); 1899 if (UseSSE >= 2) { 1900 __ push_d(); 1901 __ pop_d(xmm0); 1902 } 1903 break; 1904 case Bytecodes::_f2i: 1905 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs 1906 // as it returns 0 for any NaN. 1907 if (UseSSE >= 1) { 1908 __ push_f(xmm0); 1909 } else { 1910 __ push(rcx); // reserve space for argument 1911 __ fstp_s(at_rsp()); // pass float argument on stack 1912 } 1913 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1914 break; 1915 case Bytecodes::_f2l: 1916 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs 1917 // as it returns 0 for any NaN. 1918 if (UseSSE >= 1) { 1919 __ push_f(xmm0); 1920 } else { 1921 __ push(rcx); // reserve space for argument 1922 __ fstp_s(at_rsp()); // pass float argument on stack 1923 } 1924 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1925 break; 1926 case Bytecodes::_f2d: 1927 if (UseSSE < 1) { 1928 /* nothing to do */ 1929 } else if (UseSSE == 1) { 1930 __ push_f(xmm0); 1931 __ pop_f(); 1932 } else { // UseSSE >= 2 1933 __ cvtss2sd(xmm0, xmm0); 1934 } 1935 break; 1936 case Bytecodes::_d2i: 1937 if (UseSSE >= 2) { 1938 __ push_d(xmm0); 1939 } else { 1940 __ push(rcx); // reserve space for argument 1941 __ push(rcx); 1942 __ fstp_d(at_rsp()); // pass double argument on stack 1943 } 1944 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2); 1945 break; 1946 case Bytecodes::_d2l: 1947 if (UseSSE >= 2) { 1948 __ push_d(xmm0); 1949 } else { 1950 __ push(rcx); // reserve space for argument 1951 __ push(rcx); 1952 __ fstp_d(at_rsp()); // pass double argument on stack 1953 } 1954 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2); 1955 break; 1956 case Bytecodes::_d2f: 1957 if (UseSSE <= 1) { 1958 __ push(rcx); // reserve space for f2ieee() 1959 __ f2ieee(); // truncate to float size 1960 __ pop(rcx); // adjust rsp 1961 if (UseSSE == 1) { 1962 // The cvtsd2ss instruction is not available if UseSSE==1, therefore 1963 // the conversion is performed using the FPU in this case. 1964 __ push_f(); 1965 __ pop_f(xmm0); 1966 } 1967 } else { // UseSSE >= 2 1968 __ cvtsd2ss(xmm0, xmm0); 1969 } 1970 break; 1971 default : 1972 ShouldNotReachHere(); 1973 } 1974 #endif 1975 } 1976 1977 void TemplateTable::lcmp() { 1978 transition(ltos, itos); 1979 #ifdef _LP64 1980 Label done; 1981 __ pop_l(rdx); 1982 __ cmpq(rdx, rax); 1983 __ movl(rax, -1); 1984 __ jccb(Assembler::less, done); 1985 __ setb(Assembler::notEqual, rax); 1986 __ movzbl(rax, rax); 1987 __ bind(done); 1988 #else 1989 1990 // y = rdx:rax 1991 __ pop_l(rbx, rcx); // get x = rcx:rbx 1992 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y) 1993 __ mov(rax, rcx); 1994 #endif 1995 } 1996 1997 void TemplateTable::float_cmp(bool is_float, int unordered_result) { 1998 if ((is_float && UseSSE >= 1) || 1999 (!is_float && UseSSE >= 2)) { 2000 Label done; 2001 if (is_float) { 2002 // XXX get rid of pop here, use ... reg, mem32 2003 __ pop_f(xmm1); 2004 __ ucomiss(xmm1, xmm0); 2005 } else { 2006 // XXX get rid of pop here, use ... reg, mem64 2007 __ pop_d(xmm1); 2008 __ ucomisd(xmm1, xmm0); 2009 } 2010 if (unordered_result < 0) { 2011 __ movl(rax, -1); 2012 __ jccb(Assembler::parity, done); 2013 __ jccb(Assembler::below, done); 2014 __ setb(Assembler::notEqual, rdx); 2015 __ movzbl(rax, rdx); 2016 } else { 2017 __ movl(rax, 1); 2018 __ jccb(Assembler::parity, done); 2019 __ jccb(Assembler::above, done); 2020 __ movl(rax, 0); 2021 __ jccb(Assembler::equal, done); 2022 __ decrementl(rax); 2023 } 2024 __ bind(done); 2025 } else { 2026 #ifdef _LP64 2027 ShouldNotReachHere(); 2028 #else 2029 if (is_float) { 2030 __ fld_s(at_rsp()); 2031 } else { 2032 __ fld_d(at_rsp()); 2033 __ pop(rdx); 2034 } 2035 __ pop(rcx); 2036 __ fcmp2int(rax, unordered_result < 0); 2037 #endif // _LP64 2038 } 2039 } 2040 2041 void TemplateTable::branch(bool is_jsr, bool is_wide) { 2042 __ get_method(rcx); // rcx holds method 2043 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx 2044 // holds bumped taken count 2045 2046 const ByteSize be_offset = MethodCounters::backedge_counter_offset() + 2047 InvocationCounter::counter_offset(); 2048 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() + 2049 InvocationCounter::counter_offset(); 2050 2051 // Load up edx with the branch displacement 2052 if (is_wide) { 2053 __ movl(rdx, at_bcp(1)); 2054 } else { 2055 __ load_signed_short(rdx, at_bcp(1)); 2056 } 2057 __ bswapl(rdx); 2058 2059 if (!is_wide) { 2060 __ sarl(rdx, 16); 2061 } 2062 LP64_ONLY(__ movl2ptr(rdx, rdx)); 2063 2064 // Handle all the JSR stuff here, then exit. 2065 // It's much shorter and cleaner than intermingling with the non-JSR 2066 // normal-branch stuff occurring below. 2067 if (is_jsr) { 2068 // Pre-load the next target bytecode into rbx 2069 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0)); 2070 2071 // compute return address as bci in rax 2072 __ lea(rax, at_bcp((is_wide ? 5 : 3) - 2073 in_bytes(ConstMethod::codes_offset()))); 2074 __ subptr(rax, Address(rcx, Method::const_offset())); 2075 // Adjust the bcp in r13 by the displacement in rdx 2076 __ addptr(rbcp, rdx); 2077 // jsr returns atos that is not an oop 2078 __ push_i(rax); 2079 __ dispatch_only(vtos); 2080 return; 2081 } 2082 2083 // Normal (non-jsr) branch handling 2084 2085 // Adjust the bcp in r13 by the displacement in rdx 2086 __ addptr(rbcp, rdx); 2087 2088 assert(UseLoopCounter || !UseOnStackReplacement, 2089 "on-stack-replacement requires loop counters"); 2090 Label backedge_counter_overflow; 2091 Label profile_method; 2092 Label dispatch; 2093 if (UseLoopCounter) { 2094 // increment backedge counter for backward branches 2095 // rax: MDO 2096 // rbx: MDO bumped taken-count 2097 // rcx: method 2098 // rdx: target offset 2099 // r13: target bcp 2100 // r14: locals pointer 2101 __ testl(rdx, rdx); // check if forward or backward branch 2102 __ jcc(Assembler::positive, dispatch); // count only if backward branch 2103 2104 // check if MethodCounters exists 2105 Label has_counters; 2106 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 2107 __ testptr(rax, rax); 2108 __ jcc(Assembler::notZero, has_counters); 2109 __ push(rdx); 2110 __ push(rcx); 2111 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), 2112 rcx); 2113 __ pop(rcx); 2114 __ pop(rdx); 2115 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 2116 __ jcc(Assembler::zero, dispatch); 2117 __ bind(has_counters); 2118 2119 if (TieredCompilation) { 2120 Label no_mdo; 2121 int increment = InvocationCounter::count_increment; 2122 if (ProfileInterpreter) { 2123 // Are we profiling? 2124 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset()))); 2125 __ testptr(rbx, rbx); 2126 __ jccb(Assembler::zero, no_mdo); 2127 // Increment the MDO backedge counter 2128 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) + 2129 in_bytes(InvocationCounter::counter_offset())); 2130 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset())); 2131 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 2132 rax, false, Assembler::zero, &backedge_counter_overflow); 2133 __ jmp(dispatch); 2134 } 2135 __ bind(no_mdo); 2136 // Increment backedge counter in MethodCounters* 2137 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 2138 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset())); 2139 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask, 2140 rax, false, Assembler::zero, &backedge_counter_overflow); 2141 } else { // not TieredCompilation 2142 // increment counter 2143 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 2144 __ movl(rax, Address(rcx, be_offset)); // load backedge counter 2145 __ incrementl(rax, InvocationCounter::count_increment); // increment counter 2146 __ movl(Address(rcx, be_offset), rax); // store counter 2147 2148 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter 2149 2150 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits 2151 __ addl(rax, Address(rcx, be_offset)); // add both counters 2152 2153 if (ProfileInterpreter) { 2154 // Test to see if we should create a method data oop 2155 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset()))); 2156 __ jcc(Assembler::less, dispatch); 2157 2158 // if no method data exists, go to profile method 2159 __ test_method_data_pointer(rax, profile_method); 2160 2161 if (UseOnStackReplacement) { 2162 // check for overflow against rbx which is the MDO taken count 2163 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 2164 __ jcc(Assembler::below, dispatch); 2165 2166 // When ProfileInterpreter is on, the backedge_count comes 2167 // from the MethodData*, which value does not get reset on 2168 // the call to frequency_counter_overflow(). To avoid 2169 // excessive calls to the overflow routine while the method is 2170 // being compiled, add a second test to make sure the overflow 2171 // function is called only once every overflow_frequency. 2172 const int overflow_frequency = 1024; 2173 __ andl(rbx, overflow_frequency - 1); 2174 __ jcc(Assembler::zero, backedge_counter_overflow); 2175 2176 } 2177 } else { 2178 if (UseOnStackReplacement) { 2179 // check for overflow against rax, which is the sum of the 2180 // counters 2181 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 2182 __ jcc(Assembler::aboveEqual, backedge_counter_overflow); 2183 2184 } 2185 } 2186 } 2187 __ bind(dispatch); 2188 } 2189 2190 // Pre-load the next target bytecode into rbx 2191 __ load_unsigned_byte(rbx, Address(rbcp, 0)); 2192 2193 // continue with the bytecode @ target 2194 // rax: return bci for jsr's, unused otherwise 2195 // rbx: target bytecode 2196 // r13: target bcp 2197 __ dispatch_only(vtos); 2198 2199 if (UseLoopCounter) { 2200 if (ProfileInterpreter) { 2201 // Out-of-line code to allocate method data oop. 2202 __ bind(profile_method); 2203 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 2204 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode 2205 __ set_method_data_pointer_for_bcp(); 2206 __ jmp(dispatch); 2207 } 2208 2209 if (UseOnStackReplacement) { 2210 // invocation counter overflow 2211 __ bind(backedge_counter_overflow); 2212 __ negptr(rdx); 2213 __ addptr(rdx, rbcp); // branch bcp 2214 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp) 2215 __ call_VM(noreg, 2216 CAST_FROM_FN_PTR(address, 2217 InterpreterRuntime::frequency_counter_overflow), 2218 rdx); 2219 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode 2220 2221 // rax: osr nmethod (osr ok) or NULL (osr not possible) 2222 // rbx: target bytecode 2223 // rdx: scratch 2224 // r14: locals pointer 2225 // r13: bcp 2226 __ testptr(rax, rax); // test result 2227 __ jcc(Assembler::zero, dispatch); // no osr if null 2228 // nmethod may have been invalidated (VM may block upon call_VM return) 2229 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use); 2230 __ jcc(Assembler::notEqual, dispatch); 2231 2232 // We have the address of an on stack replacement routine in rax 2233 // We need to prepare to execute the OSR method. First we must 2234 // migrate the locals and monitors off of the stack. 2235 2236 LP64_ONLY(__ mov(r13, rax)); // save the nmethod 2237 NOT_LP64(__ mov(rbx, rax)); // save the nmethod 2238 NOT_LP64(__ get_thread(rcx)); 2239 2240 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 2241 2242 // rax is OSR buffer, move it to expected parameter location 2243 LP64_ONLY(__ mov(j_rarg0, rax)); 2244 NOT_LP64(__ mov(rcx, rax)); 2245 // We use j_rarg definitions here so that registers don't conflict as parameter 2246 // registers change across platforms as we are in the midst of a calling 2247 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters. 2248 2249 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi); 2250 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx); 2251 2252 2253 // pop the interpreter frame 2254 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 2255 __ leave(); // remove frame anchor 2256 __ pop(retaddr); // get return address 2257 __ mov(rsp, sender_sp); // set sp to sender sp 2258 // Ensure compiled code always sees stack at proper alignment 2259 __ andptr(rsp, -(StackAlignmentInBytes)); 2260 2261 // unlike x86 we need no specialized return from compiled code 2262 // to the interpreter or the call stub. 2263 2264 // push the return address 2265 __ push(retaddr); 2266 2267 // and begin the OSR nmethod 2268 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset()))); 2269 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset()))); 2270 } 2271 } 2272 } 2273 2274 void TemplateTable::if_0cmp(Condition cc) { 2275 transition(itos, vtos); 2276 // assume branch is more often taken than not (loops use backward branches) 2277 Label not_taken; 2278 __ testl(rax, rax); 2279 __ jcc(j_not(cc), not_taken); 2280 branch(false, false); 2281 __ bind(not_taken); 2282 __ profile_not_taken_branch(rax); 2283 } 2284 2285 void TemplateTable::if_icmp(Condition cc) { 2286 transition(itos, vtos); 2287 // assume branch is more often taken than not (loops use backward branches) 2288 Label not_taken; 2289 __ pop_i(rdx); 2290 __ cmpl(rdx, rax); 2291 __ jcc(j_not(cc), not_taken); 2292 branch(false, false); 2293 __ bind(not_taken); 2294 __ profile_not_taken_branch(rax); 2295 } 2296 2297 void TemplateTable::if_nullcmp(Condition cc) { 2298 transition(atos, vtos); 2299 // assume branch is more often taken than not (loops use backward branches) 2300 Label not_taken; 2301 __ testptr(rax, rax); 2302 __ jcc(j_not(cc), not_taken); 2303 branch(false, false); 2304 __ bind(not_taken); 2305 __ profile_not_taken_branch(rax); 2306 } 2307 2308 void TemplateTable::if_acmp(Condition cc) { 2309 transition(atos, vtos); 2310 // assume branch is more often taken than not (loops use backward branches) 2311 Label not_taken; 2312 __ pop_ptr(rdx); 2313 __ cmpptr(rdx, rax); 2314 __ jcc(j_not(cc), not_taken); 2315 branch(false, false); 2316 __ bind(not_taken); 2317 __ profile_not_taken_branch(rax); 2318 } 2319 2320 void TemplateTable::ret() { 2321 transition(vtos, vtos); 2322 locals_index(rbx); 2323 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp 2324 NOT_LP64(__ movptr(rbx, iaddress(rbx))); 2325 __ profile_ret(rbx, rcx); 2326 __ get_method(rax); 2327 __ movptr(rbcp, Address(rax, Method::const_offset())); 2328 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, 2329 ConstMethod::codes_offset())); 2330 __ dispatch_next(vtos); 2331 } 2332 2333 void TemplateTable::wide_ret() { 2334 transition(vtos, vtos); 2335 locals_index_wide(rbx); 2336 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp 2337 __ profile_ret(rbx, rcx); 2338 __ get_method(rax); 2339 __ movptr(rbcp, Address(rax, Method::const_offset())); 2340 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset())); 2341 __ dispatch_next(vtos); 2342 } 2343 2344 void TemplateTable::tableswitch() { 2345 Label default_case, continue_execution; 2346 transition(itos, vtos); 2347 2348 // align r13/rsi 2349 __ lea(rbx, at_bcp(BytesPerInt)); 2350 __ andptr(rbx, -BytesPerInt); 2351 // load lo & hi 2352 __ movl(rcx, Address(rbx, BytesPerInt)); 2353 __ movl(rdx, Address(rbx, 2 * BytesPerInt)); 2354 __ bswapl(rcx); 2355 __ bswapl(rdx); 2356 // check against lo & hi 2357 __ cmpl(rax, rcx); 2358 __ jcc(Assembler::less, default_case); 2359 __ cmpl(rax, rdx); 2360 __ jcc(Assembler::greater, default_case); 2361 // lookup dispatch offset 2362 __ subl(rax, rcx); 2363 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt)); 2364 __ profile_switch_case(rax, rbx, rcx); 2365 // continue execution 2366 __ bind(continue_execution); 2367 __ bswapl(rdx); 2368 LP64_ONLY(__ movl2ptr(rdx, rdx)); 2369 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1)); 2370 __ addptr(rbcp, rdx); 2371 __ dispatch_only(vtos); 2372 // handle default 2373 __ bind(default_case); 2374 __ profile_switch_default(rax); 2375 __ movl(rdx, Address(rbx, 0)); 2376 __ jmp(continue_execution); 2377 } 2378 2379 void TemplateTable::lookupswitch() { 2380 transition(itos, itos); 2381 __ stop("lookupswitch bytecode should have been rewritten"); 2382 } 2383 2384 void TemplateTable::fast_linearswitch() { 2385 transition(itos, vtos); 2386 Label loop_entry, loop, found, continue_execution; 2387 // bswap rax so we can avoid bswapping the table entries 2388 __ bswapl(rax); 2389 // align r13 2390 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of 2391 // this instruction (change offsets 2392 // below) 2393 __ andptr(rbx, -BytesPerInt); 2394 // set counter 2395 __ movl(rcx, Address(rbx, BytesPerInt)); 2396 __ bswapl(rcx); 2397 __ jmpb(loop_entry); 2398 // table search 2399 __ bind(loop); 2400 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt)); 2401 __ jcc(Assembler::equal, found); 2402 __ bind(loop_entry); 2403 __ decrementl(rcx); 2404 __ jcc(Assembler::greaterEqual, loop); 2405 // default case 2406 __ profile_switch_default(rax); 2407 __ movl(rdx, Address(rbx, 0)); 2408 __ jmp(continue_execution); 2409 // entry found -> get offset 2410 __ bind(found); 2411 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt)); 2412 __ profile_switch_case(rcx, rax, rbx); 2413 // continue execution 2414 __ bind(continue_execution); 2415 __ bswapl(rdx); 2416 __ movl2ptr(rdx, rdx); 2417 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1)); 2418 __ addptr(rbcp, rdx); 2419 __ dispatch_only(vtos); 2420 } 2421 2422 void TemplateTable::fast_binaryswitch() { 2423 transition(itos, vtos); 2424 // Implementation using the following core algorithm: 2425 // 2426 // int binary_search(int key, LookupswitchPair* array, int n) { 2427 // // Binary search according to "Methodik des Programmierens" by 2428 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 2429 // int i = 0; 2430 // int j = n; 2431 // while (i+1 < j) { 2432 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 2433 // // with Q: for all i: 0 <= i < n: key < a[i] 2434 // // where a stands for the array and assuming that the (inexisting) 2435 // // element a[n] is infinitely big. 2436 // int h = (i + j) >> 1; 2437 // // i < h < j 2438 // if (key < array[h].fast_match()) { 2439 // j = h; 2440 // } else { 2441 // i = h; 2442 // } 2443 // } 2444 // // R: a[i] <= key < a[i+1] or Q 2445 // // (i.e., if key is within array, i is the correct index) 2446 // return i; 2447 // } 2448 2449 // Register allocation 2450 const Register key = rax; // already set (tosca) 2451 const Register array = rbx; 2452 const Register i = rcx; 2453 const Register j = rdx; 2454 const Register h = rdi; 2455 const Register temp = rsi; 2456 2457 // Find array start 2458 NOT_LP64(__ save_bcp()); 2459 2460 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to 2461 // get rid of this 2462 // instruction (change 2463 // offsets below) 2464 __ andptr(array, -BytesPerInt); 2465 2466 // Initialize i & j 2467 __ xorl(i, i); // i = 0; 2468 __ movl(j, Address(array, -BytesPerInt)); // j = length(array); 2469 2470 // Convert j into native byteordering 2471 __ bswapl(j); 2472 2473 // And start 2474 Label entry; 2475 __ jmp(entry); 2476 2477 // binary search loop 2478 { 2479 Label loop; 2480 __ bind(loop); 2481 // int h = (i + j) >> 1; 2482 __ leal(h, Address(i, j, Address::times_1)); // h = i + j; 2483 __ sarl(h, 1); // h = (i + j) >> 1; 2484 // if (key < array[h].fast_match()) { 2485 // j = h; 2486 // } else { 2487 // i = h; 2488 // } 2489 // Convert array[h].match to native byte-ordering before compare 2490 __ movl(temp, Address(array, h, Address::times_8)); 2491 __ bswapl(temp); 2492 __ cmpl(key, temp); 2493 // j = h if (key < array[h].fast_match()) 2494 __ cmov32(Assembler::less, j, h); 2495 // i = h if (key >= array[h].fast_match()) 2496 __ cmov32(Assembler::greaterEqual, i, h); 2497 // while (i+1 < j) 2498 __ bind(entry); 2499 __ leal(h, Address(i, 1)); // i+1 2500 __ cmpl(h, j); // i+1 < j 2501 __ jcc(Assembler::less, loop); 2502 } 2503 2504 // end of binary search, result index is i (must check again!) 2505 Label default_case; 2506 // Convert array[i].match to native byte-ordering before compare 2507 __ movl(temp, Address(array, i, Address::times_8)); 2508 __ bswapl(temp); 2509 __ cmpl(key, temp); 2510 __ jcc(Assembler::notEqual, default_case); 2511 2512 // entry found -> j = offset 2513 __ movl(j , Address(array, i, Address::times_8, BytesPerInt)); 2514 __ profile_switch_case(i, key, array); 2515 __ bswapl(j); 2516 LP64_ONLY(__ movslq(j, j)); 2517 2518 NOT_LP64(__ restore_bcp()); 2519 NOT_LP64(__ restore_locals()); // restore rdi 2520 2521 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1)); 2522 __ addptr(rbcp, j); 2523 __ dispatch_only(vtos); 2524 2525 // default case -> j = default offset 2526 __ bind(default_case); 2527 __ profile_switch_default(i); 2528 __ movl(j, Address(array, -2 * BytesPerInt)); 2529 __ bswapl(j); 2530 LP64_ONLY(__ movslq(j, j)); 2531 2532 NOT_LP64(__ restore_bcp()); 2533 NOT_LP64(__ restore_locals()); 2534 2535 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1)); 2536 __ addptr(rbcp, j); 2537 __ dispatch_only(vtos); 2538 } 2539 2540 void TemplateTable::_return(TosState state) { 2541 transition(state, state); 2542 2543 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax); 2544 2545 assert(_desc->calls_vm(), 2546 "inconsistent calls_vm information"); // call in remove_activation 2547 2548 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 2549 assert(state == vtos, "only valid state"); 2550 __ movptr(robj, aaddress(0)); 2551 __ load_klass(rdi, robj); 2552 __ movl(rdi, Address(rdi, Klass::access_flags_offset())); 2553 __ testl(rdi, JVM_ACC_HAS_FINALIZER); 2554 Label skip_register_finalizer; 2555 __ jcc(Assembler::zero, skip_register_finalizer); 2556 2557 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj); 2558 2559 __ bind(skip_register_finalizer); 2560 } 2561 2562 __ remove_activation(state, rbcp); 2563 __ jmp(rbcp); 2564 } 2565 2566 // ---------------------------------------------------------------------------- 2567 // Volatile variables demand their effects be made known to all CPU's 2568 // in order. Store buffers on most chips allow reads & writes to 2569 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode 2570 // without some kind of memory barrier (i.e., it's not sufficient that 2571 // the interpreter does not reorder volatile references, the hardware 2572 // also must not reorder them). 2573 // 2574 // According to the new Java Memory Model (JMM): 2575 // (1) All volatiles are serialized wrt to each other. ALSO reads & 2576 // writes act as aquire & release, so: 2577 // (2) A read cannot let unrelated NON-volatile memory refs that 2578 // happen after the read float up to before the read. It's OK for 2579 // non-volatile memory refs that happen before the volatile read to 2580 // float down below it. 2581 // (3) Similar a volatile write cannot let unrelated NON-volatile 2582 // memory refs that happen BEFORE the write float down to after the 2583 // write. It's OK for non-volatile memory refs that happen after the 2584 // volatile write to float up before it. 2585 // 2586 // We only put in barriers around volatile refs (they are expensive), 2587 // not _between_ memory refs (that would require us to track the 2588 // flavor of the previous memory refs). Requirements (2) and (3) 2589 // require some barriers before volatile stores and after volatile 2590 // loads. These nearly cover requirement (1) but miss the 2591 // volatile-store-volatile-load case. This final case is placed after 2592 // volatile-stores although it could just as well go before 2593 // volatile-loads. 2594 2595 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) { 2596 // Helper function to insert a is-volatile test and memory barrier 2597 if(!os::is_MP()) return; // Not needed on single CPU 2598 __ membar(order_constraint); 2599 } 2600 2601 void TemplateTable::resolve_cache_and_index(int byte_no, 2602 Register Rcache, 2603 Register index, 2604 size_t index_size) { 2605 const Register temp = rbx; 2606 assert_different_registers(Rcache, index, temp); 2607 2608 Label resolved; 2609 2610 Bytecodes::Code code = bytecode(); 2611 switch (code) { 2612 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break; 2613 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break; 2614 } 2615 2616 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 2617 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size); 2618 __ cmpl(temp, code); // have we resolved this bytecode? 2619 __ jcc(Assembler::equal, resolved); 2620 2621 // resolve first time through 2622 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache); 2623 __ movl(temp, code); 2624 __ call_VM(noreg, entry, temp); 2625 // Update registers with resolved info 2626 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2627 __ bind(resolved); 2628 } 2629 2630 // The cache and index registers must be set before call 2631 void TemplateTable::load_field_cp_cache_entry(Register obj, 2632 Register cache, 2633 Register index, 2634 Register off, 2635 Register flags, 2636 bool is_static = false) { 2637 assert_different_registers(cache, index, flags, off); 2638 2639 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2640 // Field offset 2641 __ movptr(off, Address(cache, index, Address::times_ptr, 2642 in_bytes(cp_base_offset + 2643 ConstantPoolCacheEntry::f2_offset()))); 2644 // Flags 2645 __ movl(flags, Address(cache, index, Address::times_ptr, 2646 in_bytes(cp_base_offset + 2647 ConstantPoolCacheEntry::flags_offset()))); 2648 2649 // klass overwrite register 2650 if (is_static) { 2651 __ movptr(obj, Address(cache, index, Address::times_ptr, 2652 in_bytes(cp_base_offset + 2653 ConstantPoolCacheEntry::f1_offset()))); 2654 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 2655 __ movptr(obj, Address(obj, mirror_offset)); 2656 } 2657 } 2658 2659 void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 2660 Register method, 2661 Register itable_index, 2662 Register flags, 2663 bool is_invokevirtual, 2664 bool is_invokevfinal, /*unused*/ 2665 bool is_invokedynamic) { 2666 // setup registers 2667 const Register cache = rcx; 2668 const Register index = rdx; 2669 assert_different_registers(method, flags); 2670 assert_different_registers(method, cache, index); 2671 assert_different_registers(itable_index, flags); 2672 assert_different_registers(itable_index, cache, index); 2673 // determine constant pool cache field offsets 2674 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant"); 2675 const int method_offset = in_bytes( 2676 ConstantPoolCache::base_offset() + 2677 ((byte_no == f2_byte) 2678 ? ConstantPoolCacheEntry::f2_offset() 2679 : ConstantPoolCacheEntry::f1_offset())); 2680 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() + 2681 ConstantPoolCacheEntry::flags_offset()); 2682 // access constant pool cache fields 2683 const int index_offset = in_bytes(ConstantPoolCache::base_offset() + 2684 ConstantPoolCacheEntry::f2_offset()); 2685 2686 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2)); 2687 resolve_cache_and_index(byte_no, cache, index, index_size); 2688 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset)); 2689 2690 if (itable_index != noreg) { 2691 // pick up itable or appendix index from f2 also: 2692 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset)); 2693 } 2694 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset)); 2695 } 2696 2697 // The registers cache and index expected to be set before call. 2698 // Correct values of the cache and index registers are preserved. 2699 void TemplateTable::jvmti_post_field_access(Register cache, 2700 Register index, 2701 bool is_static, 2702 bool has_tos) { 2703 if (JvmtiExport::can_post_field_access()) { 2704 // Check to see if a field access watch has been set before we take 2705 // the time to call into the VM. 2706 Label L1; 2707 assert_different_registers(cache, index, rax); 2708 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2709 __ testl(rax,rax); 2710 __ jcc(Assembler::zero, L1); 2711 2712 // cache entry pointer 2713 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 2714 __ shll(index, LogBytesPerWord); 2715 __ addptr(cache, index); 2716 if (is_static) { 2717 __ xorptr(rax, rax); // NULL object reference 2718 } else { 2719 __ pop(atos); // Get the object 2720 __ verify_oop(rax); 2721 __ push(atos); // Restore stack state 2722 } 2723 // rax,: object pointer or NULL 2724 // cache: cache entry pointer 2725 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), 2726 rax, cache); 2727 __ get_cache_and_index_at_bcp(cache, index, 1); 2728 __ bind(L1); 2729 } 2730 } 2731 2732 void TemplateTable::pop_and_check_object(Register r) { 2733 __ pop_ptr(r); 2734 __ null_check(r); // for field access must check obj. 2735 __ verify_oop(r); 2736 } 2737 2738 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) { 2739 transition(vtos, vtos); 2740 2741 const Register cache = rcx; 2742 const Register index = rdx; 2743 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2744 const Register off = rbx; 2745 const Register flags = rax; 2746 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them 2747 2748 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2749 jvmti_post_field_access(cache, index, is_static, false); 2750 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2751 2752 if (!is_static) pop_and_check_object(obj); 2753 2754 const Address field(obj, off, Address::times_1, 0*wordSize); 2755 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize)); 2756 2757 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2758 2759 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2760 // Make sure we don't need to mask edx after the above shift 2761 assert(btos == 0, "change code, btos != 0"); 2762 2763 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 2764 2765 __ jcc(Assembler::notZero, notByte); 2766 // btos 2767 __ load_signed_byte(rax, field); 2768 __ push(btos); 2769 // Rewrite bytecode to be faster 2770 if (!is_static && rc == may_rewrite) { 2771 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx); 2772 } 2773 __ jmp(Done); 2774 2775 __ bind(notByte); 2776 __ cmpl(flags, atos); 2777 __ jcc(Assembler::notEqual, notObj); 2778 // atos 2779 __ load_heap_oop(rax, field); 2780 __ push(atos); 2781 if (!is_static && rc == may_rewrite) { 2782 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx); 2783 } 2784 __ jmp(Done); 2785 2786 __ bind(notObj); 2787 __ cmpl(flags, itos); 2788 __ jcc(Assembler::notEqual, notInt); 2789 // itos 2790 __ movl(rax, field); 2791 __ push(itos); 2792 // Rewrite bytecode to be faster 2793 if (!is_static && rc == may_rewrite) { 2794 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx); 2795 } 2796 __ jmp(Done); 2797 2798 __ bind(notInt); 2799 __ cmpl(flags, ctos); 2800 __ jcc(Assembler::notEqual, notChar); 2801 // ctos 2802 __ load_unsigned_short(rax, field); 2803 __ push(ctos); 2804 // Rewrite bytecode to be faster 2805 if (!is_static && rc == may_rewrite) { 2806 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx); 2807 } 2808 __ jmp(Done); 2809 2810 __ bind(notChar); 2811 __ cmpl(flags, stos); 2812 __ jcc(Assembler::notEqual, notShort); 2813 // stos 2814 __ load_signed_short(rax, field); 2815 __ push(stos); 2816 // Rewrite bytecode to be faster 2817 if (!is_static && rc == may_rewrite) { 2818 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx); 2819 } 2820 __ jmp(Done); 2821 2822 __ bind(notShort); 2823 __ cmpl(flags, ltos); 2824 __ jcc(Assembler::notEqual, notLong); 2825 // ltos 2826 2827 #ifndef _LP64 2828 // Generate code as if volatile. There just aren't enough registers to 2829 // save that information and this code is faster than the test. 2830 __ fild_d(field); // Must load atomically 2831 __ subptr(rsp,2*wordSize); // Make space for store 2832 __ fistp_d(Address(rsp,0)); 2833 __ pop(rax); 2834 __ pop(rdx); 2835 #else 2836 __ movq(rax, field); 2837 #endif 2838 2839 __ push(ltos); 2840 // Rewrite bytecode to be faster 2841 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx)); 2842 __ jmp(Done); 2843 2844 __ bind(notLong); 2845 __ cmpl(flags, ftos); 2846 __ jcc(Assembler::notEqual, notFloat); 2847 // ftos 2848 2849 __ load_float(field); 2850 __ push(ftos); 2851 // Rewrite bytecode to be faster 2852 if (!is_static && rc == may_rewrite) { 2853 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx); 2854 } 2855 __ jmp(Done); 2856 2857 __ bind(notFloat); 2858 #ifdef ASSERT 2859 __ cmpl(flags, dtos); 2860 __ jcc(Assembler::notEqual, notDouble); 2861 #endif 2862 // dtos 2863 __ load_double(field); 2864 __ push(dtos); 2865 // Rewrite bytecode to be faster 2866 if (!is_static && rc == may_rewrite) { 2867 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx); 2868 } 2869 #ifdef ASSERT 2870 __ jmp(Done); 2871 2872 2873 __ bind(notDouble); 2874 __ stop("Bad state"); 2875 #endif 2876 2877 __ bind(Done); 2878 // [jk] not needed currently 2879 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad | 2880 // Assembler::LoadStore)); 2881 } 2882 2883 void TemplateTable::getfield(int byte_no) { 2884 getfield_or_static(byte_no, false); 2885 } 2886 2887 void TemplateTable::nofast_getfield(int byte_no) { 2888 getfield_or_static(byte_no, false, may_not_rewrite); 2889 } 2890 2891 void TemplateTable::getstatic(int byte_no) { 2892 getfield_or_static(byte_no, true); 2893 } 2894 2895 2896 // The registers cache and index expected to be set before call. 2897 // The function may destroy various registers, just not the cache and index registers. 2898 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 2899 2900 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax); 2901 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2902 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2903 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx); 2904 2905 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2906 2907 if (JvmtiExport::can_post_field_modification()) { 2908 // Check to see if a field modification watch has been set before 2909 // we take the time to call into the VM. 2910 Label L1; 2911 assert_different_registers(cache, index, rax); 2912 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2913 __ testl(rax, rax); 2914 __ jcc(Assembler::zero, L1); 2915 2916 __ get_cache_and_index_at_bcp(robj, RDX, 1); 2917 2918 2919 if (is_static) { 2920 // Life is simple. Null out the object pointer. 2921 __ xorl(RBX, RBX); 2922 2923 } else { 2924 // Life is harder. The stack holds the value on top, followed by 2925 // the object. We don't know the size of the value, though; it 2926 // could be one or two words depending on its type. As a result, 2927 // we must find the type to determine where the object is. 2928 #ifndef _LP64 2929 Label two_word, valsize_known; 2930 #endif 2931 __ movl(RCX, Address(robj, RDX, 2932 Address::times_ptr, 2933 in_bytes(cp_base_offset + 2934 ConstantPoolCacheEntry::flags_offset()))); 2935 NOT_LP64(__ mov(rbx, rsp)); 2936 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift); 2937 2938 // Make sure we don't need to mask rcx after the above shift 2939 ConstantPoolCacheEntry::verify_tos_state_shift(); 2940 #ifdef _LP64 2941 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue 2942 __ cmpl(c_rarg3, ltos); 2943 __ cmovptr(Assembler::equal, 2944 c_rarg1, at_tos_p2()); // ltos (two word jvalue) 2945 __ cmpl(c_rarg3, dtos); 2946 __ cmovptr(Assembler::equal, 2947 c_rarg1, at_tos_p2()); // dtos (two word jvalue) 2948 #else 2949 __ cmpl(rcx, ltos); 2950 __ jccb(Assembler::equal, two_word); 2951 __ cmpl(rcx, dtos); 2952 __ jccb(Assembler::equal, two_word); 2953 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos) 2954 __ jmpb(valsize_known); 2955 2956 __ bind(two_word); 2957 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue 2958 2959 __ bind(valsize_known); 2960 // setup object pointer 2961 __ movptr(rbx, Address(rbx, 0)); 2962 #endif 2963 } 2964 // cache entry pointer 2965 __ addptr(robj, in_bytes(cp_base_offset)); 2966 __ shll(RDX, LogBytesPerWord); 2967 __ addptr(robj, RDX); 2968 // object (tos) 2969 __ mov(RCX, rsp); 2970 // c_rarg1: object pointer set up above (NULL if static) 2971 // c_rarg2: cache entry pointer 2972 // c_rarg3: jvalue object on the stack 2973 __ call_VM(noreg, 2974 CAST_FROM_FN_PTR(address, 2975 InterpreterRuntime::post_field_modification), 2976 RBX, robj, RCX); 2977 __ get_cache_and_index_at_bcp(cache, index, 1); 2978 __ bind(L1); 2979 } 2980 } 2981 2982 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) { 2983 transition(vtos, vtos); 2984 2985 const Register cache = rcx; 2986 const Register index = rdx; 2987 const Register obj = rcx; 2988 const Register off = rbx; 2989 const Register flags = rax; 2990 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2991 2992 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2993 jvmti_post_field_mod(cache, index, is_static); 2994 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2995 2996 // [jk] not needed currently 2997 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore | 2998 // Assembler::StoreStore)); 2999 3000 Label notVolatile, Done; 3001 __ movl(rdx, flags); 3002 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3003 __ andl(rdx, 0x1); 3004 3005 // field addresses 3006 const Address field(obj, off, Address::times_1, 0*wordSize); 3007 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);) 3008 3009 Label notByte, notInt, notShort, notChar, 3010 notLong, notFloat, notObj, notDouble; 3011 3012 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 3013 3014 assert(btos == 0, "change code, btos != 0"); 3015 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 3016 __ jcc(Assembler::notZero, notByte); 3017 3018 // btos 3019 { 3020 __ pop(btos); 3021 if (!is_static) pop_and_check_object(obj); 3022 __ movb(field, rax); 3023 if (!is_static && rc == may_rewrite) { 3024 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no); 3025 } 3026 __ jmp(Done); 3027 } 3028 3029 __ bind(notByte); 3030 __ cmpl(flags, atos); 3031 __ jcc(Assembler::notEqual, notObj); 3032 3033 // atos 3034 { 3035 __ pop(atos); 3036 if (!is_static) pop_and_check_object(obj); 3037 // Store into the field 3038 do_oop_store(_masm, field, rax, _bs->kind(), false); 3039 if (!is_static && rc == may_rewrite) { 3040 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no); 3041 } 3042 __ jmp(Done); 3043 } 3044 3045 __ bind(notObj); 3046 __ cmpl(flags, itos); 3047 __ jcc(Assembler::notEqual, notInt); 3048 3049 // itos 3050 { 3051 __ pop(itos); 3052 if (!is_static) pop_and_check_object(obj); 3053 __ movl(field, rax); 3054 if (!is_static && rc == may_rewrite) { 3055 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no); 3056 } 3057 __ jmp(Done); 3058 } 3059 3060 __ bind(notInt); 3061 __ cmpl(flags, ctos); 3062 __ jcc(Assembler::notEqual, notChar); 3063 3064 // ctos 3065 { 3066 __ pop(ctos); 3067 if (!is_static) pop_and_check_object(obj); 3068 __ movw(field, rax); 3069 if (!is_static && rc == may_rewrite) { 3070 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no); 3071 } 3072 __ jmp(Done); 3073 } 3074 3075 __ bind(notChar); 3076 __ cmpl(flags, stos); 3077 __ jcc(Assembler::notEqual, notShort); 3078 3079 // stos 3080 { 3081 __ pop(stos); 3082 if (!is_static) pop_and_check_object(obj); 3083 __ movw(field, rax); 3084 if (!is_static && rc == may_rewrite) { 3085 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no); 3086 } 3087 __ jmp(Done); 3088 } 3089 3090 __ bind(notShort); 3091 __ cmpl(flags, ltos); 3092 __ jcc(Assembler::notEqual, notLong); 3093 3094 // ltos 3095 #ifdef _LP64 3096 { 3097 __ pop(ltos); 3098 if (!is_static) pop_and_check_object(obj); 3099 __ movq(field, rax); 3100 if (!is_static && rc == may_rewrite) { 3101 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no); 3102 } 3103 __ jmp(Done); 3104 } 3105 #else 3106 { 3107 Label notVolatileLong; 3108 __ testl(rdx, rdx); 3109 __ jcc(Assembler::zero, notVolatileLong); 3110 3111 __ pop(ltos); // overwrites rdx, do this after testing volatile. 3112 if (!is_static) pop_and_check_object(obj); 3113 3114 // Replace with real volatile test 3115 __ push(rdx); 3116 __ push(rax); // Must update atomically with FIST 3117 __ fild_d(Address(rsp,0)); // So load into FPU register 3118 __ fistp_d(field); // and put into memory atomically 3119 __ addptr(rsp, 2*wordSize); 3120 // volatile_barrier(); 3121 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3122 Assembler::StoreStore)); 3123 // Don't rewrite volatile version 3124 __ jmp(notVolatile); 3125 3126 __ bind(notVolatileLong); 3127 3128 __ pop(ltos); // overwrites rdx 3129 if (!is_static) pop_and_check_object(obj); 3130 __ movptr(hi, rdx); 3131 __ movptr(field, rax); 3132 // Don't rewrite to _fast_lputfield for potential volatile case. 3133 __ jmp(notVolatile); 3134 } 3135 #endif // _LP64 3136 3137 __ bind(notLong); 3138 __ cmpl(flags, ftos); 3139 __ jcc(Assembler::notEqual, notFloat); 3140 3141 // ftos 3142 { 3143 __ pop(ftos); 3144 if (!is_static) pop_and_check_object(obj); 3145 __ store_float(field); 3146 if (!is_static && rc == may_rewrite) { 3147 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no); 3148 } 3149 __ jmp(Done); 3150 } 3151 3152 __ bind(notFloat); 3153 #ifdef ASSERT 3154 __ cmpl(flags, dtos); 3155 __ jcc(Assembler::notEqual, notDouble); 3156 #endif 3157 3158 // dtos 3159 { 3160 __ pop(dtos); 3161 if (!is_static) pop_and_check_object(obj); 3162 __ store_double(field); 3163 if (!is_static && rc == may_rewrite) { 3164 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no); 3165 } 3166 } 3167 3168 #ifdef ASSERT 3169 __ jmp(Done); 3170 3171 __ bind(notDouble); 3172 __ stop("Bad state"); 3173 #endif 3174 3175 __ bind(Done); 3176 3177 // Check for volatile store 3178 __ testl(rdx, rdx); 3179 __ jcc(Assembler::zero, notVolatile); 3180 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3181 Assembler::StoreStore)); 3182 __ bind(notVolatile); 3183 } 3184 3185 void TemplateTable::putfield(int byte_no) { 3186 putfield_or_static(byte_no, false); 3187 } 3188 3189 void TemplateTable::nofast_putfield(int byte_no) { 3190 putfield_or_static(byte_no, false, may_not_rewrite); 3191 } 3192 3193 void TemplateTable::putstatic(int byte_no) { 3194 putfield_or_static(byte_no, true); 3195 } 3196 3197 void TemplateTable::jvmti_post_fast_field_mod() { 3198 3199 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 3200 3201 if (JvmtiExport::can_post_field_modification()) { 3202 // Check to see if a field modification watch has been set before 3203 // we take the time to call into the VM. 3204 Label L2; 3205 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 3206 __ testl(scratch, scratch); 3207 __ jcc(Assembler::zero, L2); 3208 __ pop_ptr(rbx); // copy the object pointer from tos 3209 __ verify_oop(rbx); 3210 __ push_ptr(rbx); // put the object pointer back on tos 3211 // Save tos values before call_VM() clobbers them. Since we have 3212 // to do it for every data type, we use the saved values as the 3213 // jvalue object. 3214 switch (bytecode()) { // load values into the jvalue object 3215 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break; 3216 case Bytecodes::_fast_bputfield: // fall through 3217 case Bytecodes::_fast_sputfield: // fall through 3218 case Bytecodes::_fast_cputfield: // fall through 3219 case Bytecodes::_fast_iputfield: __ push_i(rax); break; 3220 case Bytecodes::_fast_dputfield: __ push(dtos); break; 3221 case Bytecodes::_fast_fputfield: __ push(ftos); break; 3222 case Bytecodes::_fast_lputfield: __ push_l(rax); break; 3223 3224 default: 3225 ShouldNotReachHere(); 3226 } 3227 __ mov(scratch, rsp); // points to jvalue on the stack 3228 // access constant pool cache entry 3229 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1)); 3230 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1)); 3231 __ verify_oop(rbx); 3232 // rbx: object pointer copied above 3233 // c_rarg2: cache entry pointer 3234 // c_rarg3: jvalue object on the stack 3235 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3)); 3236 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx)); 3237 3238 switch (bytecode()) { // restore tos values 3239 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break; 3240 case Bytecodes::_fast_bputfield: // fall through 3241 case Bytecodes::_fast_sputfield: // fall through 3242 case Bytecodes::_fast_cputfield: // fall through 3243 case Bytecodes::_fast_iputfield: __ pop_i(rax); break; 3244 case Bytecodes::_fast_dputfield: __ pop(dtos); break; 3245 case Bytecodes::_fast_fputfield: __ pop(ftos); break; 3246 case Bytecodes::_fast_lputfield: __ pop_l(rax); break; 3247 } 3248 __ bind(L2); 3249 } 3250 } 3251 3252 void TemplateTable::fast_storefield(TosState state) { 3253 transition(state, vtos); 3254 3255 ByteSize base = ConstantPoolCache::base_offset(); 3256 3257 jvmti_post_fast_field_mod(); 3258 3259 // access constant pool cache 3260 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 3261 3262 // test for volatile with rdx but rdx is tos register for lputfield. 3263 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, 3264 in_bytes(base + 3265 ConstantPoolCacheEntry::flags_offset()))); 3266 3267 // replace index with field offset from cache entry 3268 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, 3269 in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 3270 3271 // [jk] not needed currently 3272 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore | 3273 // Assembler::StoreStore)); 3274 3275 Label notVolatile; 3276 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3277 __ andl(rdx, 0x1); 3278 3279 // Get object from stack 3280 pop_and_check_object(rcx); 3281 3282 // field address 3283 const Address field(rcx, rbx, Address::times_1); 3284 3285 // access field 3286 switch (bytecode()) { 3287 case Bytecodes::_fast_aputfield: 3288 do_oop_store(_masm, field, rax, _bs->kind(), false); 3289 break; 3290 case Bytecodes::_fast_lputfield: 3291 #ifdef _LP64 3292 __ movq(field, rax); 3293 #else 3294 __ stop("should not be rewritten"); 3295 #endif 3296 break; 3297 case Bytecodes::_fast_iputfield: 3298 __ movl(field, rax); 3299 break; 3300 case Bytecodes::_fast_bputfield: 3301 __ movb(field, rax); 3302 break; 3303 case Bytecodes::_fast_sputfield: 3304 // fall through 3305 case Bytecodes::_fast_cputfield: 3306 __ movw(field, rax); 3307 break; 3308 case Bytecodes::_fast_fputfield: 3309 __ store_float(field); 3310 break; 3311 case Bytecodes::_fast_dputfield: 3312 __ store_double(field); 3313 break; 3314 default: 3315 ShouldNotReachHere(); 3316 } 3317 3318 // Check for volatile store 3319 __ testl(rdx, rdx); 3320 __ jcc(Assembler::zero, notVolatile); 3321 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3322 Assembler::StoreStore)); 3323 __ bind(notVolatile); 3324 } 3325 3326 void TemplateTable::fast_accessfield(TosState state) { 3327 transition(atos, state); 3328 3329 // Do the JVMTI work here to avoid disturbing the register state below 3330 if (JvmtiExport::can_post_field_access()) { 3331 // Check to see if a field access watch has been set before we 3332 // take the time to call into the VM. 3333 Label L1; 3334 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 3335 __ testl(rcx, rcx); 3336 __ jcc(Assembler::zero, L1); 3337 // access constant pool cache entry 3338 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1)); 3339 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1)); 3340 __ verify_oop(rax); 3341 __ push_ptr(rax); // save object pointer before call_VM() clobbers it 3342 LP64_ONLY(__ mov(c_rarg1, rax)); 3343 // c_rarg1: object pointer copied above 3344 // c_rarg2: cache entry pointer 3345 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2)); 3346 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx)); 3347 __ pop_ptr(rax); // restore object pointer 3348 __ bind(L1); 3349 } 3350 3351 // access constant pool cache 3352 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 3353 // replace index with field offset from cache entry 3354 // [jk] not needed currently 3355 // if (os::is_MP()) { 3356 // __ movl(rdx, Address(rcx, rbx, Address::times_8, 3357 // in_bytes(ConstantPoolCache::base_offset() + 3358 // ConstantPoolCacheEntry::flags_offset()))); 3359 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3360 // __ andl(rdx, 0x1); 3361 // } 3362 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, 3363 in_bytes(ConstantPoolCache::base_offset() + 3364 ConstantPoolCacheEntry::f2_offset()))); 3365 3366 // rax: object 3367 __ verify_oop(rax); 3368 __ null_check(rax); 3369 Address field(rax, rbx, Address::times_1); 3370 3371 // access field 3372 switch (bytecode()) { 3373 case Bytecodes::_fast_agetfield: 3374 __ load_heap_oop(rax, field); 3375 __ verify_oop(rax); 3376 break; 3377 case Bytecodes::_fast_lgetfield: 3378 #ifdef _LP64 3379 __ movq(rax, field); 3380 #else 3381 __ stop("should not be rewritten"); 3382 #endif 3383 break; 3384 case Bytecodes::_fast_igetfield: 3385 __ movl(rax, field); 3386 break; 3387 case Bytecodes::_fast_bgetfield: 3388 __ movsbl(rax, field); 3389 break; 3390 case Bytecodes::_fast_sgetfield: 3391 __ load_signed_short(rax, field); 3392 break; 3393 case Bytecodes::_fast_cgetfield: 3394 __ load_unsigned_short(rax, field); 3395 break; 3396 case Bytecodes::_fast_fgetfield: 3397 __ load_float(field); 3398 break; 3399 case Bytecodes::_fast_dgetfield: 3400 __ load_double(field); 3401 break; 3402 default: 3403 ShouldNotReachHere(); 3404 } 3405 // [jk] not needed currently 3406 // if (os::is_MP()) { 3407 // Label notVolatile; 3408 // __ testl(rdx, rdx); 3409 // __ jcc(Assembler::zero, notVolatile); 3410 // __ membar(Assembler::LoadLoad); 3411 // __ bind(notVolatile); 3412 //}; 3413 } 3414 3415 void TemplateTable::fast_xaccess(TosState state) { 3416 transition(vtos, state); 3417 3418 // get receiver 3419 __ movptr(rax, aaddress(0)); 3420 // access constant pool cache 3421 __ get_cache_and_index_at_bcp(rcx, rdx, 2); 3422 __ movptr(rbx, 3423 Address(rcx, rdx, Address::times_ptr, 3424 in_bytes(ConstantPoolCache::base_offset() + 3425 ConstantPoolCacheEntry::f2_offset()))); 3426 // make sure exception is reported in correct bcp range (getfield is 3427 // next instruction) 3428 __ increment(rbcp); 3429 __ null_check(rax); 3430 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize); 3431 switch (state) { 3432 case itos: 3433 __ movl(rax, field); 3434 break; 3435 case atos: 3436 __ load_heap_oop(rax, field); 3437 __ verify_oop(rax); 3438 break; 3439 case ftos: 3440 __ load_float(field); 3441 break; 3442 default: 3443 ShouldNotReachHere(); 3444 } 3445 3446 // [jk] not needed currently 3447 // if (os::is_MP()) { 3448 // Label notVolatile; 3449 // __ movl(rdx, Address(rcx, rdx, Address::times_8, 3450 // in_bytes(ConstantPoolCache::base_offset() + 3451 // ConstantPoolCacheEntry::flags_offset()))); 3452 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3453 // __ testl(rdx, 0x1); 3454 // __ jcc(Assembler::zero, notVolatile); 3455 // __ membar(Assembler::LoadLoad); 3456 // __ bind(notVolatile); 3457 // } 3458 3459 __ decrement(rbcp); 3460 } 3461 3462 //----------------------------------------------------------------------------- 3463 // Calls 3464 3465 void TemplateTable::count_calls(Register method, Register temp) { 3466 // implemented elsewhere 3467 ShouldNotReachHere(); 3468 } 3469 3470 void TemplateTable::prepare_invoke(int byte_no, 3471 Register method, // linked method (or i-klass) 3472 Register index, // itable index, MethodType, etc. 3473 Register recv, // if caller wants to see it 3474 Register flags // if caller wants to test it 3475 ) { 3476 // determine flags 3477 const Bytecodes::Code code = bytecode(); 3478 const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 3479 const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 3480 const bool is_invokehandle = code == Bytecodes::_invokehandle; 3481 const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 3482 const bool is_invokespecial = code == Bytecodes::_invokespecial; 3483 const bool load_receiver = (recv != noreg); 3484 const bool save_flags = (flags != noreg); 3485 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), ""); 3486 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal"); 3487 assert(flags == noreg || flags == rdx, ""); 3488 assert(recv == noreg || recv == rcx, ""); 3489 3490 // setup registers & access constant pool cache 3491 if (recv == noreg) recv = rcx; 3492 if (flags == noreg) flags = rdx; 3493 assert_different_registers(method, index, recv, flags); 3494 3495 // save 'interpreter return address' 3496 __ save_bcp(); 3497 3498 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 3499 3500 // maybe push appendix to arguments (just before return address) 3501 if (is_invokedynamic || is_invokehandle) { 3502 Label L_no_push; 3503 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift)); 3504 __ jcc(Assembler::zero, L_no_push); 3505 // Push the appendix as a trailing parameter. 3506 // This must be done before we get the receiver, 3507 // since the parameter_size includes it. 3508 __ push(rbx); 3509 __ mov(rbx, index); 3510 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0"); 3511 __ load_resolved_reference_at_index(index, rbx); 3512 __ pop(rbx); 3513 __ push(index); // push appendix (MethodType, CallSite, etc.) 3514 __ bind(L_no_push); 3515 } 3516 3517 // load receiver if needed (after appendix is pushed so parameter size is correct) 3518 // Note: no return address pushed yet 3519 if (load_receiver) { 3520 __ movl(recv, flags); 3521 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask); 3522 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address 3523 const int receiver_is_at_end = -1; // back off one slot to get receiver 3524 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end); 3525 __ movptr(recv, recv_addr); 3526 __ verify_oop(recv); 3527 } 3528 3529 if (save_flags) { 3530 __ movl(rbcp, flags); 3531 } 3532 3533 // compute return type 3534 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 3535 // Make sure we don't need to mask flags after the above shift 3536 ConstantPoolCacheEntry::verify_tos_state_shift(); 3537 // load return address 3538 { 3539 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code); 3540 ExternalAddress table(table_addr); 3541 LP64_ONLY(__ lea(rscratch1, table)); 3542 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr))); 3543 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)))); 3544 } 3545 3546 // push return address 3547 __ push(flags); 3548 3549 // Restore flags value from the constant pool cache, and restore rsi 3550 // for later null checks. r13 is the bytecode pointer 3551 if (save_flags) { 3552 __ movl(flags, rbcp); 3553 __ restore_bcp(); 3554 } 3555 } 3556 3557 void TemplateTable::invokevirtual_helper(Register index, 3558 Register recv, 3559 Register flags) { 3560 // Uses temporary registers rax, rdx 3561 assert_different_registers(index, recv, rax, rdx); 3562 assert(index == rbx, ""); 3563 assert(recv == rcx, ""); 3564 3565 // Test for an invoke of a final method 3566 Label notFinal; 3567 __ movl(rax, flags); 3568 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift)); 3569 __ jcc(Assembler::zero, notFinal); 3570 3571 const Register method = index; // method must be rbx 3572 assert(method == rbx, 3573 "Method* must be rbx for interpreter calling convention"); 3574 3575 // do the call - the index is actually the method to call 3576 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method* 3577 3578 // It's final, need a null check here! 3579 __ null_check(recv); 3580 3581 // profile this call 3582 __ profile_final_call(rax); 3583 __ profile_arguments_type(rax, method, rbcp, true); 3584 3585 __ jump_from_interpreted(method, rax); 3586 3587 __ bind(notFinal); 3588 3589 // get receiver klass 3590 __ null_check(recv, oopDesc::klass_offset_in_bytes()); 3591 __ load_klass(rax, recv); 3592 3593 // profile this call 3594 __ profile_virtual_call(rax, rlocals, rdx); 3595 // get target Method* & entry point 3596 __ lookup_virtual_method(rax, index, method); 3597 __ profile_arguments_type(rdx, method, rbcp, true); 3598 __ jump_from_interpreted(method, rdx); 3599 } 3600 3601 void TemplateTable::invokevirtual(int byte_no) { 3602 transition(vtos, vtos); 3603 assert(byte_no == f2_byte, "use this argument"); 3604 prepare_invoke(byte_no, 3605 rbx, // method or vtable index 3606 noreg, // unused itable index 3607 rcx, rdx); // recv, flags 3608 3609 // rbx: index 3610 // rcx: receiver 3611 // rdx: flags 3612 3613 invokevirtual_helper(rbx, rcx, rdx); 3614 } 3615 3616 void TemplateTable::invokespecial(int byte_no) { 3617 transition(vtos, vtos); 3618 assert(byte_no == f1_byte, "use this argument"); 3619 prepare_invoke(byte_no, rbx, noreg, // get f1 Method* 3620 rcx); // get receiver also for null check 3621 __ verify_oop(rcx); 3622 __ null_check(rcx); 3623 // do the call 3624 __ profile_call(rax); 3625 __ profile_arguments_type(rax, rbx, rbcp, false); 3626 __ jump_from_interpreted(rbx, rax); 3627 } 3628 3629 void TemplateTable::invokestatic(int byte_no) { 3630 transition(vtos, vtos); 3631 assert(byte_no == f1_byte, "use this argument"); 3632 prepare_invoke(byte_no, rbx); // get f1 Method* 3633 // do the call 3634 __ profile_call(rax); 3635 __ profile_arguments_type(rax, rbx, rbcp, false); 3636 __ jump_from_interpreted(rbx, rax); 3637 } 3638 3639 3640 void TemplateTable::fast_invokevfinal(int byte_no) { 3641 transition(vtos, vtos); 3642 assert(byte_no == f2_byte, "use this argument"); 3643 __ stop("fast_invokevfinal not used on x86"); 3644 } 3645 3646 3647 void TemplateTable::invokeinterface(int byte_no) { 3648 transition(vtos, vtos); 3649 assert(byte_no == f1_byte, "use this argument"); 3650 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index 3651 rcx, rdx); // recv, flags 3652 3653 // rax: interface klass (from f1) 3654 // rbx: itable index (from f2) 3655 // rcx: receiver 3656 // rdx: flags 3657 3658 // Special case of invokeinterface called for virtual method of 3659 // java.lang.Object. See cpCacheOop.cpp for details. 3660 // This code isn't produced by javac, but could be produced by 3661 // another compliant java compiler. 3662 Label notMethod; 3663 __ movl(rlocals, rdx); 3664 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift)); 3665 3666 __ jcc(Assembler::zero, notMethod); 3667 3668 invokevirtual_helper(rbx, rcx, rdx); 3669 __ bind(notMethod); 3670 3671 // Get receiver klass into rdx - also a null check 3672 __ restore_locals(); // restore r14 3673 __ null_check(rcx, oopDesc::klass_offset_in_bytes()); 3674 __ load_klass(rdx, rcx); 3675 3676 // profile this call 3677 __ profile_virtual_call(rdx, rbcp, rlocals); 3678 3679 Label no_such_interface, no_such_method; 3680 3681 __ lookup_interface_method(// inputs: rec. class, interface, itable index 3682 rdx, rax, rbx, 3683 // outputs: method, scan temp. reg 3684 rbx, rbcp, 3685 no_such_interface); 3686 3687 // rbx: Method* to call 3688 // rcx: receiver 3689 // Check for abstract method error 3690 // Note: This should be done more efficiently via a throw_abstract_method_error 3691 // interpreter entry point and a conditional jump to it in case of a null 3692 // method. 3693 __ testptr(rbx, rbx); 3694 __ jcc(Assembler::zero, no_such_method); 3695 3696 __ profile_arguments_type(rdx, rbx, rbcp, true); 3697 3698 // do the call 3699 // rcx: receiver 3700 // rbx,: Method* 3701 __ jump_from_interpreted(rbx, rdx); 3702 __ should_not_reach_here(); 3703 3704 // exception handling code follows... 3705 // note: must restore interpreter registers to canonical 3706 // state for exception handling to work correctly! 3707 3708 __ bind(no_such_method); 3709 // throw exception 3710 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3711 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed) 3712 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3713 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 3714 // the call_VM checks for exception, so we should never return here. 3715 __ should_not_reach_here(); 3716 3717 __ bind(no_such_interface); 3718 // throw exception 3719 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3720 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed) 3721 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3722 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3723 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3724 // the call_VM checks for exception, so we should never return here. 3725 __ should_not_reach_here(); 3726 } 3727 3728 void TemplateTable::invokehandle(int byte_no) { 3729 transition(vtos, vtos); 3730 assert(byte_no == f1_byte, "use this argument"); 3731 const Register rbx_method = rbx; 3732 const Register rax_mtype = rax; 3733 const Register rcx_recv = rcx; 3734 const Register rdx_flags = rdx; 3735 3736 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv); 3737 __ verify_method_ptr(rbx_method); 3738 __ verify_oop(rcx_recv); 3739 __ null_check(rcx_recv); 3740 3741 // rax: MethodType object (from cpool->resolved_references[f1], if necessary) 3742 // rbx: MH.invokeExact_MT method (from f2) 3743 3744 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke 3745 3746 // FIXME: profile the LambdaForm also 3747 __ profile_final_call(rax); 3748 __ profile_arguments_type(rdx, rbx_method, rbcp, true); 3749 3750 __ jump_from_interpreted(rbx_method, rdx); 3751 } 3752 3753 void TemplateTable::invokedynamic(int byte_no) { 3754 transition(vtos, vtos); 3755 assert(byte_no == f1_byte, "use this argument"); 3756 3757 const Register rbx_method = rbx; 3758 const Register rax_callsite = rax; 3759 3760 prepare_invoke(byte_no, rbx_method, rax_callsite); 3761 3762 // rax: CallSite object (from cpool->resolved_references[f1]) 3763 // rbx: MH.linkToCallSite method (from f2) 3764 3765 // Note: rax_callsite is already pushed by prepare_invoke 3766 3767 // %%% should make a type profile for any invokedynamic that takes a ref argument 3768 // profile this call 3769 __ profile_call(rbcp); 3770 __ profile_arguments_type(rdx, rbx_method, rbcp, false); 3771 3772 __ verify_oop(rax_callsite); 3773 3774 __ jump_from_interpreted(rbx_method, rdx); 3775 } 3776 3777 //----------------------------------------------------------------------------- 3778 // Allocation 3779 3780 void TemplateTable::_new() { 3781 transition(vtos, atos); 3782 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3783 Label slow_case; 3784 Label slow_case_no_pop; 3785 Label done; 3786 Label initialize_header; 3787 Label initialize_object; // including clearing the fields 3788 Label allocate_shared; 3789 3790 __ get_cpool_and_tags(rcx, rax); 3791 3792 // Make sure the class we're about to instantiate has been resolved. 3793 // This is done before loading InstanceKlass to be consistent with the order 3794 // how Constant Pool is updated (see ConstantPool::klass_at_put) 3795 const int tags_offset = Array<u1>::base_offset_in_bytes(); 3796 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class); 3797 __ jcc(Assembler::notEqual, slow_case_no_pop); 3798 3799 // get InstanceKlass 3800 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool))); 3801 __ push(rcx); // save the contexts of klass for initializing the header 3802 3803 // make sure klass is initialized & doesn't have finalizer 3804 // make sure klass is fully initialized 3805 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized); 3806 __ jcc(Assembler::notEqual, slow_case); 3807 3808 // get instance_size in InstanceKlass (scaled to a count of bytes) 3809 __ movl(rdx, Address(rcx, Klass::layout_helper_offset())); 3810 // test to see if it has a finalizer or is malformed in some way 3811 __ testl(rdx, Klass::_lh_instance_slow_path_bit); 3812 __ jcc(Assembler::notZero, slow_case); 3813 3814 // 3815 // Allocate the instance 3816 // 1) Try to allocate in the TLAB 3817 // 2) if fail and the object is large allocate in the shared Eden 3818 // 3) if the above fails (or is not applicable), go to a slow case 3819 // (creates a new TLAB, etc.) 3820 3821 const bool allow_shared_alloc = 3822 Universe::heap()->supports_inline_contig_alloc(); 3823 3824 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 3825 #ifndef _LP64 3826 if (UseTLAB || allow_shared_alloc) { 3827 __ get_thread(thread); 3828 } 3829 #endif // _LP64 3830 3831 if (UseTLAB) { 3832 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset()))); 3833 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3834 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset()))); 3835 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case); 3836 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx); 3837 if (ZeroTLAB) { 3838 // the fields have been already cleared 3839 __ jmp(initialize_header); 3840 } else { 3841 // initialize both the header and fields 3842 __ jmp(initialize_object); 3843 } 3844 } 3845 3846 // Allocation in the shared Eden, if allowed. 3847 // 3848 // rdx: instance size in bytes 3849 if (allow_shared_alloc) { 3850 __ bind(allocate_shared); 3851 3852 ExternalAddress heap_top((address)Universe::heap()->top_addr()); 3853 ExternalAddress heap_end((address)Universe::heap()->end_addr()); 3854 3855 Label retry; 3856 __ bind(retry); 3857 __ movptr(rax, heap_top); 3858 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3859 __ cmpptr(rbx, heap_end); 3860 __ jcc(Assembler::above, slow_case); 3861 3862 // Compare rax, with the top addr, and if still equal, store the new 3863 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was 3864 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs. 3865 // 3866 // rax,: object begin 3867 // rbx,: object end 3868 // rdx: instance size in bytes 3869 __ locked_cmpxchgptr(rbx, heap_top); 3870 3871 // if someone beat us on the allocation, try again, otherwise continue 3872 __ jcc(Assembler::notEqual, retry); 3873 3874 __ incr_allocated_bytes(thread, rdx, 0); 3875 } 3876 3877 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 3878 // The object is initialized before the header. If the object size is 3879 // zero, go directly to the header initialization. 3880 __ bind(initialize_object); 3881 __ decrement(rdx, sizeof(oopDesc)); 3882 __ jcc(Assembler::zero, initialize_header); 3883 3884 // Initialize topmost object field, divide rdx by 8, check if odd and 3885 // test if zero. 3886 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) 3887 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd 3888 3889 // rdx must have been multiple of 8 3890 #ifdef ASSERT 3891 // make sure rdx was multiple of 8 3892 Label L; 3893 // Ignore partial flag stall after shrl() since it is debug VM 3894 __ jccb(Assembler::carryClear, L); 3895 __ stop("object size is not multiple of 2 - adjust this code"); 3896 __ bind(L); 3897 // rdx must be > 0, no extra check needed here 3898 #endif 3899 3900 // initialize remaining object fields: rdx was a multiple of 8 3901 { Label loop; 3902 __ bind(loop); 3903 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx); 3904 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx)); 3905 __ decrement(rdx); 3906 __ jcc(Assembler::notZero, loop); 3907 } 3908 3909 // initialize object header only. 3910 __ bind(initialize_header); 3911 if (UseBiasedLocking) { 3912 __ pop(rcx); // get saved klass back in the register. 3913 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset())); 3914 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx); 3915 } else { 3916 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), 3917 (intptr_t)markOopDesc::prototype()); // header 3918 __ pop(rcx); // get saved klass back in the register. 3919 } 3920 #ifdef _LP64 3921 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code) 3922 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops 3923 #endif 3924 __ store_klass(rax, rcx); // klass 3925 3926 { 3927 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0); 3928 // Trigger dtrace event for fastpath 3929 __ push(atos); 3930 __ call_VM_leaf( 3931 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); 3932 __ pop(atos); 3933 } 3934 3935 __ jmp(done); 3936 } 3937 3938 // slow case 3939 __ bind(slow_case); 3940 __ pop(rcx); // restore stack pointer to what it was when we came in. 3941 __ bind(slow_case_no_pop); 3942 3943 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax); 3944 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 3945 3946 __ get_constant_pool(rarg1); 3947 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1); 3948 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2); 3949 __ verify_oop(rax); 3950 3951 // continue 3952 __ bind(done); 3953 } 3954 3955 void TemplateTable::newarray() { 3956 transition(itos, atos); 3957 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 3958 __ load_unsigned_byte(rarg1, at_bcp(1)); 3959 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), 3960 rarg1, rax); 3961 } 3962 3963 void TemplateTable::anewarray() { 3964 transition(itos, atos); 3965 3966 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 3967 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 3968 3969 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1); 3970 __ get_constant_pool(rarg1); 3971 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), 3972 rarg1, rarg2, rax); 3973 } 3974 3975 void TemplateTable::arraylength() { 3976 transition(atos, itos); 3977 __ null_check(rax, arrayOopDesc::length_offset_in_bytes()); 3978 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes())); 3979 } 3980 3981 void TemplateTable::checkcast() { 3982 transition(atos, atos); 3983 Label done, is_null, ok_is_subtype, quicked, resolved; 3984 __ testptr(rax, rax); // object is in rax 3985 __ jcc(Assembler::zero, is_null); 3986 3987 // Get cpool & tags index 3988 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array 3989 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index 3990 // See if bytecode has already been quicked 3991 __ cmpb(Address(rdx, rbx, 3992 Address::times_1, 3993 Array<u1>::base_offset_in_bytes()), 3994 JVM_CONSTANT_Class); 3995 __ jcc(Assembler::equal, quicked); 3996 __ push(atos); // save receiver for result, and for GC 3997 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 3998 3999 // vm_result_2 has metadata result 4000 #ifndef _LP64 4001 // borrow rdi from locals 4002 __ get_thread(rdi); 4003 __ get_vm_result_2(rax, rdi); 4004 __ restore_locals(); 4005 #else 4006 __ get_vm_result_2(rax, r15_thread); 4007 #endif 4008 4009 __ pop_ptr(rdx); // restore receiver 4010 __ jmpb(resolved); 4011 4012 // Get superklass in rax and subklass in rbx 4013 __ bind(quicked); 4014 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check 4015 __ movptr(rax, Address(rcx, rbx, 4016 Address::times_ptr, sizeof(ConstantPool))); 4017 4018 __ bind(resolved); 4019 __ load_klass(rbx, rdx); 4020 4021 // Generate subtype check. Blows rcx, rdi. Object in rdx. 4022 // Superklass in rax. Subklass in rbx. 4023 __ gen_subtype_check(rbx, ok_is_subtype); 4024 4025 // Come here on failure 4026 __ push_ptr(rdx); 4027 // object is at TOS 4028 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry)); 4029 4030 // Come here on success 4031 __ bind(ok_is_subtype); 4032 __ mov(rax, rdx); // Restore object in rdx 4033 4034 // Collect counts on whether this check-cast sees NULLs a lot or not. 4035 if (ProfileInterpreter) { 4036 __ jmp(done); 4037 __ bind(is_null); 4038 __ profile_null_seen(rcx); 4039 } else { 4040 __ bind(is_null); // same as 'done' 4041 } 4042 __ bind(done); 4043 } 4044 4045 void TemplateTable::instanceof() { 4046 transition(atos, itos); 4047 Label done, is_null, ok_is_subtype, quicked, resolved; 4048 __ testptr(rax, rax); 4049 __ jcc(Assembler::zero, is_null); 4050 4051 // Get cpool & tags index 4052 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array 4053 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index 4054 // See if bytecode has already been quicked 4055 __ cmpb(Address(rdx, rbx, 4056 Address::times_1, 4057 Array<u1>::base_offset_in_bytes()), 4058 JVM_CONSTANT_Class); 4059 __ jcc(Assembler::equal, quicked); 4060 4061 __ push(atos); // save receiver for result, and for GC 4062 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 4063 // vm_result_2 has metadata result 4064 4065 #ifndef _LP64 4066 // borrow rdi from locals 4067 __ get_thread(rdi); 4068 __ get_vm_result_2(rax, rdi); 4069 __ restore_locals(); 4070 #else 4071 __ get_vm_result_2(rax, r15_thread); 4072 #endif 4073 4074 __ pop_ptr(rdx); // restore receiver 4075 __ verify_oop(rdx); 4076 __ load_klass(rdx, rdx); 4077 __ jmpb(resolved); 4078 4079 // Get superklass in rax and subklass in rdx 4080 __ bind(quicked); 4081 __ load_klass(rdx, rax); 4082 __ movptr(rax, Address(rcx, rbx, 4083 Address::times_ptr, sizeof(ConstantPool))); 4084 4085 __ bind(resolved); 4086 4087 // Generate subtype check. Blows rcx, rdi 4088 // Superklass in rax. Subklass in rdx. 4089 __ gen_subtype_check(rdx, ok_is_subtype); 4090 4091 // Come here on failure 4092 __ xorl(rax, rax); 4093 __ jmpb(done); 4094 // Come here on success 4095 __ bind(ok_is_subtype); 4096 __ movl(rax, 1); 4097 4098 // Collect counts on whether this test sees NULLs a lot or not. 4099 if (ProfileInterpreter) { 4100 __ jmp(done); 4101 __ bind(is_null); 4102 __ profile_null_seen(rcx); 4103 } else { 4104 __ bind(is_null); // same as 'done' 4105 } 4106 __ bind(done); 4107 // rax = 0: obj == NULL or obj is not an instanceof the specified klass 4108 // rax = 1: obj != NULL and obj is an instanceof the specified klass 4109 } 4110 4111 4112 //---------------------------------------------------------------------------------------------------- 4113 // Breakpoints 4114 void TemplateTable::_breakpoint() { 4115 // Note: We get here even if we are single stepping.. 4116 // jbug insists on setting breakpoints at every bytecode 4117 // even if we are in single step mode. 4118 4119 transition(vtos, vtos); 4120 4121 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 4122 4123 // get the unpatched byte code 4124 __ get_method(rarg); 4125 __ call_VM(noreg, 4126 CAST_FROM_FN_PTR(address, 4127 InterpreterRuntime::get_original_bytecode_at), 4128 rarg, rbcp); 4129 __ mov(rbx, rax); // why? 4130 4131 // post the breakpoint event 4132 __ get_method(rarg); 4133 __ call_VM(noreg, 4134 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), 4135 rarg, rbcp); 4136 4137 // complete the execution of original bytecode 4138 __ dispatch_only_normal(vtos); 4139 } 4140 4141 //----------------------------------------------------------------------------- 4142 // Exceptions 4143 4144 void TemplateTable::athrow() { 4145 transition(atos, vtos); 4146 __ null_check(rax); 4147 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 4148 } 4149 4150 //----------------------------------------------------------------------------- 4151 // Synchronization 4152 // 4153 // Note: monitorenter & exit are symmetric routines; which is reflected 4154 // in the assembly code structure as well 4155 // 4156 // Stack layout: 4157 // 4158 // [expressions ] <--- rsp = expression stack top 4159 // .. 4160 // [expressions ] 4161 // [monitor entry] <--- monitor block top = expression stack bot 4162 // .. 4163 // [monitor entry] 4164 // [frame data ] <--- monitor block bot 4165 // ... 4166 // [saved rbp ] <--- rbp 4167 void TemplateTable::monitorenter() { 4168 transition(atos, vtos); 4169 4170 // check for NULL object 4171 __ null_check(rax); 4172 4173 const Address monitor_block_top( 4174 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 4175 const Address monitor_block_bot( 4176 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 4177 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 4178 4179 Label allocated; 4180 4181 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 4182 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx); 4183 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 4184 4185 // initialize entry pointer 4186 __ xorl(rmon, rmon); // points to free slot or NULL 4187 4188 // find a free slot in the monitor block (result in rmon) 4189 { 4190 Label entry, loop, exit; 4191 __ movptr(rtop, monitor_block_top); // points to current entry, 4192 // starting with top-most entry 4193 __ lea(rbot, monitor_block_bot); // points to word before bottom 4194 // of monitor block 4195 __ jmpb(entry); 4196 4197 __ bind(loop); 4198 // check if current entry is used 4199 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD); 4200 // if not used then remember entry in rmon 4201 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr 4202 // check if current entry is for same object 4203 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes())); 4204 // if same object then stop searching 4205 __ jccb(Assembler::equal, exit); 4206 // otherwise advance to next entry 4207 __ addptr(rtop, entry_size); 4208 __ bind(entry); 4209 // check if bottom reached 4210 __ cmpptr(rtop, rbot); 4211 // if not at bottom then check this entry 4212 __ jcc(Assembler::notEqual, loop); 4213 __ bind(exit); 4214 } 4215 4216 __ testptr(rmon, rmon); // check if a slot has been found 4217 __ jcc(Assembler::notZero, allocated); // if found, continue with that one 4218 4219 // allocate one if there's no free slot 4220 { 4221 Label entry, loop; 4222 // 1. compute new pointers // rsp: old expression stack top 4223 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom 4224 __ subptr(rsp, entry_size); // move expression stack top 4225 __ subptr(rmon, entry_size); // move expression stack bottom 4226 __ mov(rtop, rsp); // set start value for copy loop 4227 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom 4228 __ jmp(entry); 4229 // 2. move expression stack contents 4230 __ bind(loop); 4231 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack 4232 // word from old location 4233 __ movptr(Address(rtop, 0), rbot); // and store it at new location 4234 __ addptr(rtop, wordSize); // advance to next word 4235 __ bind(entry); 4236 __ cmpptr(rtop, rmon); // check if bottom reached 4237 __ jcc(Assembler::notEqual, loop); // if not at bottom then 4238 // copy next word 4239 } 4240 4241 // call run-time routine 4242 // rmon: points to monitor entry 4243 __ bind(allocated); 4244 4245 // Increment bcp to point to the next bytecode, so exception 4246 // handling for async. exceptions work correctly. 4247 // The object has already been poped from the stack, so the 4248 // expression stack looks correct. 4249 __ increment(rbcp); 4250 4251 // store object 4252 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax); 4253 __ lock_object(rmon); 4254 4255 // check to make sure this monitor doesn't cause stack overflow after locking 4256 __ save_bcp(); // in case of exception 4257 __ generate_stack_overflow_check(0); 4258 4259 // The bcp has already been incremented. Just need to dispatch to 4260 // next instruction. 4261 __ dispatch_next(vtos); 4262 } 4263 4264 void TemplateTable::monitorexit() { 4265 transition(atos, vtos); 4266 4267 // check for NULL object 4268 __ null_check(rax); 4269 4270 const Address monitor_block_top( 4271 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 4272 const Address monitor_block_bot( 4273 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 4274 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 4275 4276 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 4277 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx); 4278 4279 Label found; 4280 4281 // find matching slot 4282 { 4283 Label entry, loop; 4284 __ movptr(rtop, monitor_block_top); // points to current entry, 4285 // starting with top-most entry 4286 __ lea(rbot, monitor_block_bot); // points to word before bottom 4287 // of monitor block 4288 __ jmpb(entry); 4289 4290 __ bind(loop); 4291 // check if current entry is for same object 4292 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes())); 4293 // if same object then stop searching 4294 __ jcc(Assembler::equal, found); 4295 // otherwise advance to next entry 4296 __ addptr(rtop, entry_size); 4297 __ bind(entry); 4298 // check if bottom reached 4299 __ cmpptr(rtop, rbot); 4300 // if not at bottom then check this entry 4301 __ jcc(Assembler::notEqual, loop); 4302 } 4303 4304 // error handling. Unlocking was not block-structured 4305 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 4306 InterpreterRuntime::throw_illegal_monitor_state_exception)); 4307 __ should_not_reach_here(); 4308 4309 // call run-time routine 4310 __ bind(found); 4311 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps) 4312 __ unlock_object(rtop); 4313 __ pop_ptr(rax); // discard object 4314 } 4315 4316 // Wide instructions 4317 void TemplateTable::wide() { 4318 transition(vtos, vtos); 4319 __ load_unsigned_byte(rbx, at_bcp(1)); 4320 ExternalAddress wtable((address)Interpreter::_wentry_point); 4321 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr))); 4322 // Note: the rbcp increment step is part of the individual wide bytecode implementations 4323 } 4324 4325 // Multi arrays 4326 void TemplateTable::multianewarray() { 4327 transition(vtos, atos); 4328 4329 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax); 4330 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions 4331 // last dim is on top of stack; we want address of first one: 4332 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize 4333 // the latter wordSize to point to the beginning of the array. 4334 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize)); 4335 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg); 4336 __ load_unsigned_byte(rbx, at_bcp(3)); 4337 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts 4338 } 4339 #endif /* !CC_INTERP */ 4340