1 /* 2 * Copyright (c) 1997, 2017, 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 "gc/shared/barrierSet.hpp" 28 #include "gc/shared/barrierSetCodeGen.hpp" 29 #include "interpreter/interpreter.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/interp_masm.hpp" 32 #include "interpreter/templateTable.hpp" 33 #include "memory/universe.inline.hpp" 34 #include "oops/methodData.hpp" 35 #include "oops/objArrayKlass.hpp" 36 #include "oops/oop.inline.hpp" 37 #include "prims/methodHandles.hpp" 38 #include "runtime/sharedRuntime.hpp" 39 #include "runtime/stubRoutines.hpp" 40 #include "runtime/synchronizer.hpp" 41 #include "utilities/macros.hpp" 42 43 #define __ _masm-> 44 45 // Global Register Names 46 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi); 47 static const 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 dst, 153 Register val, 154 DecoratorSet decorators) { 155 assert(val == noreg || val == rax, "parameter is just for looks"); 156 BarrierSetCodeGen *code_gen = Universe::heap()->barrier_set()->code_gen(); 157 code_gen->store_at(_masm, decorators, T_OBJECT, dst, val, /*tmp1*/ rdx, /*tmp2*/ rbx); 158 } 159 160 static void do_oop_load(InterpreterMacroAssembler* _masm, 161 Address src, 162 Register dst, 163 DecoratorSet decorators) { 164 BarrierSetCodeGen *code_gen = Universe::heap()->barrier_set()->code_gen(); 165 code_gen->load_at(_masm, decorators, T_OBJECT, dst, src, /*tmp1*/ rdx, /*tmp2*/ rbx); 166 } 167 168 Address TemplateTable::at_bcp(int offset) { 169 assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 170 return Address(rbcp, offset); 171 } 172 173 174 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg, 175 Register temp_reg, bool load_bc_into_bc_reg/*=true*/, 176 int byte_no) { 177 if (!RewriteBytecodes) return; 178 Label L_patch_done; 179 180 switch (bc) { 181 case Bytecodes::_fast_aputfield: 182 case Bytecodes::_fast_bputfield: 183 case Bytecodes::_fast_zputfield: 184 case Bytecodes::_fast_cputfield: 185 case Bytecodes::_fast_dputfield: 186 case Bytecodes::_fast_fputfield: 187 case Bytecodes::_fast_iputfield: 188 case Bytecodes::_fast_lputfield: 189 case Bytecodes::_fast_sputfield: 190 { 191 // We skip bytecode quickening for putfield instructions when 192 // the put_code written to the constant pool cache is zero. 193 // This is required so that every execution of this instruction 194 // calls out to InterpreterRuntime::resolve_get_put to do 195 // additional, required work. 196 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 197 assert(load_bc_into_bc_reg, "we use bc_reg as temp"); 198 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1); 199 __ movl(bc_reg, bc); 200 __ cmpl(temp_reg, (int) 0); 201 __ jcc(Assembler::zero, L_patch_done); // don't patch 202 } 203 break; 204 default: 205 assert(byte_no == -1, "sanity"); 206 // the pair bytecodes have already done the load. 207 if (load_bc_into_bc_reg) { 208 __ movl(bc_reg, bc); 209 } 210 } 211 212 if (JvmtiExport::can_post_breakpoint()) { 213 Label L_fast_patch; 214 // if a breakpoint is present we can't rewrite the stream directly 215 __ movzbl(temp_reg, at_bcp(0)); 216 __ cmpl(temp_reg, Bytecodes::_breakpoint); 217 __ jcc(Assembler::notEqual, L_fast_patch); 218 __ get_method(temp_reg); 219 // Let breakpoint table handling rewrite to quicker bytecode 220 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg); 221 #ifndef ASSERT 222 __ jmpb(L_patch_done); 223 #else 224 __ jmp(L_patch_done); 225 #endif 226 __ bind(L_fast_patch); 227 } 228 229 #ifdef ASSERT 230 Label L_okay; 231 __ load_unsigned_byte(temp_reg, at_bcp(0)); 232 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc)); 233 __ jcc(Assembler::equal, L_okay); 234 __ cmpl(temp_reg, bc_reg); 235 __ jcc(Assembler::equal, L_okay); 236 __ stop("patching the wrong bytecode"); 237 __ bind(L_okay); 238 #endif 239 240 // patch bytecode 241 __ movb(at_bcp(0), bc_reg); 242 __ bind(L_patch_done); 243 } 244 // Individual instructions 245 246 247 void TemplateTable::nop() { 248 transition(vtos, vtos); 249 // nothing to do 250 } 251 252 void TemplateTable::shouldnotreachhere() { 253 transition(vtos, vtos); 254 __ stop("shouldnotreachhere bytecode"); 255 } 256 257 void TemplateTable::aconst_null() { 258 transition(vtos, atos); 259 __ xorl(rax, rax); 260 } 261 262 void TemplateTable::iconst(int value) { 263 transition(vtos, itos); 264 if (value == 0) { 265 __ xorl(rax, rax); 266 } else { 267 __ movl(rax, value); 268 } 269 } 270 271 void TemplateTable::lconst(int value) { 272 transition(vtos, ltos); 273 if (value == 0) { 274 __ xorl(rax, rax); 275 } else { 276 __ movl(rax, value); 277 } 278 #ifndef _LP64 279 assert(value >= 0, "check this code"); 280 __ xorptr(rdx, rdx); 281 #endif 282 } 283 284 285 286 void TemplateTable::fconst(int value) { 287 transition(vtos, ftos); 288 if (UseSSE >= 1) { 289 static float one = 1.0f, two = 2.0f; 290 switch (value) { 291 case 0: 292 __ xorps(xmm0, xmm0); 293 break; 294 case 1: 295 __ movflt(xmm0, ExternalAddress((address) &one)); 296 break; 297 case 2: 298 __ movflt(xmm0, ExternalAddress((address) &two)); 299 break; 300 default: 301 ShouldNotReachHere(); 302 break; 303 } 304 } else { 305 #ifdef _LP64 306 ShouldNotReachHere(); 307 #else 308 if (value == 0) { __ fldz(); 309 } else if (value == 1) { __ fld1(); 310 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here 311 } else { ShouldNotReachHere(); 312 } 313 #endif // _LP64 314 } 315 } 316 317 void TemplateTable::dconst(int value) { 318 transition(vtos, dtos); 319 if (UseSSE >= 2) { 320 static double one = 1.0; 321 switch (value) { 322 case 0: 323 __ xorpd(xmm0, xmm0); 324 break; 325 case 1: 326 __ movdbl(xmm0, ExternalAddress((address) &one)); 327 break; 328 default: 329 ShouldNotReachHere(); 330 break; 331 } 332 } else { 333 #ifdef _LP64 334 ShouldNotReachHere(); 335 #else 336 if (value == 0) { __ fldz(); 337 } else if (value == 1) { __ fld1(); 338 } else { ShouldNotReachHere(); 339 } 340 #endif 341 } 342 } 343 344 void TemplateTable::bipush() { 345 transition(vtos, itos); 346 __ load_signed_byte(rax, at_bcp(1)); 347 } 348 349 void TemplateTable::sipush() { 350 transition(vtos, itos); 351 __ load_unsigned_short(rax, at_bcp(1)); 352 __ bswapl(rax); 353 __ sarl(rax, 16); 354 } 355 356 void TemplateTable::ldc(bool wide) { 357 transition(vtos, vtos); 358 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1); 359 Label call_ldc, notFloat, notClass, Done; 360 361 if (wide) { 362 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 363 } else { 364 __ load_unsigned_byte(rbx, at_bcp(1)); 365 } 366 367 __ get_cpool_and_tags(rcx, rax); 368 const int base_offset = ConstantPool::header_size() * wordSize; 369 const int tags_offset = Array<u1>::base_offset_in_bytes(); 370 371 // get type 372 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset)); 373 374 // unresolved class - get the resolved class 375 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass); 376 __ jccb(Assembler::equal, call_ldc); 377 378 // unresolved class in error state - call into runtime to throw the error 379 // from the first resolution attempt 380 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError); 381 __ jccb(Assembler::equal, call_ldc); 382 383 // resolved class - need to call vm to get java mirror of the class 384 __ cmpl(rdx, JVM_CONSTANT_Class); 385 __ jcc(Assembler::notEqual, notClass); 386 387 __ bind(call_ldc); 388 389 __ movl(rarg, wide); 390 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg); 391 392 __ push(atos); 393 __ jmp(Done); 394 395 __ bind(notClass); 396 __ cmpl(rdx, JVM_CONSTANT_Float); 397 __ jccb(Assembler::notEqual, notFloat); 398 399 // ftos 400 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset)); 401 __ push(ftos); 402 __ jmp(Done); 403 404 __ bind(notFloat); 405 #ifdef ASSERT 406 { 407 Label L; 408 __ cmpl(rdx, JVM_CONSTANT_Integer); 409 __ jcc(Assembler::equal, L); 410 // String and Object are rewritten to fast_aldc 411 __ stop("unexpected tag type in ldc"); 412 __ bind(L); 413 } 414 #endif 415 // itos JVM_CONSTANT_Integer only 416 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 417 __ push(itos); 418 __ bind(Done); 419 } 420 421 // Fast path for caching oop constants. 422 void TemplateTable::fast_aldc(bool wide) { 423 transition(vtos, atos); 424 425 Register result = rax; 426 Register tmp = rdx; 427 int index_size = wide ? sizeof(u2) : sizeof(u1); 428 429 Label resolved; 430 431 // We are resolved if the resolved reference cache entry contains a 432 // non-null object (String, MethodType, etc.) 433 assert_different_registers(result, tmp); 434 __ get_cache_index_at_bcp(tmp, 1, index_size); 435 __ load_resolved_reference_at_index(result, tmp); 436 __ testl(result, result); 437 __ jcc(Assembler::notZero, resolved); 438 439 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); 440 441 // first time invocation - must resolve first 442 __ movl(tmp, (int)bytecode()); 443 __ call_VM(result, entry, tmp); 444 445 __ bind(resolved); 446 447 if (VerifyOops) { 448 __ verify_oop(result); 449 } 450 } 451 452 void TemplateTable::ldc2_w() { 453 transition(vtos, vtos); 454 Label Long, Done; 455 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 456 457 __ get_cpool_and_tags(rcx, rax); 458 const int base_offset = ConstantPool::header_size() * wordSize; 459 const int tags_offset = Array<u1>::base_offset_in_bytes(); 460 461 // get type 462 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), 463 JVM_CONSTANT_Double); 464 __ jccb(Assembler::notEqual, Long); 465 466 // dtos 467 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset)); 468 __ push(dtos); 469 470 __ jmpb(Done); 471 __ bind(Long); 472 473 // ltos 474 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize)); 475 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize))); 476 __ push(ltos); 477 478 __ bind(Done); 479 } 480 481 void TemplateTable::locals_index(Register reg, int offset) { 482 __ load_unsigned_byte(reg, at_bcp(offset)); 483 __ negptr(reg); 484 } 485 486 void TemplateTable::iload() { 487 iload_internal(); 488 } 489 490 void TemplateTable::nofast_iload() { 491 iload_internal(may_not_rewrite); 492 } 493 494 void TemplateTable::iload_internal(RewriteControl rc) { 495 transition(vtos, itos); 496 if (RewriteFrequentPairs && rc == may_rewrite) { 497 Label rewrite, done; 498 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 499 LP64_ONLY(assert(rbx != bc, "register damaged")); 500 501 // get next byte 502 __ load_unsigned_byte(rbx, 503 at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 504 // if _iload, wait to rewrite to iload2. We only want to rewrite the 505 // last two iloads in a pair. Comparing against fast_iload means that 506 // the next bytecode is neither an iload or a caload, and therefore 507 // an iload pair. 508 __ cmpl(rbx, Bytecodes::_iload); 509 __ jcc(Assembler::equal, done); 510 511 __ cmpl(rbx, Bytecodes::_fast_iload); 512 __ movl(bc, Bytecodes::_fast_iload2); 513 514 __ jccb(Assembler::equal, rewrite); 515 516 // if _caload, rewrite to fast_icaload 517 __ cmpl(rbx, Bytecodes::_caload); 518 __ movl(bc, Bytecodes::_fast_icaload); 519 __ jccb(Assembler::equal, rewrite); 520 521 // rewrite so iload doesn't check again. 522 __ movl(bc, Bytecodes::_fast_iload); 523 524 // rewrite 525 // bc: fast bytecode 526 __ bind(rewrite); 527 patch_bytecode(Bytecodes::_iload, bc, rbx, false); 528 __ bind(done); 529 } 530 531 // Get the local value into tos 532 locals_index(rbx); 533 __ movl(rax, iaddress(rbx)); 534 } 535 536 void TemplateTable::fast_iload2() { 537 transition(vtos, itos); 538 locals_index(rbx); 539 __ movl(rax, iaddress(rbx)); 540 __ push(itos); 541 locals_index(rbx, 3); 542 __ movl(rax, iaddress(rbx)); 543 } 544 545 void TemplateTable::fast_iload() { 546 transition(vtos, itos); 547 locals_index(rbx); 548 __ movl(rax, iaddress(rbx)); 549 } 550 551 void TemplateTable::lload() { 552 transition(vtos, ltos); 553 locals_index(rbx); 554 __ movptr(rax, laddress(rbx)); 555 NOT_LP64(__ movl(rdx, haddress(rbx))); 556 } 557 558 void TemplateTable::fload() { 559 transition(vtos, ftos); 560 locals_index(rbx); 561 __ load_float(faddress(rbx)); 562 } 563 564 void TemplateTable::dload() { 565 transition(vtos, dtos); 566 locals_index(rbx); 567 __ load_double(daddress(rbx)); 568 } 569 570 void TemplateTable::aload() { 571 transition(vtos, atos); 572 locals_index(rbx); 573 __ movptr(rax, aaddress(rbx)); 574 } 575 576 void TemplateTable::locals_index_wide(Register reg) { 577 __ load_unsigned_short(reg, at_bcp(2)); 578 __ bswapl(reg); 579 __ shrl(reg, 16); 580 __ negptr(reg); 581 } 582 583 void TemplateTable::wide_iload() { 584 transition(vtos, itos); 585 locals_index_wide(rbx); 586 __ movl(rax, iaddress(rbx)); 587 } 588 589 void TemplateTable::wide_lload() { 590 transition(vtos, ltos); 591 locals_index_wide(rbx); 592 __ movptr(rax, laddress(rbx)); 593 NOT_LP64(__ movl(rdx, haddress(rbx))); 594 } 595 596 void TemplateTable::wide_fload() { 597 transition(vtos, ftos); 598 locals_index_wide(rbx); 599 __ load_float(faddress(rbx)); 600 } 601 602 void TemplateTable::wide_dload() { 603 transition(vtos, dtos); 604 locals_index_wide(rbx); 605 __ load_double(daddress(rbx)); 606 } 607 608 void TemplateTable::wide_aload() { 609 transition(vtos, atos); 610 locals_index_wide(rbx); 611 __ movptr(rax, aaddress(rbx)); 612 } 613 614 void TemplateTable::index_check(Register array, Register index) { 615 // Pop ptr into array 616 __ pop_ptr(array); 617 index_check_without_pop(array, index); 618 } 619 620 void TemplateTable::index_check_without_pop(Register array, Register index) { 621 // destroys rbx 622 // check array 623 __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 624 // sign extend index for use by indexed load 625 __ movl2ptr(index, index); 626 // check index 627 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes())); 628 if (index != rbx) { 629 // ??? convention: move aberrant index into rbx for exception message 630 assert(rbx != array, "different registers"); 631 __ movl(rbx, index); 632 } 633 __ jump_cc(Assembler::aboveEqual, 634 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry)); 635 } 636 637 638 void TemplateTable::iaload() { 639 transition(itos, itos); 640 // rax: index 641 // rdx: array 642 index_check(rdx, rax); // kills rbx 643 __ movl(rax, Address(rdx, rax, 644 Address::times_4, 645 arrayOopDesc::base_offset_in_bytes(T_INT))); 646 } 647 648 void TemplateTable::laload() { 649 transition(itos, ltos); 650 // rax: index 651 // rdx: array 652 index_check(rdx, rax); // kills rbx 653 NOT_LP64(__ mov(rbx, rax)); 654 // rbx,: index 655 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize)); 656 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize))); 657 } 658 659 660 661 void TemplateTable::faload() { 662 transition(itos, ftos); 663 // rax: index 664 // rdx: array 665 index_check(rdx, rax); // kills rbx 666 __ load_float(Address(rdx, rax, 667 Address::times_4, 668 arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 669 } 670 671 void TemplateTable::daload() { 672 transition(itos, dtos); 673 // rax: index 674 // rdx: array 675 index_check(rdx, rax); // kills rbx 676 __ load_double(Address(rdx, rax, 677 Address::times_8, 678 arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 679 } 680 681 void TemplateTable::aaload() { 682 transition(itos, atos); 683 // rax: index 684 // rdx: array 685 index_check(rdx, rax); // kills rbx 686 do_oop_load(_masm, 687 Address(rdx, rax, 688 UseCompressedOops ? Address::times_4 : Address::times_ptr, 689 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), 690 rax, 691 ACCESS_ON_HEAP | ACCESS_ON_ARRAY); 692 } 693 694 void TemplateTable::baload() { 695 transition(itos, itos); 696 // rax: index 697 // rdx: array 698 index_check(rdx, rax); // kills rbx 699 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 700 } 701 702 void TemplateTable::caload() { 703 transition(itos, itos); 704 // rax: index 705 // rdx: array 706 index_check(rdx, rax); // kills rbx 707 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 708 } 709 710 // iload followed by caload frequent pair 711 void TemplateTable::fast_icaload() { 712 transition(vtos, itos); 713 // load index out of locals 714 locals_index(rbx); 715 __ movl(rax, iaddress(rbx)); 716 717 // rax: index 718 // rdx: array 719 index_check(rdx, rax); // kills rbx 720 __ load_unsigned_short(rax, 721 Address(rdx, rax, 722 Address::times_2, 723 arrayOopDesc::base_offset_in_bytes(T_CHAR))); 724 } 725 726 727 void TemplateTable::saload() { 728 transition(itos, itos); 729 // rax: index 730 // rdx: array 731 index_check(rdx, rax); // kills rbx 732 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 733 } 734 735 void TemplateTable::iload(int n) { 736 transition(vtos, itos); 737 __ movl(rax, iaddress(n)); 738 } 739 740 void TemplateTable::lload(int n) { 741 transition(vtos, ltos); 742 __ movptr(rax, laddress(n)); 743 NOT_LP64(__ movptr(rdx, haddress(n))); 744 } 745 746 void TemplateTable::fload(int n) { 747 transition(vtos, ftos); 748 __ load_float(faddress(n)); 749 } 750 751 void TemplateTable::dload(int n) { 752 transition(vtos, dtos); 753 __ load_double(daddress(n)); 754 } 755 756 void TemplateTable::aload(int n) { 757 transition(vtos, atos); 758 __ movptr(rax, aaddress(n)); 759 } 760 761 void TemplateTable::aload_0() { 762 aload_0_internal(); 763 } 764 765 void TemplateTable::nofast_aload_0() { 766 aload_0_internal(may_not_rewrite); 767 } 768 769 void TemplateTable::aload_0_internal(RewriteControl rc) { 770 transition(vtos, atos); 771 // According to bytecode histograms, the pairs: 772 // 773 // _aload_0, _fast_igetfield 774 // _aload_0, _fast_agetfield 775 // _aload_0, _fast_fgetfield 776 // 777 // occur frequently. If RewriteFrequentPairs is set, the (slow) 778 // _aload_0 bytecode checks if the next bytecode is either 779 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then 780 // rewrites the current bytecode into a pair bytecode; otherwise it 781 // rewrites the current bytecode into _fast_aload_0 that doesn't do 782 // the pair check anymore. 783 // 784 // Note: If the next bytecode is _getfield, the rewrite must be 785 // delayed, otherwise we may miss an opportunity for a pair. 786 // 787 // Also rewrite frequent pairs 788 // aload_0, aload_1 789 // aload_0, iload_1 790 // These bytecodes with a small amount of code are most profitable 791 // to rewrite 792 if (RewriteFrequentPairs && rc == may_rewrite) { 793 Label rewrite, done; 794 795 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 796 LP64_ONLY(assert(rbx != bc, "register damaged")); 797 798 // get next byte 799 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 800 801 // if _getfield then wait with rewrite 802 __ cmpl(rbx, Bytecodes::_getfield); 803 __ jcc(Assembler::equal, done); 804 805 // if _igetfield then rewrite to _fast_iaccess_0 806 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 807 __ cmpl(rbx, Bytecodes::_fast_igetfield); 808 __ movl(bc, Bytecodes::_fast_iaccess_0); 809 __ jccb(Assembler::equal, rewrite); 810 811 // if _agetfield then rewrite to _fast_aaccess_0 812 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 813 __ cmpl(rbx, Bytecodes::_fast_agetfield); 814 __ movl(bc, Bytecodes::_fast_aaccess_0); 815 __ jccb(Assembler::equal, rewrite); 816 817 // if _fgetfield then rewrite to _fast_faccess_0 818 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 819 __ cmpl(rbx, Bytecodes::_fast_fgetfield); 820 __ movl(bc, Bytecodes::_fast_faccess_0); 821 __ jccb(Assembler::equal, rewrite); 822 823 // else rewrite to _fast_aload0 824 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 825 __ movl(bc, Bytecodes::_fast_aload_0); 826 827 // rewrite 828 // bc: fast bytecode 829 __ bind(rewrite); 830 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false); 831 832 __ bind(done); 833 } 834 835 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop). 836 aload(0); 837 } 838 839 void TemplateTable::istore() { 840 transition(itos, vtos); 841 locals_index(rbx); 842 __ movl(iaddress(rbx), rax); 843 } 844 845 846 void TemplateTable::lstore() { 847 transition(ltos, vtos); 848 locals_index(rbx); 849 __ movptr(laddress(rbx), rax); 850 NOT_LP64(__ movptr(haddress(rbx), rdx)); 851 } 852 853 void TemplateTable::fstore() { 854 transition(ftos, vtos); 855 locals_index(rbx); 856 __ store_float(faddress(rbx)); 857 } 858 859 void TemplateTable::dstore() { 860 transition(dtos, vtos); 861 locals_index(rbx); 862 __ store_double(daddress(rbx)); 863 } 864 865 void TemplateTable::astore() { 866 transition(vtos, vtos); 867 __ pop_ptr(rax); 868 locals_index(rbx); 869 __ movptr(aaddress(rbx), rax); 870 } 871 872 void TemplateTable::wide_istore() { 873 transition(vtos, vtos); 874 __ pop_i(); 875 locals_index_wide(rbx); 876 __ movl(iaddress(rbx), rax); 877 } 878 879 void TemplateTable::wide_lstore() { 880 transition(vtos, vtos); 881 NOT_LP64(__ pop_l(rax, rdx)); 882 LP64_ONLY(__ pop_l()); 883 locals_index_wide(rbx); 884 __ movptr(laddress(rbx), rax); 885 NOT_LP64(__ movl(haddress(rbx), rdx)); 886 } 887 888 void TemplateTable::wide_fstore() { 889 #ifdef _LP64 890 transition(vtos, vtos); 891 __ pop_f(xmm0); 892 locals_index_wide(rbx); 893 __ movflt(faddress(rbx), xmm0); 894 #else 895 wide_istore(); 896 #endif 897 } 898 899 void TemplateTable::wide_dstore() { 900 #ifdef _LP64 901 transition(vtos, vtos); 902 __ pop_d(xmm0); 903 locals_index_wide(rbx); 904 __ movdbl(daddress(rbx), xmm0); 905 #else 906 wide_lstore(); 907 #endif 908 } 909 910 void TemplateTable::wide_astore() { 911 transition(vtos, vtos); 912 __ pop_ptr(rax); 913 locals_index_wide(rbx); 914 __ movptr(aaddress(rbx), rax); 915 } 916 917 void TemplateTable::iastore() { 918 transition(itos, vtos); 919 __ pop_i(rbx); 920 // rax: value 921 // rbx: index 922 // rdx: array 923 index_check(rdx, rbx); // prefer index in rbx 924 __ movl(Address(rdx, rbx, 925 Address::times_4, 926 arrayOopDesc::base_offset_in_bytes(T_INT)), 927 rax); 928 } 929 930 void TemplateTable::lastore() { 931 transition(ltos, vtos); 932 __ pop_i(rbx); 933 // rax,: low(value) 934 // rcx: array 935 // rdx: high(value) 936 index_check(rcx, rbx); // prefer index in rbx, 937 // rbx,: index 938 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax); 939 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx)); 940 } 941 942 943 void TemplateTable::fastore() { 944 transition(ftos, vtos); 945 __ pop_i(rbx); 946 // value is in UseSSE >= 1 ? xmm0 : ST(0) 947 // rbx: index 948 // rdx: array 949 index_check(rdx, rbx); // prefer index in rbx 950 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 951 } 952 953 void TemplateTable::dastore() { 954 transition(dtos, vtos); 955 __ pop_i(rbx); 956 // value is in UseSSE >= 2 ? xmm0 : ST(0) 957 // rbx: index 958 // rdx: array 959 index_check(rdx, rbx); // prefer index in rbx 960 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 961 } 962 963 void TemplateTable::aastore() { 964 Label is_null, ok_is_subtype, done; 965 transition(vtos, vtos); 966 // stack: ..., array, index, value 967 __ movptr(rax, at_tos()); // value 968 __ movl(rcx, at_tos_p1()); // index 969 __ movptr(rdx, at_tos_p2()); // array 970 971 Address element_address(rdx, rcx, 972 UseCompressedOops? Address::times_4 : Address::times_ptr, 973 arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 974 975 index_check_without_pop(rdx, rcx); // kills rbx 976 __ testptr(rax, rax); 977 __ jcc(Assembler::zero, is_null); 978 979 // Move subklass into rbx 980 __ load_klass(rbx, rax); 981 // Move superklass into rax 982 __ load_klass(rax, rdx); 983 __ movptr(rax, Address(rax, 984 ObjArrayKlass::element_klass_offset())); 985 // Compress array + index*oopSize + 12 into a single register. Frees rcx. 986 __ lea(rdx, element_address); 987 988 // Generate subtype check. Blows rcx, rdi 989 // Superklass in rax. Subklass in rbx. 990 __ gen_subtype_check(rbx, ok_is_subtype); 991 992 // Come here on failure 993 // object is at TOS 994 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry)); 995 996 // Come here on success 997 __ bind(ok_is_subtype); 998 999 // Get the value we will store 1000 __ movptr(rax, at_tos()); 1001 // Now store using the appropriate barrier 1002 do_oop_store(_masm, Address(rdx, 0), rax, ACCESS_ON_HEAP); 1003 __ jmp(done); 1004 1005 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx] 1006 __ bind(is_null); 1007 __ profile_null_seen(rbx); 1008 1009 // Store a NULL 1010 do_oop_store(_masm, element_address, noreg, ACCESS_ON_HEAP); 1011 1012 // Pop stack arguments 1013 __ bind(done); 1014 __ addptr(rsp, 3 * Interpreter::stackElementSize); 1015 } 1016 1017 void TemplateTable::bastore() { 1018 transition(itos, vtos); 1019 __ pop_i(rbx); 1020 // rax: value 1021 // rbx: index 1022 // rdx: array 1023 index_check(rdx, rbx); // prefer index in rbx 1024 // Need to check whether array is boolean or byte 1025 // since both types share the bastore bytecode. 1026 __ load_klass(rcx, rdx); 1027 __ movl(rcx, Address(rcx, Klass::layout_helper_offset())); 1028 int diffbit = Klass::layout_helper_boolean_diffbit(); 1029 __ testl(rcx, diffbit); 1030 Label L_skip; 1031 __ jccb(Assembler::zero, L_skip); 1032 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1 1033 __ bind(L_skip); 1034 __ movb(Address(rdx, rbx, 1035 Address::times_1, 1036 arrayOopDesc::base_offset_in_bytes(T_BYTE)), 1037 rax); 1038 } 1039 1040 void TemplateTable::castore() { 1041 transition(itos, vtos); 1042 __ pop_i(rbx); 1043 // rax: value 1044 // rbx: index 1045 // rdx: array 1046 index_check(rdx, rbx); // prefer index in rbx 1047 __ movw(Address(rdx, rbx, 1048 Address::times_2, 1049 arrayOopDesc::base_offset_in_bytes(T_CHAR)), 1050 rax); 1051 } 1052 1053 1054 void TemplateTable::sastore() { 1055 castore(); 1056 } 1057 1058 void TemplateTable::istore(int n) { 1059 transition(itos, vtos); 1060 __ movl(iaddress(n), rax); 1061 } 1062 1063 void TemplateTable::lstore(int n) { 1064 transition(ltos, vtos); 1065 __ movptr(laddress(n), rax); 1066 NOT_LP64(__ movptr(haddress(n), rdx)); 1067 } 1068 1069 void TemplateTable::fstore(int n) { 1070 transition(ftos, vtos); 1071 __ store_float(faddress(n)); 1072 } 1073 1074 void TemplateTable::dstore(int n) { 1075 transition(dtos, vtos); 1076 __ store_double(daddress(n)); 1077 } 1078 1079 1080 void TemplateTable::astore(int n) { 1081 transition(vtos, vtos); 1082 __ pop_ptr(rax); 1083 __ movptr(aaddress(n), rax); 1084 } 1085 1086 void TemplateTable::pop() { 1087 transition(vtos, vtos); 1088 __ addptr(rsp, Interpreter::stackElementSize); 1089 } 1090 1091 void TemplateTable::pop2() { 1092 transition(vtos, vtos); 1093 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1094 } 1095 1096 1097 void TemplateTable::dup() { 1098 transition(vtos, vtos); 1099 __ load_ptr(0, rax); 1100 __ push_ptr(rax); 1101 // stack: ..., a, a 1102 } 1103 1104 void TemplateTable::dup_x1() { 1105 transition(vtos, vtos); 1106 // stack: ..., a, b 1107 __ load_ptr( 0, rax); // load b 1108 __ load_ptr( 1, rcx); // load a 1109 __ store_ptr(1, rax); // store b 1110 __ store_ptr(0, rcx); // store a 1111 __ push_ptr(rax); // push b 1112 // stack: ..., b, a, b 1113 } 1114 1115 void TemplateTable::dup_x2() { 1116 transition(vtos, vtos); 1117 // stack: ..., a, b, c 1118 __ load_ptr( 0, rax); // load c 1119 __ load_ptr( 2, rcx); // load a 1120 __ store_ptr(2, rax); // store c in a 1121 __ push_ptr(rax); // push c 1122 // stack: ..., c, b, c, c 1123 __ load_ptr( 2, rax); // load b 1124 __ store_ptr(2, rcx); // store a in b 1125 // stack: ..., c, a, c, c 1126 __ store_ptr(1, rax); // store b in c 1127 // stack: ..., c, a, b, c 1128 } 1129 1130 void TemplateTable::dup2() { 1131 transition(vtos, vtos); 1132 // stack: ..., a, b 1133 __ load_ptr(1, rax); // load a 1134 __ push_ptr(rax); // push a 1135 __ load_ptr(1, rax); // load b 1136 __ push_ptr(rax); // push b 1137 // stack: ..., a, b, a, b 1138 } 1139 1140 1141 void TemplateTable::dup2_x1() { 1142 transition(vtos, vtos); 1143 // stack: ..., a, b, c 1144 __ load_ptr( 0, rcx); // load c 1145 __ load_ptr( 1, rax); // load b 1146 __ push_ptr(rax); // push b 1147 __ push_ptr(rcx); // push c 1148 // stack: ..., a, b, c, b, c 1149 __ store_ptr(3, rcx); // store c in b 1150 // stack: ..., a, c, c, b, c 1151 __ load_ptr( 4, rcx); // load a 1152 __ store_ptr(2, rcx); // store a in 2nd c 1153 // stack: ..., a, c, a, b, c 1154 __ store_ptr(4, rax); // store b in a 1155 // stack: ..., b, c, a, b, c 1156 } 1157 1158 void TemplateTable::dup2_x2() { 1159 transition(vtos, vtos); 1160 // stack: ..., a, b, c, d 1161 __ load_ptr( 0, rcx); // load d 1162 __ load_ptr( 1, rax); // load c 1163 __ push_ptr(rax); // push c 1164 __ push_ptr(rcx); // push d 1165 // stack: ..., a, b, c, d, c, d 1166 __ load_ptr( 4, rax); // load b 1167 __ store_ptr(2, rax); // store b in d 1168 __ store_ptr(4, rcx); // store d in b 1169 // stack: ..., a, d, c, b, c, d 1170 __ load_ptr( 5, rcx); // load a 1171 __ load_ptr( 3, rax); // load c 1172 __ store_ptr(3, rcx); // store a in c 1173 __ store_ptr(5, rax); // store c in a 1174 // stack: ..., c, d, a, b, c, d 1175 } 1176 1177 void TemplateTable::swap() { 1178 transition(vtos, vtos); 1179 // stack: ..., a, b 1180 __ load_ptr( 1, rcx); // load a 1181 __ load_ptr( 0, rax); // load b 1182 __ store_ptr(0, rcx); // store a in b 1183 __ store_ptr(1, rax); // store b in a 1184 // stack: ..., b, a 1185 } 1186 1187 void TemplateTable::iop2(Operation op) { 1188 transition(itos, itos); 1189 switch (op) { 1190 case add : __ pop_i(rdx); __ addl (rax, rdx); break; 1191 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break; 1192 case mul : __ pop_i(rdx); __ imull(rax, rdx); break; 1193 case _and : __ pop_i(rdx); __ andl (rax, rdx); break; 1194 case _or : __ pop_i(rdx); __ orl (rax, rdx); break; 1195 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break; 1196 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break; 1197 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break; 1198 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break; 1199 default : ShouldNotReachHere(); 1200 } 1201 } 1202 1203 void TemplateTable::lop2(Operation op) { 1204 transition(ltos, ltos); 1205 #ifdef _LP64 1206 switch (op) { 1207 case add : __ pop_l(rdx); __ addptr(rax, rdx); break; 1208 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break; 1209 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break; 1210 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break; 1211 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break; 1212 default : ShouldNotReachHere(); 1213 } 1214 #else 1215 __ pop_l(rbx, rcx); 1216 switch (op) { 1217 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break; 1218 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx); 1219 __ mov (rax, rbx); __ mov (rdx, rcx); break; 1220 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break; 1221 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break; 1222 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break; 1223 default : ShouldNotReachHere(); 1224 } 1225 #endif 1226 } 1227 1228 void TemplateTable::idiv() { 1229 transition(itos, itos); 1230 __ movl(rcx, rax); 1231 __ pop_i(rax); 1232 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If 1233 // they are not equal, one could do a normal division (no correction 1234 // needed), which may speed up this implementation for the common case. 1235 // (see also JVM spec., p.243 & p.271) 1236 __ corrected_idivl(rcx); 1237 } 1238 1239 void TemplateTable::irem() { 1240 transition(itos, itos); 1241 __ movl(rcx, rax); 1242 __ pop_i(rax); 1243 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If 1244 // they are not equal, one could do a normal division (no correction 1245 // needed), which may speed up this implementation for the common case. 1246 // (see also JVM spec., p.243 & p.271) 1247 __ corrected_idivl(rcx); 1248 __ movl(rax, rdx); 1249 } 1250 1251 void TemplateTable::lmul() { 1252 transition(ltos, ltos); 1253 #ifdef _LP64 1254 __ pop_l(rdx); 1255 __ imulq(rax, rdx); 1256 #else 1257 __ pop_l(rbx, rcx); 1258 __ push(rcx); __ push(rbx); 1259 __ push(rdx); __ push(rax); 1260 __ lmul(2 * wordSize, 0); 1261 __ addptr(rsp, 4 * wordSize); // take off temporaries 1262 #endif 1263 } 1264 1265 void TemplateTable::ldiv() { 1266 transition(ltos, ltos); 1267 #ifdef _LP64 1268 __ mov(rcx, rax); 1269 __ pop_l(rax); 1270 // generate explicit div0 check 1271 __ testq(rcx, rcx); 1272 __ jump_cc(Assembler::zero, 1273 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1274 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If 1275 // they are not equal, one could do a normal division (no correction 1276 // needed), which may speed up this implementation for the common case. 1277 // (see also JVM spec., p.243 & p.271) 1278 __ corrected_idivq(rcx); // kills rbx 1279 #else 1280 __ pop_l(rbx, rcx); 1281 __ push(rcx); __ push(rbx); 1282 __ push(rdx); __ push(rax); 1283 // check if y = 0 1284 __ orl(rax, rdx); 1285 __ jump_cc(Assembler::zero, 1286 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1287 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv)); 1288 __ addptr(rsp, 4 * wordSize); // take off temporaries 1289 #endif 1290 } 1291 1292 void TemplateTable::lrem() { 1293 transition(ltos, ltos); 1294 #ifdef _LP64 1295 __ mov(rcx, rax); 1296 __ pop_l(rax); 1297 __ testq(rcx, rcx); 1298 __ jump_cc(Assembler::zero, 1299 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1300 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If 1301 // they are not equal, one could do a normal division (no correction 1302 // needed), which may speed up this implementation for the common case. 1303 // (see also JVM spec., p.243 & p.271) 1304 __ corrected_idivq(rcx); // kills rbx 1305 __ mov(rax, rdx); 1306 #else 1307 __ pop_l(rbx, rcx); 1308 __ push(rcx); __ push(rbx); 1309 __ push(rdx); __ push(rax); 1310 // check if y = 0 1311 __ orl(rax, rdx); 1312 __ jump_cc(Assembler::zero, 1313 ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1314 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem)); 1315 __ addptr(rsp, 4 * wordSize); 1316 #endif 1317 } 1318 1319 void TemplateTable::lshl() { 1320 transition(itos, ltos); 1321 __ movl(rcx, rax); // get shift count 1322 #ifdef _LP64 1323 __ pop_l(rax); // get shift value 1324 __ shlq(rax); 1325 #else 1326 __ pop_l(rax, rdx); // get shift value 1327 __ lshl(rdx, rax); 1328 #endif 1329 } 1330 1331 void TemplateTable::lshr() { 1332 #ifdef _LP64 1333 transition(itos, ltos); 1334 __ movl(rcx, rax); // get shift count 1335 __ pop_l(rax); // get shift value 1336 __ sarq(rax); 1337 #else 1338 transition(itos, ltos); 1339 __ mov(rcx, rax); // get shift count 1340 __ pop_l(rax, rdx); // get shift value 1341 __ lshr(rdx, rax, true); 1342 #endif 1343 } 1344 1345 void TemplateTable::lushr() { 1346 transition(itos, ltos); 1347 #ifdef _LP64 1348 __ movl(rcx, rax); // get shift count 1349 __ pop_l(rax); // get shift value 1350 __ shrq(rax); 1351 #else 1352 __ mov(rcx, rax); // get shift count 1353 __ pop_l(rax, rdx); // get shift value 1354 __ lshr(rdx, rax); 1355 #endif 1356 } 1357 1358 void TemplateTable::fop2(Operation op) { 1359 transition(ftos, ftos); 1360 1361 if (UseSSE >= 1) { 1362 switch (op) { 1363 case add: 1364 __ addss(xmm0, at_rsp()); 1365 __ addptr(rsp, Interpreter::stackElementSize); 1366 break; 1367 case sub: 1368 __ movflt(xmm1, xmm0); 1369 __ pop_f(xmm0); 1370 __ subss(xmm0, xmm1); 1371 break; 1372 case mul: 1373 __ mulss(xmm0, at_rsp()); 1374 __ addptr(rsp, Interpreter::stackElementSize); 1375 break; 1376 case div: 1377 __ movflt(xmm1, xmm0); 1378 __ pop_f(xmm0); 1379 __ divss(xmm0, xmm1); 1380 break; 1381 case rem: 1382 // On x86_64 platforms the SharedRuntime::frem method is called to perform the 1383 // modulo operation. The frem method calls the function 1384 // double fmod(double x, double y) in math.h. The documentation of fmod states: 1385 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN 1386 // (signalling or quiet) is returned. 1387 // 1388 // On x86_32 platforms the FPU is used to perform the modulo operation. The 1389 // reason is that on 32-bit Windows the sign of modulo operations diverges from 1390 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f). 1391 // The fprem instruction used on x86_32 is functionally equivalent to 1392 // SharedRuntime::frem in that it returns a NaN. 1393 #ifdef _LP64 1394 __ movflt(xmm1, xmm0); 1395 __ pop_f(xmm0); 1396 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2); 1397 #else 1398 __ push_f(xmm0); 1399 __ pop_f(); 1400 __ fld_s(at_rsp()); 1401 __ fremr(rax); 1402 __ f2ieee(); 1403 __ pop(rax); // pop second operand off the stack 1404 __ push_f(); 1405 __ pop_f(xmm0); 1406 #endif 1407 break; 1408 default: 1409 ShouldNotReachHere(); 1410 break; 1411 } 1412 } else { 1413 #ifdef _LP64 1414 ShouldNotReachHere(); 1415 #else 1416 switch (op) { 1417 case add: __ fadd_s (at_rsp()); break; 1418 case sub: __ fsubr_s(at_rsp()); break; 1419 case mul: __ fmul_s (at_rsp()); break; 1420 case div: __ fdivr_s(at_rsp()); break; 1421 case rem: __ fld_s (at_rsp()); __ fremr(rax); break; 1422 default : ShouldNotReachHere(); 1423 } 1424 __ f2ieee(); 1425 __ pop(rax); // pop second operand off the stack 1426 #endif // _LP64 1427 } 1428 } 1429 1430 void TemplateTable::dop2(Operation op) { 1431 transition(dtos, dtos); 1432 if (UseSSE >= 2) { 1433 switch (op) { 1434 case add: 1435 __ addsd(xmm0, at_rsp()); 1436 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1437 break; 1438 case sub: 1439 __ movdbl(xmm1, xmm0); 1440 __ pop_d(xmm0); 1441 __ subsd(xmm0, xmm1); 1442 break; 1443 case mul: 1444 __ mulsd(xmm0, at_rsp()); 1445 __ addptr(rsp, 2 * Interpreter::stackElementSize); 1446 break; 1447 case div: 1448 __ movdbl(xmm1, xmm0); 1449 __ pop_d(xmm0); 1450 __ divsd(xmm0, xmm1); 1451 break; 1452 case rem: 1453 // Similar to fop2(), the modulo operation is performed using the 1454 // SharedRuntime::drem method (on x86_64 platforms) or using the 1455 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2(). 1456 #ifdef _LP64 1457 __ movdbl(xmm1, xmm0); 1458 __ pop_d(xmm0); 1459 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2); 1460 #else 1461 __ push_d(xmm0); 1462 __ pop_d(); 1463 __ fld_d(at_rsp()); 1464 __ fremr(rax); 1465 __ d2ieee(); 1466 __ pop(rax); 1467 __ pop(rdx); 1468 __ push_d(); 1469 __ pop_d(xmm0); 1470 #endif 1471 break; 1472 default: 1473 ShouldNotReachHere(); 1474 break; 1475 } 1476 } else { 1477 #ifdef _LP64 1478 ShouldNotReachHere(); 1479 #else 1480 switch (op) { 1481 case add: __ fadd_d (at_rsp()); break; 1482 case sub: __ fsubr_d(at_rsp()); break; 1483 case mul: { 1484 Label L_strict; 1485 Label L_join; 1486 const Address access_flags (rcx, Method::access_flags_offset()); 1487 __ get_method(rcx); 1488 __ movl(rcx, access_flags); 1489 __ testl(rcx, JVM_ACC_STRICT); 1490 __ jccb(Assembler::notZero, L_strict); 1491 __ fmul_d (at_rsp()); 1492 __ jmpb(L_join); 1493 __ bind(L_strict); 1494 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1495 __ fmulp(); 1496 __ fmul_d (at_rsp()); 1497 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1498 __ fmulp(); 1499 __ bind(L_join); 1500 break; 1501 } 1502 case div: { 1503 Label L_strict; 1504 Label L_join; 1505 const Address access_flags (rcx, Method::access_flags_offset()); 1506 __ get_method(rcx); 1507 __ movl(rcx, access_flags); 1508 __ testl(rcx, JVM_ACC_STRICT); 1509 __ jccb(Assembler::notZero, L_strict); 1510 __ fdivr_d(at_rsp()); 1511 __ jmp(L_join); 1512 __ bind(L_strict); 1513 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1514 __ fmul_d (at_rsp()); 1515 __ fdivrp(); 1516 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1517 __ fmulp(); 1518 __ bind(L_join); 1519 break; 1520 } 1521 case rem: __ fld_d (at_rsp()); __ fremr(rax); break; 1522 default : ShouldNotReachHere(); 1523 } 1524 __ d2ieee(); 1525 // Pop double precision number from rsp. 1526 __ pop(rax); 1527 __ pop(rdx); 1528 #endif 1529 } 1530 } 1531 1532 void TemplateTable::ineg() { 1533 transition(itos, itos); 1534 __ negl(rax); 1535 } 1536 1537 void TemplateTable::lneg() { 1538 transition(ltos, ltos); 1539 LP64_ONLY(__ negq(rax)); 1540 NOT_LP64(__ lneg(rdx, rax)); 1541 } 1542 1543 // Note: 'double' and 'long long' have 32-bits alignment on x86. 1544 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) { 1545 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address 1546 // of 128-bits operands for SSE instructions. 1547 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF))); 1548 // Store the value to a 128-bits operand. 1549 operand[0] = lo; 1550 operand[1] = hi; 1551 return operand; 1552 } 1553 1554 // Buffer for 128-bits masks used by SSE instructions. 1555 static jlong float_signflip_pool[2*2]; 1556 static jlong double_signflip_pool[2*2]; 1557 1558 void TemplateTable::fneg() { 1559 transition(ftos, ftos); 1560 if (UseSSE >= 1) { 1561 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000)); 1562 __ xorps(xmm0, ExternalAddress((address) float_signflip)); 1563 } else { 1564 LP64_ONLY(ShouldNotReachHere()); 1565 NOT_LP64(__ fchs()); 1566 } 1567 } 1568 1569 void TemplateTable::dneg() { 1570 transition(dtos, dtos); 1571 if (UseSSE >= 2) { 1572 static jlong *double_signflip = 1573 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000)); 1574 __ xorpd(xmm0, ExternalAddress((address) double_signflip)); 1575 } else { 1576 #ifdef _LP64 1577 ShouldNotReachHere(); 1578 #else 1579 __ fchs(); 1580 #endif 1581 } 1582 } 1583 1584 void TemplateTable::iinc() { 1585 transition(vtos, vtos); 1586 __ load_signed_byte(rdx, at_bcp(2)); // get constant 1587 locals_index(rbx); 1588 __ addl(iaddress(rbx), rdx); 1589 } 1590 1591 void TemplateTable::wide_iinc() { 1592 transition(vtos, vtos); 1593 __ movl(rdx, at_bcp(4)); // get constant 1594 locals_index_wide(rbx); 1595 __ bswapl(rdx); // swap bytes & sign-extend constant 1596 __ sarl(rdx, 16); 1597 __ addl(iaddress(rbx), rdx); 1598 // Note: should probably use only one movl to get both 1599 // the index and the constant -> fix this 1600 } 1601 1602 void TemplateTable::convert() { 1603 #ifdef _LP64 1604 // Checking 1605 #ifdef ASSERT 1606 { 1607 TosState tos_in = ilgl; 1608 TosState tos_out = ilgl; 1609 switch (bytecode()) { 1610 case Bytecodes::_i2l: // fall through 1611 case Bytecodes::_i2f: // fall through 1612 case Bytecodes::_i2d: // fall through 1613 case Bytecodes::_i2b: // fall through 1614 case Bytecodes::_i2c: // fall through 1615 case Bytecodes::_i2s: tos_in = itos; break; 1616 case Bytecodes::_l2i: // fall through 1617 case Bytecodes::_l2f: // fall through 1618 case Bytecodes::_l2d: tos_in = ltos; break; 1619 case Bytecodes::_f2i: // fall through 1620 case Bytecodes::_f2l: // fall through 1621 case Bytecodes::_f2d: tos_in = ftos; break; 1622 case Bytecodes::_d2i: // fall through 1623 case Bytecodes::_d2l: // fall through 1624 case Bytecodes::_d2f: tos_in = dtos; break; 1625 default : ShouldNotReachHere(); 1626 } 1627 switch (bytecode()) { 1628 case Bytecodes::_l2i: // fall through 1629 case Bytecodes::_f2i: // fall through 1630 case Bytecodes::_d2i: // fall through 1631 case Bytecodes::_i2b: // fall through 1632 case Bytecodes::_i2c: // fall through 1633 case Bytecodes::_i2s: tos_out = itos; break; 1634 case Bytecodes::_i2l: // fall through 1635 case Bytecodes::_f2l: // fall through 1636 case Bytecodes::_d2l: tos_out = ltos; break; 1637 case Bytecodes::_i2f: // fall through 1638 case Bytecodes::_l2f: // fall through 1639 case Bytecodes::_d2f: tos_out = ftos; break; 1640 case Bytecodes::_i2d: // fall through 1641 case Bytecodes::_l2d: // fall through 1642 case Bytecodes::_f2d: tos_out = dtos; break; 1643 default : ShouldNotReachHere(); 1644 } 1645 transition(tos_in, tos_out); 1646 } 1647 #endif // ASSERT 1648 1649 static const int64_t is_nan = 0x8000000000000000L; 1650 1651 // Conversion 1652 switch (bytecode()) { 1653 case Bytecodes::_i2l: 1654 __ movslq(rax, rax); 1655 break; 1656 case Bytecodes::_i2f: 1657 __ cvtsi2ssl(xmm0, rax); 1658 break; 1659 case Bytecodes::_i2d: 1660 __ cvtsi2sdl(xmm0, rax); 1661 break; 1662 case Bytecodes::_i2b: 1663 __ movsbl(rax, rax); 1664 break; 1665 case Bytecodes::_i2c: 1666 __ movzwl(rax, rax); 1667 break; 1668 case Bytecodes::_i2s: 1669 __ movswl(rax, rax); 1670 break; 1671 case Bytecodes::_l2i: 1672 __ movl(rax, rax); 1673 break; 1674 case Bytecodes::_l2f: 1675 __ cvtsi2ssq(xmm0, rax); 1676 break; 1677 case Bytecodes::_l2d: 1678 __ cvtsi2sdq(xmm0, rax); 1679 break; 1680 case Bytecodes::_f2i: 1681 { 1682 Label L; 1683 __ cvttss2sil(rax, xmm0); 1684 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow? 1685 __ jcc(Assembler::notEqual, L); 1686 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1687 __ bind(L); 1688 } 1689 break; 1690 case Bytecodes::_f2l: 1691 { 1692 Label L; 1693 __ cvttss2siq(rax, xmm0); 1694 // NaN or overflow/underflow? 1695 __ cmp64(rax, ExternalAddress((address) &is_nan)); 1696 __ jcc(Assembler::notEqual, L); 1697 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1698 __ bind(L); 1699 } 1700 break; 1701 case Bytecodes::_f2d: 1702 __ cvtss2sd(xmm0, xmm0); 1703 break; 1704 case Bytecodes::_d2i: 1705 { 1706 Label L; 1707 __ cvttsd2sil(rax, xmm0); 1708 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow? 1709 __ jcc(Assembler::notEqual, L); 1710 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1); 1711 __ bind(L); 1712 } 1713 break; 1714 case Bytecodes::_d2l: 1715 { 1716 Label L; 1717 __ cvttsd2siq(rax, xmm0); 1718 // NaN or overflow/underflow? 1719 __ cmp64(rax, ExternalAddress((address) &is_nan)); 1720 __ jcc(Assembler::notEqual, L); 1721 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1); 1722 __ bind(L); 1723 } 1724 break; 1725 case Bytecodes::_d2f: 1726 __ cvtsd2ss(xmm0, xmm0); 1727 break; 1728 default: 1729 ShouldNotReachHere(); 1730 } 1731 #else 1732 // Checking 1733 #ifdef ASSERT 1734 { TosState tos_in = ilgl; 1735 TosState tos_out = ilgl; 1736 switch (bytecode()) { 1737 case Bytecodes::_i2l: // fall through 1738 case Bytecodes::_i2f: // fall through 1739 case Bytecodes::_i2d: // fall through 1740 case Bytecodes::_i2b: // fall through 1741 case Bytecodes::_i2c: // fall through 1742 case Bytecodes::_i2s: tos_in = itos; break; 1743 case Bytecodes::_l2i: // fall through 1744 case Bytecodes::_l2f: // fall through 1745 case Bytecodes::_l2d: tos_in = ltos; break; 1746 case Bytecodes::_f2i: // fall through 1747 case Bytecodes::_f2l: // fall through 1748 case Bytecodes::_f2d: tos_in = ftos; break; 1749 case Bytecodes::_d2i: // fall through 1750 case Bytecodes::_d2l: // fall through 1751 case Bytecodes::_d2f: tos_in = dtos; break; 1752 default : ShouldNotReachHere(); 1753 } 1754 switch (bytecode()) { 1755 case Bytecodes::_l2i: // fall through 1756 case Bytecodes::_f2i: // fall through 1757 case Bytecodes::_d2i: // fall through 1758 case Bytecodes::_i2b: // fall through 1759 case Bytecodes::_i2c: // fall through 1760 case Bytecodes::_i2s: tos_out = itos; break; 1761 case Bytecodes::_i2l: // fall through 1762 case Bytecodes::_f2l: // fall through 1763 case Bytecodes::_d2l: tos_out = ltos; break; 1764 case Bytecodes::_i2f: // fall through 1765 case Bytecodes::_l2f: // fall through 1766 case Bytecodes::_d2f: tos_out = ftos; break; 1767 case Bytecodes::_i2d: // fall through 1768 case Bytecodes::_l2d: // fall through 1769 case Bytecodes::_f2d: tos_out = dtos; break; 1770 default : ShouldNotReachHere(); 1771 } 1772 transition(tos_in, tos_out); 1773 } 1774 #endif // ASSERT 1775 1776 // Conversion 1777 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation) 1778 switch (bytecode()) { 1779 case Bytecodes::_i2l: 1780 __ extend_sign(rdx, rax); 1781 break; 1782 case Bytecodes::_i2f: 1783 if (UseSSE >= 1) { 1784 __ cvtsi2ssl(xmm0, rax); 1785 } else { 1786 __ push(rax); // store int on tos 1787 __ fild_s(at_rsp()); // load int to ST0 1788 __ f2ieee(); // truncate to float size 1789 __ pop(rcx); // adjust rsp 1790 } 1791 break; 1792 case Bytecodes::_i2d: 1793 if (UseSSE >= 2) { 1794 __ cvtsi2sdl(xmm0, rax); 1795 } else { 1796 __ push(rax); // add one slot for d2ieee() 1797 __ push(rax); // store int on tos 1798 __ fild_s(at_rsp()); // load int to ST0 1799 __ d2ieee(); // truncate to double size 1800 __ pop(rcx); // adjust rsp 1801 __ pop(rcx); 1802 } 1803 break; 1804 case Bytecodes::_i2b: 1805 __ shll(rax, 24); // truncate upper 24 bits 1806 __ sarl(rax, 24); // and sign-extend byte 1807 LP64_ONLY(__ movsbl(rax, rax)); 1808 break; 1809 case Bytecodes::_i2c: 1810 __ andl(rax, 0xFFFF); // truncate upper 16 bits 1811 LP64_ONLY(__ movzwl(rax, rax)); 1812 break; 1813 case Bytecodes::_i2s: 1814 __ shll(rax, 16); // truncate upper 16 bits 1815 __ sarl(rax, 16); // and sign-extend short 1816 LP64_ONLY(__ movswl(rax, rax)); 1817 break; 1818 case Bytecodes::_l2i: 1819 /* nothing to do */ 1820 break; 1821 case Bytecodes::_l2f: 1822 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert 1823 // 64-bit long values to floats. On 32-bit platforms it is not possible 1824 // to use that instruction with 64-bit operands, therefore the FPU is 1825 // used to perform the conversion. 1826 __ push(rdx); // store long on tos 1827 __ push(rax); 1828 __ fild_d(at_rsp()); // load long to ST0 1829 __ f2ieee(); // truncate to float size 1830 __ pop(rcx); // adjust rsp 1831 __ pop(rcx); 1832 if (UseSSE >= 1) { 1833 __ push_f(); 1834 __ pop_f(xmm0); 1835 } 1836 break; 1837 case Bytecodes::_l2d: 1838 // On 32-bit platforms the FPU is used for conversion because on 1839 // 32-bit platforms it is not not possible to use the cvtsi2sdq 1840 // instruction with 64-bit operands. 1841 __ push(rdx); // store long on tos 1842 __ push(rax); 1843 __ fild_d(at_rsp()); // load long to ST0 1844 __ d2ieee(); // truncate to double size 1845 __ pop(rcx); // adjust rsp 1846 __ pop(rcx); 1847 if (UseSSE >= 2) { 1848 __ push_d(); 1849 __ pop_d(xmm0); 1850 } 1851 break; 1852 case Bytecodes::_f2i: 1853 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs 1854 // as it returns 0 for any NaN. 1855 if (UseSSE >= 1) { 1856 __ push_f(xmm0); 1857 } else { 1858 __ push(rcx); // reserve space for argument 1859 __ fstp_s(at_rsp()); // pass float argument on stack 1860 } 1861 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1862 break; 1863 case Bytecodes::_f2l: 1864 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs 1865 // as it returns 0 for any NaN. 1866 if (UseSSE >= 1) { 1867 __ push_f(xmm0); 1868 } else { 1869 __ push(rcx); // reserve space for argument 1870 __ fstp_s(at_rsp()); // pass float argument on stack 1871 } 1872 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1873 break; 1874 case Bytecodes::_f2d: 1875 if (UseSSE < 1) { 1876 /* nothing to do */ 1877 } else if (UseSSE == 1) { 1878 __ push_f(xmm0); 1879 __ pop_f(); 1880 } else { // UseSSE >= 2 1881 __ cvtss2sd(xmm0, xmm0); 1882 } 1883 break; 1884 case Bytecodes::_d2i: 1885 if (UseSSE >= 2) { 1886 __ push_d(xmm0); 1887 } else { 1888 __ push(rcx); // reserve space for argument 1889 __ push(rcx); 1890 __ fstp_d(at_rsp()); // pass double argument on stack 1891 } 1892 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2); 1893 break; 1894 case Bytecodes::_d2l: 1895 if (UseSSE >= 2) { 1896 __ push_d(xmm0); 1897 } else { 1898 __ push(rcx); // reserve space for argument 1899 __ push(rcx); 1900 __ fstp_d(at_rsp()); // pass double argument on stack 1901 } 1902 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2); 1903 break; 1904 case Bytecodes::_d2f: 1905 if (UseSSE <= 1) { 1906 __ push(rcx); // reserve space for f2ieee() 1907 __ f2ieee(); // truncate to float size 1908 __ pop(rcx); // adjust rsp 1909 if (UseSSE == 1) { 1910 // The cvtsd2ss instruction is not available if UseSSE==1, therefore 1911 // the conversion is performed using the FPU in this case. 1912 __ push_f(); 1913 __ pop_f(xmm0); 1914 } 1915 } else { // UseSSE >= 2 1916 __ cvtsd2ss(xmm0, xmm0); 1917 } 1918 break; 1919 default : 1920 ShouldNotReachHere(); 1921 } 1922 #endif 1923 } 1924 1925 void TemplateTable::lcmp() { 1926 transition(ltos, itos); 1927 #ifdef _LP64 1928 Label done; 1929 __ pop_l(rdx); 1930 __ cmpq(rdx, rax); 1931 __ movl(rax, -1); 1932 __ jccb(Assembler::less, done); 1933 __ setb(Assembler::notEqual, rax); 1934 __ movzbl(rax, rax); 1935 __ bind(done); 1936 #else 1937 1938 // y = rdx:rax 1939 __ pop_l(rbx, rcx); // get x = rcx:rbx 1940 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y) 1941 __ mov(rax, rcx); 1942 #endif 1943 } 1944 1945 void TemplateTable::float_cmp(bool is_float, int unordered_result) { 1946 if ((is_float && UseSSE >= 1) || 1947 (!is_float && UseSSE >= 2)) { 1948 Label done; 1949 if (is_float) { 1950 // XXX get rid of pop here, use ... reg, mem32 1951 __ pop_f(xmm1); 1952 __ ucomiss(xmm1, xmm0); 1953 } else { 1954 // XXX get rid of pop here, use ... reg, mem64 1955 __ pop_d(xmm1); 1956 __ ucomisd(xmm1, xmm0); 1957 } 1958 if (unordered_result < 0) { 1959 __ movl(rax, -1); 1960 __ jccb(Assembler::parity, done); 1961 __ jccb(Assembler::below, done); 1962 __ setb(Assembler::notEqual, rdx); 1963 __ movzbl(rax, rdx); 1964 } else { 1965 __ movl(rax, 1); 1966 __ jccb(Assembler::parity, done); 1967 __ jccb(Assembler::above, done); 1968 __ movl(rax, 0); 1969 __ jccb(Assembler::equal, done); 1970 __ decrementl(rax); 1971 } 1972 __ bind(done); 1973 } else { 1974 #ifdef _LP64 1975 ShouldNotReachHere(); 1976 #else 1977 if (is_float) { 1978 __ fld_s(at_rsp()); 1979 } else { 1980 __ fld_d(at_rsp()); 1981 __ pop(rdx); 1982 } 1983 __ pop(rcx); 1984 __ fcmp2int(rax, unordered_result < 0); 1985 #endif // _LP64 1986 } 1987 } 1988 1989 void TemplateTable::branch(bool is_jsr, bool is_wide) { 1990 __ get_method(rcx); // rcx holds method 1991 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx 1992 // holds bumped taken count 1993 1994 const ByteSize be_offset = MethodCounters::backedge_counter_offset() + 1995 InvocationCounter::counter_offset(); 1996 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() + 1997 InvocationCounter::counter_offset(); 1998 1999 // Load up edx with the branch displacement 2000 if (is_wide) { 2001 __ movl(rdx, at_bcp(1)); 2002 } else { 2003 __ load_signed_short(rdx, at_bcp(1)); 2004 } 2005 __ bswapl(rdx); 2006 2007 if (!is_wide) { 2008 __ sarl(rdx, 16); 2009 } 2010 LP64_ONLY(__ movl2ptr(rdx, rdx)); 2011 2012 // Handle all the JSR stuff here, then exit. 2013 // It's much shorter and cleaner than intermingling with the non-JSR 2014 // normal-branch stuff occurring below. 2015 if (is_jsr) { 2016 // Pre-load the next target bytecode into rbx 2017 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0)); 2018 2019 // compute return address as bci in rax 2020 __ lea(rax, at_bcp((is_wide ? 5 : 3) - 2021 in_bytes(ConstMethod::codes_offset()))); 2022 __ subptr(rax, Address(rcx, Method::const_offset())); 2023 // Adjust the bcp in r13 by the displacement in rdx 2024 __ addptr(rbcp, rdx); 2025 // jsr returns atos that is not an oop 2026 __ push_i(rax); 2027 __ dispatch_only(vtos); 2028 return; 2029 } 2030 2031 // Normal (non-jsr) branch handling 2032 2033 // Adjust the bcp in r13 by the displacement in rdx 2034 __ addptr(rbcp, rdx); 2035 2036 assert(UseLoopCounter || !UseOnStackReplacement, 2037 "on-stack-replacement requires loop counters"); 2038 Label backedge_counter_overflow; 2039 Label profile_method; 2040 Label dispatch; 2041 if (UseLoopCounter) { 2042 // increment backedge counter for backward branches 2043 // rax: MDO 2044 // rbx: MDO bumped taken-count 2045 // rcx: method 2046 // rdx: target offset 2047 // r13: target bcp 2048 // r14: locals pointer 2049 __ testl(rdx, rdx); // check if forward or backward branch 2050 __ jcc(Assembler::positive, dispatch); // count only if backward branch 2051 2052 // check if MethodCounters exists 2053 Label has_counters; 2054 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 2055 __ testptr(rax, rax); 2056 __ jcc(Assembler::notZero, has_counters); 2057 __ push(rdx); 2058 __ push(rcx); 2059 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), 2060 rcx); 2061 __ pop(rcx); 2062 __ pop(rdx); 2063 __ movptr(rax, Address(rcx, Method::method_counters_offset())); 2064 __ testptr(rax, rax); 2065 __ jcc(Assembler::zero, dispatch); 2066 __ bind(has_counters); 2067 2068 if (TieredCompilation) { 2069 Label no_mdo; 2070 int increment = InvocationCounter::count_increment; 2071 if (ProfileInterpreter) { 2072 // Are we profiling? 2073 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset()))); 2074 __ testptr(rbx, rbx); 2075 __ jccb(Assembler::zero, no_mdo); 2076 // Increment the MDO backedge counter 2077 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) + 2078 in_bytes(InvocationCounter::counter_offset())); 2079 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset())); 2080 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 2081 rax, false, Assembler::zero, &backedge_counter_overflow); 2082 __ jmp(dispatch); 2083 } 2084 __ bind(no_mdo); 2085 // Increment backedge counter in MethodCounters* 2086 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 2087 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset())); 2088 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask, 2089 rax, false, Assembler::zero, &backedge_counter_overflow); 2090 } else { // not TieredCompilation 2091 // increment counter 2092 __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 2093 __ movl(rax, Address(rcx, be_offset)); // load backedge counter 2094 __ incrementl(rax, InvocationCounter::count_increment); // increment counter 2095 __ movl(Address(rcx, be_offset), rax); // store counter 2096 2097 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter 2098 2099 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits 2100 __ addl(rax, Address(rcx, be_offset)); // add both counters 2101 2102 if (ProfileInterpreter) { 2103 // Test to see if we should create a method data oop 2104 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset()))); 2105 __ jcc(Assembler::less, dispatch); 2106 2107 // if no method data exists, go to profile method 2108 __ test_method_data_pointer(rax, profile_method); 2109 2110 if (UseOnStackReplacement) { 2111 // check for overflow against rbx which is the MDO taken count 2112 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 2113 __ jcc(Assembler::below, dispatch); 2114 2115 // When ProfileInterpreter is on, the backedge_count comes 2116 // from the MethodData*, which value does not get reset on 2117 // the call to frequency_counter_overflow(). To avoid 2118 // excessive calls to the overflow routine while the method is 2119 // being compiled, add a second test to make sure the overflow 2120 // function is called only once every overflow_frequency. 2121 const int overflow_frequency = 1024; 2122 __ andl(rbx, overflow_frequency - 1); 2123 __ jcc(Assembler::zero, backedge_counter_overflow); 2124 2125 } 2126 } else { 2127 if (UseOnStackReplacement) { 2128 // check for overflow against rax, which is the sum of the 2129 // counters 2130 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 2131 __ jcc(Assembler::aboveEqual, backedge_counter_overflow); 2132 2133 } 2134 } 2135 } 2136 __ bind(dispatch); 2137 } 2138 2139 // Pre-load the next target bytecode into rbx 2140 __ load_unsigned_byte(rbx, Address(rbcp, 0)); 2141 2142 // continue with the bytecode @ target 2143 // rax: return bci for jsr's, unused otherwise 2144 // rbx: target bytecode 2145 // r13: target bcp 2146 __ dispatch_only(vtos); 2147 2148 if (UseLoopCounter) { 2149 if (ProfileInterpreter) { 2150 // Out-of-line code to allocate method data oop. 2151 __ bind(profile_method); 2152 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 2153 __ set_method_data_pointer_for_bcp(); 2154 __ jmp(dispatch); 2155 } 2156 2157 if (UseOnStackReplacement) { 2158 // invocation counter overflow 2159 __ bind(backedge_counter_overflow); 2160 __ negptr(rdx); 2161 __ addptr(rdx, rbcp); // branch bcp 2162 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp) 2163 __ call_VM(noreg, 2164 CAST_FROM_FN_PTR(address, 2165 InterpreterRuntime::frequency_counter_overflow), 2166 rdx); 2167 2168 // rax: osr nmethod (osr ok) or NULL (osr not possible) 2169 // rdx: scratch 2170 // r14: locals pointer 2171 // r13: bcp 2172 __ testptr(rax, rax); // test result 2173 __ jcc(Assembler::zero, dispatch); // no osr if null 2174 // nmethod may have been invalidated (VM may block upon call_VM return) 2175 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use); 2176 __ jcc(Assembler::notEqual, dispatch); 2177 2178 // We have the address of an on stack replacement routine in rax. 2179 // In preparation of invoking it, first we must migrate the locals 2180 // and monitors from off the interpreter frame on the stack. 2181 // Ensure to save the osr nmethod over the migration call, 2182 // it will be preserved in rbx. 2183 __ mov(rbx, rax); 2184 2185 NOT_LP64(__ get_thread(rcx)); 2186 2187 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 2188 2189 // rax is OSR buffer, move it to expected parameter location 2190 LP64_ONLY(__ mov(j_rarg0, rax)); 2191 NOT_LP64(__ mov(rcx, rax)); 2192 // We use j_rarg definitions here so that registers don't conflict as parameter 2193 // registers change across platforms as we are in the midst of a calling 2194 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters. 2195 2196 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi); 2197 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx); 2198 2199 // pop the interpreter frame 2200 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 2201 __ leave(); // remove frame anchor 2202 __ pop(retaddr); // get return address 2203 __ mov(rsp, sender_sp); // set sp to sender sp 2204 // Ensure compiled code always sees stack at proper alignment 2205 __ andptr(rsp, -(StackAlignmentInBytes)); 2206 2207 // unlike x86 we need no specialized return from compiled code 2208 // to the interpreter or the call stub. 2209 2210 // push the return address 2211 __ push(retaddr); 2212 2213 // and begin the OSR nmethod 2214 __ jmp(Address(rbx, nmethod::osr_entry_point_offset())); 2215 } 2216 } 2217 } 2218 2219 void TemplateTable::if_0cmp(Condition cc) { 2220 transition(itos, vtos); 2221 // assume branch is more often taken than not (loops use backward branches) 2222 Label not_taken; 2223 __ testl(rax, rax); 2224 __ jcc(j_not(cc), not_taken); 2225 branch(false, false); 2226 __ bind(not_taken); 2227 __ profile_not_taken_branch(rax); 2228 } 2229 2230 void TemplateTable::if_icmp(Condition cc) { 2231 transition(itos, vtos); 2232 // assume branch is more often taken than not (loops use backward branches) 2233 Label not_taken; 2234 __ pop_i(rdx); 2235 __ cmpl(rdx, rax); 2236 __ jcc(j_not(cc), not_taken); 2237 branch(false, false); 2238 __ bind(not_taken); 2239 __ profile_not_taken_branch(rax); 2240 } 2241 2242 void TemplateTable::if_nullcmp(Condition cc) { 2243 transition(atos, vtos); 2244 // assume branch is more often taken than not (loops use backward branches) 2245 Label not_taken; 2246 __ testptr(rax, rax); 2247 __ jcc(j_not(cc), not_taken); 2248 branch(false, false); 2249 __ bind(not_taken); 2250 __ profile_not_taken_branch(rax); 2251 } 2252 2253 void TemplateTable::if_acmp(Condition cc) { 2254 transition(atos, vtos); 2255 // assume branch is more often taken than not (loops use backward branches) 2256 Label not_taken; 2257 __ pop_ptr(rdx); 2258 __ cmpptr(rdx, rax); 2259 __ jcc(j_not(cc), not_taken); 2260 branch(false, false); 2261 __ bind(not_taken); 2262 __ profile_not_taken_branch(rax); 2263 } 2264 2265 void TemplateTable::ret() { 2266 transition(vtos, vtos); 2267 locals_index(rbx); 2268 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp 2269 NOT_LP64(__ movptr(rbx, iaddress(rbx))); 2270 __ profile_ret(rbx, rcx); 2271 __ get_method(rax); 2272 __ movptr(rbcp, Address(rax, Method::const_offset())); 2273 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, 2274 ConstMethod::codes_offset())); 2275 __ dispatch_next(vtos); 2276 } 2277 2278 void TemplateTable::wide_ret() { 2279 transition(vtos, vtos); 2280 locals_index_wide(rbx); 2281 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp 2282 __ profile_ret(rbx, rcx); 2283 __ get_method(rax); 2284 __ movptr(rbcp, Address(rax, Method::const_offset())); 2285 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset())); 2286 __ dispatch_next(vtos); 2287 } 2288 2289 void TemplateTable::tableswitch() { 2290 Label default_case, continue_execution; 2291 transition(itos, vtos); 2292 2293 // align r13/rsi 2294 __ lea(rbx, at_bcp(BytesPerInt)); 2295 __ andptr(rbx, -BytesPerInt); 2296 // load lo & hi 2297 __ movl(rcx, Address(rbx, BytesPerInt)); 2298 __ movl(rdx, Address(rbx, 2 * BytesPerInt)); 2299 __ bswapl(rcx); 2300 __ bswapl(rdx); 2301 // check against lo & hi 2302 __ cmpl(rax, rcx); 2303 __ jcc(Assembler::less, default_case); 2304 __ cmpl(rax, rdx); 2305 __ jcc(Assembler::greater, default_case); 2306 // lookup dispatch offset 2307 __ subl(rax, rcx); 2308 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt)); 2309 __ profile_switch_case(rax, rbx, rcx); 2310 // continue execution 2311 __ bind(continue_execution); 2312 __ bswapl(rdx); 2313 LP64_ONLY(__ movl2ptr(rdx, rdx)); 2314 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1)); 2315 __ addptr(rbcp, rdx); 2316 __ dispatch_only(vtos); 2317 // handle default 2318 __ bind(default_case); 2319 __ profile_switch_default(rax); 2320 __ movl(rdx, Address(rbx, 0)); 2321 __ jmp(continue_execution); 2322 } 2323 2324 void TemplateTable::lookupswitch() { 2325 transition(itos, itos); 2326 __ stop("lookupswitch bytecode should have been rewritten"); 2327 } 2328 2329 void TemplateTable::fast_linearswitch() { 2330 transition(itos, vtos); 2331 Label loop_entry, loop, found, continue_execution; 2332 // bswap rax so we can avoid bswapping the table entries 2333 __ bswapl(rax); 2334 // align r13 2335 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of 2336 // this instruction (change offsets 2337 // below) 2338 __ andptr(rbx, -BytesPerInt); 2339 // set counter 2340 __ movl(rcx, Address(rbx, BytesPerInt)); 2341 __ bswapl(rcx); 2342 __ jmpb(loop_entry); 2343 // table search 2344 __ bind(loop); 2345 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt)); 2346 __ jcc(Assembler::equal, found); 2347 __ bind(loop_entry); 2348 __ decrementl(rcx); 2349 __ jcc(Assembler::greaterEqual, loop); 2350 // default case 2351 __ profile_switch_default(rax); 2352 __ movl(rdx, Address(rbx, 0)); 2353 __ jmp(continue_execution); 2354 // entry found -> get offset 2355 __ bind(found); 2356 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt)); 2357 __ profile_switch_case(rcx, rax, rbx); 2358 // continue execution 2359 __ bind(continue_execution); 2360 __ bswapl(rdx); 2361 __ movl2ptr(rdx, rdx); 2362 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1)); 2363 __ addptr(rbcp, rdx); 2364 __ dispatch_only(vtos); 2365 } 2366 2367 void TemplateTable::fast_binaryswitch() { 2368 transition(itos, vtos); 2369 // Implementation using the following core algorithm: 2370 // 2371 // int binary_search(int key, LookupswitchPair* array, int n) { 2372 // // Binary search according to "Methodik des Programmierens" by 2373 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 2374 // int i = 0; 2375 // int j = n; 2376 // while (i+1 < j) { 2377 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 2378 // // with Q: for all i: 0 <= i < n: key < a[i] 2379 // // where a stands for the array and assuming that the (inexisting) 2380 // // element a[n] is infinitely big. 2381 // int h = (i + j) >> 1; 2382 // // i < h < j 2383 // if (key < array[h].fast_match()) { 2384 // j = h; 2385 // } else { 2386 // i = h; 2387 // } 2388 // } 2389 // // R: a[i] <= key < a[i+1] or Q 2390 // // (i.e., if key is within array, i is the correct index) 2391 // return i; 2392 // } 2393 2394 // Register allocation 2395 const Register key = rax; // already set (tosca) 2396 const Register array = rbx; 2397 const Register i = rcx; 2398 const Register j = rdx; 2399 const Register h = rdi; 2400 const Register temp = rsi; 2401 2402 // Find array start 2403 NOT_LP64(__ save_bcp()); 2404 2405 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to 2406 // get rid of this 2407 // instruction (change 2408 // offsets below) 2409 __ andptr(array, -BytesPerInt); 2410 2411 // Initialize i & j 2412 __ xorl(i, i); // i = 0; 2413 __ movl(j, Address(array, -BytesPerInt)); // j = length(array); 2414 2415 // Convert j into native byteordering 2416 __ bswapl(j); 2417 2418 // And start 2419 Label entry; 2420 __ jmp(entry); 2421 2422 // binary search loop 2423 { 2424 Label loop; 2425 __ bind(loop); 2426 // int h = (i + j) >> 1; 2427 __ leal(h, Address(i, j, Address::times_1)); // h = i + j; 2428 __ sarl(h, 1); // h = (i + j) >> 1; 2429 // if (key < array[h].fast_match()) { 2430 // j = h; 2431 // } else { 2432 // i = h; 2433 // } 2434 // Convert array[h].match to native byte-ordering before compare 2435 __ movl(temp, Address(array, h, Address::times_8)); 2436 __ bswapl(temp); 2437 __ cmpl(key, temp); 2438 // j = h if (key < array[h].fast_match()) 2439 __ cmov32(Assembler::less, j, h); 2440 // i = h if (key >= array[h].fast_match()) 2441 __ cmov32(Assembler::greaterEqual, i, h); 2442 // while (i+1 < j) 2443 __ bind(entry); 2444 __ leal(h, Address(i, 1)); // i+1 2445 __ cmpl(h, j); // i+1 < j 2446 __ jcc(Assembler::less, loop); 2447 } 2448 2449 // end of binary search, result index is i (must check again!) 2450 Label default_case; 2451 // Convert array[i].match to native byte-ordering before compare 2452 __ movl(temp, Address(array, i, Address::times_8)); 2453 __ bswapl(temp); 2454 __ cmpl(key, temp); 2455 __ jcc(Assembler::notEqual, default_case); 2456 2457 // entry found -> j = offset 2458 __ movl(j , Address(array, i, Address::times_8, BytesPerInt)); 2459 __ profile_switch_case(i, key, array); 2460 __ bswapl(j); 2461 LP64_ONLY(__ movslq(j, j)); 2462 2463 NOT_LP64(__ restore_bcp()); 2464 NOT_LP64(__ restore_locals()); // restore rdi 2465 2466 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1)); 2467 __ addptr(rbcp, j); 2468 __ dispatch_only(vtos); 2469 2470 // default case -> j = default offset 2471 __ bind(default_case); 2472 __ profile_switch_default(i); 2473 __ movl(j, Address(array, -2 * BytesPerInt)); 2474 __ bswapl(j); 2475 LP64_ONLY(__ movslq(j, j)); 2476 2477 NOT_LP64(__ restore_bcp()); 2478 NOT_LP64(__ restore_locals()); 2479 2480 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1)); 2481 __ addptr(rbcp, j); 2482 __ dispatch_only(vtos); 2483 } 2484 2485 void TemplateTable::_return(TosState state) { 2486 transition(state, state); 2487 2488 assert(_desc->calls_vm(), 2489 "inconsistent calls_vm information"); // call in remove_activation 2490 2491 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 2492 assert(state == vtos, "only valid state"); 2493 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax); 2494 __ movptr(robj, aaddress(0)); 2495 __ load_klass(rdi, robj); 2496 __ movl(rdi, Address(rdi, Klass::access_flags_offset())); 2497 __ testl(rdi, JVM_ACC_HAS_FINALIZER); 2498 Label skip_register_finalizer; 2499 __ jcc(Assembler::zero, skip_register_finalizer); 2500 2501 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj); 2502 2503 __ bind(skip_register_finalizer); 2504 } 2505 2506 // Narrow result if state is itos but result type is smaller. 2507 // Need to narrow in the return bytecode rather than in generate_return_entry 2508 // since compiled code callers expect the result to already be narrowed. 2509 if (state == itos) { 2510 __ narrow(rax); 2511 } 2512 __ remove_activation(state, rbcp); 2513 2514 __ jmp(rbcp); 2515 } 2516 2517 // ---------------------------------------------------------------------------- 2518 // Volatile variables demand their effects be made known to all CPU's 2519 // in order. Store buffers on most chips allow reads & writes to 2520 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode 2521 // without some kind of memory barrier (i.e., it's not sufficient that 2522 // the interpreter does not reorder volatile references, the hardware 2523 // also must not reorder them). 2524 // 2525 // According to the new Java Memory Model (JMM): 2526 // (1) All volatiles are serialized wrt to each other. ALSO reads & 2527 // writes act as aquire & release, so: 2528 // (2) A read cannot let unrelated NON-volatile memory refs that 2529 // happen after the read float up to before the read. It's OK for 2530 // non-volatile memory refs that happen before the volatile read to 2531 // float down below it. 2532 // (3) Similar a volatile write cannot let unrelated NON-volatile 2533 // memory refs that happen BEFORE the write float down to after the 2534 // write. It's OK for non-volatile memory refs that happen after the 2535 // volatile write to float up before it. 2536 // 2537 // We only put in barriers around volatile refs (they are expensive), 2538 // not _between_ memory refs (that would require us to track the 2539 // flavor of the previous memory refs). Requirements (2) and (3) 2540 // require some barriers before volatile stores and after volatile 2541 // loads. These nearly cover requirement (1) but miss the 2542 // volatile-store-volatile-load case. This final case is placed after 2543 // volatile-stores although it could just as well go before 2544 // volatile-loads. 2545 2546 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) { 2547 // Helper function to insert a is-volatile test and memory barrier 2548 if(!os::is_MP()) return; // Not needed on single CPU 2549 __ membar(order_constraint); 2550 } 2551 2552 void TemplateTable::resolve_cache_and_index(int byte_no, 2553 Register Rcache, 2554 Register index, 2555 size_t index_size) { 2556 const Register temp = rbx; 2557 assert_different_registers(Rcache, index, temp); 2558 2559 Label resolved; 2560 2561 Bytecodes::Code code = bytecode(); 2562 switch (code) { 2563 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break; 2564 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break; 2565 } 2566 2567 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 2568 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size); 2569 __ cmpl(temp, code); // have we resolved this bytecode? 2570 __ jcc(Assembler::equal, resolved); 2571 2572 // resolve first time through 2573 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache); 2574 __ movl(temp, code); 2575 __ call_VM(noreg, entry, temp); 2576 // Update registers with resolved info 2577 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2578 __ bind(resolved); 2579 } 2580 2581 // The cache and index registers must be set before call 2582 void TemplateTable::load_field_cp_cache_entry(Register obj, 2583 Register cache, 2584 Register index, 2585 Register off, 2586 Register flags, 2587 bool is_static = false) { 2588 assert_different_registers(cache, index, flags, off); 2589 2590 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2591 // Field offset 2592 __ movptr(off, Address(cache, index, Address::times_ptr, 2593 in_bytes(cp_base_offset + 2594 ConstantPoolCacheEntry::f2_offset()))); 2595 // Flags 2596 __ movl(flags, Address(cache, index, Address::times_ptr, 2597 in_bytes(cp_base_offset + 2598 ConstantPoolCacheEntry::flags_offset()))); 2599 2600 // klass overwrite register 2601 if (is_static) { 2602 __ movptr(obj, Address(cache, index, Address::times_ptr, 2603 in_bytes(cp_base_offset + 2604 ConstantPoolCacheEntry::f1_offset()))); 2605 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 2606 __ movptr(obj, Address(obj, mirror_offset)); 2607 } 2608 } 2609 2610 void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 2611 Register method, 2612 Register itable_index, 2613 Register flags, 2614 bool is_invokevirtual, 2615 bool is_invokevfinal, /*unused*/ 2616 bool is_invokedynamic) { 2617 // setup registers 2618 const Register cache = rcx; 2619 const Register index = rdx; 2620 assert_different_registers(method, flags); 2621 assert_different_registers(method, cache, index); 2622 assert_different_registers(itable_index, flags); 2623 assert_different_registers(itable_index, cache, index); 2624 // determine constant pool cache field offsets 2625 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant"); 2626 const int method_offset = in_bytes( 2627 ConstantPoolCache::base_offset() + 2628 ((byte_no == f2_byte) 2629 ? ConstantPoolCacheEntry::f2_offset() 2630 : ConstantPoolCacheEntry::f1_offset())); 2631 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() + 2632 ConstantPoolCacheEntry::flags_offset()); 2633 // access constant pool cache fields 2634 const int index_offset = in_bytes(ConstantPoolCache::base_offset() + 2635 ConstantPoolCacheEntry::f2_offset()); 2636 2637 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2)); 2638 resolve_cache_and_index(byte_no, cache, index, index_size); 2639 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset)); 2640 2641 if (itable_index != noreg) { 2642 // pick up itable or appendix index from f2 also: 2643 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset)); 2644 } 2645 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset)); 2646 } 2647 2648 // The registers cache and index expected to be set before call. 2649 // Correct values of the cache and index registers are preserved. 2650 void TemplateTable::jvmti_post_field_access(Register cache, 2651 Register index, 2652 bool is_static, 2653 bool has_tos) { 2654 if (JvmtiExport::can_post_field_access()) { 2655 // Check to see if a field access watch has been set before we take 2656 // the time to call into the VM. 2657 Label L1; 2658 assert_different_registers(cache, index, rax); 2659 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2660 __ testl(rax,rax); 2661 __ jcc(Assembler::zero, L1); 2662 2663 // cache entry pointer 2664 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 2665 __ shll(index, LogBytesPerWord); 2666 __ addptr(cache, index); 2667 if (is_static) { 2668 __ xorptr(rax, rax); // NULL object reference 2669 } else { 2670 __ pop(atos); // Get the object 2671 __ verify_oop(rax); 2672 __ push(atos); // Restore stack state 2673 } 2674 // rax,: object pointer or NULL 2675 // cache: cache entry pointer 2676 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), 2677 rax, cache); 2678 __ get_cache_and_index_at_bcp(cache, index, 1); 2679 __ bind(L1); 2680 } 2681 } 2682 2683 void TemplateTable::pop_and_check_object(Register r) { 2684 __ pop_ptr(r); 2685 __ null_check(r); // for field access must check obj. 2686 __ verify_oop(r); 2687 } 2688 2689 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) { 2690 transition(vtos, vtos); 2691 2692 const Register cache = rcx; 2693 const Register index = rdx; 2694 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2695 const Register off = rbx; 2696 const Register flags = rax; 2697 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them 2698 2699 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2700 jvmti_post_field_access(cache, index, is_static, false); 2701 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2702 2703 if (!is_static) pop_and_check_object(obj); 2704 2705 const Address field(obj, off, Address::times_1, 0*wordSize); 2706 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize)); 2707 2708 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 2709 2710 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2711 // Make sure we don't need to mask edx after the above shift 2712 assert(btos == 0, "change code, btos != 0"); 2713 2714 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 2715 2716 __ jcc(Assembler::notZero, notByte); 2717 // btos 2718 __ load_signed_byte(rax, field); 2719 __ push(btos); 2720 // Rewrite bytecode to be faster 2721 if (!is_static && rc == may_rewrite) { 2722 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx); 2723 } 2724 __ jmp(Done); 2725 2726 __ bind(notByte); 2727 __ cmpl(flags, ztos); 2728 __ jcc(Assembler::notEqual, notBool); 2729 2730 // ztos (same code as btos) 2731 __ load_signed_byte(rax, field); 2732 __ push(ztos); 2733 // Rewrite bytecode to be faster 2734 if (!is_static && rc == may_rewrite) { 2735 // use btos rewriting, no truncating to t/f bit is needed for getfield. 2736 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx); 2737 } 2738 __ jmp(Done); 2739 2740 __ bind(notBool); 2741 __ cmpl(flags, atos); 2742 __ jcc(Assembler::notEqual, notObj); 2743 // atos 2744 __ load_heap_oop(rax, field); 2745 __ push(atos); 2746 if (!is_static && rc == may_rewrite) { 2747 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx); 2748 } 2749 __ jmp(Done); 2750 2751 __ bind(notObj); 2752 __ cmpl(flags, itos); 2753 __ jcc(Assembler::notEqual, notInt); 2754 // itos 2755 __ movl(rax, field); 2756 __ push(itos); 2757 // Rewrite bytecode to be faster 2758 if (!is_static && rc == may_rewrite) { 2759 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx); 2760 } 2761 __ jmp(Done); 2762 2763 __ bind(notInt); 2764 __ cmpl(flags, ctos); 2765 __ jcc(Assembler::notEqual, notChar); 2766 // ctos 2767 __ load_unsigned_short(rax, field); 2768 __ push(ctos); 2769 // Rewrite bytecode to be faster 2770 if (!is_static && rc == may_rewrite) { 2771 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx); 2772 } 2773 __ jmp(Done); 2774 2775 __ bind(notChar); 2776 __ cmpl(flags, stos); 2777 __ jcc(Assembler::notEqual, notShort); 2778 // stos 2779 __ load_signed_short(rax, field); 2780 __ push(stos); 2781 // Rewrite bytecode to be faster 2782 if (!is_static && rc == may_rewrite) { 2783 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx); 2784 } 2785 __ jmp(Done); 2786 2787 __ bind(notShort); 2788 __ cmpl(flags, ltos); 2789 __ jcc(Assembler::notEqual, notLong); 2790 // ltos 2791 2792 #ifndef _LP64 2793 // Generate code as if volatile. There just aren't enough registers to 2794 // save that information and this code is faster than the test. 2795 __ fild_d(field); // Must load atomically 2796 __ subptr(rsp,2*wordSize); // Make space for store 2797 __ fistp_d(Address(rsp,0)); 2798 __ pop(rax); 2799 __ pop(rdx); 2800 #else 2801 __ movq(rax, field); 2802 #endif 2803 2804 __ push(ltos); 2805 // Rewrite bytecode to be faster 2806 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx)); 2807 __ jmp(Done); 2808 2809 __ bind(notLong); 2810 __ cmpl(flags, ftos); 2811 __ jcc(Assembler::notEqual, notFloat); 2812 // ftos 2813 2814 __ load_float(field); 2815 __ push(ftos); 2816 // Rewrite bytecode to be faster 2817 if (!is_static && rc == may_rewrite) { 2818 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx); 2819 } 2820 __ jmp(Done); 2821 2822 __ bind(notFloat); 2823 #ifdef ASSERT 2824 __ cmpl(flags, dtos); 2825 __ jcc(Assembler::notEqual, notDouble); 2826 #endif 2827 // dtos 2828 __ load_double(field); 2829 __ push(dtos); 2830 // Rewrite bytecode to be faster 2831 if (!is_static && rc == may_rewrite) { 2832 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx); 2833 } 2834 #ifdef ASSERT 2835 __ jmp(Done); 2836 2837 2838 __ bind(notDouble); 2839 __ stop("Bad state"); 2840 #endif 2841 2842 __ bind(Done); 2843 // [jk] not needed currently 2844 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad | 2845 // Assembler::LoadStore)); 2846 } 2847 2848 void TemplateTable::getfield(int byte_no) { 2849 getfield_or_static(byte_no, false); 2850 } 2851 2852 void TemplateTable::nofast_getfield(int byte_no) { 2853 getfield_or_static(byte_no, false, may_not_rewrite); 2854 } 2855 2856 void TemplateTable::getstatic(int byte_no) { 2857 getfield_or_static(byte_no, true); 2858 } 2859 2860 2861 // The registers cache and index expected to be set before call. 2862 // The function may destroy various registers, just not the cache and index registers. 2863 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 2864 2865 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax); 2866 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2867 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2868 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx); 2869 2870 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2871 2872 if (JvmtiExport::can_post_field_modification()) { 2873 // Check to see if a field modification watch has been set before 2874 // we take the time to call into the VM. 2875 Label L1; 2876 assert_different_registers(cache, index, rax); 2877 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2878 __ testl(rax, rax); 2879 __ jcc(Assembler::zero, L1); 2880 2881 __ get_cache_and_index_at_bcp(robj, RDX, 1); 2882 2883 2884 if (is_static) { 2885 // Life is simple. Null out the object pointer. 2886 __ xorl(RBX, RBX); 2887 2888 } else { 2889 // Life is harder. The stack holds the value on top, followed by 2890 // the object. We don't know the size of the value, though; it 2891 // could be one or two words depending on its type. As a result, 2892 // we must find the type to determine where the object is. 2893 #ifndef _LP64 2894 Label two_word, valsize_known; 2895 #endif 2896 __ movl(RCX, Address(robj, RDX, 2897 Address::times_ptr, 2898 in_bytes(cp_base_offset + 2899 ConstantPoolCacheEntry::flags_offset()))); 2900 NOT_LP64(__ mov(rbx, rsp)); 2901 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift); 2902 2903 // Make sure we don't need to mask rcx after the above shift 2904 ConstantPoolCacheEntry::verify_tos_state_shift(); 2905 #ifdef _LP64 2906 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue 2907 __ cmpl(c_rarg3, ltos); 2908 __ cmovptr(Assembler::equal, 2909 c_rarg1, at_tos_p2()); // ltos (two word jvalue) 2910 __ cmpl(c_rarg3, dtos); 2911 __ cmovptr(Assembler::equal, 2912 c_rarg1, at_tos_p2()); // dtos (two word jvalue) 2913 #else 2914 __ cmpl(rcx, ltos); 2915 __ jccb(Assembler::equal, two_word); 2916 __ cmpl(rcx, dtos); 2917 __ jccb(Assembler::equal, two_word); 2918 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos) 2919 __ jmpb(valsize_known); 2920 2921 __ bind(two_word); 2922 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue 2923 2924 __ bind(valsize_known); 2925 // setup object pointer 2926 __ movptr(rbx, Address(rbx, 0)); 2927 #endif 2928 } 2929 // cache entry pointer 2930 __ addptr(robj, in_bytes(cp_base_offset)); 2931 __ shll(RDX, LogBytesPerWord); 2932 __ addptr(robj, RDX); 2933 // object (tos) 2934 __ mov(RCX, rsp); 2935 // c_rarg1: object pointer set up above (NULL if static) 2936 // c_rarg2: cache entry pointer 2937 // c_rarg3: jvalue object on the stack 2938 __ call_VM(noreg, 2939 CAST_FROM_FN_PTR(address, 2940 InterpreterRuntime::post_field_modification), 2941 RBX, robj, RCX); 2942 __ get_cache_and_index_at_bcp(cache, index, 1); 2943 __ bind(L1); 2944 } 2945 } 2946 2947 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) { 2948 transition(vtos, vtos); 2949 2950 const Register cache = rcx; 2951 const Register index = rdx; 2952 const Register obj = rcx; 2953 const Register off = rbx; 2954 const Register flags = rax; 2955 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 2956 2957 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2958 jvmti_post_field_mod(cache, index, is_static); 2959 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2960 2961 // [jk] not needed currently 2962 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore | 2963 // Assembler::StoreStore)); 2964 2965 Label notVolatile, Done; 2966 __ movl(rdx, flags); 2967 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 2968 __ andl(rdx, 0x1); 2969 2970 // field addresses 2971 const Address field(obj, off, Address::times_1, 0*wordSize); 2972 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);) 2973 2974 Label notByte, notBool, notInt, notShort, notChar, 2975 notLong, notFloat, notObj, notDouble; 2976 2977 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 2978 2979 assert(btos == 0, "change code, btos != 0"); 2980 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 2981 __ jcc(Assembler::notZero, notByte); 2982 2983 // btos 2984 { 2985 __ pop(btos); 2986 if (!is_static) pop_and_check_object(obj); 2987 __ movb(field, rax); 2988 if (!is_static && rc == may_rewrite) { 2989 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no); 2990 } 2991 __ jmp(Done); 2992 } 2993 2994 __ bind(notByte); 2995 __ cmpl(flags, ztos); 2996 __ jcc(Assembler::notEqual, notBool); 2997 2998 // ztos 2999 { 3000 __ pop(ztos); 3001 if (!is_static) pop_and_check_object(obj); 3002 __ andl(rax, 0x1); 3003 __ movb(field, rax); 3004 if (!is_static && rc == may_rewrite) { 3005 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no); 3006 } 3007 __ jmp(Done); 3008 } 3009 3010 __ bind(notBool); 3011 __ cmpl(flags, atos); 3012 __ jcc(Assembler::notEqual, notObj); 3013 3014 // atos 3015 { 3016 __ pop(atos); 3017 if (!is_static) pop_and_check_object(obj); 3018 // Store into the field 3019 do_oop_store(_masm, field, rax, ACCESS_ON_HEAP | ACCESS_ON_ARRAY); 3020 if (!is_static && rc == may_rewrite) { 3021 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no); 3022 } 3023 __ jmp(Done); 3024 } 3025 3026 __ bind(notObj); 3027 __ cmpl(flags, itos); 3028 __ jcc(Assembler::notEqual, notInt); 3029 3030 // itos 3031 { 3032 __ pop(itos); 3033 if (!is_static) pop_and_check_object(obj); 3034 __ movl(field, rax); 3035 if (!is_static && rc == may_rewrite) { 3036 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no); 3037 } 3038 __ jmp(Done); 3039 } 3040 3041 __ bind(notInt); 3042 __ cmpl(flags, ctos); 3043 __ jcc(Assembler::notEqual, notChar); 3044 3045 // ctos 3046 { 3047 __ pop(ctos); 3048 if (!is_static) pop_and_check_object(obj); 3049 __ movw(field, rax); 3050 if (!is_static && rc == may_rewrite) { 3051 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no); 3052 } 3053 __ jmp(Done); 3054 } 3055 3056 __ bind(notChar); 3057 __ cmpl(flags, stos); 3058 __ jcc(Assembler::notEqual, notShort); 3059 3060 // stos 3061 { 3062 __ pop(stos); 3063 if (!is_static) pop_and_check_object(obj); 3064 __ movw(field, rax); 3065 if (!is_static && rc == may_rewrite) { 3066 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no); 3067 } 3068 __ jmp(Done); 3069 } 3070 3071 __ bind(notShort); 3072 __ cmpl(flags, ltos); 3073 __ jcc(Assembler::notEqual, notLong); 3074 3075 // ltos 3076 #ifdef _LP64 3077 { 3078 __ pop(ltos); 3079 if (!is_static) pop_and_check_object(obj); 3080 __ movq(field, rax); 3081 if (!is_static && rc == may_rewrite) { 3082 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no); 3083 } 3084 __ jmp(Done); 3085 } 3086 #else 3087 { 3088 Label notVolatileLong; 3089 __ testl(rdx, rdx); 3090 __ jcc(Assembler::zero, notVolatileLong); 3091 3092 __ pop(ltos); // overwrites rdx, do this after testing volatile. 3093 if (!is_static) pop_and_check_object(obj); 3094 3095 // Replace with real volatile test 3096 __ push(rdx); 3097 __ push(rax); // Must update atomically with FIST 3098 __ fild_d(Address(rsp,0)); // So load into FPU register 3099 __ fistp_d(field); // and put into memory atomically 3100 __ addptr(rsp, 2*wordSize); 3101 // volatile_barrier(); 3102 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3103 Assembler::StoreStore)); 3104 // Don't rewrite volatile version 3105 __ jmp(notVolatile); 3106 3107 __ bind(notVolatileLong); 3108 3109 __ pop(ltos); // overwrites rdx 3110 if (!is_static) pop_and_check_object(obj); 3111 __ movptr(hi, rdx); 3112 __ movptr(field, rax); 3113 // Don't rewrite to _fast_lputfield for potential volatile case. 3114 __ jmp(notVolatile); 3115 } 3116 #endif // _LP64 3117 3118 __ bind(notLong); 3119 __ cmpl(flags, ftos); 3120 __ jcc(Assembler::notEqual, notFloat); 3121 3122 // ftos 3123 { 3124 __ pop(ftos); 3125 if (!is_static) pop_and_check_object(obj); 3126 __ store_float(field); 3127 if (!is_static && rc == may_rewrite) { 3128 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no); 3129 } 3130 __ jmp(Done); 3131 } 3132 3133 __ bind(notFloat); 3134 #ifdef ASSERT 3135 __ cmpl(flags, dtos); 3136 __ jcc(Assembler::notEqual, notDouble); 3137 #endif 3138 3139 // dtos 3140 { 3141 __ pop(dtos); 3142 if (!is_static) pop_and_check_object(obj); 3143 __ store_double(field); 3144 if (!is_static && rc == may_rewrite) { 3145 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no); 3146 } 3147 } 3148 3149 #ifdef ASSERT 3150 __ jmp(Done); 3151 3152 __ bind(notDouble); 3153 __ stop("Bad state"); 3154 #endif 3155 3156 __ bind(Done); 3157 3158 // Check for volatile store 3159 __ testl(rdx, rdx); 3160 __ jcc(Assembler::zero, notVolatile); 3161 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3162 Assembler::StoreStore)); 3163 __ bind(notVolatile); 3164 } 3165 3166 void TemplateTable::putfield(int byte_no) { 3167 putfield_or_static(byte_no, false); 3168 } 3169 3170 void TemplateTable::nofast_putfield(int byte_no) { 3171 putfield_or_static(byte_no, false, may_not_rewrite); 3172 } 3173 3174 void TemplateTable::putstatic(int byte_no) { 3175 putfield_or_static(byte_no, true); 3176 } 3177 3178 void TemplateTable::jvmti_post_fast_field_mod() { 3179 3180 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 3181 3182 if (JvmtiExport::can_post_field_modification()) { 3183 // Check to see if a field modification watch has been set before 3184 // we take the time to call into the VM. 3185 Label L2; 3186 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 3187 __ testl(scratch, scratch); 3188 __ jcc(Assembler::zero, L2); 3189 __ pop_ptr(rbx); // copy the object pointer from tos 3190 __ verify_oop(rbx); 3191 __ push_ptr(rbx); // put the object pointer back on tos 3192 // Save tos values before call_VM() clobbers them. Since we have 3193 // to do it for every data type, we use the saved values as the 3194 // jvalue object. 3195 switch (bytecode()) { // load values into the jvalue object 3196 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break; 3197 case Bytecodes::_fast_bputfield: // fall through 3198 case Bytecodes::_fast_zputfield: // fall through 3199 case Bytecodes::_fast_sputfield: // fall through 3200 case Bytecodes::_fast_cputfield: // fall through 3201 case Bytecodes::_fast_iputfield: __ push_i(rax); break; 3202 case Bytecodes::_fast_dputfield: __ push(dtos); break; 3203 case Bytecodes::_fast_fputfield: __ push(ftos); break; 3204 case Bytecodes::_fast_lputfield: __ push_l(rax); break; 3205 3206 default: 3207 ShouldNotReachHere(); 3208 } 3209 __ mov(scratch, rsp); // points to jvalue on the stack 3210 // access constant pool cache entry 3211 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1)); 3212 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1)); 3213 __ verify_oop(rbx); 3214 // rbx: object pointer copied above 3215 // c_rarg2: cache entry pointer 3216 // c_rarg3: jvalue object on the stack 3217 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3)); 3218 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx)); 3219 3220 switch (bytecode()) { // restore tos values 3221 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break; 3222 case Bytecodes::_fast_bputfield: // fall through 3223 case Bytecodes::_fast_zputfield: // fall through 3224 case Bytecodes::_fast_sputfield: // fall through 3225 case Bytecodes::_fast_cputfield: // fall through 3226 case Bytecodes::_fast_iputfield: __ pop_i(rax); break; 3227 case Bytecodes::_fast_dputfield: __ pop(dtos); break; 3228 case Bytecodes::_fast_fputfield: __ pop(ftos); break; 3229 case Bytecodes::_fast_lputfield: __ pop_l(rax); break; 3230 } 3231 __ bind(L2); 3232 } 3233 } 3234 3235 void TemplateTable::fast_storefield(TosState state) { 3236 transition(state, vtos); 3237 3238 ByteSize base = ConstantPoolCache::base_offset(); 3239 3240 jvmti_post_fast_field_mod(); 3241 3242 // access constant pool cache 3243 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 3244 3245 // test for volatile with rdx but rdx is tos register for lputfield. 3246 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, 3247 in_bytes(base + 3248 ConstantPoolCacheEntry::flags_offset()))); 3249 3250 // replace index with field offset from cache entry 3251 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, 3252 in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 3253 3254 // [jk] not needed currently 3255 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore | 3256 // Assembler::StoreStore)); 3257 3258 Label notVolatile; 3259 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3260 __ andl(rdx, 0x1); 3261 3262 // Get object from stack 3263 pop_and_check_object(rcx); 3264 3265 // field address 3266 const Address field(rcx, rbx, Address::times_1); 3267 3268 // access field 3269 switch (bytecode()) { 3270 case Bytecodes::_fast_aputfield: 3271 do_oop_store(_masm, field, rax, ACCESS_ON_HEAP | ACCESS_ON_ARRAY); 3272 break; 3273 case Bytecodes::_fast_lputfield: 3274 #ifdef _LP64 3275 __ movq(field, rax); 3276 #else 3277 __ stop("should not be rewritten"); 3278 #endif 3279 break; 3280 case Bytecodes::_fast_iputfield: 3281 __ movl(field, rax); 3282 break; 3283 case Bytecodes::_fast_zputfield: 3284 __ andl(rax, 0x1); // boolean is true if LSB is 1 3285 // fall through to bputfield 3286 case Bytecodes::_fast_bputfield: 3287 __ movb(field, rax); 3288 break; 3289 case Bytecodes::_fast_sputfield: 3290 // fall through 3291 case Bytecodes::_fast_cputfield: 3292 __ movw(field, rax); 3293 break; 3294 case Bytecodes::_fast_fputfield: 3295 __ store_float(field); 3296 break; 3297 case Bytecodes::_fast_dputfield: 3298 __ store_double(field); 3299 break; 3300 default: 3301 ShouldNotReachHere(); 3302 } 3303 3304 // Check for volatile store 3305 __ testl(rdx, rdx); 3306 __ jcc(Assembler::zero, notVolatile); 3307 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 3308 Assembler::StoreStore)); 3309 __ bind(notVolatile); 3310 } 3311 3312 void TemplateTable::fast_accessfield(TosState state) { 3313 transition(atos, state); 3314 3315 // Do the JVMTI work here to avoid disturbing the register state below 3316 if (JvmtiExport::can_post_field_access()) { 3317 // Check to see if a field access watch has been set before we 3318 // take the time to call into the VM. 3319 Label L1; 3320 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 3321 __ testl(rcx, rcx); 3322 __ jcc(Assembler::zero, L1); 3323 // access constant pool cache entry 3324 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1)); 3325 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1)); 3326 __ verify_oop(rax); 3327 __ push_ptr(rax); // save object pointer before call_VM() clobbers it 3328 LP64_ONLY(__ mov(c_rarg1, rax)); 3329 // c_rarg1: object pointer copied above 3330 // c_rarg2: cache entry pointer 3331 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2)); 3332 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx)); 3333 __ pop_ptr(rax); // restore object pointer 3334 __ bind(L1); 3335 } 3336 3337 // access constant pool cache 3338 __ get_cache_and_index_at_bcp(rcx, rbx, 1); 3339 // replace index with field offset from cache entry 3340 // [jk] not needed currently 3341 // if (os::is_MP()) { 3342 // __ movl(rdx, Address(rcx, rbx, Address::times_8, 3343 // in_bytes(ConstantPoolCache::base_offset() + 3344 // ConstantPoolCacheEntry::flags_offset()))); 3345 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3346 // __ andl(rdx, 0x1); 3347 // } 3348 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, 3349 in_bytes(ConstantPoolCache::base_offset() + 3350 ConstantPoolCacheEntry::f2_offset()))); 3351 3352 // rax: object 3353 __ verify_oop(rax); 3354 __ null_check(rax); 3355 Address field(rax, rbx, Address::times_1); 3356 3357 // access field 3358 switch (bytecode()) { 3359 case Bytecodes::_fast_agetfield: 3360 __ load_heap_oop(rax, field); 3361 __ verify_oop(rax); 3362 break; 3363 case Bytecodes::_fast_lgetfield: 3364 #ifdef _LP64 3365 __ movq(rax, field); 3366 #else 3367 __ stop("should not be rewritten"); 3368 #endif 3369 break; 3370 case Bytecodes::_fast_igetfield: 3371 __ movl(rax, field); 3372 break; 3373 case Bytecodes::_fast_bgetfield: 3374 __ movsbl(rax, field); 3375 break; 3376 case Bytecodes::_fast_sgetfield: 3377 __ load_signed_short(rax, field); 3378 break; 3379 case Bytecodes::_fast_cgetfield: 3380 __ load_unsigned_short(rax, field); 3381 break; 3382 case Bytecodes::_fast_fgetfield: 3383 __ load_float(field); 3384 break; 3385 case Bytecodes::_fast_dgetfield: 3386 __ load_double(field); 3387 break; 3388 default: 3389 ShouldNotReachHere(); 3390 } 3391 // [jk] not needed currently 3392 // if (os::is_MP()) { 3393 // Label notVolatile; 3394 // __ testl(rdx, rdx); 3395 // __ jcc(Assembler::zero, notVolatile); 3396 // __ membar(Assembler::LoadLoad); 3397 // __ bind(notVolatile); 3398 //}; 3399 } 3400 3401 void TemplateTable::fast_xaccess(TosState state) { 3402 transition(vtos, state); 3403 3404 // get receiver 3405 __ movptr(rax, aaddress(0)); 3406 // access constant pool cache 3407 __ get_cache_and_index_at_bcp(rcx, rdx, 2); 3408 __ movptr(rbx, 3409 Address(rcx, rdx, Address::times_ptr, 3410 in_bytes(ConstantPoolCache::base_offset() + 3411 ConstantPoolCacheEntry::f2_offset()))); 3412 // make sure exception is reported in correct bcp range (getfield is 3413 // next instruction) 3414 __ increment(rbcp); 3415 __ null_check(rax); 3416 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize); 3417 switch (state) { 3418 case itos: 3419 __ movl(rax, field); 3420 break; 3421 case atos: 3422 __ load_heap_oop(rax, field); 3423 __ verify_oop(rax); 3424 break; 3425 case ftos: 3426 __ load_float(field); 3427 break; 3428 default: 3429 ShouldNotReachHere(); 3430 } 3431 3432 // [jk] not needed currently 3433 // if (os::is_MP()) { 3434 // Label notVolatile; 3435 // __ movl(rdx, Address(rcx, rdx, Address::times_8, 3436 // in_bytes(ConstantPoolCache::base_offset() + 3437 // ConstantPoolCacheEntry::flags_offset()))); 3438 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 3439 // __ testl(rdx, 0x1); 3440 // __ jcc(Assembler::zero, notVolatile); 3441 // __ membar(Assembler::LoadLoad); 3442 // __ bind(notVolatile); 3443 // } 3444 3445 __ decrement(rbcp); 3446 } 3447 3448 //----------------------------------------------------------------------------- 3449 // Calls 3450 3451 void TemplateTable::count_calls(Register method, Register temp) { 3452 // implemented elsewhere 3453 ShouldNotReachHere(); 3454 } 3455 3456 void TemplateTable::prepare_invoke(int byte_no, 3457 Register method, // linked method (or i-klass) 3458 Register index, // itable index, MethodType, etc. 3459 Register recv, // if caller wants to see it 3460 Register flags // if caller wants to test it 3461 ) { 3462 // determine flags 3463 const Bytecodes::Code code = bytecode(); 3464 const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 3465 const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 3466 const bool is_invokehandle = code == Bytecodes::_invokehandle; 3467 const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 3468 const bool is_invokespecial = code == Bytecodes::_invokespecial; 3469 const bool load_receiver = (recv != noreg); 3470 const bool save_flags = (flags != noreg); 3471 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), ""); 3472 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal"); 3473 assert(flags == noreg || flags == rdx, ""); 3474 assert(recv == noreg || recv == rcx, ""); 3475 3476 // setup registers & access constant pool cache 3477 if (recv == noreg) recv = rcx; 3478 if (flags == noreg) flags = rdx; 3479 assert_different_registers(method, index, recv, flags); 3480 3481 // save 'interpreter return address' 3482 __ save_bcp(); 3483 3484 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 3485 3486 // maybe push appendix to arguments (just before return address) 3487 if (is_invokedynamic || is_invokehandle) { 3488 Label L_no_push; 3489 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift)); 3490 __ jcc(Assembler::zero, L_no_push); 3491 // Push the appendix as a trailing parameter. 3492 // This must be done before we get the receiver, 3493 // since the parameter_size includes it. 3494 __ push(rbx); 3495 __ mov(rbx, index); 3496 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0"); 3497 __ load_resolved_reference_at_index(index, rbx); 3498 __ pop(rbx); 3499 __ push(index); // push appendix (MethodType, CallSite, etc.) 3500 __ bind(L_no_push); 3501 } 3502 3503 // load receiver if needed (after appendix is pushed so parameter size is correct) 3504 // Note: no return address pushed yet 3505 if (load_receiver) { 3506 __ movl(recv, flags); 3507 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask); 3508 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address 3509 const int receiver_is_at_end = -1; // back off one slot to get receiver 3510 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end); 3511 __ movptr(recv, recv_addr); 3512 __ verify_oop(recv); 3513 } 3514 3515 if (save_flags) { 3516 __ movl(rbcp, flags); 3517 } 3518 3519 // compute return type 3520 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 3521 // Make sure we don't need to mask flags after the above shift 3522 ConstantPoolCacheEntry::verify_tos_state_shift(); 3523 // load return address 3524 { 3525 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code); 3526 ExternalAddress table(table_addr); 3527 LP64_ONLY(__ lea(rscratch1, table)); 3528 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr))); 3529 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)))); 3530 } 3531 3532 // push return address 3533 __ push(flags); 3534 3535 // Restore flags value from the constant pool cache, and restore rsi 3536 // for later null checks. r13 is the bytecode pointer 3537 if (save_flags) { 3538 __ movl(flags, rbcp); 3539 __ restore_bcp(); 3540 } 3541 } 3542 3543 void TemplateTable::invokevirtual_helper(Register index, 3544 Register recv, 3545 Register flags) { 3546 // Uses temporary registers rax, rdx 3547 assert_different_registers(index, recv, rax, rdx); 3548 assert(index == rbx, ""); 3549 assert(recv == rcx, ""); 3550 3551 // Test for an invoke of a final method 3552 Label notFinal; 3553 __ movl(rax, flags); 3554 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift)); 3555 __ jcc(Assembler::zero, notFinal); 3556 3557 const Register method = index; // method must be rbx 3558 assert(method == rbx, 3559 "Method* must be rbx for interpreter calling convention"); 3560 3561 // do the call - the index is actually the method to call 3562 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method* 3563 3564 // It's final, need a null check here! 3565 __ null_check(recv); 3566 3567 // profile this call 3568 __ profile_final_call(rax); 3569 __ profile_arguments_type(rax, method, rbcp, true); 3570 3571 __ jump_from_interpreted(method, rax); 3572 3573 __ bind(notFinal); 3574 3575 // get receiver klass 3576 __ null_check(recv, oopDesc::klass_offset_in_bytes()); 3577 __ load_klass(rax, recv); 3578 3579 // profile this call 3580 __ profile_virtual_call(rax, rlocals, rdx); 3581 // get target Method* & entry point 3582 __ lookup_virtual_method(rax, index, method); 3583 __ profile_called_method(method, rdx, rbcp); 3584 3585 __ profile_arguments_type(rdx, method, rbcp, true); 3586 __ jump_from_interpreted(method, rdx); 3587 } 3588 3589 void TemplateTable::invokevirtual(int byte_no) { 3590 transition(vtos, vtos); 3591 assert(byte_no == f2_byte, "use this argument"); 3592 prepare_invoke(byte_no, 3593 rbx, // method or vtable index 3594 noreg, // unused itable index 3595 rcx, rdx); // recv, flags 3596 3597 // rbx: index 3598 // rcx: receiver 3599 // rdx: flags 3600 3601 invokevirtual_helper(rbx, rcx, rdx); 3602 } 3603 3604 void TemplateTable::invokespecial(int byte_no) { 3605 transition(vtos, vtos); 3606 assert(byte_no == f1_byte, "use this argument"); 3607 prepare_invoke(byte_no, rbx, noreg, // get f1 Method* 3608 rcx); // get receiver also for null check 3609 __ verify_oop(rcx); 3610 __ null_check(rcx); 3611 // do the call 3612 __ profile_call(rax); 3613 __ profile_arguments_type(rax, rbx, rbcp, false); 3614 __ jump_from_interpreted(rbx, rax); 3615 } 3616 3617 void TemplateTable::invokestatic(int byte_no) { 3618 transition(vtos, vtos); 3619 assert(byte_no == f1_byte, "use this argument"); 3620 prepare_invoke(byte_no, rbx); // get f1 Method* 3621 // do the call 3622 __ profile_call(rax); 3623 __ profile_arguments_type(rax, rbx, rbcp, false); 3624 __ jump_from_interpreted(rbx, rax); 3625 } 3626 3627 3628 void TemplateTable::fast_invokevfinal(int byte_no) { 3629 transition(vtos, vtos); 3630 assert(byte_no == f2_byte, "use this argument"); 3631 __ stop("fast_invokevfinal not used on x86"); 3632 } 3633 3634 3635 void TemplateTable::invokeinterface(int byte_no) { 3636 transition(vtos, vtos); 3637 assert(byte_no == f1_byte, "use this argument"); 3638 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index 3639 rcx, rdx); // recv, flags 3640 3641 // rax: interface klass (from f1) 3642 // rbx: itable index (from f2) 3643 // rcx: receiver 3644 // rdx: flags 3645 3646 // Special case of invokeinterface called for virtual method of 3647 // java.lang.Object. See cpCacheOop.cpp for details. 3648 // This code isn't produced by javac, but could be produced by 3649 // another compliant java compiler. 3650 Label notMethod; 3651 __ movl(rlocals, rdx); 3652 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift)); 3653 3654 __ jcc(Assembler::zero, notMethod); 3655 3656 invokevirtual_helper(rbx, rcx, rdx); 3657 __ bind(notMethod); 3658 3659 // Get receiver klass into rdx - also a null check 3660 __ restore_locals(); // restore r14 3661 __ null_check(rcx, oopDesc::klass_offset_in_bytes()); 3662 __ load_klass(rdx, rcx); 3663 3664 // profile this call 3665 __ profile_virtual_call(rdx, rbcp, rlocals); 3666 3667 Label no_such_interface, no_such_method; 3668 3669 __ lookup_interface_method(// inputs: rec. class, interface, itable index 3670 rdx, rax, rbx, 3671 // outputs: method, scan temp. reg 3672 rbx, rbcp, 3673 no_such_interface); 3674 3675 // rbx: Method* to call 3676 // rcx: receiver 3677 // Check for abstract method error 3678 // Note: This should be done more efficiently via a throw_abstract_method_error 3679 // interpreter entry point and a conditional jump to it in case of a null 3680 // method. 3681 __ testptr(rbx, rbx); 3682 __ jcc(Assembler::zero, no_such_method); 3683 3684 __ profile_called_method(rbx, rbcp, rdx); 3685 __ profile_arguments_type(rdx, rbx, rbcp, true); 3686 3687 // do the call 3688 // rcx: receiver 3689 // rbx,: Method* 3690 __ jump_from_interpreted(rbx, rdx); 3691 __ should_not_reach_here(); 3692 3693 // exception handling code follows... 3694 // note: must restore interpreter registers to canonical 3695 // state for exception handling to work correctly! 3696 3697 __ bind(no_such_method); 3698 // throw exception 3699 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3700 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed) 3701 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3702 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 3703 // the call_VM checks for exception, so we should never return here. 3704 __ should_not_reach_here(); 3705 3706 __ bind(no_such_interface); 3707 // throw exception 3708 __ pop(rbx); // pop return address (pushed by prepare_invoke) 3709 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed) 3710 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3711 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3712 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3713 // the call_VM checks for exception, so we should never return here. 3714 __ should_not_reach_here(); 3715 } 3716 3717 void TemplateTable::invokehandle(int byte_no) { 3718 transition(vtos, vtos); 3719 assert(byte_no == f1_byte, "use this argument"); 3720 const Register rbx_method = rbx; 3721 const Register rax_mtype = rax; 3722 const Register rcx_recv = rcx; 3723 const Register rdx_flags = rdx; 3724 3725 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv); 3726 __ verify_method_ptr(rbx_method); 3727 __ verify_oop(rcx_recv); 3728 __ null_check(rcx_recv); 3729 3730 // rax: MethodType object (from cpool->resolved_references[f1], if necessary) 3731 // rbx: MH.invokeExact_MT method (from f2) 3732 3733 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke 3734 3735 // FIXME: profile the LambdaForm also 3736 __ profile_final_call(rax); 3737 __ profile_arguments_type(rdx, rbx_method, rbcp, true); 3738 3739 __ jump_from_interpreted(rbx_method, rdx); 3740 } 3741 3742 void TemplateTable::invokedynamic(int byte_no) { 3743 transition(vtos, vtos); 3744 assert(byte_no == f1_byte, "use this argument"); 3745 3746 const Register rbx_method = rbx; 3747 const Register rax_callsite = rax; 3748 3749 prepare_invoke(byte_no, rbx_method, rax_callsite); 3750 3751 // rax: CallSite object (from cpool->resolved_references[f1]) 3752 // rbx: MH.linkToCallSite method (from f2) 3753 3754 // Note: rax_callsite is already pushed by prepare_invoke 3755 3756 // %%% should make a type profile for any invokedynamic that takes a ref argument 3757 // profile this call 3758 __ profile_call(rbcp); 3759 __ profile_arguments_type(rdx, rbx_method, rbcp, false); 3760 3761 __ verify_oop(rax_callsite); 3762 3763 __ jump_from_interpreted(rbx_method, rdx); 3764 } 3765 3766 //----------------------------------------------------------------------------- 3767 // Allocation 3768 3769 void TemplateTable::_new() { 3770 transition(vtos, atos); 3771 __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 3772 Label slow_case; 3773 Label slow_case_no_pop; 3774 Label done; 3775 Label initialize_header; 3776 Label initialize_object; // including clearing the fields 3777 Label allocate_shared; 3778 3779 __ get_cpool_and_tags(rcx, rax); 3780 3781 // Make sure the class we're about to instantiate has been resolved. 3782 // This is done before loading InstanceKlass to be consistent with the order 3783 // how Constant Pool is updated (see ConstantPool::klass_at_put) 3784 const int tags_offset = Array<u1>::base_offset_in_bytes(); 3785 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class); 3786 __ jcc(Assembler::notEqual, slow_case_no_pop); 3787 3788 // get InstanceKlass 3789 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool))); 3790 __ push(rcx); // save the contexts of klass for initializing the header 3791 3792 // make sure klass is initialized & doesn't have finalizer 3793 // make sure klass is fully initialized 3794 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized); 3795 __ jcc(Assembler::notEqual, slow_case); 3796 3797 // get instance_size in InstanceKlass (scaled to a count of bytes) 3798 __ movl(rdx, Address(rcx, Klass::layout_helper_offset())); 3799 // test to see if it has a finalizer or is malformed in some way 3800 __ testl(rdx, Klass::_lh_instance_slow_path_bit); 3801 __ jcc(Assembler::notZero, slow_case); 3802 3803 // 3804 // Allocate the instance 3805 // 1) Try to allocate in the TLAB 3806 // 2) if fail and the object is large allocate in the shared Eden 3807 // 3) if the above fails (or is not applicable), go to a slow case 3808 // (creates a new TLAB, etc.) 3809 3810 const bool allow_shared_alloc = 3811 Universe::heap()->supports_inline_contig_alloc(); 3812 3813 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 3814 #ifndef _LP64 3815 if (UseTLAB || allow_shared_alloc) { 3816 __ get_thread(thread); 3817 } 3818 #endif // _LP64 3819 3820 if (UseTLAB) { 3821 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset()))); 3822 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3823 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset()))); 3824 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case); 3825 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx); 3826 if (ZeroTLAB) { 3827 // the fields have been already cleared 3828 __ jmp(initialize_header); 3829 } else { 3830 // initialize both the header and fields 3831 __ jmp(initialize_object); 3832 } 3833 } 3834 3835 // Allocation in the shared Eden, if allowed. 3836 // 3837 // rdx: instance size in bytes 3838 if (allow_shared_alloc) { 3839 __ bind(allocate_shared); 3840 3841 ExternalAddress heap_top((address)Universe::heap()->top_addr()); 3842 ExternalAddress heap_end((address)Universe::heap()->end_addr()); 3843 3844 Label retry; 3845 __ bind(retry); 3846 __ movptr(rax, heap_top); 3847 __ lea(rbx, Address(rax, rdx, Address::times_1)); 3848 __ cmpptr(rbx, heap_end); 3849 __ jcc(Assembler::above, slow_case); 3850 3851 // Compare rax, with the top addr, and if still equal, store the new 3852 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was 3853 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs. 3854 // 3855 // rax,: object begin 3856 // rbx,: object end 3857 // rdx: instance size in bytes 3858 __ locked_cmpxchgptr(rbx, heap_top); 3859 3860 // if someone beat us on the allocation, try again, otherwise continue 3861 __ jcc(Assembler::notEqual, retry); 3862 3863 __ incr_allocated_bytes(thread, rdx, 0); 3864 } 3865 3866 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 3867 // The object is initialized before the header. If the object size is 3868 // zero, go directly to the header initialization. 3869 __ bind(initialize_object); 3870 __ decrement(rdx, sizeof(oopDesc)); 3871 __ jcc(Assembler::zero, initialize_header); 3872 3873 // Initialize topmost object field, divide rdx by 8, check if odd and 3874 // test if zero. 3875 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) 3876 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd 3877 3878 // rdx must have been multiple of 8 3879 #ifdef ASSERT 3880 // make sure rdx was multiple of 8 3881 Label L; 3882 // Ignore partial flag stall after shrl() since it is debug VM 3883 __ jccb(Assembler::carryClear, L); 3884 __ stop("object size is not multiple of 2 - adjust this code"); 3885 __ bind(L); 3886 // rdx must be > 0, no extra check needed here 3887 #endif 3888 3889 // initialize remaining object fields: rdx was a multiple of 8 3890 { Label loop; 3891 __ bind(loop); 3892 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx); 3893 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx)); 3894 __ decrement(rdx); 3895 __ jcc(Assembler::notZero, loop); 3896 } 3897 3898 // initialize object header only. 3899 __ bind(initialize_header); 3900 if (UseBiasedLocking) { 3901 __ pop(rcx); // get saved klass back in the register. 3902 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset())); 3903 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx); 3904 } else { 3905 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), 3906 (intptr_t)markOopDesc::prototype()); // header 3907 __ pop(rcx); // get saved klass back in the register. 3908 } 3909 #ifdef _LP64 3910 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code) 3911 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops 3912 #endif 3913 __ store_klass(rax, rcx); // klass 3914 3915 { 3916 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0); 3917 // Trigger dtrace event for fastpath 3918 __ push(atos); 3919 __ call_VM_leaf( 3920 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); 3921 __ pop(atos); 3922 } 3923 3924 __ jmp(done); 3925 } 3926 3927 // slow case 3928 __ bind(slow_case); 3929 __ pop(rcx); // restore stack pointer to what it was when we came in. 3930 __ bind(slow_case_no_pop); 3931 3932 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax); 3933 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 3934 3935 __ get_constant_pool(rarg1); 3936 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1); 3937 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2); 3938 __ verify_oop(rax); 3939 3940 // continue 3941 __ bind(done); 3942 } 3943 3944 void TemplateTable::newarray() { 3945 transition(itos, atos); 3946 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 3947 __ load_unsigned_byte(rarg1, at_bcp(1)); 3948 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), 3949 rarg1, rax); 3950 } 3951 3952 void TemplateTable::anewarray() { 3953 transition(itos, atos); 3954 3955 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 3956 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 3957 3958 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1); 3959 __ get_constant_pool(rarg1); 3960 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), 3961 rarg1, rarg2, rax); 3962 } 3963 3964 void TemplateTable::arraylength() { 3965 transition(atos, itos); 3966 __ null_check(rax, arrayOopDesc::length_offset_in_bytes()); 3967 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes())); 3968 } 3969 3970 void TemplateTable::checkcast() { 3971 transition(atos, atos); 3972 Label done, is_null, ok_is_subtype, quicked, resolved; 3973 __ testptr(rax, rax); // object is in rax 3974 __ jcc(Assembler::zero, is_null); 3975 3976 // Get cpool & tags index 3977 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array 3978 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index 3979 // See if bytecode has already been quicked 3980 __ cmpb(Address(rdx, rbx, 3981 Address::times_1, 3982 Array<u1>::base_offset_in_bytes()), 3983 JVM_CONSTANT_Class); 3984 __ jcc(Assembler::equal, quicked); 3985 __ push(atos); // save receiver for result, and for GC 3986 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 3987 3988 // vm_result_2 has metadata result 3989 #ifndef _LP64 3990 // borrow rdi from locals 3991 __ get_thread(rdi); 3992 __ get_vm_result_2(rax, rdi); 3993 __ restore_locals(); 3994 #else 3995 __ get_vm_result_2(rax, r15_thread); 3996 #endif 3997 3998 __ pop_ptr(rdx); // restore receiver 3999 __ jmpb(resolved); 4000 4001 // Get superklass in rax and subklass in rbx 4002 __ bind(quicked); 4003 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check 4004 __ movptr(rax, Address(rcx, rbx, 4005 Address::times_ptr, sizeof(ConstantPool))); 4006 4007 __ bind(resolved); 4008 __ load_klass(rbx, rdx); 4009 4010 // Generate subtype check. Blows rcx, rdi. Object in rdx. 4011 // Superklass in rax. Subklass in rbx. 4012 __ gen_subtype_check(rbx, ok_is_subtype); 4013 4014 // Come here on failure 4015 __ push_ptr(rdx); 4016 // object is at TOS 4017 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry)); 4018 4019 // Come here on success 4020 __ bind(ok_is_subtype); 4021 __ mov(rax, rdx); // Restore object in rdx 4022 4023 // Collect counts on whether this check-cast sees NULLs a lot or not. 4024 if (ProfileInterpreter) { 4025 __ jmp(done); 4026 __ bind(is_null); 4027 __ profile_null_seen(rcx); 4028 } else { 4029 __ bind(is_null); // same as 'done' 4030 } 4031 __ bind(done); 4032 } 4033 4034 void TemplateTable::instanceof() { 4035 transition(atos, itos); 4036 Label done, is_null, ok_is_subtype, quicked, resolved; 4037 __ testptr(rax, rax); 4038 __ jcc(Assembler::zero, is_null); 4039 4040 // Get cpool & tags index 4041 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array 4042 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index 4043 // See if bytecode has already been quicked 4044 __ cmpb(Address(rdx, rbx, 4045 Address::times_1, 4046 Array<u1>::base_offset_in_bytes()), 4047 JVM_CONSTANT_Class); 4048 __ jcc(Assembler::equal, quicked); 4049 4050 __ push(atos); // save receiver for result, and for GC 4051 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 4052 // vm_result_2 has metadata result 4053 4054 #ifndef _LP64 4055 // borrow rdi from locals 4056 __ get_thread(rdi); 4057 __ get_vm_result_2(rax, rdi); 4058 __ restore_locals(); 4059 #else 4060 __ get_vm_result_2(rax, r15_thread); 4061 #endif 4062 4063 __ pop_ptr(rdx); // restore receiver 4064 __ verify_oop(rdx); 4065 __ load_klass(rdx, rdx); 4066 __ jmpb(resolved); 4067 4068 // Get superklass in rax and subklass in rdx 4069 __ bind(quicked); 4070 __ load_klass(rdx, rax); 4071 __ movptr(rax, Address(rcx, rbx, 4072 Address::times_ptr, sizeof(ConstantPool))); 4073 4074 __ bind(resolved); 4075 4076 // Generate subtype check. Blows rcx, rdi 4077 // Superklass in rax. Subklass in rdx. 4078 __ gen_subtype_check(rdx, ok_is_subtype); 4079 4080 // Come here on failure 4081 __ xorl(rax, rax); 4082 __ jmpb(done); 4083 // Come here on success 4084 __ bind(ok_is_subtype); 4085 __ movl(rax, 1); 4086 4087 // Collect counts on whether this test sees NULLs a lot or not. 4088 if (ProfileInterpreter) { 4089 __ jmp(done); 4090 __ bind(is_null); 4091 __ profile_null_seen(rcx); 4092 } else { 4093 __ bind(is_null); // same as 'done' 4094 } 4095 __ bind(done); 4096 // rax = 0: obj == NULL or obj is not an instanceof the specified klass 4097 // rax = 1: obj != NULL and obj is an instanceof the specified klass 4098 } 4099 4100 4101 //---------------------------------------------------------------------------------------------------- 4102 // Breakpoints 4103 void TemplateTable::_breakpoint() { 4104 // Note: We get here even if we are single stepping.. 4105 // jbug insists on setting breakpoints at every bytecode 4106 // even if we are in single step mode. 4107 4108 transition(vtos, vtos); 4109 4110 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 4111 4112 // get the unpatched byte code 4113 __ get_method(rarg); 4114 __ call_VM(noreg, 4115 CAST_FROM_FN_PTR(address, 4116 InterpreterRuntime::get_original_bytecode_at), 4117 rarg, rbcp); 4118 __ mov(rbx, rax); // why? 4119 4120 // post the breakpoint event 4121 __ get_method(rarg); 4122 __ call_VM(noreg, 4123 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), 4124 rarg, rbcp); 4125 4126 // complete the execution of original bytecode 4127 __ dispatch_only_normal(vtos); 4128 } 4129 4130 //----------------------------------------------------------------------------- 4131 // Exceptions 4132 4133 void TemplateTable::athrow() { 4134 transition(atos, vtos); 4135 __ null_check(rax); 4136 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 4137 } 4138 4139 //----------------------------------------------------------------------------- 4140 // Synchronization 4141 // 4142 // Note: monitorenter & exit are symmetric routines; which is reflected 4143 // in the assembly code structure as well 4144 // 4145 // Stack layout: 4146 // 4147 // [expressions ] <--- rsp = expression stack top 4148 // .. 4149 // [expressions ] 4150 // [monitor entry] <--- monitor block top = expression stack bot 4151 // .. 4152 // [monitor entry] 4153 // [frame data ] <--- monitor block bot 4154 // ... 4155 // [saved rbp ] <--- rbp 4156 void TemplateTable::monitorenter() { 4157 transition(atos, vtos); 4158 4159 // check for NULL object 4160 __ null_check(rax); 4161 4162 const Address monitor_block_top( 4163 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 4164 const Address monitor_block_bot( 4165 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 4166 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 4167 4168 Label allocated; 4169 4170 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 4171 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx); 4172 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 4173 4174 // initialize entry pointer 4175 __ xorl(rmon, rmon); // points to free slot or NULL 4176 4177 // find a free slot in the monitor block (result in rmon) 4178 { 4179 Label entry, loop, exit; 4180 __ movptr(rtop, monitor_block_top); // points to current entry, 4181 // starting with top-most entry 4182 __ lea(rbot, monitor_block_bot); // points to word before bottom 4183 // of monitor block 4184 __ jmpb(entry); 4185 4186 __ bind(loop); 4187 // check if current entry is used 4188 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD); 4189 // if not used then remember entry in rmon 4190 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr 4191 // check if current entry is for same object 4192 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes())); 4193 // if same object then stop searching 4194 __ jccb(Assembler::equal, exit); 4195 // otherwise advance to next entry 4196 __ addptr(rtop, entry_size); 4197 __ bind(entry); 4198 // check if bottom reached 4199 __ cmpptr(rtop, rbot); 4200 // if not at bottom then check this entry 4201 __ jcc(Assembler::notEqual, loop); 4202 __ bind(exit); 4203 } 4204 4205 __ testptr(rmon, rmon); // check if a slot has been found 4206 __ jcc(Assembler::notZero, allocated); // if found, continue with that one 4207 4208 // allocate one if there's no free slot 4209 { 4210 Label entry, loop; 4211 // 1. compute new pointers // rsp: old expression stack top 4212 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom 4213 __ subptr(rsp, entry_size); // move expression stack top 4214 __ subptr(rmon, entry_size); // move expression stack bottom 4215 __ mov(rtop, rsp); // set start value for copy loop 4216 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom 4217 __ jmp(entry); 4218 // 2. move expression stack contents 4219 __ bind(loop); 4220 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack 4221 // word from old location 4222 __ movptr(Address(rtop, 0), rbot); // and store it at new location 4223 __ addptr(rtop, wordSize); // advance to next word 4224 __ bind(entry); 4225 __ cmpptr(rtop, rmon); // check if bottom reached 4226 __ jcc(Assembler::notEqual, loop); // if not at bottom then 4227 // copy next word 4228 } 4229 4230 // call run-time routine 4231 // rmon: points to monitor entry 4232 __ bind(allocated); 4233 4234 // Increment bcp to point to the next bytecode, so exception 4235 // handling for async. exceptions work correctly. 4236 // The object has already been poped from the stack, so the 4237 // expression stack looks correct. 4238 __ increment(rbcp); 4239 4240 // store object 4241 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax); 4242 __ lock_object(rmon); 4243 4244 // check to make sure this monitor doesn't cause stack overflow after locking 4245 __ save_bcp(); // in case of exception 4246 __ generate_stack_overflow_check(0); 4247 4248 // The bcp has already been incremented. Just need to dispatch to 4249 // next instruction. 4250 __ dispatch_next(vtos); 4251 } 4252 4253 void TemplateTable::monitorexit() { 4254 transition(atos, vtos); 4255 4256 // check for NULL object 4257 __ null_check(rax); 4258 4259 const Address monitor_block_top( 4260 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 4261 const Address monitor_block_bot( 4262 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 4263 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 4264 4265 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 4266 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx); 4267 4268 Label found; 4269 4270 // find matching slot 4271 { 4272 Label entry, loop; 4273 __ movptr(rtop, monitor_block_top); // points to current entry, 4274 // starting with top-most entry 4275 __ lea(rbot, monitor_block_bot); // points to word before bottom 4276 // of monitor block 4277 __ jmpb(entry); 4278 4279 __ bind(loop); 4280 // check if current entry is for same object 4281 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes())); 4282 // if same object then stop searching 4283 __ jcc(Assembler::equal, found); 4284 // otherwise advance to next entry 4285 __ addptr(rtop, entry_size); 4286 __ bind(entry); 4287 // check if bottom reached 4288 __ cmpptr(rtop, rbot); 4289 // if not at bottom then check this entry 4290 __ jcc(Assembler::notEqual, loop); 4291 } 4292 4293 // error handling. Unlocking was not block-structured 4294 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 4295 InterpreterRuntime::throw_illegal_monitor_state_exception)); 4296 __ should_not_reach_here(); 4297 4298 // call run-time routine 4299 __ bind(found); 4300 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps) 4301 __ unlock_object(rtop); 4302 __ pop_ptr(rax); // discard object 4303 } 4304 4305 // Wide instructions 4306 void TemplateTable::wide() { 4307 transition(vtos, vtos); 4308 __ load_unsigned_byte(rbx, at_bcp(1)); 4309 ExternalAddress wtable((address)Interpreter::_wentry_point); 4310 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr))); 4311 // Note: the rbcp increment step is part of the individual wide bytecode implementations 4312 } 4313 4314 // Multi arrays 4315 void TemplateTable::multianewarray() { 4316 transition(vtos, atos); 4317 4318 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax); 4319 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions 4320 // last dim is on top of stack; we want address of first one: 4321 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize 4322 // the latter wordSize to point to the beginning of the array. 4323 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize)); 4324 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg); 4325 __ load_unsigned_byte(rbx, at_bcp(3)); 4326 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts 4327 }