1 /*
   2  * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "interpreter/templateTable.hpp"
  30 #include "memory/universe.inline.hpp"
  31 #include "oops/methodData.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "oops/oop.inline.hpp"
  34 #include "prims/methodHandles.hpp"
  35 #include "runtime/sharedRuntime.hpp"
  36 #include "runtime/stubRoutines.hpp"
  37 #include "runtime/synchronizer.hpp"
  38 #include "utilities/macros.hpp"
  39 
  40 #ifndef CC_INTERP
  41 #define __ _masm->
  42 
  43 //----------------------------------------------------------------------------------------------------
  44 // Platform-dependent initialization
  45 
  46 void TemplateTable::pd_initialize() {
  47   // No i486 specific initialization
  48 }
  49 
  50 //----------------------------------------------------------------------------------------------------
  51 // Address computation
  52 
  53 // local variables
  54 static inline Address iaddress(int n)            {
  55   return Address(rdi, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n)            { return iaddress(n + 1); }
  59 static inline Address haddress(int n)            { return iaddress(n + 0); }
  60 static inline Address faddress(int n)            { return iaddress(n); }
  61 static inline Address daddress(int n)            { return laddress(n); }
  62 static inline Address aaddress(int n)            { return iaddress(n); }
  63 
  64 static inline Address iaddress(Register r)       {
  65   return Address(rdi, r, Interpreter::stackElementScale());
  66 }
  67 static inline Address laddress(Register r)       {
  68   return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
  69 }
  70 static inline Address haddress(Register r)       {
  71   return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  72 }
  73 
  74 static inline Address faddress(Register r)       { return iaddress(r); }
  75 static inline Address daddress(Register r)       { return laddress(r); }
  76 static inline Address aaddress(Register r)       { return iaddress(r); }
  77 
  78 // expression stack
  79 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
  80 // data beyond the rsp which is potentially unsafe in an MT environment;
  81 // an interrupt may overwrite that data.)
  82 static inline Address at_rsp   () {
  83   return Address(rsp, 0);
  84 }
  85 
  86 // At top of Java expression stack which may be different than rsp().  It
  87 // isn't for category 1 objects.
  88 static inline Address at_tos   () {
  89   Address tos = Address(rsp,  Interpreter::expr_offset_in_bytes(0));
  90   return tos;
  91 }
  92 
  93 static inline Address at_tos_p1() {
  94   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
  95 }
  96 
  97 static inline Address at_tos_p2() {
  98   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
  99 }
 100 
 101 // Condition conversion
 102 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 103   switch (cc) {
 104     case TemplateTable::equal        : return Assembler::notEqual;
 105     case TemplateTable::not_equal    : return Assembler::equal;
 106     case TemplateTable::less         : return Assembler::greaterEqual;
 107     case TemplateTable::less_equal   : return Assembler::greater;
 108     case TemplateTable::greater      : return Assembler::lessEqual;
 109     case TemplateTable::greater_equal: return Assembler::less;
 110   }
 111   ShouldNotReachHere();
 112   return Assembler::zero;
 113 }
 114 
 115 
 116 //----------------------------------------------------------------------------------------------------
 117 // Miscelaneous helper routines
 118 
 119 // Store an oop (or NULL) at the address described by obj.
 120 // If val == noreg this means store a NULL
 121 
 122 static void do_oop_store(InterpreterMacroAssembler* _masm,
 123                          Address obj,
 124                          Register val,
 125                          BarrierSet::Name barrier,
 126                          bool precise) {
 127   assert(val == noreg || val == rax, "parameter is just for looks");
 128   switch (barrier) {
 129 #if INCLUDE_ALL_GCS
 130     case BarrierSet::G1SATBCT:
 131     case BarrierSet::G1SATBCTLogging:
 132       {
 133         // flatten object address if needed
 134         // We do it regardless of precise because we need the registers
 135         if (obj.index() == noreg && obj.disp() == 0) {
 136           if (obj.base() != rdx) {
 137             __ movl(rdx, obj.base());
 138           }
 139         } else {
 140           __ leal(rdx, obj);
 141         }
 142         __ get_thread(rcx);
 143         __ save_bcp();
 144         __ g1_write_barrier_pre(rdx /* obj */,
 145                                 rbx /* pre_val */,
 146                                 rcx /* thread */,
 147                                 rsi /* tmp */,
 148                                 val != noreg /* tosca_live */,
 149                                 false /* expand_call */);
 150 
 151         // Do the actual store
 152         // noreg means NULL
 153         if (val == noreg) {
 154           __ movptr(Address(rdx, 0), NULL_WORD);
 155           // No post barrier for NULL
 156         } else {
 157           __ movl(Address(rdx, 0), val);
 158           __ g1_write_barrier_post(rdx /* store_adr */,
 159                                    val /* new_val */,
 160                                    rcx /* thread */,
 161                                    rbx /* tmp */,
 162                                    rsi /* tmp2 */);
 163         }
 164         __ restore_bcp();
 165 
 166       }
 167       break;
 168 #endif // INCLUDE_ALL_GCS
 169     case BarrierSet::CardTableModRef:
 170     case BarrierSet::CardTableExtension:
 171       {
 172         if (val == noreg) {
 173           __ movptr(obj, NULL_WORD);
 174         } else {
 175           __ movl(obj, val);
 176           // flatten object address if needed
 177           if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
 178             __ store_check(obj.base());
 179           } else {
 180             __ leal(rdx, obj);
 181             __ store_check(rdx);
 182           }
 183         }
 184       }
 185       break;
 186     case BarrierSet::ModRef:
 187     case BarrierSet::Epsilon:
 188     case BarrierSet::Other:
 189       if (val == noreg) {
 190         __ movptr(obj, NULL_WORD);
 191       } else {
 192         __ movl(obj, val);
 193       }
 194       break;
 195     default      :
 196       ShouldNotReachHere();
 197 
 198   }
 199 }
 200 
 201 Address TemplateTable::at_bcp(int offset) {
 202   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 203   return Address(rsi, offset);
 204 }
 205 
 206 
 207 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 208                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 209                                    int byte_no) {
 210   if (!RewriteBytecodes)  return;
 211   Label L_patch_done;
 212 
 213   switch (bc) {
 214   case Bytecodes::_fast_aputfield:
 215   case Bytecodes::_fast_bputfield:
 216   case Bytecodes::_fast_zputfield:
 217   case Bytecodes::_fast_cputfield:
 218   case Bytecodes::_fast_dputfield:
 219   case Bytecodes::_fast_fputfield:
 220   case Bytecodes::_fast_iputfield:
 221   case Bytecodes::_fast_lputfield:
 222   case Bytecodes::_fast_sputfield:
 223     {
 224       // We skip bytecode quickening for putfield instructions when
 225       // the put_code written to the constant pool cache is zero.
 226       // This is required so that every execution of this instruction
 227       // calls out to InterpreterRuntime::resolve_get_put to do
 228       // additional, required work.
 229       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 230       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 231       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
 232       __ movl(bc_reg, bc);
 233       __ cmpl(temp_reg, (int) 0);
 234       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 235     }
 236     break;
 237   default:
 238     assert(byte_no == -1, "sanity");
 239     // the pair bytecodes have already done the load.
 240     if (load_bc_into_bc_reg) {
 241       __ movl(bc_reg, bc);
 242     }
 243   }
 244 
 245   if (JvmtiExport::can_post_breakpoint()) {
 246     Label L_fast_patch;
 247     // if a breakpoint is present we can't rewrite the stream directly
 248     __ movzbl(temp_reg, at_bcp(0));
 249     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 250     __ jcc(Assembler::notEqual, L_fast_patch);
 251     __ get_method(temp_reg);
 252     // Let breakpoint table handling rewrite to quicker bytecode
 253     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rsi, bc_reg);
 254 #ifndef ASSERT
 255     __ jmpb(L_patch_done);
 256 #else
 257     __ jmp(L_patch_done);
 258 #endif
 259     __ bind(L_fast_patch);
 260   }
 261 
 262 #ifdef ASSERT
 263   Label L_okay;
 264   __ load_unsigned_byte(temp_reg, at_bcp(0));
 265   __ cmpl(temp_reg, (int)Bytecodes::java_code(bc));
 266   __ jccb(Assembler::equal, L_okay);
 267   __ cmpl(temp_reg, bc_reg);
 268   __ jcc(Assembler::equal, L_okay);
 269   __ stop("patching the wrong bytecode");
 270   __ bind(L_okay);
 271 #endif
 272 
 273   // patch bytecode
 274   __ movb(at_bcp(0), bc_reg);
 275   __ bind(L_patch_done);
 276 }
 277 
 278 //----------------------------------------------------------------------------------------------------
 279 // Individual instructions
 280 
 281 void TemplateTable::nop() {
 282   transition(vtos, vtos);
 283   // nothing to do
 284 }
 285 
 286 void TemplateTable::shouldnotreachhere() {
 287   transition(vtos, vtos);
 288   __ stop("shouldnotreachhere bytecode");
 289 }
 290 
 291 
 292 
 293 void TemplateTable::aconst_null() {
 294   transition(vtos, atos);
 295   __ xorptr(rax, rax);
 296 }
 297 
 298 
 299 void TemplateTable::iconst(int value) {
 300   transition(vtos, itos);
 301   if (value == 0) {
 302     __ xorptr(rax, rax);
 303   } else {
 304     __ movptr(rax, value);
 305   }
 306 }
 307 
 308 
 309 void TemplateTable::lconst(int value) {
 310   transition(vtos, ltos);
 311   if (value == 0) {
 312     __ xorptr(rax, rax);
 313   } else {
 314     __ movptr(rax, value);
 315   }
 316   assert(value >= 0, "check this code");
 317   __ xorptr(rdx, rdx);
 318 }
 319 
 320 
 321 void TemplateTable::fconst(int value) {
 322   transition(vtos, ftos);
 323          if (value == 0) { __ fldz();
 324   } else if (value == 1) { __ fld1();
 325   } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 326   } else                 { ShouldNotReachHere();
 327   }
 328 }
 329 
 330 
 331 void TemplateTable::dconst(int value) {
 332   transition(vtos, dtos);
 333          if (value == 0) { __ fldz();
 334   } else if (value == 1) { __ fld1();
 335   } else                 { ShouldNotReachHere();
 336   }
 337 }
 338 
 339 
 340 void TemplateTable::bipush() {
 341   transition(vtos, itos);
 342   __ load_signed_byte(rax, at_bcp(1));
 343 }
 344 
 345 
 346 void TemplateTable::sipush() {
 347   transition(vtos, itos);
 348   __ load_unsigned_short(rax, at_bcp(1));
 349   __ bswapl(rax);
 350   __ sarl(rax, 16);
 351 }
 352 
 353 void TemplateTable::ldc(bool wide) {
 354   transition(vtos, vtos);
 355   Label call_ldc, notFloat, notClass, Done;
 356 
 357   if (wide) {
 358     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 359   } else {
 360     __ load_unsigned_byte(rbx, at_bcp(1));
 361   }
 362   __ get_cpool_and_tags(rcx, rax);
 363   const int base_offset = ConstantPool::header_size() * wordSize;
 364   const int tags_offset = Array<u1>::base_offset_in_bytes();
 365 
 366   // get type
 367   __ xorptr(rdx, rdx);
 368   __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 369 
 370   // unresolved class - get the resolved class
 371   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 372   __ jccb(Assembler::equal, call_ldc);
 373 
 374   // unresolved class in error (resolution failed) - call into runtime
 375   // so that the same error from first resolution attempt is thrown.
 376   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 377   __ jccb(Assembler::equal, call_ldc);
 378 
 379   // resolved class - need to call vm to get java mirror of the class
 380   __ cmpl(rdx, JVM_CONSTANT_Class);
 381   __ jcc(Assembler::notEqual, notClass);
 382 
 383   __ bind(call_ldc);
 384   __ movl(rcx, wide);
 385   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
 386   __ push(atos);
 387   __ jmp(Done);
 388 
 389   __ bind(notClass);
 390   __ cmpl(rdx, JVM_CONSTANT_Float);
 391   __ jccb(Assembler::notEqual, notFloat);
 392   // ftos
 393   __ fld_s(    Address(rcx, rbx, Address::times_ptr, base_offset));
 394   __ push(ftos);
 395   __ jmp(Done);
 396 
 397   __ bind(notFloat);
 398 #ifdef ASSERT
 399   { Label L;
 400     __ cmpl(rdx, JVM_CONSTANT_Integer);
 401     __ jcc(Assembler::equal, L);
 402     // String and Object are rewritten to fast_aldc
 403     __ stop("unexpected tag type in ldc");
 404     __ bind(L);
 405   }
 406 #endif
 407   // itos JVM_CONSTANT_Integer only
 408   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 409   __ push(itos);
 410   __ bind(Done);
 411 }
 412 
 413 // Fast path for caching oop constants.
 414 void TemplateTable::fast_aldc(bool wide) {
 415   transition(vtos, atos);
 416 
 417   Register result = rax;
 418   Register tmp = rdx;
 419   int index_size = wide ? sizeof(u2) : sizeof(u1);
 420 
 421   Label resolved;
 422 
 423   // We are resolved if the resolved reference cache entry contains a
 424   // non-null object (String, MethodType, etc.)
 425   assert_different_registers(result, tmp);
 426   __ get_cache_index_at_bcp(tmp, 1, index_size);
 427   __ load_resolved_reference_at_index(result, tmp);
 428   __ testl(result, result);
 429   __ jcc(Assembler::notZero, resolved);
 430 
 431   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 432 
 433   // first time invocation - must resolve first
 434   __ movl(tmp, (int)bytecode());
 435   __ call_VM(result, entry, tmp);
 436 
 437   __ bind(resolved);
 438 
 439   if (VerifyOops) {
 440     __ verify_oop(result);
 441   }
 442 }
 443 
 444 void TemplateTable::ldc2_w() {
 445   transition(vtos, vtos);
 446   Label Long, Done;
 447   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 448 
 449   __ get_cpool_and_tags(rcx, rax);
 450   const int base_offset = ConstantPool::header_size() * wordSize;
 451   const int tags_offset = Array<u1>::base_offset_in_bytes();
 452 
 453   // get type
 454   __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
 455   __ jccb(Assembler::notEqual, Long);
 456   // dtos
 457   __ fld_d(    Address(rcx, rbx, Address::times_ptr, base_offset));
 458   __ push(dtos);
 459   __ jmpb(Done);
 460 
 461   __ bind(Long);
 462   // ltos
 463   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 464   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 465 
 466   __ push(ltos);
 467 
 468   __ bind(Done);
 469 }
 470 
 471 
 472 void TemplateTable::locals_index(Register reg, int offset) {
 473   __ load_unsigned_byte(reg, at_bcp(offset));
 474   __ negptr(reg);
 475 }
 476 
 477 
 478 void TemplateTable::iload() {
 479   transition(vtos, itos);
 480   if (RewriteFrequentPairs) {
 481     Label rewrite, done;
 482 
 483     // get next byte
 484     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 485     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 486     // last two iloads in a pair.  Comparing against fast_iload means that
 487     // the next bytecode is neither an iload or a caload, and therefore
 488     // an iload pair.
 489     __ cmpl(rbx, Bytecodes::_iload);
 490     __ jcc(Assembler::equal, done);
 491 
 492     __ cmpl(rbx, Bytecodes::_fast_iload);
 493     __ movl(rcx, Bytecodes::_fast_iload2);
 494     __ jccb(Assembler::equal, rewrite);
 495 
 496     // if _caload, rewrite to fast_icaload
 497     __ cmpl(rbx, Bytecodes::_caload);
 498     __ movl(rcx, Bytecodes::_fast_icaload);
 499     __ jccb(Assembler::equal, rewrite);
 500 
 501     // rewrite so iload doesn't check again.
 502     __ movl(rcx, Bytecodes::_fast_iload);
 503 
 504     // rewrite
 505     // rcx: fast bytecode
 506     __ bind(rewrite);
 507     patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
 508     __ bind(done);
 509   }
 510 
 511   // Get the local value into tos
 512   locals_index(rbx);
 513   __ movl(rax, iaddress(rbx));
 514 }
 515 
 516 
 517 void TemplateTable::fast_iload2() {
 518   transition(vtos, itos);
 519   locals_index(rbx);
 520   __ movl(rax, iaddress(rbx));
 521   __ push(itos);
 522   locals_index(rbx, 3);
 523   __ movl(rax, iaddress(rbx));
 524 }
 525 
 526 void TemplateTable::fast_iload() {
 527   transition(vtos, itos);
 528   locals_index(rbx);
 529   __ movl(rax, iaddress(rbx));
 530 }
 531 
 532 
 533 void TemplateTable::lload() {
 534   transition(vtos, ltos);
 535   locals_index(rbx);
 536   __ movptr(rax, laddress(rbx));
 537   NOT_LP64(__ movl(rdx, haddress(rbx)));
 538 }
 539 
 540 
 541 void TemplateTable::fload() {
 542   transition(vtos, ftos);
 543   locals_index(rbx);
 544   __ fld_s(faddress(rbx));
 545 }
 546 
 547 
 548 void TemplateTable::dload() {
 549   transition(vtos, dtos);
 550   locals_index(rbx);
 551   __ fld_d(daddress(rbx));
 552 }
 553 
 554 
 555 void TemplateTable::aload() {
 556   transition(vtos, atos);
 557   locals_index(rbx);
 558   __ movptr(rax, aaddress(rbx));
 559 }
 560 
 561 
 562 void TemplateTable::locals_index_wide(Register reg) {
 563   __ load_unsigned_short(reg, at_bcp(2));
 564   __ bswapl(reg);
 565   __ shrl(reg, 16);
 566   __ negptr(reg);
 567 }
 568 
 569 
 570 void TemplateTable::wide_iload() {
 571   transition(vtos, itos);
 572   locals_index_wide(rbx);
 573   __ movl(rax, iaddress(rbx));
 574 }
 575 
 576 
 577 void TemplateTable::wide_lload() {
 578   transition(vtos, ltos);
 579   locals_index_wide(rbx);
 580   __ movptr(rax, laddress(rbx));
 581   NOT_LP64(__ movl(rdx, haddress(rbx)));
 582 }
 583 
 584 
 585 void TemplateTable::wide_fload() {
 586   transition(vtos, ftos);
 587   locals_index_wide(rbx);
 588   __ fld_s(faddress(rbx));
 589 }
 590 
 591 
 592 void TemplateTable::wide_dload() {
 593   transition(vtos, dtos);
 594   locals_index_wide(rbx);
 595   __ fld_d(daddress(rbx));
 596 }
 597 
 598 
 599 void TemplateTable::wide_aload() {
 600   transition(vtos, atos);
 601   locals_index_wide(rbx);
 602   __ movptr(rax, aaddress(rbx));
 603 }
 604 
 605 void TemplateTable::index_check(Register array, Register index) {
 606   // Pop ptr into array
 607   __ pop_ptr(array);
 608   index_check_without_pop(array, index);
 609 }
 610 
 611 void TemplateTable::index_check_without_pop(Register array, Register index) {
 612   // destroys rbx,
 613   // check array
 614   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 615   LP64_ONLY(__ movslq(index, index));
 616   // check index
 617   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 618   if (index != rbx) {
 619     // ??? convention: move aberrant index into rbx, for exception message
 620     assert(rbx != array, "different registers");
 621     __ mov(rbx, index);
 622   }
 623   __ jump_cc(Assembler::aboveEqual,
 624              ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 625 }
 626 
 627 
 628 void TemplateTable::iaload() {
 629   transition(itos, itos);
 630   // rdx: array
 631   index_check(rdx, rax);  // kills rbx,
 632   // rax,: index
 633   __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
 634 }
 635 
 636 
 637 void TemplateTable::laload() {
 638   transition(itos, ltos);
 639   // rax,: index
 640   // rdx: array
 641   index_check(rdx, rax);
 642   __ mov(rbx, rax);
 643   // rbx,: index
 644   __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
 645   NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
 646 }
 647 
 648 
 649 void TemplateTable::faload() {
 650   transition(itos, ftos);
 651   // rdx: array
 652   index_check(rdx, rax);  // kills rbx,
 653   // rax,: index
 654   __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 655 }
 656 
 657 
 658 void TemplateTable::daload() {
 659   transition(itos, dtos);
 660   // rdx: array
 661   index_check(rdx, rax);  // kills rbx,
 662   // rax,: index
 663   __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 664 }
 665 
 666 
 667 void TemplateTable::aaload() {
 668   transition(itos, atos);
 669   // rdx: array
 670   index_check(rdx, rax);  // kills rbx,
 671   // rax,: index
 672   __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 673 }
 674 
 675 
 676 void TemplateTable::baload() {
 677   transition(itos, itos);
 678   // rdx: array
 679   index_check(rdx, rax);  // kills rbx,
 680   // rax,: index
 681   // can do better code for P5 - fix this at some point
 682   __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 683   __ mov(rax, rbx);
 684 }
 685 
 686 
 687 void TemplateTable::caload() {
 688   transition(itos, itos);
 689   // rdx: array
 690   index_check(rdx, rax);  // kills rbx,
 691   // rax,: index
 692   // can do better code for P5 - may want to improve this at some point
 693   __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 694   __ mov(rax, rbx);
 695 }
 696 
 697 // iload followed by caload frequent pair
 698 void TemplateTable::fast_icaload() {
 699   transition(vtos, itos);
 700   // load index out of locals
 701   locals_index(rbx);
 702   __ movl(rax, iaddress(rbx));
 703 
 704   // rdx: array
 705   index_check(rdx, rax);
 706   // rax,: index
 707   __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 708   __ mov(rax, rbx);
 709 }
 710 
 711 void TemplateTable::saload() {
 712   transition(itos, itos);
 713   // rdx: array
 714   index_check(rdx, rax);  // kills rbx,
 715   // rax,: index
 716   // can do better code for P5 - may want to improve this at some point
 717   __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 718   __ mov(rax, rbx);
 719 }
 720 
 721 
 722 void TemplateTable::iload(int n) {
 723   transition(vtos, itos);
 724   __ movl(rax, iaddress(n));
 725 }
 726 
 727 
 728 void TemplateTable::lload(int n) {
 729   transition(vtos, ltos);
 730   __ movptr(rax, laddress(n));
 731   NOT_LP64(__ movptr(rdx, haddress(n)));
 732 }
 733 
 734 
 735 void TemplateTable::fload(int n) {
 736   transition(vtos, ftos);
 737   __ fld_s(faddress(n));
 738 }
 739 
 740 
 741 void TemplateTable::dload(int n) {
 742   transition(vtos, dtos);
 743   __ fld_d(daddress(n));
 744 }
 745 
 746 
 747 void TemplateTable::aload(int n) {
 748   transition(vtos, atos);
 749   __ movptr(rax, aaddress(n));
 750 }
 751 
 752 
 753 void TemplateTable::aload_0() {
 754   transition(vtos, atos);
 755   // According to bytecode histograms, the pairs:
 756   //
 757   // _aload_0, _fast_igetfield
 758   // _aload_0, _fast_agetfield
 759   // _aload_0, _fast_fgetfield
 760   //
 761   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 762   // bytecode checks if the next bytecode is either _fast_igetfield,
 763   // _fast_agetfield or _fast_fgetfield and then rewrites the
 764   // current bytecode into a pair bytecode; otherwise it rewrites the current
 765   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 766   //
 767   // Note: If the next bytecode is _getfield, the rewrite must be delayed,
 768   //       otherwise we may miss an opportunity for a pair.
 769   //
 770   // Also rewrite frequent pairs
 771   //   aload_0, aload_1
 772   //   aload_0, iload_1
 773   // These bytecodes with a small amount of code are most profitable to rewrite
 774   if (RewriteFrequentPairs) {
 775     Label rewrite, done;
 776     // get next byte
 777     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 778 
 779     // do actual aload_0
 780     aload(0);
 781 
 782     // if _getfield then wait with rewrite
 783     __ cmpl(rbx, Bytecodes::_getfield);
 784     __ jcc(Assembler::equal, done);
 785 
 786     // if _igetfield then reqrite to _fast_iaccess_0
 787     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 788     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 789     __ movl(rcx, Bytecodes::_fast_iaccess_0);
 790     __ jccb(Assembler::equal, rewrite);
 791 
 792     // if _agetfield then reqrite to _fast_aaccess_0
 793     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 794     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 795     __ movl(rcx, Bytecodes::_fast_aaccess_0);
 796     __ jccb(Assembler::equal, rewrite);
 797 
 798     // if _fgetfield then reqrite to _fast_faccess_0
 799     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 800     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 801     __ movl(rcx, Bytecodes::_fast_faccess_0);
 802     __ jccb(Assembler::equal, rewrite);
 803 
 804     // else rewrite to _fast_aload0
 805     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 806     __ movl(rcx, Bytecodes::_fast_aload_0);
 807 
 808     // rewrite
 809     // rcx: fast bytecode
 810     __ bind(rewrite);
 811     patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
 812 
 813     __ bind(done);
 814   } else {
 815     aload(0);
 816   }
 817 }
 818 
 819 void TemplateTable::istore() {
 820   transition(itos, vtos);
 821   locals_index(rbx);
 822   __ movl(iaddress(rbx), rax);
 823 }
 824 
 825 
 826 void TemplateTable::lstore() {
 827   transition(ltos, vtos);
 828   locals_index(rbx);
 829   __ movptr(laddress(rbx), rax);
 830   NOT_LP64(__ movptr(haddress(rbx), rdx));
 831 }
 832 
 833 
 834 void TemplateTable::fstore() {
 835   transition(ftos, vtos);
 836   locals_index(rbx);
 837   __ fstp_s(faddress(rbx));
 838 }
 839 
 840 
 841 void TemplateTable::dstore() {
 842   transition(dtos, vtos);
 843   locals_index(rbx);
 844   __ fstp_d(daddress(rbx));
 845 }
 846 
 847 
 848 void TemplateTable::astore() {
 849   transition(vtos, vtos);
 850   __ pop_ptr(rax);
 851   locals_index(rbx);
 852   __ movptr(aaddress(rbx), rax);
 853 }
 854 
 855 
 856 void TemplateTable::wide_istore() {
 857   transition(vtos, vtos);
 858   __ pop_i(rax);
 859   locals_index_wide(rbx);
 860   __ movl(iaddress(rbx), rax);
 861 }
 862 
 863 
 864 void TemplateTable::wide_lstore() {
 865   transition(vtos, vtos);
 866   __ pop_l(rax, rdx);
 867   locals_index_wide(rbx);
 868   __ movptr(laddress(rbx), rax);
 869   NOT_LP64(__ movl(haddress(rbx), rdx));
 870 }
 871 
 872 
 873 void TemplateTable::wide_fstore() {
 874   wide_istore();
 875 }
 876 
 877 
 878 void TemplateTable::wide_dstore() {
 879   wide_lstore();
 880 }
 881 
 882 
 883 void TemplateTable::wide_astore() {
 884   transition(vtos, vtos);
 885   __ pop_ptr(rax);
 886   locals_index_wide(rbx);
 887   __ movptr(aaddress(rbx), rax);
 888 }
 889 
 890 
 891 void TemplateTable::iastore() {
 892   transition(itos, vtos);
 893   __ pop_i(rbx);
 894   // rax,: value
 895   // rdx: array
 896   index_check(rdx, rbx);  // prefer index in rbx,
 897   // rbx,: index
 898   __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
 899 }
 900 
 901 
 902 void TemplateTable::lastore() {
 903   transition(ltos, vtos);
 904   __ pop_i(rbx);
 905   // rax,: low(value)
 906   // rcx: array
 907   // rdx: high(value)
 908   index_check(rcx, rbx);  // prefer index in rbx,
 909   // rbx,: index
 910   __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
 911   NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
 912 }
 913 
 914 
 915 void TemplateTable::fastore() {
 916   transition(ftos, vtos);
 917   __ pop_i(rbx);
 918   // rdx: array
 919   // st0: value
 920   index_check(rdx, rbx);  // prefer index in rbx,
 921   // rbx,: index
 922   __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 923 }
 924 
 925 
 926 void TemplateTable::dastore() {
 927   transition(dtos, vtos);
 928   __ pop_i(rbx);
 929   // rdx: array
 930   // st0: value
 931   index_check(rdx, rbx);  // prefer index in rbx,
 932   // rbx,: index
 933   __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 934 }
 935 
 936 
 937 void TemplateTable::aastore() {
 938   Label is_null, ok_is_subtype, done;
 939   transition(vtos, vtos);
 940   // stack: ..., array, index, value
 941   __ movptr(rax, at_tos());     // Value
 942   __ movl(rcx, at_tos_p1());  // Index
 943   __ movptr(rdx, at_tos_p2());  // Array
 944 
 945   Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 946   index_check_without_pop(rdx, rcx);      // kills rbx,
 947   // do array store check - check for NULL value first
 948   __ testptr(rax, rax);
 949   __ jcc(Assembler::zero, is_null);
 950 
 951   // Move subklass into EBX
 952   __ load_klass(rbx, rax);
 953   // Move superklass into EAX
 954   __ load_klass(rax, rdx);
 955   __ movptr(rax, Address(rax, ObjArrayKlass::element_klass_offset()));
 956   // Compress array+index*wordSize+12 into a single register.  Frees ECX.
 957   __ lea(rdx, element_address);
 958 
 959   // Generate subtype check.  Blows ECX.  Resets EDI to locals.
 960   // Superklass in EAX.  Subklass in EBX.
 961   __ gen_subtype_check( rbx, ok_is_subtype );
 962 
 963   // Come here on failure
 964   // object is at TOS
 965   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
 966 
 967   // Come here on success
 968   __ bind(ok_is_subtype);
 969 
 970   // Get the value to store
 971   __ movptr(rax, at_rsp());
 972   // and store it with appropriate barrier
 973   do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
 974 
 975   __ jmp(done);
 976 
 977   // Have a NULL in EAX, EDX=array, ECX=index.  Store NULL at ary[idx]
 978   __ bind(is_null);
 979   __ profile_null_seen(rbx);
 980 
 981   // Store NULL, (noreg means NULL to do_oop_store)
 982   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
 983 
 984   // Pop stack arguments
 985   __ bind(done);
 986   __ addptr(rsp, 3 * Interpreter::stackElementSize);
 987 }
 988 
 989 
 990 void TemplateTable::bastore() {
 991   transition(itos, vtos);
 992   __ pop_i(rbx);
 993   // rax: value
 994   // rbx: index
 995   // rdx: array
 996   index_check(rdx, rbx);  // prefer index in rbx
 997   // Need to check whether array is boolean or byte
 998   // since both types share the bastore bytecode.
 999   __ load_klass(rcx, rdx);
1000   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1001   int diffbit = Klass::layout_helper_boolean_diffbit();
1002   __ testl(rcx, diffbit);
1003   Label L_skip;
1004   __ jccb(Assembler::zero, L_skip);
1005   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1006   __ bind(L_skip);
1007  __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
1008 }
1009 
1010 
1011 void TemplateTable::castore() {
1012   transition(itos, vtos);
1013   __ pop_i(rbx);
1014   // rax,: value
1015   // rdx: array
1016   index_check(rdx, rbx);  // prefer index in rbx,
1017   // rbx,: index
1018   __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
1019 }
1020 
1021 
1022 void TemplateTable::sastore() {
1023   castore();
1024 }
1025 
1026 
1027 void TemplateTable::istore(int n) {
1028   transition(itos, vtos);
1029   __ movl(iaddress(n), rax);
1030 }
1031 
1032 
1033 void TemplateTable::lstore(int n) {
1034   transition(ltos, vtos);
1035   __ movptr(laddress(n), rax);
1036   NOT_LP64(__ movptr(haddress(n), rdx));
1037 }
1038 
1039 
1040 void TemplateTable::fstore(int n) {
1041   transition(ftos, vtos);
1042   __ fstp_s(faddress(n));
1043 }
1044 
1045 
1046 void TemplateTable::dstore(int n) {
1047   transition(dtos, vtos);
1048   __ fstp_d(daddress(n));
1049 }
1050 
1051 
1052 void TemplateTable::astore(int n) {
1053   transition(vtos, vtos);
1054   __ pop_ptr(rax);
1055   __ movptr(aaddress(n), rax);
1056 }
1057 
1058 
1059 void TemplateTable::pop() {
1060   transition(vtos, vtos);
1061   __ addptr(rsp, Interpreter::stackElementSize);
1062 }
1063 
1064 
1065 void TemplateTable::pop2() {
1066   transition(vtos, vtos);
1067   __ addptr(rsp, 2*Interpreter::stackElementSize);
1068 }
1069 
1070 
1071 void TemplateTable::dup() {
1072   transition(vtos, vtos);
1073   // stack: ..., a
1074   __ load_ptr(0, rax);
1075   __ push_ptr(rax);
1076   // stack: ..., a, a
1077 }
1078 
1079 
1080 void TemplateTable::dup_x1() {
1081   transition(vtos, vtos);
1082   // stack: ..., a, b
1083   __ load_ptr( 0, rax);  // load b
1084   __ load_ptr( 1, rcx);  // load a
1085   __ store_ptr(1, rax);  // store b
1086   __ store_ptr(0, rcx);  // store a
1087   __ push_ptr(rax);      // push b
1088   // stack: ..., b, a, b
1089 }
1090 
1091 
1092 void TemplateTable::dup_x2() {
1093   transition(vtos, vtos);
1094   // stack: ..., a, b, c
1095   __ load_ptr( 0, rax);  // load c
1096   __ load_ptr( 2, rcx);  // load a
1097   __ store_ptr(2, rax);  // store c in a
1098   __ push_ptr(rax);      // push c
1099   // stack: ..., c, b, c, c
1100   __ load_ptr( 2, rax);  // load b
1101   __ store_ptr(2, rcx);  // store a in b
1102   // stack: ..., c, a, c, c
1103   __ store_ptr(1, rax);  // store b in c
1104   // stack: ..., c, a, b, c
1105 }
1106 
1107 
1108 void TemplateTable::dup2() {
1109   transition(vtos, vtos);
1110   // stack: ..., a, b
1111   __ load_ptr(1, rax);  // load a
1112   __ push_ptr(rax);     // push a
1113   __ load_ptr(1, rax);  // load b
1114   __ push_ptr(rax);     // push b
1115   // stack: ..., a, b, a, b
1116 }
1117 
1118 
1119 void TemplateTable::dup2_x1() {
1120   transition(vtos, vtos);
1121   // stack: ..., a, b, c
1122   __ load_ptr( 0, rcx);  // load c
1123   __ load_ptr( 1, rax);  // load b
1124   __ push_ptr(rax);      // push b
1125   __ push_ptr(rcx);      // push c
1126   // stack: ..., a, b, c, b, c
1127   __ store_ptr(3, rcx);  // store c in b
1128   // stack: ..., a, c, c, b, c
1129   __ load_ptr( 4, rcx);  // load a
1130   __ store_ptr(2, rcx);  // store a in 2nd c
1131   // stack: ..., a, c, a, b, c
1132   __ store_ptr(4, rax);  // store b in a
1133   // stack: ..., b, c, a, b, c
1134   // stack: ..., b, c, a, b, c
1135 }
1136 
1137 
1138 void TemplateTable::dup2_x2() {
1139   transition(vtos, vtos);
1140   // stack: ..., a, b, c, d
1141   __ load_ptr( 0, rcx);  // load d
1142   __ load_ptr( 1, rax);  // load c
1143   __ push_ptr(rax);      // push c
1144   __ push_ptr(rcx);      // push d
1145   // stack: ..., a, b, c, d, c, d
1146   __ load_ptr( 4, rax);  // load b
1147   __ store_ptr(2, rax);  // store b in d
1148   __ store_ptr(4, rcx);  // store d in b
1149   // stack: ..., a, d, c, b, c, d
1150   __ load_ptr( 5, rcx);  // load a
1151   __ load_ptr( 3, rax);  // load c
1152   __ store_ptr(3, rcx);  // store a in c
1153   __ store_ptr(5, rax);  // store c in a
1154   // stack: ..., c, d, a, b, c, d
1155   // stack: ..., c, d, a, b, c, d
1156 }
1157 
1158 
1159 void TemplateTable::swap() {
1160   transition(vtos, vtos);
1161   // stack: ..., a, b
1162   __ load_ptr( 1, rcx);  // load a
1163   __ load_ptr( 0, rax);  // load b
1164   __ store_ptr(0, rcx);  // store a in b
1165   __ store_ptr(1, rax);  // store b in a
1166   // stack: ..., b, a
1167 }
1168 
1169 
1170 void TemplateTable::iop2(Operation op) {
1171   transition(itos, itos);
1172   switch (op) {
1173     case add  :                   __ pop_i(rdx); __ addl (rax, rdx); break;
1174     case sub  : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1175     case mul  :                   __ pop_i(rdx); __ imull(rax, rdx); break;
1176     case _and :                   __ pop_i(rdx); __ andl (rax, rdx); break;
1177     case _or  :                   __ pop_i(rdx); __ orl  (rax, rdx); break;
1178     case _xor :                   __ pop_i(rdx); __ xorl (rax, rdx); break;
1179     case shl  : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1180     case shr  : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1181     case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1182     default   : ShouldNotReachHere();
1183   }
1184 }
1185 
1186 
1187 void TemplateTable::lop2(Operation op) {
1188   transition(ltos, ltos);
1189   __ pop_l(rbx, rcx);
1190   switch (op) {
1191     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1192     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1193                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1194     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1195     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1196     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1197     default   : ShouldNotReachHere();
1198   }
1199 }
1200 
1201 
1202 void TemplateTable::idiv() {
1203   transition(itos, itos);
1204   __ mov(rcx, rax);
1205   __ pop_i(rax);
1206   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1207   //       they are not equal, one could do a normal division (no correction
1208   //       needed), which may speed up this implementation for the common case.
1209   //       (see also JVM spec., p.243 & p.271)
1210   __ corrected_idivl(rcx);
1211 }
1212 
1213 
1214 void TemplateTable::irem() {
1215   transition(itos, itos);
1216   __ mov(rcx, rax);
1217   __ pop_i(rax);
1218   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1219   //       they are not equal, one could do a normal division (no correction
1220   //       needed), which may speed up this implementation for the common case.
1221   //       (see also JVM spec., p.243 & p.271)
1222   __ corrected_idivl(rcx);
1223   __ mov(rax, rdx);
1224 }
1225 
1226 
1227 void TemplateTable::lmul() {
1228   transition(ltos, ltos);
1229   __ pop_l(rbx, rcx);
1230   __ push(rcx); __ push(rbx);
1231   __ push(rdx); __ push(rax);
1232   __ lmul(2 * wordSize, 0);
1233   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1234 }
1235 
1236 
1237 void TemplateTable::ldiv() {
1238   transition(ltos, ltos);
1239   __ pop_l(rbx, rcx);
1240   __ push(rcx); __ push(rbx);
1241   __ push(rdx); __ push(rax);
1242   // check if y = 0
1243   __ orl(rax, rdx);
1244   __ jump_cc(Assembler::zero,
1245              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1246   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1247   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1248 }
1249 
1250 
1251 void TemplateTable::lrem() {
1252   transition(ltos, ltos);
1253   __ pop_l(rbx, rcx);
1254   __ push(rcx); __ push(rbx);
1255   __ push(rdx); __ push(rax);
1256   // check if y = 0
1257   __ orl(rax, rdx);
1258   __ jump_cc(Assembler::zero,
1259              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1260   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1261   __ addptr(rsp, 4 * wordSize);
1262 }
1263 
1264 
1265 void TemplateTable::lshl() {
1266   transition(itos, ltos);
1267   __ movl(rcx, rax);                             // get shift count
1268   __ pop_l(rax, rdx);                            // get shift value
1269   __ lshl(rdx, rax);
1270 }
1271 
1272 
1273 void TemplateTable::lshr() {
1274   transition(itos, ltos);
1275   __ mov(rcx, rax);                              // get shift count
1276   __ pop_l(rax, rdx);                            // get shift value
1277   __ lshr(rdx, rax, true);
1278 }
1279 
1280 
1281 void TemplateTable::lushr() {
1282   transition(itos, ltos);
1283   __ mov(rcx, rax);                              // get shift count
1284   __ pop_l(rax, rdx);                            // get shift value
1285   __ lshr(rdx, rax);
1286 }
1287 
1288 
1289 void TemplateTable::fop2(Operation op) {
1290   transition(ftos, ftos);
1291   switch (op) {
1292     case add: __ fadd_s (at_rsp());                break;
1293     case sub: __ fsubr_s(at_rsp());                break;
1294     case mul: __ fmul_s (at_rsp());                break;
1295     case div: __ fdivr_s(at_rsp());                break;
1296     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1297     default : ShouldNotReachHere();
1298   }
1299   __ f2ieee();
1300   __ pop(rax);  // pop float thing off
1301 }
1302 
1303 
1304 void TemplateTable::dop2(Operation op) {
1305   transition(dtos, dtos);
1306 
1307   switch (op) {
1308     case add: __ fadd_d (at_rsp());                break;
1309     case sub: __ fsubr_d(at_rsp());                break;
1310     case mul: {
1311       Label L_strict;
1312       Label L_join;
1313       const Address access_flags      (rcx, Method::access_flags_offset());
1314       __ get_method(rcx);
1315       __ movl(rcx, access_flags);
1316       __ testl(rcx, JVM_ACC_STRICT);
1317       __ jccb(Assembler::notZero, L_strict);
1318       __ fmul_d (at_rsp());
1319       __ jmpb(L_join);
1320       __ bind(L_strict);
1321       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1322       __ fmulp();
1323       __ fmul_d (at_rsp());
1324       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1325       __ fmulp();
1326       __ bind(L_join);
1327       break;
1328     }
1329     case div: {
1330       Label L_strict;
1331       Label L_join;
1332       const Address access_flags      (rcx, Method::access_flags_offset());
1333       __ get_method(rcx);
1334       __ movl(rcx, access_flags);
1335       __ testl(rcx, JVM_ACC_STRICT);
1336       __ jccb(Assembler::notZero, L_strict);
1337       __ fdivr_d(at_rsp());
1338       __ jmp(L_join);
1339       __ bind(L_strict);
1340       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1341       __ fmul_d (at_rsp());
1342       __ fdivrp();
1343       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1344       __ fmulp();
1345       __ bind(L_join);
1346       break;
1347     }
1348     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1349     default : ShouldNotReachHere();
1350   }
1351   __ d2ieee();
1352   // Pop double precision number from rsp.
1353   __ pop(rax);
1354   __ pop(rdx);
1355 }
1356 
1357 
1358 void TemplateTable::ineg() {
1359   transition(itos, itos);
1360   __ negl(rax);
1361 }
1362 
1363 
1364 void TemplateTable::lneg() {
1365   transition(ltos, ltos);
1366   __ lneg(rdx, rax);
1367 }
1368 
1369 
1370 void TemplateTable::fneg() {
1371   transition(ftos, ftos);
1372   __ fchs();
1373 }
1374 
1375 
1376 void TemplateTable::dneg() {
1377   transition(dtos, dtos);
1378   __ fchs();
1379 }
1380 
1381 
1382 void TemplateTable::iinc() {
1383   transition(vtos, vtos);
1384   __ load_signed_byte(rdx, at_bcp(2));           // get constant
1385   locals_index(rbx);
1386   __ addl(iaddress(rbx), rdx);
1387 }
1388 
1389 
1390 void TemplateTable::wide_iinc() {
1391   transition(vtos, vtos);
1392   __ movl(rdx, at_bcp(4));                       // get constant
1393   locals_index_wide(rbx);
1394   __ bswapl(rdx);                                 // swap bytes & sign-extend constant
1395   __ sarl(rdx, 16);
1396   __ addl(iaddress(rbx), rdx);
1397   // Note: should probably use only one movl to get both
1398   //       the index and the constant -> fix this
1399 }
1400 
1401 
1402 void TemplateTable::convert() {
1403   // Checking
1404 #ifdef ASSERT
1405   { TosState tos_in  = ilgl;
1406     TosState tos_out = ilgl;
1407     switch (bytecode()) {
1408       case Bytecodes::_i2l: // fall through
1409       case Bytecodes::_i2f: // fall through
1410       case Bytecodes::_i2d: // fall through
1411       case Bytecodes::_i2b: // fall through
1412       case Bytecodes::_i2c: // fall through
1413       case Bytecodes::_i2s: tos_in = itos; break;
1414       case Bytecodes::_l2i: // fall through
1415       case Bytecodes::_l2f: // fall through
1416       case Bytecodes::_l2d: tos_in = ltos; break;
1417       case Bytecodes::_f2i: // fall through
1418       case Bytecodes::_f2l: // fall through
1419       case Bytecodes::_f2d: tos_in = ftos; break;
1420       case Bytecodes::_d2i: // fall through
1421       case Bytecodes::_d2l: // fall through
1422       case Bytecodes::_d2f: tos_in = dtos; break;
1423       default             : ShouldNotReachHere();
1424     }
1425     switch (bytecode()) {
1426       case Bytecodes::_l2i: // fall through
1427       case Bytecodes::_f2i: // fall through
1428       case Bytecodes::_d2i: // fall through
1429       case Bytecodes::_i2b: // fall through
1430       case Bytecodes::_i2c: // fall through
1431       case Bytecodes::_i2s: tos_out = itos; break;
1432       case Bytecodes::_i2l: // fall through
1433       case Bytecodes::_f2l: // fall through
1434       case Bytecodes::_d2l: tos_out = ltos; break;
1435       case Bytecodes::_i2f: // fall through
1436       case Bytecodes::_l2f: // fall through
1437       case Bytecodes::_d2f: tos_out = ftos; break;
1438       case Bytecodes::_i2d: // fall through
1439       case Bytecodes::_l2d: // fall through
1440       case Bytecodes::_f2d: tos_out = dtos; break;
1441       default             : ShouldNotReachHere();
1442     }
1443     transition(tos_in, tos_out);
1444   }
1445 #endif // ASSERT
1446 
1447   // Conversion
1448   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1449   switch (bytecode()) {
1450     case Bytecodes::_i2l:
1451       __ extend_sign(rdx, rax);
1452       break;
1453     case Bytecodes::_i2f:
1454       __ push(rax);          // store int on tos
1455       __ fild_s(at_rsp());   // load int to ST0
1456       __ f2ieee();           // truncate to float size
1457       __ pop(rcx);           // adjust rsp
1458       break;
1459     case Bytecodes::_i2d:
1460       __ push(rax);          // add one slot for d2ieee()
1461       __ push(rax);          // store int on tos
1462       __ fild_s(at_rsp());   // load int to ST0
1463       __ d2ieee();           // truncate to double size
1464       __ pop(rcx);           // adjust rsp
1465       __ pop(rcx);
1466       break;
1467     case Bytecodes::_i2b:
1468       __ shll(rax, 24);      // truncate upper 24 bits
1469       __ sarl(rax, 24);      // and sign-extend byte
1470       LP64_ONLY(__ movsbl(rax, rax));
1471       break;
1472     case Bytecodes::_i2c:
1473       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1474       LP64_ONLY(__ movzwl(rax, rax));
1475       break;
1476     case Bytecodes::_i2s:
1477       __ shll(rax, 16);      // truncate upper 16 bits
1478       __ sarl(rax, 16);      // and sign-extend short
1479       LP64_ONLY(__ movswl(rax, rax));
1480       break;
1481     case Bytecodes::_l2i:
1482       /* nothing to do */
1483       break;
1484     case Bytecodes::_l2f:
1485       __ push(rdx);          // store long on tos
1486       __ push(rax);
1487       __ fild_d(at_rsp());   // load long to ST0
1488       __ f2ieee();           // truncate to float size
1489       __ pop(rcx);           // adjust rsp
1490       __ pop(rcx);
1491       break;
1492     case Bytecodes::_l2d:
1493       __ push(rdx);          // store long on tos
1494       __ push(rax);
1495       __ fild_d(at_rsp());   // load long to ST0
1496       __ d2ieee();           // truncate to double size
1497       __ pop(rcx);           // adjust rsp
1498       __ pop(rcx);
1499       break;
1500     case Bytecodes::_f2i:
1501       __ push(rcx);          // reserve space for argument
1502       __ fstp_s(at_rsp());   // pass float argument on stack
1503       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1504       break;
1505     case Bytecodes::_f2l:
1506       __ push(rcx);          // reserve space for argument
1507       __ fstp_s(at_rsp());   // pass float argument on stack
1508       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1509       break;
1510     case Bytecodes::_f2d:
1511       /* nothing to do */
1512       break;
1513     case Bytecodes::_d2i:
1514       __ push(rcx);          // reserve space for argument
1515       __ push(rcx);
1516       __ fstp_d(at_rsp());   // pass double argument on stack
1517       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1518       break;
1519     case Bytecodes::_d2l:
1520       __ push(rcx);          // reserve space for argument
1521       __ push(rcx);
1522       __ fstp_d(at_rsp());   // pass double argument on stack
1523       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1524       break;
1525     case Bytecodes::_d2f:
1526       __ push(rcx);          // reserve space for f2ieee()
1527       __ f2ieee();           // truncate to float size
1528       __ pop(rcx);           // adjust rsp
1529       break;
1530     default             :
1531       ShouldNotReachHere();
1532   }
1533 }
1534 
1535 
1536 void TemplateTable::lcmp() {
1537   transition(ltos, itos);
1538   // y = rdx:rax
1539   __ pop_l(rbx, rcx);             // get x = rcx:rbx
1540   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1541   __ mov(rax, rcx);
1542 }
1543 
1544 
1545 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1546   if (is_float) {
1547     __ fld_s(at_rsp());
1548   } else {
1549     __ fld_d(at_rsp());
1550     __ pop(rdx);
1551   }
1552   __ pop(rcx);
1553   __ fcmp2int(rax, unordered_result < 0);
1554 }
1555 
1556 
1557 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1558   __ get_method(rcx);           // ECX holds method
1559   __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1560 
1561   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1562                              InvocationCounter::counter_offset();
1563   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1564                               InvocationCounter::counter_offset();
1565 
1566   // Load up EDX with the branch displacement
1567   if (is_wide) {
1568     __ movl(rdx, at_bcp(1));
1569   } else {
1570     __ load_signed_short(rdx, at_bcp(1));
1571   }
1572   __ bswapl(rdx);
1573   if (!is_wide) __ sarl(rdx, 16);
1574   LP64_ONLY(__ movslq(rdx, rdx));
1575 
1576 
1577   // Handle all the JSR stuff here, then exit.
1578   // It's much shorter and cleaner than intermingling with the
1579   // non-JSR normal-branch stuff occurring below.
1580   if (is_jsr) {
1581     // Pre-load the next target bytecode into EBX
1582     __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1583 
1584     // compute return address as bci in rax,
1585     __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(ConstMethod::codes_offset())));
1586     __ subptr(rax, Address(rcx, Method::const_offset()));
1587     // Adjust the bcp in RSI by the displacement in EDX
1588     __ addptr(rsi, rdx);
1589     // Push return address
1590     __ push_i(rax);
1591     // jsr returns vtos
1592     __ dispatch_only_noverify(vtos);
1593     return;
1594   }
1595 
1596   // Normal (non-jsr) branch handling
1597 
1598   // Adjust the bcp in RSI by the displacement in EDX
1599   __ addptr(rsi, rdx);
1600 
1601   assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1602   Label backedge_counter_overflow;
1603   Label profile_method;
1604   Label dispatch;
1605   if (UseLoopCounter) {
1606     // increment backedge counter for backward branches
1607     // rax,: MDO
1608     // rbx,: MDO bumped taken-count
1609     // rcx: method
1610     // rdx: target offset
1611     // rsi: target bcp
1612     // rdi: locals pointer
1613     __ testl(rdx, rdx);             // check if forward or backward branch
1614     __ jcc(Assembler::positive, dispatch); // count only if backward branch
1615 
1616     // check if MethodCounters exists
1617     Label has_counters;
1618     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1619     __ testptr(rax, rax);
1620     __ jcc(Assembler::notZero, has_counters);
1621     __ push(rdx);
1622     __ push(rcx);
1623     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1624                rcx);
1625     __ pop(rcx);
1626     __ pop(rdx);
1627     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1628     __ testptr(rax, rax);
1629     __ jcc(Assembler::zero, dispatch);
1630     __ bind(has_counters);
1631 
1632     if (TieredCompilation) {
1633       Label no_mdo;
1634       int increment = InvocationCounter::count_increment;
1635       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1636       if (ProfileInterpreter) {
1637         // Are we profiling?
1638         __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
1639         __ testptr(rbx, rbx);
1640         __ jccb(Assembler::zero, no_mdo);
1641         // Increment the MDO backedge counter
1642         const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
1643                                                 in_bytes(InvocationCounter::counter_offset()));
1644         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1645                                    rax, false, Assembler::zero, &backedge_counter_overflow);
1646         __ jmp(dispatch);
1647       }
1648       __ bind(no_mdo);
1649       // Increment backedge counter in MethodCounters*
1650       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1651       __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1652                                  rax, false, Assembler::zero, &backedge_counter_overflow);
1653     } else {
1654       // increment counter
1655       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1656       __ movl(rax, Address(rcx, be_offset));        // load backedge counter
1657       __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1658       __ movl(Address(rcx, be_offset), rax);        // store counter
1659 
1660       __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
1661 
1662       __ andl(rax, InvocationCounter::count_mask_value);     // and the status bits
1663       __ addl(rax, Address(rcx, be_offset));        // add both counters
1664 
1665       if (ProfileInterpreter) {
1666         // Test to see if we should create a method data oop
1667         __ cmp32(rax,
1668                  ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1669         __ jcc(Assembler::less, dispatch);
1670 
1671         // if no method data exists, go to profile method
1672         __ test_method_data_pointer(rax, profile_method);
1673 
1674         if (UseOnStackReplacement) {
1675           // check for overflow against rbx, which is the MDO taken count
1676           __ cmp32(rbx,
1677                    ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1678           __ jcc(Assembler::below, dispatch);
1679 
1680           // When ProfileInterpreter is on, the backedge_count comes from the
1681           // MethodData*, which value does not get reset on the call to
1682           // frequency_counter_overflow().  To avoid excessive calls to the overflow
1683           // routine while the method is being compiled, add a second test to make
1684           // sure the overflow function is called only once every overflow_frequency.
1685           const int overflow_frequency = 1024;
1686           __ andptr(rbx, overflow_frequency-1);
1687           __ jcc(Assembler::zero, backedge_counter_overflow);
1688         }
1689       } else {
1690         if (UseOnStackReplacement) {
1691           // check for overflow against rax, which is the sum of the counters
1692           __ cmp32(rax,
1693                    ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1694           __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1695 
1696         }
1697       }
1698     }
1699     __ bind(dispatch);
1700   }
1701 
1702   // Pre-load the next target bytecode into EBX
1703   __ load_unsigned_byte(rbx, Address(rsi, 0));
1704 
1705   // continue with the bytecode @ target
1706   // rax,: return bci for jsr's, unused otherwise
1707   // rbx,: target bytecode
1708   // rsi: target bcp
1709   __ dispatch_only(vtos);
1710 
1711   if (UseLoopCounter) {
1712     if (ProfileInterpreter) {
1713       // Out-of-line code to allocate method data oop.
1714       __ bind(profile_method);
1715       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1716       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1717       __ set_method_data_pointer_for_bcp();
1718       __ jmp(dispatch);
1719     }
1720 
1721     if (UseOnStackReplacement) {
1722 
1723       // invocation counter overflow
1724       __ bind(backedge_counter_overflow);
1725       __ negptr(rdx);
1726       __ addptr(rdx, rsi);        // branch bcp
1727       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1728       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1729 
1730       // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1731       // rbx,: target bytecode
1732       // rdx: scratch
1733       // rdi: locals pointer
1734       // rsi: bcp
1735       __ testptr(rax, rax);                      // test result
1736       __ jcc(Assembler::zero, dispatch);         // no osr if null
1737       // nmethod may have been invalidated (VM may block upon call_VM return)
1738       __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1739       __ cmpl(rcx, InvalidOSREntryBci);
1740       __ jcc(Assembler::equal, dispatch);
1741 
1742       // We have the address of an on stack replacement routine in rax,
1743       // We need to prepare to execute the OSR method. First we must
1744       // migrate the locals and monitors off of the stack.
1745 
1746       __ mov(rbx, rax);                             // save the nmethod
1747 
1748       const Register thread = rcx;
1749       __ get_thread(thread);
1750       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1751       // rax, is OSR buffer, move it to expected parameter location
1752       __ mov(rcx, rax);
1753 
1754       // pop the interpreter frame
1755       __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1756       __ leave();                                // remove frame anchor
1757       __ pop(rdi);                               // get return address
1758       __ mov(rsp, rdx);                          // set sp to sender sp
1759 
1760       // Align stack pointer for compiled code (note that caller is
1761       // responsible for undoing this fixup by remembering the old SP
1762       // in an rbp,-relative location)
1763       __ andptr(rsp, -(StackAlignmentInBytes));
1764 
1765       // push the (possibly adjusted) return address
1766       __ push(rdi);
1767 
1768       // and begin the OSR nmethod
1769       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1770     }
1771   }
1772 }
1773 
1774 
1775 void TemplateTable::if_0cmp(Condition cc) {
1776   transition(itos, vtos);
1777   // assume branch is more often taken than not (loops use backward branches)
1778   Label not_taken;
1779   __ testl(rax, rax);
1780   __ jcc(j_not(cc), not_taken);
1781   branch(false, false);
1782   __ bind(not_taken);
1783   __ profile_not_taken_branch(rax);
1784 }
1785 
1786 
1787 void TemplateTable::if_icmp(Condition cc) {
1788   transition(itos, vtos);
1789   // assume branch is more often taken than not (loops use backward branches)
1790   Label not_taken;
1791   __ pop_i(rdx);
1792   __ cmpl(rdx, rax);
1793   __ jcc(j_not(cc), not_taken);
1794   branch(false, false);
1795   __ bind(not_taken);
1796   __ profile_not_taken_branch(rax);
1797 }
1798 
1799 
1800 void TemplateTable::if_nullcmp(Condition cc) {
1801   transition(atos, vtos);
1802   // assume branch is more often taken than not (loops use backward branches)
1803   Label not_taken;
1804   __ testptr(rax, rax);
1805   __ jcc(j_not(cc), not_taken);
1806   branch(false, false);
1807   __ bind(not_taken);
1808   __ profile_not_taken_branch(rax);
1809 }
1810 
1811 
1812 void TemplateTable::if_acmp(Condition cc) {
1813   transition(atos, vtos);
1814   // assume branch is more often taken than not (loops use backward branches)
1815   Label not_taken;
1816   __ pop_ptr(rdx);
1817   __ cmpptr(rdx, rax);
1818   __ jcc(j_not(cc), not_taken);
1819   branch(false, false);
1820   __ bind(not_taken);
1821   __ profile_not_taken_branch(rax);
1822 }
1823 
1824 
1825 void TemplateTable::ret() {
1826   transition(vtos, vtos);
1827   locals_index(rbx);
1828   __ movptr(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1829   __ profile_ret(rbx, rcx);
1830   __ get_method(rax);
1831   __ movptr(rsi, Address(rax, Method::const_offset()));
1832   __ lea(rsi, Address(rsi, rbx, Address::times_1,
1833                       ConstMethod::codes_offset()));
1834   __ dispatch_next(vtos);
1835 }
1836 
1837 
1838 void TemplateTable::wide_ret() {
1839   transition(vtos, vtos);
1840   locals_index_wide(rbx);
1841   __ movptr(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1842   __ profile_ret(rbx, rcx);
1843   __ get_method(rax);
1844   __ movptr(rsi, Address(rax, Method::const_offset()));
1845   __ lea(rsi, Address(rsi, rbx, Address::times_1, ConstMethod::codes_offset()));
1846   __ dispatch_next(vtos);
1847 }
1848 
1849 
1850 void TemplateTable::tableswitch() {
1851   Label default_case, continue_execution;
1852   transition(itos, vtos);
1853   // align rsi
1854   __ lea(rbx, at_bcp(wordSize));
1855   __ andptr(rbx, -wordSize);
1856   // load lo & hi
1857   __ movl(rcx, Address(rbx, 1 * wordSize));
1858   __ movl(rdx, Address(rbx, 2 * wordSize));
1859   __ bswapl(rcx);
1860   __ bswapl(rdx);
1861   // check against lo & hi
1862   __ cmpl(rax, rcx);
1863   __ jccb(Assembler::less, default_case);
1864   __ cmpl(rax, rdx);
1865   __ jccb(Assembler::greater, default_case);
1866   // lookup dispatch offset
1867   __ subl(rax, rcx);
1868   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1869   __ profile_switch_case(rax, rbx, rcx);
1870   // continue execution
1871   __ bind(continue_execution);
1872   __ bswapl(rdx);
1873   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1874   __ addptr(rsi, rdx);
1875   __ dispatch_only(vtos);
1876   // handle default
1877   __ bind(default_case);
1878   __ profile_switch_default(rax);
1879   __ movl(rdx, Address(rbx, 0));
1880   __ jmp(continue_execution);
1881 }
1882 
1883 
1884 void TemplateTable::lookupswitch() {
1885   transition(itos, itos);
1886   __ stop("lookupswitch bytecode should have been rewritten");
1887 }
1888 
1889 
1890 void TemplateTable::fast_linearswitch() {
1891   transition(itos, vtos);
1892   Label loop_entry, loop, found, continue_execution;
1893   // bswapl rax, so we can avoid bswapping the table entries
1894   __ bswapl(rax);
1895   // align rsi
1896   __ lea(rbx, at_bcp(wordSize));                // btw: should be able to get rid of this instruction (change offsets below)
1897   __ andptr(rbx, -wordSize);
1898   // set counter
1899   __ movl(rcx, Address(rbx, wordSize));
1900   __ bswapl(rcx);
1901   __ jmpb(loop_entry);
1902   // table search
1903   __ bind(loop);
1904   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1905   __ jccb(Assembler::equal, found);
1906   __ bind(loop_entry);
1907   __ decrementl(rcx);
1908   __ jcc(Assembler::greaterEqual, loop);
1909   // default case
1910   __ profile_switch_default(rax);
1911   __ movl(rdx, Address(rbx, 0));
1912   __ jmpb(continue_execution);
1913   // entry found -> get offset
1914   __ bind(found);
1915   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1916   __ profile_switch_case(rcx, rax, rbx);
1917   // continue execution
1918   __ bind(continue_execution);
1919   __ bswapl(rdx);
1920   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1921   __ addptr(rsi, rdx);
1922   __ dispatch_only(vtos);
1923 }
1924 
1925 
1926 void TemplateTable::fast_binaryswitch() {
1927   transition(itos, vtos);
1928   // Implementation using the following core algorithm:
1929   //
1930   // int binary_search(int key, LookupswitchPair* array, int n) {
1931   //   // Binary search according to "Methodik des Programmierens" by
1932   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1933   //   int i = 0;
1934   //   int j = n;
1935   //   while (i+1 < j) {
1936   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1937   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1938   //     // where a stands for the array and assuming that the (inexisting)
1939   //     // element a[n] is infinitely big.
1940   //     int h = (i + j) >> 1;
1941   //     // i < h < j
1942   //     if (key < array[h].fast_match()) {
1943   //       j = h;
1944   //     } else {
1945   //       i = h;
1946   //     }
1947   //   }
1948   //   // R: a[i] <= key < a[i+1] or Q
1949   //   // (i.e., if key is within array, i is the correct index)
1950   //   return i;
1951   // }
1952 
1953   // register allocation
1954   const Register key   = rax;                    // already set (tosca)
1955   const Register array = rbx;
1956   const Register i     = rcx;
1957   const Register j     = rdx;
1958   const Register h     = rdi;                    // needs to be restored
1959   const Register temp  = rsi;
1960   // setup array
1961   __ save_bcp();
1962 
1963   __ lea(array, at_bcp(3*wordSize));             // btw: should be able to get rid of this instruction (change offsets below)
1964   __ andptr(array, -wordSize);
1965   // initialize i & j
1966   __ xorl(i, i);                                 // i = 0;
1967   __ movl(j, Address(array, -wordSize));         // j = length(array);
1968   // Convert j into native byteordering
1969   __ bswapl(j);
1970   // and start
1971   Label entry;
1972   __ jmp(entry);
1973 
1974   // binary search loop
1975   { Label loop;
1976     __ bind(loop);
1977     // int h = (i + j) >> 1;
1978     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1979     __ sarl(h, 1);                               // h = (i + j) >> 1;
1980     // if (key < array[h].fast_match()) {
1981     //   j = h;
1982     // } else {
1983     //   i = h;
1984     // }
1985     // Convert array[h].match to native byte-ordering before compare
1986     __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1987     __ bswapl(temp);
1988     __ cmpl(key, temp);
1989     // j = h if (key <  array[h].fast_match())
1990     __ cmov32(Assembler::less        , j, h);
1991     // i = h if (key >= array[h].fast_match())
1992     __ cmov32(Assembler::greaterEqual, i, h);
1993     // while (i+1 < j)
1994     __ bind(entry);
1995     __ leal(h, Address(i, 1));                   // i+1
1996     __ cmpl(h, j);                               // i+1 < j
1997     __ jcc(Assembler::less, loop);
1998   }
1999 
2000   // end of binary search, result index is i (must check again!)
2001   Label default_case;
2002   // Convert array[i].match to native byte-ordering before compare
2003   __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
2004   __ bswapl(temp);
2005   __ cmpl(key, temp);
2006   __ jcc(Assembler::notEqual, default_case);
2007 
2008   // entry found -> j = offset
2009   __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
2010   __ profile_switch_case(i, key, array);
2011   __ bswapl(j);
2012   LP64_ONLY(__ movslq(j, j));
2013   __ restore_bcp();
2014   __ restore_locals();                           // restore rdi
2015   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2016 
2017   __ addptr(rsi, j);
2018   __ dispatch_only(vtos);
2019 
2020   // default case -> j = default offset
2021   __ bind(default_case);
2022   __ profile_switch_default(i);
2023   __ movl(j, Address(array, -2*wordSize));
2024   __ bswapl(j);
2025   LP64_ONLY(__ movslq(j, j));
2026   __ restore_bcp();
2027   __ restore_locals();                           // restore rdi
2028   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2029   __ addptr(rsi, j);
2030   __ dispatch_only(vtos);
2031 }
2032 
2033 
2034 void TemplateTable::_return(TosState state) {
2035   transition(state, state);
2036   assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2037 
2038   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2039     assert(state == vtos, "only valid state");
2040     __ movptr(rax, aaddress(0));
2041     __ load_klass(rdi, rax);
2042     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2043     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2044     Label skip_register_finalizer;
2045     __ jcc(Assembler::zero, skip_register_finalizer);
2046 
2047     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2048 
2049     __ bind(skip_register_finalizer);
2050   }
2051 
2052   // Narrow result if state is itos but result type is smaller.
2053   // Need to narrow in the return bytecode rather than in generate_return_entry
2054   // since compiled code callers expect the result to already be narrowed.
2055   if (state == itos) {
2056     __ narrow(rax);
2057   }
2058   __ remove_activation(state, rsi);
2059 
2060   __ jmp(rsi);
2061 }
2062 
2063 
2064 // ----------------------------------------------------------------------------
2065 // Volatile variables demand their effects be made known to all CPU's in
2066 // order.  Store buffers on most chips allow reads & writes to reorder; the
2067 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2068 // memory barrier (i.e., it's not sufficient that the interpreter does not
2069 // reorder volatile references, the hardware also must not reorder them).
2070 //
2071 // According to the new Java Memory Model (JMM):
2072 // (1) All volatiles are serialized wrt to each other.
2073 // ALSO reads & writes act as aquire & release, so:
2074 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2075 // the read float up to before the read.  It's OK for non-volatile memory refs
2076 // that happen before the volatile read to float down below it.
2077 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2078 // that happen BEFORE the write float down to after the write.  It's OK for
2079 // non-volatile memory refs that happen after the volatile write to float up
2080 // before it.
2081 //
2082 // We only put in barriers around volatile refs (they are expensive), not
2083 // _between_ memory refs (that would require us to track the flavor of the
2084 // previous memory refs).  Requirements (2) and (3) require some barriers
2085 // before volatile stores and after volatile loads.  These nearly cover
2086 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2087 // case is placed after volatile-stores although it could just as well go
2088 // before volatile-loads.
2089 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2090   // Helper function to insert a is-volatile test and memory barrier
2091   if( !os::is_MP() ) return;    // Not needed on single CPU
2092   __ membar(order_constraint);
2093 }
2094 
2095 void TemplateTable::resolve_cache_and_index(int byte_no,
2096                                             Register Rcache,
2097                                             Register index,
2098                                             size_t index_size) {
2099   const Register temp = rbx;
2100   assert_different_registers(Rcache, index, temp);
2101 
2102   Label resolved;
2103     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2104     __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2105     __ cmpl(temp, (int) bytecode());  // have we resolved this bytecode?
2106     __ jcc(Assembler::equal, resolved);
2107 
2108   // resolve first time through
2109   address entry;
2110   switch (bytecode()) {
2111     case Bytecodes::_getstatic      : // fall through
2112     case Bytecodes::_putstatic      : // fall through
2113     case Bytecodes::_getfield       : // fall through
2114     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);        break;
2115     case Bytecodes::_invokevirtual  : // fall through
2116     case Bytecodes::_invokespecial  : // fall through
2117     case Bytecodes::_invokestatic   : // fall through
2118     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);         break;
2119     case Bytecodes::_invokehandle   : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);   break;
2120     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
2121     default:
2122       fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2123       break;
2124   }
2125   __ movl(temp, (int)bytecode());
2126   __ call_VM(noreg, entry, temp);
2127   // Update registers with resolved info
2128   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2129   __ bind(resolved);
2130 }
2131 
2132 
2133 // The cache and index registers must be set before call
2134 void TemplateTable::load_field_cp_cache_entry(Register obj,
2135                                               Register cache,
2136                                               Register index,
2137                                               Register off,
2138                                               Register flags,
2139                                               bool is_static = false) {
2140   assert_different_registers(cache, index, flags, off);
2141 
2142   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2143   // Field offset
2144   __ movptr(off, Address(cache, index, Address::times_ptr,
2145                          in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2146   // Flags
2147   __ movl(flags, Address(cache, index, Address::times_ptr,
2148            in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2149 
2150   // klass overwrite register
2151   if (is_static) {
2152     __ movptr(obj, Address(cache, index, Address::times_ptr,
2153                            in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2154     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2155     __ movptr(obj, Address(obj, mirror_offset));
2156   }
2157 }
2158 
2159 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2160                                                Register method,
2161                                                Register itable_index,
2162                                                Register flags,
2163                                                bool is_invokevirtual,
2164                                                bool is_invokevfinal, /*unused*/
2165                                                bool is_invokedynamic) {
2166   // setup registers
2167   const Register cache = rcx;
2168   const Register index = rdx;
2169   assert_different_registers(method, flags);
2170   assert_different_registers(method, cache, index);
2171   assert_different_registers(itable_index, flags);
2172   assert_different_registers(itable_index, cache, index);
2173   // determine constant pool cache field offsets
2174   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2175   const int method_offset = in_bytes(
2176     ConstantPoolCache::base_offset() +
2177       ((byte_no == f2_byte)
2178        ? ConstantPoolCacheEntry::f2_offset()
2179        : ConstantPoolCacheEntry::f1_offset()));
2180   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2181                                     ConstantPoolCacheEntry::flags_offset());
2182   // access constant pool cache fields
2183   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2184                                     ConstantPoolCacheEntry::f2_offset());
2185 
2186   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2187   resolve_cache_and_index(byte_no, cache, index, index_size);
2188     __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2189 
2190   if (itable_index != noreg) {
2191     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2192   }
2193   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2194 }
2195 
2196 
2197 // The registers cache and index expected to be set before call.
2198 // Correct values of the cache and index registers are preserved.
2199 void TemplateTable::jvmti_post_field_access(Register cache,
2200                                             Register index,
2201                                             bool is_static,
2202                                             bool has_tos) {
2203   if (JvmtiExport::can_post_field_access()) {
2204     // Check to see if a field access watch has been set before we take
2205     // the time to call into the VM.
2206     Label L1;
2207     assert_different_registers(cache, index, rax);
2208     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2209     __ testl(rax,rax);
2210     __ jcc(Assembler::zero, L1);
2211 
2212     // cache entry pointer
2213     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2214     __ shll(index, LogBytesPerWord);
2215     __ addptr(cache, index);
2216     if (is_static) {
2217       __ xorptr(rax, rax);      // NULL object reference
2218     } else {
2219       __ pop(atos);         // Get the object
2220       __ verify_oop(rax);
2221       __ push(atos);        // Restore stack state
2222     }
2223     // rax,:   object pointer or NULL
2224     // cache: cache entry pointer
2225     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2226                rax, cache);
2227     __ get_cache_and_index_at_bcp(cache, index, 1);
2228     __ bind(L1);
2229   }
2230 }
2231 
2232 void TemplateTable::pop_and_check_object(Register r) {
2233   __ pop_ptr(r);
2234   __ null_check(r);  // for field access must check obj.
2235   __ verify_oop(r);
2236 }
2237 
2238 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2239   transition(vtos, vtos);
2240 
2241   const Register cache = rcx;
2242   const Register index = rdx;
2243   const Register obj   = rcx;
2244   const Register off   = rbx;
2245   const Register flags = rax;
2246 
2247   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2248   jvmti_post_field_access(cache, index, is_static, false);
2249   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2250 
2251   if (!is_static) pop_and_check_object(obj);
2252 
2253   const Address lo(obj, off, Address::times_1, 0*wordSize);
2254   const Address hi(obj, off, Address::times_1, 1*wordSize);
2255 
2256   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2257 
2258   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2259   assert(btos == 0, "change code, btos != 0");
2260   // btos
2261   __ andptr(flags, ConstantPoolCacheEntry::tos_state_mask);
2262   __ jcc(Assembler::notZero, notByte);
2263 
2264   __ load_signed_byte(rax, lo );
2265   __ push(btos);
2266   // Rewrite bytecode to be faster
2267   if (!is_static) {
2268     patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2269   }
2270   __ jmp(Done);
2271 
2272   __ bind(notByte);
2273 
2274   __ cmpl(flags, ztos);
2275   __ jcc(Assembler::notEqual, notBool);
2276 
2277   // ztos (same code as btos)
2278   __ load_signed_byte(rax, lo);
2279   __ push(ztos);
2280   // Rewrite bytecode to be faster
2281   if (!is_static) {
2282     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2283     patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2284   }
2285   __ jmp(Done);
2286 
2287   __ bind(notBool);
2288 
2289   // itos
2290   __ cmpl(flags, itos );
2291   __ jcc(Assembler::notEqual, notInt);
2292 
2293   __ movl(rax, lo );
2294   __ push(itos);
2295   // Rewrite bytecode to be faster
2296   if (!is_static) {
2297     patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2298   }
2299   __ jmp(Done);
2300 
2301   __ bind(notInt);
2302   // atos
2303   __ cmpl(flags, atos );
2304   __ jcc(Assembler::notEqual, notObj);
2305 
2306   __ movl(rax, lo );
2307   __ push(atos);
2308   if (!is_static) {
2309     patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2310   }
2311   __ jmp(Done);
2312 
2313   __ bind(notObj);
2314   // ctos
2315   __ cmpl(flags, ctos );
2316   __ jcc(Assembler::notEqual, notChar);
2317 
2318   __ load_unsigned_short(rax, lo );
2319   __ push(ctos);
2320   if (!is_static) {
2321     patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2322   }
2323   __ jmp(Done);
2324 
2325   __ bind(notChar);
2326   // stos
2327   __ cmpl(flags, stos );
2328   __ jcc(Assembler::notEqual, notShort);
2329 
2330   __ load_signed_short(rax, lo );
2331   __ push(stos);
2332   if (!is_static) {
2333     patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2334   }
2335   __ jmp(Done);
2336 
2337   __ bind(notShort);
2338   // ltos
2339   __ cmpl(flags, ltos );
2340   __ jcc(Assembler::notEqual, notLong);
2341 
2342   // Generate code as if volatile.  There just aren't enough registers to
2343   // save that information and this code is faster than the test.
2344   __ fild_d(lo);                // Must load atomically
2345   __ subptr(rsp,2*wordSize);    // Make space for store
2346   __ fistp_d(Address(rsp,0));
2347   __ pop(rax);
2348   __ pop(rdx);
2349 
2350   __ push(ltos);
2351   // Don't rewrite to _fast_lgetfield for potential volatile case.
2352   __ jmp(Done);
2353 
2354   __ bind(notLong);
2355   // ftos
2356   __ cmpl(flags, ftos );
2357   __ jcc(Assembler::notEqual, notFloat);
2358 
2359   __ fld_s(lo);
2360   __ push(ftos);
2361   if (!is_static) {
2362     patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2363   }
2364   __ jmp(Done);
2365 
2366   __ bind(notFloat);
2367   // dtos
2368   __ cmpl(flags, dtos );
2369   __ jcc(Assembler::notEqual, notDouble);
2370 
2371   __ fld_d(lo);
2372   __ push(dtos);
2373   if (!is_static) {
2374     patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2375   }
2376   __ jmpb(Done);
2377 
2378   __ bind(notDouble);
2379 
2380   __ stop("Bad state");
2381 
2382   __ bind(Done);
2383   // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2384   // volatile_barrier( );
2385 }
2386 
2387 
2388 void TemplateTable::getfield(int byte_no) {
2389   getfield_or_static(byte_no, false);
2390 }
2391 
2392 
2393 void TemplateTable::getstatic(int byte_no) {
2394   getfield_or_static(byte_no, true);
2395 }
2396 
2397 // The registers cache and index expected to be set before call.
2398 // The function may destroy various registers, just not the cache and index registers.
2399 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2400 
2401   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2402 
2403   if (JvmtiExport::can_post_field_modification()) {
2404     // Check to see if a field modification watch has been set before we take
2405     // the time to call into the VM.
2406     Label L1;
2407     assert_different_registers(cache, index, rax);
2408     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2409     __ testl(rax, rax);
2410     __ jcc(Assembler::zero, L1);
2411 
2412     // The cache and index registers have been already set.
2413     // This allows to eliminate this call but the cache and index
2414     // registers have to be correspondingly used after this line.
2415     __ get_cache_and_index_at_bcp(rax, rdx, 1);
2416 
2417     if (is_static) {
2418       // Life is simple.  Null out the object pointer.
2419       __ xorptr(rbx, rbx);
2420     } else {
2421       // Life is harder. The stack holds the value on top, followed by the object.
2422       // We don't know the size of the value, though; it could be one or two words
2423       // depending on its type. As a result, we must find the type to determine where
2424       // the object is.
2425       Label two_word, valsize_known;
2426       __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2427                                    ConstantPoolCacheEntry::flags_offset())));
2428       __ mov(rbx, rsp);
2429       __ shrl(rcx, ConstantPoolCacheEntry::tos_state_shift);
2430       // Make sure we don't need to mask rcx after the above shift
2431       ConstantPoolCacheEntry::verify_tos_state_shift();
2432       __ cmpl(rcx, ltos);
2433       __ jccb(Assembler::equal, two_word);
2434       __ cmpl(rcx, dtos);
2435       __ jccb(Assembler::equal, two_word);
2436       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2437       __ jmpb(valsize_known);
2438 
2439       __ bind(two_word);
2440       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2441 
2442       __ bind(valsize_known);
2443       // setup object pointer
2444       __ movptr(rbx, Address(rbx, 0));
2445     }
2446     // cache entry pointer
2447     __ addptr(rax, in_bytes(cp_base_offset));
2448     __ shll(rdx, LogBytesPerWord);
2449     __ addptr(rax, rdx);
2450     // object (tos)
2451     __ mov(rcx, rsp);
2452     // rbx,: object pointer set up above (NULL if static)
2453     // rax,: cache entry pointer
2454     // rcx: jvalue object on the stack
2455     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2456                rbx, rax, rcx);
2457     __ get_cache_and_index_at_bcp(cache, index, 1);
2458     __ bind(L1);
2459   }
2460 }
2461 
2462 
2463 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2464   transition(vtos, vtos);
2465 
2466   const Register cache = rcx;
2467   const Register index = rdx;
2468   const Register obj   = rcx;
2469   const Register off   = rbx;
2470   const Register flags = rax;
2471 
2472   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2473   jvmti_post_field_mod(cache, index, is_static);
2474   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2475 
2476   // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2477   // volatile_barrier( );
2478 
2479   Label notVolatile, Done;
2480   __ movl(rdx, flags);
2481   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2482   __ andl(rdx, 0x1);
2483 
2484   // field addresses
2485   const Address lo(obj, off, Address::times_1, 0*wordSize);
2486   const Address hi(obj, off, Address::times_1, 1*wordSize);
2487 
2488   Label notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2489 
2490   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2491   assert(btos == 0, "change code, btos != 0");
2492   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2493   __ jcc(Assembler::notZero, notByte);
2494 
2495   // btos
2496   {
2497     __ pop(btos);
2498     if (!is_static) pop_and_check_object(obj);
2499     __ movb(lo, rax);
2500     if (!is_static) {
2501       patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx, true, byte_no);
2502     }
2503     __ jmp(Done);
2504   }
2505 
2506   __ bind(notByte);
2507   __ cmpl(flags, ztos);
2508   __ jcc(Assembler::notEqual, notBool);
2509 
2510   // ztos
2511   {
2512     __ pop(ztos);
2513     if (!is_static) pop_and_check_object(obj);
2514     __ andl(rax, 0x1);
2515     __ movb(lo, rax);
2516     if (!is_static) {
2517       patch_bytecode(Bytecodes::_fast_zputfield, rcx, rbx, true, byte_no);
2518     }
2519     __ jmp(Done);
2520   }
2521 
2522   __ bind(notBool);
2523   __ cmpl(flags, itos);
2524   __ jcc(Assembler::notEqual, notInt);
2525 
2526   // itos
2527   {
2528     __ pop(itos);
2529     if (!is_static) pop_and_check_object(obj);
2530     __ movl(lo, rax);
2531     if (!is_static) {
2532       patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx, true, byte_no);
2533     }
2534     __ jmp(Done);
2535   }
2536 
2537   __ bind(notInt);
2538   __ cmpl(flags, atos);
2539   __ jcc(Assembler::notEqual, notObj);
2540 
2541   // atos
2542   {
2543     __ pop(atos);
2544     if (!is_static) pop_and_check_object(obj);
2545     do_oop_store(_masm, lo, rax, _bs->kind(), false);
2546     if (!is_static) {
2547       patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx, true, byte_no);
2548     }
2549     __ jmp(Done);
2550   }
2551 
2552   __ bind(notObj);
2553   __ cmpl(flags, ctos);
2554   __ jcc(Assembler::notEqual, notChar);
2555 
2556   // ctos
2557   {
2558     __ pop(ctos);
2559     if (!is_static) pop_and_check_object(obj);
2560     __ movw(lo, rax);
2561     if (!is_static) {
2562       patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx, true, byte_no);
2563     }
2564     __ jmp(Done);
2565   }
2566 
2567   __ bind(notChar);
2568   __ cmpl(flags, stos);
2569   __ jcc(Assembler::notEqual, notShort);
2570 
2571   // stos
2572   {
2573     __ pop(stos);
2574     if (!is_static) pop_and_check_object(obj);
2575     __ movw(lo, rax);
2576     if (!is_static) {
2577       patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx, true, byte_no);
2578     }
2579     __ jmp(Done);
2580   }
2581 
2582   __ bind(notShort);
2583   __ cmpl(flags, ltos);
2584   __ jcc(Assembler::notEqual, notLong);
2585 
2586   // ltos
2587   {
2588     Label notVolatileLong;
2589     __ testl(rdx, rdx);
2590     __ jcc(Assembler::zero, notVolatileLong);
2591 
2592     __ pop(ltos);  // overwrites rdx, do this after testing volatile.
2593     if (!is_static) pop_and_check_object(obj);
2594 
2595     // Replace with real volatile test
2596     __ push(rdx);
2597     __ push(rax);                 // Must update atomically with FIST
2598     __ fild_d(Address(rsp,0));    // So load into FPU register
2599     __ fistp_d(lo);               // and put into memory atomically
2600     __ addptr(rsp, 2*wordSize);
2601     // volatile_barrier();
2602     volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2603                                                  Assembler::StoreStore));
2604     // Don't rewrite volatile version
2605     __ jmp(notVolatile);
2606 
2607     __ bind(notVolatileLong);
2608 
2609     __ pop(ltos);  // overwrites rdx
2610     if (!is_static) pop_and_check_object(obj);
2611     NOT_LP64(__ movptr(hi, rdx));
2612     __ movptr(lo, rax);
2613     if (!is_static) {
2614       patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx, true, byte_no);
2615     }
2616     __ jmp(notVolatile);
2617   }
2618 
2619   __ bind(notLong);
2620   __ cmpl(flags, ftos);
2621   __ jcc(Assembler::notEqual, notFloat);
2622 
2623   // ftos
2624   {
2625     __ pop(ftos);
2626     if (!is_static) pop_and_check_object(obj);
2627     __ fstp_s(lo);
2628     if (!is_static) {
2629       patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx, true, byte_no);
2630     }
2631     __ jmp(Done);
2632   }
2633 
2634   __ bind(notFloat);
2635 #ifdef ASSERT
2636   __ cmpl(flags, dtos);
2637   __ jcc(Assembler::notEqual, notDouble);
2638 #endif
2639 
2640   // dtos
2641   {
2642     __ pop(dtos);
2643     if (!is_static) pop_and_check_object(obj);
2644     __ fstp_d(lo);
2645     if (!is_static) {
2646       patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx, true, byte_no);
2647     }
2648     __ jmp(Done);
2649   }
2650 
2651 #ifdef ASSERT
2652   __ bind(notDouble);
2653   __ stop("Bad state");
2654 #endif
2655 
2656   __ bind(Done);
2657 
2658   // Check for volatile store
2659   __ testl(rdx, rdx);
2660   __ jcc(Assembler::zero, notVolatile);
2661   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2662                                                Assembler::StoreStore));
2663   __ bind(notVolatile);
2664 }
2665 
2666 
2667 void TemplateTable::putfield(int byte_no) {
2668   putfield_or_static(byte_no, false);
2669 }
2670 
2671 
2672 void TemplateTable::putstatic(int byte_no) {
2673   putfield_or_static(byte_no, true);
2674 }
2675 
2676 void TemplateTable::jvmti_post_fast_field_mod() {
2677   if (JvmtiExport::can_post_field_modification()) {
2678     // Check to see if a field modification watch has been set before we take
2679     // the time to call into the VM.
2680     Label L2;
2681      __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2682      __ testl(rcx,rcx);
2683      __ jcc(Assembler::zero, L2);
2684      __ pop_ptr(rbx);               // copy the object pointer from tos
2685      __ verify_oop(rbx);
2686      __ push_ptr(rbx);              // put the object pointer back on tos
2687 
2688      // Save tos values before call_VM() clobbers them. Since we have
2689      // to do it for every data type, we use the saved values as the
2690      // jvalue object.
2691      switch (bytecode()) {          // load values into the jvalue object
2692      case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
2693      case Bytecodes::_fast_bputfield: // fall through
2694      case Bytecodes::_fast_zputfield: // fall through
2695      case Bytecodes::_fast_sputfield: // fall through
2696      case Bytecodes::_fast_cputfield: // fall through
2697      case Bytecodes::_fast_iputfield: __ push_i(rax); break;
2698      case Bytecodes::_fast_dputfield: __ push_d(); break;
2699      case Bytecodes::_fast_fputfield: __ push_f(); break;
2700      case Bytecodes::_fast_lputfield: __ push_l(rax); break;
2701 
2702      default:
2703        ShouldNotReachHere();
2704      }
2705      __ mov(rcx, rsp);              // points to jvalue on the stack
2706      // access constant pool cache entry
2707      __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2708      __ verify_oop(rbx);
2709      // rbx,: object pointer copied above
2710      // rax,: cache entry pointer
2711      // rcx: jvalue object on the stack
2712      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2713 
2714      switch (bytecode()) {             // restore tos values
2715      case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
2716      case Bytecodes::_fast_bputfield: // fall through
2717      case Bytecodes::_fast_zputfield: // fall through
2718      case Bytecodes::_fast_sputfield: // fall through
2719      case Bytecodes::_fast_cputfield: // fall through
2720      case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
2721      case Bytecodes::_fast_dputfield: __ pop_d(); break;
2722      case Bytecodes::_fast_fputfield: __ pop_f(); break;
2723      case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
2724      }
2725      __ bind(L2);
2726   }
2727 }
2728 
2729 void TemplateTable::fast_storefield(TosState state) {
2730   transition(state, vtos);
2731 
2732   ByteSize base = ConstantPoolCache::base_offset();
2733 
2734   jvmti_post_fast_field_mod();
2735 
2736   // access constant pool cache
2737   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2738 
2739   // test for volatile with rdx but rdx is tos register for lputfield.
2740   if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2741   __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2742                        ConstantPoolCacheEntry::flags_offset())));
2743 
2744   // replace index with field offset from cache entry
2745   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2746 
2747   // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2748   // volatile_barrier( );
2749 
2750   Label notVolatile, Done;
2751   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2752   __ andl(rdx, 0x1);
2753   // Check for volatile store
2754   __ testl(rdx, rdx);
2755   __ jcc(Assembler::zero, notVolatile);
2756 
2757   if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2758 
2759   // Get object from stack
2760   pop_and_check_object(rcx);
2761 
2762   // field addresses
2763   const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2764   const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2765 
2766   // access field
2767   switch (bytecode()) {
2768     case Bytecodes::_fast_zputfield: __ andl(rax, 0x1);  // boolean is true if LSB is 1
2769     // fall through to bputfield
2770     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2771     case Bytecodes::_fast_sputfield: // fall through
2772     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2773     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2774     case Bytecodes::_fast_lputfield:
2775       NOT_LP64(__ movptr(hi, rdx));
2776       __ movptr(lo, rax);
2777       break;
2778     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2779     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2780     case Bytecodes::_fast_aputfield: {
2781       do_oop_store(_masm, lo, rax, _bs->kind(), false);
2782       break;
2783     }
2784     default:
2785       ShouldNotReachHere();
2786   }
2787 
2788   Label done;
2789   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2790                                                Assembler::StoreStore));
2791   // Barriers are so large that short branch doesn't reach!
2792   __ jmp(done);
2793 
2794   // Same code as above, but don't need rdx to test for volatile.
2795   __ bind(notVolatile);
2796 
2797   if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2798 
2799   // Get object from stack
2800   pop_and_check_object(rcx);
2801 
2802   // access field
2803   switch (bytecode()) {
2804     case Bytecodes::_fast_zputfield: __ andl(rax, 0x1);  // boolean is true if LSB is 1
2805     // fall through to bputfield
2806     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2807     case Bytecodes::_fast_sputfield: // fall through
2808     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2809     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2810     case Bytecodes::_fast_lputfield:
2811       NOT_LP64(__ movptr(hi, rdx));
2812       __ movptr(lo, rax);
2813       break;
2814     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2815     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2816     case Bytecodes::_fast_aputfield: {
2817       do_oop_store(_masm, lo, rax, _bs->kind(), false);
2818       break;
2819     }
2820     default:
2821       ShouldNotReachHere();
2822   }
2823   __ bind(done);
2824 }
2825 
2826 
2827 void TemplateTable::fast_accessfield(TosState state) {
2828   transition(atos, state);
2829 
2830   // do the JVMTI work here to avoid disturbing the register state below
2831   if (JvmtiExport::can_post_field_access()) {
2832     // Check to see if a field access watch has been set before we take
2833     // the time to call into the VM.
2834     Label L1;
2835     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2836     __ testl(rcx,rcx);
2837     __ jcc(Assembler::zero, L1);
2838     // access constant pool cache entry
2839     __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2840     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
2841     __ verify_oop(rax);
2842     // rax,: object pointer copied above
2843     // rcx: cache entry pointer
2844     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2845     __ pop_ptr(rax);   // restore object pointer
2846     __ bind(L1);
2847   }
2848 
2849   // access constant pool cache
2850   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2851   // replace index with field offset from cache entry
2852   __ movptr(rbx, Address(rcx,
2853                          rbx,
2854                          Address::times_ptr,
2855                          in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2856 
2857 
2858   // rax,: object
2859   __ verify_oop(rax);
2860   __ null_check(rax);
2861   // field addresses
2862   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2863   const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2864 
2865   // access field
2866   switch (bytecode()) {
2867     case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo );                 break;
2868     case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo );      break;
2869     case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo );    break;
2870     case Bytecodes::_fast_igetfield: __ movl(rax, lo);                    break;
2871     case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten");  break;
2872     case Bytecodes::_fast_fgetfield: __ fld_s(lo);                        break;
2873     case Bytecodes::_fast_dgetfield: __ fld_d(lo);                        break;
2874     case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2875     default:
2876       ShouldNotReachHere();
2877   }
2878 
2879   // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2880   // volatile_barrier( );
2881 }
2882 
2883 void TemplateTable::fast_xaccess(TosState state) {
2884   transition(vtos, state);
2885   // get receiver
2886   __ movptr(rax, aaddress(0));
2887   // access constant pool cache
2888   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2889   __ movptr(rbx, Address(rcx,
2890                          rdx,
2891                          Address::times_ptr,
2892                          in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2893   // make sure exception is reported in correct bcp range (getfield is next instruction)
2894   __ increment(rsi);
2895   __ null_check(rax);
2896   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2897   if (state == itos) {
2898     __ movl(rax, lo);
2899   } else if (state == atos) {
2900     __ movptr(rax, lo);
2901     __ verify_oop(rax);
2902   } else if (state == ftos) {
2903     __ fld_s(lo);
2904   } else {
2905     ShouldNotReachHere();
2906   }
2907   __ decrement(rsi);
2908 }
2909 
2910 
2911 
2912 //----------------------------------------------------------------------------------------------------
2913 // Calls
2914 
2915 void TemplateTable::count_calls(Register method, Register temp) {
2916   // implemented elsewhere
2917   ShouldNotReachHere();
2918 }
2919 
2920 
2921 void TemplateTable::prepare_invoke(int byte_no,
2922                                    Register method,  // linked method (or i-klass)
2923                                    Register index,   // itable index, MethodType, etc.
2924                                    Register recv,    // if caller wants to see it
2925                                    Register flags    // if caller wants to test it
2926                                    ) {
2927   // determine flags
2928   const Bytecodes::Code code = bytecode();
2929   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
2930   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
2931   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
2932   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2933   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2934   const bool load_receiver       = (recv  != noreg);
2935   const bool save_flags          = (flags != noreg);
2936   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2937   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
2938   assert(flags == noreg || flags == rdx, "");
2939   assert(recv  == noreg || recv  == rcx, "");
2940 
2941   // setup registers & access constant pool cache
2942   if (recv  == noreg)  recv  = rcx;
2943   if (flags == noreg)  flags = rdx;
2944   assert_different_registers(method, index, recv, flags);
2945 
2946   // save 'interpreter return address'
2947   __ save_bcp();
2948 
2949   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2950 
2951   // maybe push appendix to arguments (just before return address)
2952   if (is_invokedynamic || is_invokehandle) {
2953     Label L_no_push;
2954     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
2955     __ jccb(Assembler::zero, L_no_push);
2956     // Push the appendix as a trailing parameter.
2957     // This must be done before we get the receiver,
2958     // since the parameter_size includes it.
2959     __ push(rbx);
2960     __ mov(rbx, index);
2961     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2962     __ load_resolved_reference_at_index(index, rbx);
2963     __ pop(rbx);
2964     __ push(index);  // push appendix (MethodType, CallSite, etc.)
2965     __ bind(L_no_push);
2966   }
2967 
2968   // load receiver if needed (note: no return address pushed yet)
2969   if (load_receiver) {
2970     __ movl(recv, flags);
2971     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
2972     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
2973     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
2974     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
2975     __ movptr(recv, recv_addr);
2976     __ verify_oop(recv);
2977   }
2978 
2979   if (save_flags) {
2980     __ mov(rsi, flags);
2981   }
2982 
2983   // compute return type
2984   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2985   // Make sure we don't need to mask flags after the above shift
2986   ConstantPoolCacheEntry::verify_tos_state_shift();
2987   // load return address
2988   {
2989     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2990     ExternalAddress table(table_addr);
2991     __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2992   }
2993 
2994   // push return address
2995   __ push(flags);
2996 
2997   // Restore flags value from the constant pool cache, and restore rsi
2998   // for later null checks.  rsi is the bytecode pointer
2999   if (save_flags) {
3000     __ mov(flags, rsi);
3001     __ restore_bcp();
3002   }
3003 }
3004 
3005 
3006 void TemplateTable::invokevirtual_helper(Register index,
3007                                          Register recv,
3008                                          Register flags) {
3009   // Uses temporary registers rax, rdx
3010   assert_different_registers(index, recv, rax, rdx);
3011   assert(index == rbx, "");
3012   assert(recv  == rcx, "");
3013 
3014   // Test for an invoke of a final method
3015   Label notFinal;
3016   __ movl(rax, flags);
3017   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3018   __ jcc(Assembler::zero, notFinal);
3019 
3020   const Register method = index;  // method must be rbx
3021   assert(method == rbx,
3022          "Method* must be rbx for interpreter calling convention");
3023 
3024   // do the call - the index is actually the method to call
3025   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3026 
3027   // It's final, need a null check here!
3028   __ null_check(recv);
3029 
3030   // profile this call
3031   __ profile_final_call(rax);
3032   __ profile_arguments_type(rax, method, rsi, true);
3033 
3034   __ jump_from_interpreted(method, rax);
3035 
3036   __ bind(notFinal);
3037 
3038   // get receiver klass
3039   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3040   __ load_klass(rax, recv);
3041 
3042   // profile this call
3043   __ profile_virtual_call(rax, rdi, rdx);
3044 
3045   // get target Method* & entry point
3046   __ lookup_virtual_method(rax, index, method);
3047   __ profile_arguments_type(rdx, method, rsi, true);
3048   __ jump_from_interpreted(method, rdx);
3049 }
3050 
3051 
3052 void TemplateTable::invokevirtual(int byte_no) {
3053   transition(vtos, vtos);
3054   assert(byte_no == f2_byte, "use this argument");
3055   prepare_invoke(byte_no,
3056                  rbx,    // method or vtable index
3057                  noreg,  // unused itable index
3058                  rcx, rdx); // recv, flags
3059 
3060   // rbx: index
3061   // rcx: receiver
3062   // rdx: flags
3063 
3064   invokevirtual_helper(rbx, rcx, rdx);
3065 }
3066 
3067 
3068 void TemplateTable::invokespecial(int byte_no) {
3069   transition(vtos, vtos);
3070   assert(byte_no == f1_byte, "use this argument");
3071   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3072                  rcx);  // get receiver also for null check
3073   __ verify_oop(rcx);
3074   __ null_check(rcx);
3075   // do the call
3076   __ profile_call(rax);
3077   __ profile_arguments_type(rax, rbx, rsi, false);
3078   __ jump_from_interpreted(rbx, rax);
3079 }
3080 
3081 
3082 void TemplateTable::invokestatic(int byte_no) {
3083   transition(vtos, vtos);
3084   assert(byte_no == f1_byte, "use this argument");
3085   prepare_invoke(byte_no, rbx);  // get f1 Method*
3086   // do the call
3087   __ profile_call(rax);
3088   __ profile_arguments_type(rax, rbx, rsi, false);
3089   __ jump_from_interpreted(rbx, rax);
3090 }
3091 
3092 
3093 void TemplateTable::fast_invokevfinal(int byte_no) {
3094   transition(vtos, vtos);
3095   assert(byte_no == f2_byte, "use this argument");
3096   __ stop("fast_invokevfinal not used on x86");
3097 }
3098 
3099 
3100 void TemplateTable::invokeinterface(int byte_no) {
3101   transition(vtos, vtos);
3102   assert(byte_no == f1_byte, "use this argument");
3103   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 itable index
3104                  rcx, rdx); // recv, flags
3105 
3106   // rax: interface klass (from f1)
3107   // rbx: itable index (from f2)
3108   // rcx: receiver
3109   // rdx: flags
3110 
3111   // Special case of invokeinterface called for virtual method of
3112   // java.lang.Object.  See cpCacheOop.cpp for details.
3113   // This code isn't produced by javac, but could be produced by
3114   // another compliant java compiler.
3115   Label notMethod;
3116   __ movl(rdi, rdx);
3117   __ andl(rdi, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3118   __ jcc(Assembler::zero, notMethod);
3119 
3120   invokevirtual_helper(rbx, rcx, rdx);
3121   __ bind(notMethod);
3122 
3123   // Get receiver klass into rdx - also a null check
3124   __ restore_locals();  // restore rdi
3125   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3126   __ load_klass(rdx, rcx);
3127 
3128   // profile this call
3129   __ profile_virtual_call(rdx, rsi, rdi);
3130 
3131   Label no_such_interface, no_such_method;
3132 
3133   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3134                              rdx, rax, rbx,
3135                              // outputs: method, scan temp. reg
3136                              rbx, rsi,
3137                              no_such_interface);
3138 
3139   // rbx: Method* to call
3140   // rcx: receiver
3141   // Check for abstract method error
3142   // Note: This should be done more efficiently via a throw_abstract_method_error
3143   //       interpreter entry point and a conditional jump to it in case of a null
3144   //       method.
3145   __ testptr(rbx, rbx);
3146   __ jcc(Assembler::zero, no_such_method);
3147 
3148   __ profile_arguments_type(rdx, rbx, rsi, true);
3149 
3150   // do the call
3151   // rcx: receiver
3152   // rbx,: Method*
3153   __ jump_from_interpreted(rbx, rdx);
3154   __ should_not_reach_here();
3155 
3156   // exception handling code follows...
3157   // note: must restore interpreter registers to canonical
3158   //       state for exception handling to work correctly!
3159 
3160   __ bind(no_such_method);
3161   // throw exception
3162   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3163   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
3164   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3165   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3166   // the call_VM checks for exception, so we should never return here.
3167   __ should_not_reach_here();
3168 
3169   __ bind(no_such_interface);
3170   // throw exception
3171   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3172   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
3173   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3174   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3175                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3176   // the call_VM checks for exception, so we should never return here.
3177   __ should_not_reach_here();
3178 }
3179 
3180 void TemplateTable::invokehandle(int byte_no) {
3181   transition(vtos, vtos);
3182   assert(byte_no == f1_byte, "use this argument");
3183   const Register rbx_method = rbx;
3184   const Register rax_mtype  = rax;
3185   const Register rcx_recv   = rcx;
3186   const Register rdx_flags  = rdx;
3187 
3188   if (!EnableInvokeDynamic) {
3189     // rewriter does not generate this bytecode
3190     __ should_not_reach_here();
3191     return;
3192   }
3193 
3194   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3195   __ verify_method_ptr(rbx_method);
3196   __ verify_oop(rcx_recv);
3197   __ null_check(rcx_recv);
3198 
3199   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3200   // rbx: MH.invokeExact_MT method (from f2)
3201 
3202   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3203 
3204   // FIXME: profile the LambdaForm also
3205   __ profile_final_call(rax);
3206   __ profile_arguments_type(rdx, rbx_method, rsi, true);
3207 
3208   __ jump_from_interpreted(rbx_method, rdx);
3209 }
3210 
3211 
3212 void TemplateTable::invokedynamic(int byte_no) {
3213   transition(vtos, vtos);
3214   assert(byte_no == f1_byte, "use this argument");
3215 
3216   if (!EnableInvokeDynamic) {
3217     // We should not encounter this bytecode if !EnableInvokeDynamic.
3218     // The verifier will stop it.  However, if we get past the verifier,
3219     // this will stop the thread in a reasonable way, without crashing the JVM.
3220     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3221                      InterpreterRuntime::throw_IncompatibleClassChangeError));
3222     // the call_VM checks for exception, so we should never return here.
3223     __ should_not_reach_here();
3224     return;
3225   }
3226 
3227   const Register rbx_method   = rbx;
3228   const Register rax_callsite = rax;
3229 
3230   prepare_invoke(byte_no, rbx_method, rax_callsite);
3231 
3232   // rax: CallSite object (from cpool->resolved_references[f1])
3233   // rbx: MH.linkToCallSite method (from f2)
3234 
3235   // Note:  rax_callsite is already pushed by prepare_invoke
3236 
3237   // %%% should make a type profile for any invokedynamic that takes a ref argument
3238   // profile this call
3239   __ profile_call(rsi);
3240   __ profile_arguments_type(rdx, rbx, rsi, false);
3241 
3242   __ verify_oop(rax_callsite);
3243 
3244   __ jump_from_interpreted(rbx_method, rdx);
3245 }
3246 
3247 //----------------------------------------------------------------------------------------------------
3248 // Allocation
3249 
3250 void TemplateTable::_new() {
3251   transition(vtos, atos);
3252   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3253   Label slow_case;
3254   Label slow_case_no_pop;
3255   Label done;
3256   Label initialize_header;
3257   Label initialize_object;  // including clearing the fields
3258   Label allocate_shared;
3259 
3260   __ get_cpool_and_tags(rcx, rax);
3261 
3262   // Make sure the class we're about to instantiate has been resolved.
3263   // This is done before loading InstanceKlass to be consistent with the order
3264   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3265   const int tags_offset = Array<u1>::base_offset_in_bytes();
3266   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3267   __ jcc(Assembler::notEqual, slow_case_no_pop);
3268 
3269   // get InstanceKlass
3270   __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3271   __ push(rcx);  // save the contexts of klass for initializing the header
3272 
3273   // make sure klass is initialized & doesn't have finalizer
3274   // make sure klass is fully initialized
3275   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3276   __ jcc(Assembler::notEqual, slow_case);
3277 
3278   // get instance_size in InstanceKlass (scaled to a count of bytes)
3279   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3280   // test to see if it has a finalizer or is malformed in some way
3281   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3282   __ jcc(Assembler::notZero, slow_case);
3283 
3284   //
3285   // Allocate the instance
3286   // 1) Try to allocate in the TLAB
3287   // 2) if fail and the object is large allocate in the shared Eden
3288   // 3) if the above fails (or is not applicable), go to a slow case
3289   // (creates a new TLAB, etc.)
3290 
3291   const bool allow_shared_alloc =
3292     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3293 
3294   const Register thread = rcx;
3295   if (UseTLAB || allow_shared_alloc) {
3296     __ get_thread(thread);
3297   }
3298 
3299   if (UseTLAB) {
3300     __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3301     __ lea(rbx, Address(rax, rdx, Address::times_1));
3302     __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3303     __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3304     __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3305     if (ZeroTLAB) {
3306       // the fields have been already cleared
3307       __ jmp(initialize_header);
3308     } else {
3309       // initialize both the header and fields
3310       __ jmp(initialize_object);
3311     }
3312   }
3313 
3314   // Allocation in the shared Eden, if allowed.
3315   //
3316   // rdx: instance size in bytes
3317   if (allow_shared_alloc) {
3318     __ bind(allocate_shared);
3319 
3320     ExternalAddress heap_top((address)Universe::heap()->top_addr());
3321 
3322     Label retry;
3323     __ bind(retry);
3324     __ movptr(rax, heap_top);
3325     __ lea(rbx, Address(rax, rdx, Address::times_1));
3326     __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3327     __ jcc(Assembler::above, slow_case);
3328 
3329     // Compare rax, with the top addr, and if still equal, store the new
3330     // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3331     // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3332     //
3333     // rax,: object begin
3334     // rbx,: object end
3335     // rdx: instance size in bytes
3336     __ locked_cmpxchgptr(rbx, heap_top);
3337 
3338     // if someone beat us on the allocation, try again, otherwise continue
3339     __ jcc(Assembler::notEqual, retry);
3340 
3341     __ incr_allocated_bytes(thread, rdx, 0);
3342   }
3343 
3344   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3345     // The object is initialized before the header.  If the object size is
3346     // zero, go directly to the header initialization.
3347     __ bind(initialize_object);
3348     __ decrement(rdx, sizeof(oopDesc));
3349     __ jcc(Assembler::zero, initialize_header);
3350 
3351     // Initialize topmost object field, divide rdx by 8, check if odd and
3352     // test if zero.
3353     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
3354     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3355 
3356     // rdx must have been multiple of 8
3357 #ifdef ASSERT
3358     // make sure rdx was multiple of 8
3359     Label L;
3360     // Ignore partial flag stall after shrl() since it is debug VM
3361     __ jccb(Assembler::carryClear, L);
3362     __ stop("object size is not multiple of 2 - adjust this code");
3363     __ bind(L);
3364     // rdx must be > 0, no extra check needed here
3365 #endif
3366 
3367     // initialize remaining object fields: rdx was a multiple of 8
3368     { Label loop;
3369     __ bind(loop);
3370     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3371     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3372     __ decrement(rdx);
3373     __ jcc(Assembler::notZero, loop);
3374     }
3375 
3376     // initialize object header only.
3377     __ bind(initialize_header);
3378     if (UseBiasedLocking) {
3379       __ pop(rcx);   // get saved klass back in the register.
3380       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3381       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3382     } else {
3383       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3384                 (int32_t)markOopDesc::prototype()); // header
3385       __ pop(rcx);   // get saved klass back in the register.
3386     }
3387     __ store_klass(rax, rcx);  // klass
3388 
3389     {
3390       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3391       // Trigger dtrace event for fastpath
3392       __ push(atos);
3393       __ call_VM_leaf(
3394            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3395       __ pop(atos);
3396     }
3397 
3398     __ jmp(done);
3399   }
3400 
3401   // slow case
3402   __ bind(slow_case);
3403   __ pop(rcx);   // restore stack pointer to what it was when we came in.
3404   __ bind(slow_case_no_pop);
3405   __ get_constant_pool(rax);
3406   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3407   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3408 
3409   // continue
3410   __ bind(done);
3411 }
3412 
3413 
3414 void TemplateTable::newarray() {
3415   transition(itos, atos);
3416   __ push_i(rax);                                 // make sure everything is on the stack
3417   __ load_unsigned_byte(rdx, at_bcp(1));
3418   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3419   __ pop_i(rdx);                                  // discard size
3420 }
3421 
3422 
3423 void TemplateTable::anewarray() {
3424   transition(itos, atos);
3425   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3426   __ get_constant_pool(rcx);
3427   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3428 }
3429 
3430 
3431 void TemplateTable::arraylength() {
3432   transition(atos, itos);
3433   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3434   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3435 }
3436 
3437 
3438 void TemplateTable::checkcast() {
3439   transition(atos, atos);
3440   Label done, is_null, ok_is_subtype, quicked, resolved;
3441   __ testptr(rax, rax);   // Object is in EAX
3442   __ jcc(Assembler::zero, is_null);
3443 
3444   // Get cpool & tags index
3445   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3446   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3447   // See if bytecode has already been quicked
3448   __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class);
3449   __ jcc(Assembler::equal, quicked);
3450 
3451   __ push(atos);
3452   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3453   // vm_result_2 has metadata result
3454   // borrow rdi from locals
3455   __ get_thread(rdi);
3456   __ get_vm_result_2(rax, rdi);
3457   __ restore_locals();
3458   __ pop_ptr(rdx);
3459   __ jmpb(resolved);
3460 
3461   // Get superklass in EAX and subklass in EBX
3462   __ bind(quicked);
3463   __ mov(rdx, rax);          // Save object in EDX; EAX needed for subtype check
3464   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool)));
3465 
3466   __ bind(resolved);
3467   __ load_klass(rbx, rdx);
3468 
3469   // Generate subtype check.  Blows ECX.  Resets EDI.  Object in EDX.
3470   // Superklass in EAX.  Subklass in EBX.
3471   __ gen_subtype_check( rbx, ok_is_subtype );
3472 
3473   // Come here on failure
3474   __ push(rdx);
3475   // object is at TOS
3476   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3477 
3478   // Come here on success
3479   __ bind(ok_is_subtype);
3480   __ mov(rax,rdx);           // Restore object in EDX
3481 
3482   // Collect counts on whether this check-cast sees NULLs a lot or not.
3483   if (ProfileInterpreter) {
3484     __ jmp(done);
3485     __ bind(is_null);
3486     __ profile_null_seen(rcx);
3487   } else {
3488     __ bind(is_null);   // same as 'done'
3489   }
3490   __ bind(done);
3491 }
3492 
3493 
3494 void TemplateTable::instanceof() {
3495   transition(atos, itos);
3496   Label done, is_null, ok_is_subtype, quicked, resolved;
3497   __ testptr(rax, rax);
3498   __ jcc(Assembler::zero, is_null);
3499 
3500   // Get cpool & tags index
3501   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3502   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3503   // See if bytecode has already been quicked
3504   __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class);
3505   __ jcc(Assembler::equal, quicked);
3506 
3507   __ push(atos);
3508   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3509   // vm_result_2 has metadata result
3510   // borrow rdi from locals
3511   __ get_thread(rdi);
3512   __ get_vm_result_2(rax, rdi);
3513   __ restore_locals();
3514   __ pop_ptr(rdx);
3515   __ load_klass(rdx, rdx);
3516   __ jmp(resolved);
3517 
3518   // Get superklass in EAX and subklass in EDX
3519   __ bind(quicked);
3520   __ load_klass(rdx, rax);
3521   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool)));
3522 
3523   __ bind(resolved);
3524 
3525   // Generate subtype check.  Blows ECX.  Resets EDI.
3526   // Superklass in EAX.  Subklass in EDX.
3527   __ gen_subtype_check( rdx, ok_is_subtype );
3528 
3529   // Come here on failure
3530   __ xorl(rax,rax);
3531   __ jmpb(done);
3532   // Come here on success
3533   __ bind(ok_is_subtype);
3534   __ movl(rax, 1);
3535 
3536   // Collect counts on whether this test sees NULLs a lot or not.
3537   if (ProfileInterpreter) {
3538     __ jmp(done);
3539     __ bind(is_null);
3540     __ profile_null_seen(rcx);
3541   } else {
3542     __ bind(is_null);   // same as 'done'
3543   }
3544   __ bind(done);
3545   // rax, = 0: obj == NULL or  obj is not an instanceof the specified klass
3546   // rax, = 1: obj != NULL and obj is     an instanceof the specified klass
3547 }
3548 
3549 
3550 //----------------------------------------------------------------------------------------------------
3551 // Breakpoints
3552 void TemplateTable::_breakpoint() {
3553 
3554   // Note: We get here even if we are single stepping..
3555   // jbug inists on setting breakpoints at every bytecode
3556   // even if we are in single step mode.
3557 
3558   transition(vtos, vtos);
3559 
3560   // get the unpatched byte code
3561   __ get_method(rcx);
3562   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3563   __ mov(rbx, rax);
3564 
3565   // post the breakpoint event
3566   __ get_method(rcx);
3567   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3568 
3569   // complete the execution of original bytecode
3570   __ dispatch_only_normal(vtos);
3571 }
3572 
3573 
3574 //----------------------------------------------------------------------------------------------------
3575 // Exceptions
3576 
3577 void TemplateTable::athrow() {
3578   transition(atos, vtos);
3579   __ null_check(rax);
3580   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3581 }
3582 
3583 
3584 //----------------------------------------------------------------------------------------------------
3585 // Synchronization
3586 //
3587 // Note: monitorenter & exit are symmetric routines; which is reflected
3588 //       in the assembly code structure as well
3589 //
3590 // Stack layout:
3591 //
3592 // [expressions  ] <--- rsp               = expression stack top
3593 // ..
3594 // [expressions  ]
3595 // [monitor entry] <--- monitor block top = expression stack bot
3596 // ..
3597 // [monitor entry]
3598 // [frame data   ] <--- monitor block bot
3599 // ...
3600 // [saved rbp,    ] <--- rbp,
3601 
3602 
3603 void TemplateTable::monitorenter() {
3604   transition(atos, vtos);
3605 
3606   // check for NULL object
3607   __ null_check(rax);
3608 
3609   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3610   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3611   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3612   Label allocated;
3613 
3614   // initialize entry pointer
3615   __ xorl(rdx, rdx);                             // points to free slot or NULL
3616 
3617   // find a free slot in the monitor block (result in rdx)
3618   { Label entry, loop, exit;
3619     __ movptr(rcx, monitor_block_top);           // points to current entry, starting with top-most entry
3620 
3621     __ lea(rbx, monitor_block_bot);              // points to word before bottom of monitor block
3622     __ jmpb(entry);
3623 
3624     __ bind(loop);
3625     __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);  // check if current entry is used
3626     __ cmovptr(Assembler::equal, rdx, rcx);      // if not used then remember entry in rdx
3627     __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3628     __ jccb(Assembler::equal, exit);             // if same object then stop searching
3629     __ addptr(rcx, entry_size);                  // otherwise advance to next entry
3630     __ bind(entry);
3631     __ cmpptr(rcx, rbx);                         // check if bottom reached
3632     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3633     __ bind(exit);
3634   }
3635 
3636   __ testptr(rdx, rdx);                          // check if a slot has been found
3637   __ jccb(Assembler::notZero, allocated);        // if found, continue with that one
3638 
3639   // allocate one if there's no free slot
3640   { Label entry, loop;
3641     // 1. compute new pointers                   // rsp: old expression stack top
3642     __ movptr(rdx, monitor_block_bot);           // rdx: old expression stack bottom
3643     __ subptr(rsp, entry_size);                  // move expression stack top
3644     __ subptr(rdx, entry_size);                  // move expression stack bottom
3645     __ mov(rcx, rsp);                            // set start value for copy loop
3646     __ movptr(monitor_block_bot, rdx);           // set new monitor block top
3647     __ jmp(entry);
3648     // 2. move expression stack contents
3649     __ bind(loop);
3650     __ movptr(rbx, Address(rcx, entry_size));    // load expression stack word from old location
3651     __ movptr(Address(rcx, 0), rbx);             // and store it at new location
3652     __ addptr(rcx, wordSize);                    // advance to next word
3653     __ bind(entry);
3654     __ cmpptr(rcx, rdx);                         // check if bottom reached
3655     __ jcc(Assembler::notEqual, loop);           // if not at bottom then copy next word
3656   }
3657 
3658   // call run-time routine
3659   // rdx: points to monitor entry
3660   __ bind(allocated);
3661 
3662   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3663   // The object has already been poped from the stack, so the expression stack looks correct.
3664   __ increment(rsi);
3665 
3666   __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax);     // store object
3667   __ lock_object(rdx);
3668 
3669   // check to make sure this monitor doesn't cause stack overflow after locking
3670   __ save_bcp();  // in case of exception
3671   __ generate_stack_overflow_check(0);
3672 
3673   // The bcp has already been incremented. Just need to dispatch to next instruction.
3674   __ dispatch_next(vtos);
3675 }
3676 
3677 
3678 void TemplateTable::monitorexit() {
3679   transition(atos, vtos);
3680 
3681   // check for NULL object
3682   __ null_check(rax);
3683 
3684   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3685   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3686   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3687   Label found;
3688 
3689   // find matching slot
3690   { Label entry, loop;
3691     __ movptr(rdx, monitor_block_top);           // points to current entry, starting with top-most entry
3692     __ lea(rbx, monitor_block_bot);             // points to word before bottom of monitor block
3693     __ jmpb(entry);
3694 
3695     __ bind(loop);
3696     __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3697     __ jcc(Assembler::equal, found);             // if same object then stop searching
3698     __ addptr(rdx, entry_size);                  // otherwise advance to next entry
3699     __ bind(entry);
3700     __ cmpptr(rdx, rbx);                         // check if bottom reached
3701     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3702   }
3703 
3704   // error handling. Unlocking was not block-structured
3705   Label end;
3706   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3707   __ should_not_reach_here();
3708 
3709   // call run-time routine
3710   // rcx: points to monitor entry
3711   __ bind(found);
3712   __ push_ptr(rax);                                 // make sure object is on stack (contract with oopMaps)
3713   __ unlock_object(rdx);
3714   __ pop_ptr(rax);                                  // discard object
3715   __ bind(end);
3716 }
3717 
3718 
3719 //----------------------------------------------------------------------------------------------------
3720 // Wide instructions
3721 
3722 void TemplateTable::wide() {
3723   transition(vtos, vtos);
3724   __ load_unsigned_byte(rbx, at_bcp(1));
3725   ExternalAddress wtable((address)Interpreter::_wentry_point);
3726   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3727   // Note: the rsi increment step is part of the individual wide bytecode implementations
3728 }
3729 
3730 
3731 //----------------------------------------------------------------------------------------------------
3732 // Multi arrays
3733 
3734 void TemplateTable::multianewarray() {
3735   transition(vtos, atos);
3736   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3737   // last dim is on top of stack; we want address of first one:
3738   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3739   // the latter wordSize to point to the beginning of the array.
3740   __ lea(  rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3741   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax);     // pass in rax,
3742   __ load_unsigned_byte(rbx, at_bcp(3));
3743   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
3744 }
3745 
3746 #endif /* !CC_INTERP */