1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "interpreter/interp_masm.hpp"
  30 #include "interpreter/templateTable.hpp"
  31 #include "memory/universe.inline.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/objArrayKlass.hpp"
  34 #include "oops/oop.inline.hpp"
  35 #include "prims/methodHandles.hpp"
  36 #include "runtime/sharedRuntime.hpp"
  37 #include "runtime/stubRoutines.hpp"
  38 #include "runtime/synchronizer.hpp"
  39 #include "utilities/macros.hpp"
  40 
  41 #define __ _masm->
  42 
  43 // Global Register Names
  44 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  45 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  46 
  47 // Platform-dependent initialization
  48 void TemplateTable::pd_initialize() {
  49   // No x86 specific initialization
  50 }
  51 
  52 // Address Computation: local variables
  53 static inline Address iaddress(int n) {
  54   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  55 }
  56 
  57 static inline Address laddress(int n) {
  58   return iaddress(n + 1);
  59 }
  60 
  61 #ifndef _LP64
  62 static inline Address haddress(int n) {
  63   return iaddress(n + 0);
  64 }
  65 #endif
  66 
  67 static inline Address faddress(int n) {
  68   return iaddress(n);
  69 }
  70 
  71 static inline Address daddress(int n) {
  72   return laddress(n);
  73 }
  74 
  75 static inline Address aaddress(int n) {
  76   return iaddress(n);
  77 }
  78 
  79 static inline Address iaddress(Register r) {
  80   return Address(rlocals, r, Address::times_ptr);
  81 }
  82 
  83 static inline Address laddress(Register r) {
  84   return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
  85 }
  86 
  87 #ifndef _LP64
  88 static inline Address haddress(Register r)       {
  89   return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  90 }
  91 #endif
  92 
  93 static inline Address faddress(Register r) {
  94   return iaddress(r);
  95 }
  96 
  97 static inline Address daddress(Register r) {
  98   return laddress(r);
  99 }
 100 
 101 static inline Address aaddress(Register r) {
 102   return iaddress(r);
 103 }
 104 
 105 
 106 // expression stack
 107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
 108 // data beyond the rsp which is potentially unsafe in an MT environment;
 109 // an interrupt may overwrite that data.)
 110 static inline Address at_rsp   () {
 111   return Address(rsp, 0);
 112 }
 113 
 114 // At top of Java expression stack which may be different than esp().  It
 115 // isn't for category 1 objects.
 116 static inline Address at_tos   () {
 117   return Address(rsp,  Interpreter::expr_offset_in_bytes(0));
 118 }
 119 
 120 static inline Address at_tos_p1() {
 121   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
 122 }
 123 
 124 static inline Address at_tos_p2() {
 125   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
 126 }
 127 
 128 // Condition conversion
 129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 130   switch (cc) {
 131   case TemplateTable::equal        : return Assembler::notEqual;
 132   case TemplateTable::not_equal    : return Assembler::equal;
 133   case TemplateTable::less         : return Assembler::greaterEqual;
 134   case TemplateTable::less_equal   : return Assembler::greater;
 135   case TemplateTable::greater      : return Assembler::lessEqual;
 136   case TemplateTable::greater_equal: return Assembler::less;
 137   }
 138   ShouldNotReachHere();
 139   return Assembler::zero;
 140 }
 141 
 142 
 143 
 144 // Miscelaneous helper routines
 145 // Store an oop (or NULL) at the address described by obj.
 146 // If val == noreg this means store a NULL
 147 
 148 
 149 static void do_oop_store(InterpreterMacroAssembler* _masm,
 150                          Address obj,
 151                          Register val,
 152                          BarrierSet::Name barrier,
 153                          bool precise) {
 154   assert(val == noreg || val == rax, "parameter is just for looks");
 155   switch (barrier) {
 156 #if INCLUDE_ALL_GCS
 157     case BarrierSet::G1SATBCTLogging:
 158       {
 159         // flatten object address if needed
 160         // We do it regardless of precise because we need the registers
 161         if (obj.index() == noreg && obj.disp() == 0) {
 162           if (obj.base() != rdx) {
 163             __ movptr(rdx, obj.base());
 164           }
 165         } else {
 166           __ lea(rdx, obj);
 167         }
 168 
 169         Register rtmp    = LP64_ONLY(r8)         NOT_LP64(rsi);
 170         Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 171 
 172         NOT_LP64(__ get_thread(rcx));
 173         NOT_LP64(__ save_bcp());
 174 
 175         __ g1_write_barrier_pre(rdx /* obj */,
 176                                 rbx /* pre_val */,
 177                                 rthread /* thread */,
 178                                 rtmp  /* tmp */,
 179                                 val != noreg /* tosca_live */,
 180                                 false /* expand_call */);
 181         if (val == noreg) {
 182           __ store_heap_oop_null(Address(rdx, 0));
 183         } else {
 184           // G1 barrier needs uncompressed oop for region cross check.
 185           Register new_val = val;
 186           if (UseCompressedOops) {
 187             new_val = rbx;
 188             __ movptr(new_val, val);
 189           }
 190           __ store_heap_oop(Address(rdx, 0), val);
 191           __ g1_write_barrier_post(rdx /* store_adr */,
 192                                    new_val /* new_val */,
 193                                    rthread /* thread */,
 194                                    rtmp /* tmp */,
 195                                    rbx /* tmp2 */);
 196         }
 197         NOT_LP64( __ restore_bcp());
 198       }
 199       break;
 200 #endif // INCLUDE_ALL_GCS
 201     case BarrierSet::CardTableForRS:
 202     case BarrierSet::CardTableExtension:
 203       {
 204         if (val == noreg) {
 205           __ store_heap_oop_null(obj);
 206         } else {
 207           __ store_heap_oop(obj, val);
 208           // flatten object address if needed
 209           if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
 210             __ store_check(obj.base());
 211           } else {
 212             __ lea(rdx, obj);
 213             __ store_check(rdx);
 214           }
 215         }
 216       }
 217       break;
 218     case BarrierSet::ModRef:
 219       if (val == noreg) {
 220         __ store_heap_oop_null(obj);
 221       } else {
 222         __ store_heap_oop(obj, val);
 223       }
 224       break;
 225     default      :
 226       ShouldNotReachHere();
 227 
 228   }
 229 }
 230 
 231 Address TemplateTable::at_bcp(int offset) {
 232   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 233   return Address(rbcp, offset);
 234 }
 235 
 236 
 237 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 238                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 239                                    int byte_no) {
 240   if (!RewriteBytecodes)  return;
 241   Label L_patch_done;
 242 
 243   switch (bc) {
 244   case Bytecodes::_fast_aputfield:
 245   case Bytecodes::_fast_bputfield:
 246   case Bytecodes::_fast_zputfield:
 247   case Bytecodes::_fast_cputfield:
 248   case Bytecodes::_fast_dputfield:
 249   case Bytecodes::_fast_fputfield:
 250   case Bytecodes::_fast_iputfield:
 251   case Bytecodes::_fast_lputfield:
 252   case Bytecodes::_fast_sputfield:
 253     {
 254       // We skip bytecode quickening for putfield instructions when
 255       // the put_code written to the constant pool cache is zero.
 256       // This is required so that every execution of this instruction
 257       // calls out to InterpreterRuntime::resolve_get_put to do
 258       // additional, required work.
 259       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 260       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 261       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 262       __ movl(bc_reg, bc);
 263       __ cmpl(temp_reg, (int) 0);
 264       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 265     }
 266     break;
 267   default:
 268     assert(byte_no == -1, "sanity");
 269     // the pair bytecodes have already done the load.
 270     if (load_bc_into_bc_reg) {
 271       __ movl(bc_reg, bc);
 272     }
 273   }
 274 
 275   if (JvmtiExport::can_post_breakpoint()) {
 276     Label L_fast_patch;
 277     // if a breakpoint is present we can't rewrite the stream directly
 278     __ movzbl(temp_reg, at_bcp(0));
 279     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 280     __ jcc(Assembler::notEqual, L_fast_patch);
 281     __ get_method(temp_reg);
 282     // Let breakpoint table handling rewrite to quicker bytecode
 283     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 284 #ifndef ASSERT
 285     __ jmpb(L_patch_done);
 286 #else
 287     __ jmp(L_patch_done);
 288 #endif
 289     __ bind(L_fast_patch);
 290   }
 291 
 292 #ifdef ASSERT
 293   Label L_okay;
 294   __ load_unsigned_byte(temp_reg, at_bcp(0));
 295   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 296   __ jcc(Assembler::equal, L_okay);
 297   __ cmpl(temp_reg, bc_reg);
 298   __ jcc(Assembler::equal, L_okay);
 299   __ stop("patching the wrong bytecode");
 300   __ bind(L_okay);
 301 #endif
 302 
 303   // patch bytecode
 304   __ movb(at_bcp(0), bc_reg);
 305   __ bind(L_patch_done);
 306 }
 307 // Individual instructions
 308 
 309 
 310 void TemplateTable::nop() {
 311   transition(vtos, vtos);
 312   // nothing to do
 313 }
 314 
 315 void TemplateTable::shouldnotreachhere() {
 316   transition(vtos, vtos);
 317   __ stop("shouldnotreachhere bytecode");
 318 }
 319 
 320 void TemplateTable::aconst_null() {
 321   transition(vtos, atos);
 322   __ xorl(rax, rax);
 323 }
 324 
 325 void TemplateTable::iconst(int value) {
 326   transition(vtos, itos);
 327   if (value == 0) {
 328     __ xorl(rax, rax);
 329   } else {
 330     __ movl(rax, value);
 331   }
 332 }
 333 
 334 void TemplateTable::lconst(int value) {
 335   transition(vtos, ltos);
 336   if (value == 0) {
 337     __ xorl(rax, rax);
 338   } else {
 339     __ movl(rax, value);
 340   }
 341 #ifndef _LP64
 342   assert(value >= 0, "check this code");
 343   __ xorptr(rdx, rdx);
 344 #endif
 345 }
 346 
 347 
 348 
 349 void TemplateTable::fconst(int value) {
 350   transition(vtos, ftos);
 351   if (UseSSE >= 1) {
 352     static float one = 1.0f, two = 2.0f;
 353     switch (value) {
 354     case 0:
 355       __ xorps(xmm0, xmm0);
 356       break;
 357     case 1:
 358       __ movflt(xmm0, ExternalAddress((address) &one));
 359       break;
 360     case 2:
 361       __ movflt(xmm0, ExternalAddress((address) &two));
 362       break;
 363     default:
 364       ShouldNotReachHere();
 365       break;
 366     }
 367   } else {
 368 #ifdef _LP64
 369     ShouldNotReachHere();
 370 #else
 371            if (value == 0) { __ fldz();
 372     } else if (value == 1) { __ fld1();
 373     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 374     } else                 { ShouldNotReachHere();
 375     }
 376 #endif // _LP64
 377   }
 378 }
 379 
 380 void TemplateTable::dconst(int value) {
 381   transition(vtos, dtos);
 382   if (UseSSE >= 2) {
 383     static double one = 1.0;
 384     switch (value) {
 385     case 0:
 386       __ xorpd(xmm0, xmm0);
 387       break;
 388     case 1:
 389       __ movdbl(xmm0, ExternalAddress((address) &one));
 390       break;
 391     default:
 392       ShouldNotReachHere();
 393       break;
 394     }
 395   } else {
 396 #ifdef _LP64
 397     ShouldNotReachHere();
 398 #else
 399            if (value == 0) { __ fldz();
 400     } else if (value == 1) { __ fld1();
 401     } else                 { ShouldNotReachHere();
 402     }
 403 #endif
 404   }
 405 }
 406 
 407 void TemplateTable::bipush() {
 408   transition(vtos, itos);
 409   __ load_signed_byte(rax, at_bcp(1));
 410 }
 411 
 412 void TemplateTable::sipush() {
 413   transition(vtos, itos);
 414   __ load_unsigned_short(rax, at_bcp(1));
 415   __ bswapl(rax);
 416   __ sarl(rax, 16);
 417 }
 418 
 419 void TemplateTable::ldc(bool wide) {
 420   transition(vtos, vtos);
 421   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 422   Label call_ldc, notFloat, notClass, Done;
 423 
 424   if (wide) {
 425     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 426   } else {
 427     __ load_unsigned_byte(rbx, at_bcp(1));
 428   }
 429 
 430   __ get_cpool_and_tags(rcx, rax);
 431   const int base_offset = ConstantPool::header_size() * wordSize;
 432   const int tags_offset = Array<u1>::base_offset_in_bytes();
 433 
 434   // get type
 435   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 436 
 437   // unresolved class - get the resolved class
 438   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 439   __ jccb(Assembler::equal, call_ldc);
 440 
 441   // unresolved class in error state - call into runtime to throw the error
 442   // from the first resolution attempt
 443   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 444   __ jccb(Assembler::equal, call_ldc);
 445 
 446   // resolved class - need to call vm to get java mirror of the class
 447   __ cmpl(rdx, JVM_CONSTANT_Class);
 448   __ jcc(Assembler::notEqual, notClass);
 449 
 450   __ bind(call_ldc);
 451 
 452   __ movl(rarg, wide);
 453   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 454 
 455   __ push(atos);
 456   __ jmp(Done);
 457 
 458   __ bind(notClass);
 459   __ cmpl(rdx, JVM_CONSTANT_Float);
 460   __ jccb(Assembler::notEqual, notFloat);
 461 
 462   // ftos
 463   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 464   __ push(ftos);
 465   __ jmp(Done);
 466 
 467   __ bind(notFloat);
 468 #ifdef ASSERT
 469   {
 470     Label L;
 471     __ cmpl(rdx, JVM_CONSTANT_Integer);
 472     __ jcc(Assembler::equal, L);
 473     // String and Object are rewritten to fast_aldc
 474     __ stop("unexpected tag type in ldc");
 475     __ bind(L);
 476   }
 477 #endif
 478   // itos JVM_CONSTANT_Integer only
 479   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 480   __ push(itos);
 481   __ bind(Done);
 482 }
 483 
 484 // Fast path for caching oop constants.
 485 void TemplateTable::fast_aldc(bool wide) {
 486   transition(vtos, atos);
 487 
 488   Register result = rax;
 489   Register tmp = rdx;
 490   int index_size = wide ? sizeof(u2) : sizeof(u1);
 491 
 492   Label resolved;
 493 
 494   // We are resolved if the resolved reference cache entry contains a
 495   // non-null object (String, MethodType, etc.)
 496   assert_different_registers(result, tmp);
 497   __ get_cache_index_at_bcp(tmp, 1, index_size);
 498   __ load_resolved_reference_at_index(result, tmp);
 499   __ testl(result, result);
 500   __ jcc(Assembler::notZero, resolved);
 501 
 502   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 503 
 504   // first time invocation - must resolve first
 505   __ movl(tmp, (int)bytecode());
 506   __ call_VM(result, entry, tmp);
 507 
 508   __ bind(resolved);
 509 
 510   if (VerifyOops) {
 511     __ verify_oop(result);
 512   }
 513 }
 514 
 515 void TemplateTable::ldc2_w() {
 516   transition(vtos, vtos);
 517   Label Long, Done;
 518   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 519 
 520   __ get_cpool_and_tags(rcx, rax);
 521   const int base_offset = ConstantPool::header_size() * wordSize;
 522   const int tags_offset = Array<u1>::base_offset_in_bytes();
 523 
 524   // get type
 525   __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
 526           JVM_CONSTANT_Double);
 527   __ jccb(Assembler::notEqual, Long);
 528 
 529   // dtos
 530   __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
 531   __ push(dtos);
 532 
 533   __ jmpb(Done);
 534   __ bind(Long);
 535 
 536   // ltos
 537   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 538   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 539   __ push(ltos);
 540 
 541   __ bind(Done);
 542 }
 543 
 544 void TemplateTable::locals_index(Register reg, int offset) {
 545   __ load_unsigned_byte(reg, at_bcp(offset));
 546   __ negptr(reg);
 547 }
 548 
 549 void TemplateTable::iload() {
 550   iload_internal();
 551 }
 552 
 553 void TemplateTable::nofast_iload() {
 554   iload_internal(may_not_rewrite);
 555 }
 556 
 557 void TemplateTable::iload_internal(RewriteControl rc) {
 558   transition(vtos, itos);
 559   if (RewriteFrequentPairs && rc == may_rewrite) {
 560     Label rewrite, done;
 561     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 562     LP64_ONLY(assert(rbx != bc, "register damaged"));
 563 
 564     // get next byte
 565     __ load_unsigned_byte(rbx,
 566                           at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 567     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 568     // last two iloads in a pair.  Comparing against fast_iload means that
 569     // the next bytecode is neither an iload or a caload, and therefore
 570     // an iload pair.
 571     __ cmpl(rbx, Bytecodes::_iload);
 572     __ jcc(Assembler::equal, done);
 573 
 574     __ cmpl(rbx, Bytecodes::_fast_iload);
 575     __ movl(bc, Bytecodes::_fast_iload2);
 576 
 577     __ jccb(Assembler::equal, rewrite);
 578 
 579     // if _caload, rewrite to fast_icaload
 580     __ cmpl(rbx, Bytecodes::_caload);
 581     __ movl(bc, Bytecodes::_fast_icaload);
 582     __ jccb(Assembler::equal, rewrite);
 583 
 584     // rewrite so iload doesn't check again.
 585     __ movl(bc, Bytecodes::_fast_iload);
 586 
 587     // rewrite
 588     // bc: fast bytecode
 589     __ bind(rewrite);
 590     patch_bytecode(Bytecodes::_iload, bc, rbx, false);
 591     __ bind(done);
 592   }
 593 
 594   // Get the local value into tos
 595   locals_index(rbx);
 596   __ movl(rax, iaddress(rbx));
 597 }
 598 
 599 void TemplateTable::fast_iload2() {
 600   transition(vtos, itos);
 601   locals_index(rbx);
 602   __ movl(rax, iaddress(rbx));
 603   __ push(itos);
 604   locals_index(rbx, 3);
 605   __ movl(rax, iaddress(rbx));
 606 }
 607 
 608 void TemplateTable::fast_iload() {
 609   transition(vtos, itos);
 610   locals_index(rbx);
 611   __ movl(rax, iaddress(rbx));
 612 }
 613 
 614 void TemplateTable::lload() {
 615   transition(vtos, ltos);
 616   locals_index(rbx);
 617   __ movptr(rax, laddress(rbx));
 618   NOT_LP64(__ movl(rdx, haddress(rbx)));
 619 }
 620 
 621 void TemplateTable::fload() {
 622   transition(vtos, ftos);
 623   locals_index(rbx);
 624   __ load_float(faddress(rbx));
 625 }
 626 
 627 void TemplateTable::dload() {
 628   transition(vtos, dtos);
 629   locals_index(rbx);
 630   __ load_double(daddress(rbx));
 631 }
 632 
 633 void TemplateTable::aload() {
 634   transition(vtos, atos);
 635   locals_index(rbx);
 636   __ movptr(rax, aaddress(rbx));
 637 }
 638 
 639 void TemplateTable::locals_index_wide(Register reg) {
 640   __ load_unsigned_short(reg, at_bcp(2));
 641   __ bswapl(reg);
 642   __ shrl(reg, 16);
 643   __ negptr(reg);
 644 }
 645 
 646 void TemplateTable::wide_iload() {
 647   transition(vtos, itos);
 648   locals_index_wide(rbx);
 649   __ movl(rax, iaddress(rbx));
 650 }
 651 
 652 void TemplateTable::wide_lload() {
 653   transition(vtos, ltos);
 654   locals_index_wide(rbx);
 655   __ movptr(rax, laddress(rbx));
 656   NOT_LP64(__ movl(rdx, haddress(rbx)));
 657 }
 658 
 659 void TemplateTable::wide_fload() {
 660   transition(vtos, ftos);
 661   locals_index_wide(rbx);
 662   __ load_float(faddress(rbx));
 663 }
 664 
 665 void TemplateTable::wide_dload() {
 666   transition(vtos, dtos);
 667   locals_index_wide(rbx);
 668   __ load_double(daddress(rbx));
 669 }
 670 
 671 void TemplateTable::wide_aload() {
 672   transition(vtos, atos);
 673   locals_index_wide(rbx);
 674   __ movptr(rax, aaddress(rbx));
 675 }
 676 
 677 void TemplateTable::index_check(Register array, Register index) {
 678   // Pop ptr into array
 679   __ pop_ptr(array);
 680   index_check_without_pop(array, index);
 681 }
 682 
 683 void TemplateTable::index_check_without_pop(Register array, Register index) {
 684   // destroys rbx
 685   // check array
 686   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 687   // sign extend index for use by indexed load
 688   __ movl2ptr(index, index);
 689   // check index
 690   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 691   if (index != rbx) {
 692     // ??? convention: move aberrant index into rbx for exception message
 693     assert(rbx != array, "different registers");
 694     __ movl(rbx, index);
 695   }
 696   __ jump_cc(Assembler::aboveEqual,
 697              ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 698 }
 699 
 700 
 701 void TemplateTable::iaload() {
 702   transition(itos, itos);
 703   // rax: index
 704   // rdx: array
 705   index_check(rdx, rax); // kills rbx
 706   __ movl(rax, Address(rdx, rax,
 707                        Address::times_4,
 708                        arrayOopDesc::base_offset_in_bytes(T_INT)));
 709 }
 710 
 711 void TemplateTable::laload() {
 712   transition(itos, ltos);
 713   // rax: index
 714   // rdx: array
 715   index_check(rdx, rax); // kills rbx
 716   NOT_LP64(__ mov(rbx, rax));
 717   // rbx,: index
 718   __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
 719   NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
 720 }
 721 
 722 
 723 
 724 void TemplateTable::faload() {
 725   transition(itos, ftos);
 726   // rax: index
 727   // rdx: array
 728   index_check(rdx, rax); // kills rbx
 729   __ load_float(Address(rdx, rax,
 730                         Address::times_4,
 731                         arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 732 }
 733 
 734 void TemplateTable::daload() {
 735   transition(itos, dtos);
 736   // rax: index
 737   // rdx: array
 738   index_check(rdx, rax); // kills rbx
 739   __ load_double(Address(rdx, rax,
 740                          Address::times_8,
 741                          arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 742 }
 743 
 744 void TemplateTable::aaload() {
 745   transition(itos, atos);
 746   // rax: index
 747   // rdx: array
 748   index_check(rdx, rax); // kills rbx
 749   __ load_heap_oop(rax, Address(rdx, rax,
 750                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 751                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 752 }
 753 
 754 void TemplateTable::baload() {
 755   transition(itos, itos);
 756   // rax: index
 757   // rdx: array
 758   index_check(rdx, rax); // kills rbx
 759   __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 760 }
 761 
 762 void TemplateTable::caload() {
 763   transition(itos, itos);
 764   // rax: index
 765   // rdx: array
 766   index_check(rdx, rax); // kills rbx
 767   __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 768 }
 769 
 770 // iload followed by caload frequent pair
 771 void TemplateTable::fast_icaload() {
 772   transition(vtos, itos);
 773   // load index out of locals
 774   locals_index(rbx);
 775   __ movl(rax, iaddress(rbx));
 776 
 777   // rax: index
 778   // rdx: array
 779   index_check(rdx, rax); // kills rbx
 780   __ load_unsigned_short(rax,
 781                          Address(rdx, rax,
 782                                  Address::times_2,
 783                                  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 784 }
 785 
 786 
 787 void TemplateTable::saload() {
 788   transition(itos, itos);
 789   // rax: index
 790   // rdx: array
 791   index_check(rdx, rax); // kills rbx
 792   __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 793 }
 794 
 795 void TemplateTable::iload(int n) {
 796   transition(vtos, itos);
 797   __ movl(rax, iaddress(n));
 798 }
 799 
 800 void TemplateTable::lload(int n) {
 801   transition(vtos, ltos);
 802   __ movptr(rax, laddress(n));
 803   NOT_LP64(__ movptr(rdx, haddress(n)));
 804 }
 805 
 806 void TemplateTable::fload(int n) {
 807   transition(vtos, ftos);
 808   __ load_float(faddress(n));
 809 }
 810 
 811 void TemplateTable::dload(int n) {
 812   transition(vtos, dtos);
 813   __ load_double(daddress(n));
 814 }
 815 
 816 void TemplateTable::aload(int n) {
 817   transition(vtos, atos);
 818   __ movptr(rax, aaddress(n));
 819 }
 820 
 821 void TemplateTable::aload_0() {
 822   aload_0_internal();
 823 }
 824 
 825 void TemplateTable::nofast_aload_0() {
 826   aload_0_internal(may_not_rewrite);
 827 }
 828 
 829 void TemplateTable::aload_0_internal(RewriteControl rc) {
 830   transition(vtos, atos);
 831   // According to bytecode histograms, the pairs:
 832   //
 833   // _aload_0, _fast_igetfield
 834   // _aload_0, _fast_agetfield
 835   // _aload_0, _fast_fgetfield
 836   //
 837   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 838   // _aload_0 bytecode checks if the next bytecode is either
 839   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 840   // rewrites the current bytecode into a pair bytecode; otherwise it
 841   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 842   // the pair check anymore.
 843   //
 844   // Note: If the next bytecode is _getfield, the rewrite must be
 845   //       delayed, otherwise we may miss an opportunity for a pair.
 846   //
 847   // Also rewrite frequent pairs
 848   //   aload_0, aload_1
 849   //   aload_0, iload_1
 850   // These bytecodes with a small amount of code are most profitable
 851   // to rewrite
 852   if (RewriteFrequentPairs && rc == may_rewrite) {
 853     Label rewrite, done;
 854 
 855     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 856     LP64_ONLY(assert(rbx != bc, "register damaged"));
 857 
 858     // get next byte
 859     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 860 
 861     // if _getfield then wait with rewrite
 862     __ cmpl(rbx, Bytecodes::_getfield);
 863     __ jcc(Assembler::equal, done);
 864 
 865     // if _igetfield then rewrite to _fast_iaccess_0
 866     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 867     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 868     __ movl(bc, Bytecodes::_fast_iaccess_0);
 869     __ jccb(Assembler::equal, rewrite);
 870 
 871     // if _agetfield then rewrite to _fast_aaccess_0
 872     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 873     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 874     __ movl(bc, Bytecodes::_fast_aaccess_0);
 875     __ jccb(Assembler::equal, rewrite);
 876 
 877     // if _fgetfield then rewrite to _fast_faccess_0
 878     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 879     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 880     __ movl(bc, Bytecodes::_fast_faccess_0);
 881     __ jccb(Assembler::equal, rewrite);
 882 
 883     // else rewrite to _fast_aload0
 884     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 885     __ movl(bc, Bytecodes::_fast_aload_0);
 886 
 887     // rewrite
 888     // bc: fast bytecode
 889     __ bind(rewrite);
 890     patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
 891 
 892     __ bind(done);
 893   }
 894 
 895   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 896   aload(0);
 897 }
 898 
 899 void TemplateTable::istore() {
 900   transition(itos, vtos);
 901   locals_index(rbx);
 902   __ movl(iaddress(rbx), rax);
 903 }
 904 
 905 
 906 void TemplateTable::lstore() {
 907   transition(ltos, vtos);
 908   locals_index(rbx);
 909   __ movptr(laddress(rbx), rax);
 910   NOT_LP64(__ movptr(haddress(rbx), rdx));
 911 }
 912 
 913 void TemplateTable::fstore() {
 914   transition(ftos, vtos);
 915   locals_index(rbx);
 916   __ store_float(faddress(rbx));
 917 }
 918 
 919 void TemplateTable::dstore() {
 920   transition(dtos, vtos);
 921   locals_index(rbx);
 922   __ store_double(daddress(rbx));
 923 }
 924 
 925 void TemplateTable::astore() {
 926   transition(vtos, vtos);
 927   __ pop_ptr(rax);
 928   locals_index(rbx);
 929   __ movptr(aaddress(rbx), rax);
 930 }
 931 
 932 void TemplateTable::wide_istore() {
 933   transition(vtos, vtos);
 934   __ pop_i();
 935   locals_index_wide(rbx);
 936   __ movl(iaddress(rbx), rax);
 937 }
 938 
 939 void TemplateTable::wide_lstore() {
 940   transition(vtos, vtos);
 941   NOT_LP64(__ pop_l(rax, rdx));
 942   LP64_ONLY(__ pop_l());
 943   locals_index_wide(rbx);
 944   __ movptr(laddress(rbx), rax);
 945   NOT_LP64(__ movl(haddress(rbx), rdx));
 946 }
 947 
 948 void TemplateTable::wide_fstore() {
 949 #ifdef _LP64
 950   transition(vtos, vtos);
 951   __ pop_f(xmm0);
 952   locals_index_wide(rbx);
 953   __ movflt(faddress(rbx), xmm0);
 954 #else
 955   wide_istore();
 956 #endif
 957 }
 958 
 959 void TemplateTable::wide_dstore() {
 960 #ifdef _LP64
 961   transition(vtos, vtos);
 962   __ pop_d(xmm0);
 963   locals_index_wide(rbx);
 964   __ movdbl(daddress(rbx), xmm0);
 965 #else
 966   wide_lstore();
 967 #endif
 968 }
 969 
 970 void TemplateTable::wide_astore() {
 971   transition(vtos, vtos);
 972   __ pop_ptr(rax);
 973   locals_index_wide(rbx);
 974   __ movptr(aaddress(rbx), rax);
 975 }
 976 
 977 void TemplateTable::iastore() {
 978   transition(itos, vtos);
 979   __ pop_i(rbx);
 980   // rax: value
 981   // rbx: index
 982   // rdx: array
 983   index_check(rdx, rbx); // prefer index in rbx
 984   __ movl(Address(rdx, rbx,
 985                   Address::times_4,
 986                   arrayOopDesc::base_offset_in_bytes(T_INT)),
 987           rax);
 988 }
 989 
 990 void TemplateTable::lastore() {
 991   transition(ltos, vtos);
 992   __ pop_i(rbx);
 993   // rax,: low(value)
 994   // rcx: array
 995   // rdx: high(value)
 996   index_check(rcx, rbx);  // prefer index in rbx,
 997   // rbx,: index
 998   __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
 999   NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1000 }
1001 
1002 
1003 void TemplateTable::fastore() {
1004   transition(ftos, vtos);
1005   __ pop_i(rbx);
1006   // value is in UseSSE >= 1 ? xmm0 : ST(0)
1007   // rbx:  index
1008   // rdx:  array
1009   index_check(rdx, rbx); // prefer index in rbx
1010   __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1011 }
1012 
1013 void TemplateTable::dastore() {
1014   transition(dtos, vtos);
1015   __ pop_i(rbx);
1016   // value is in UseSSE >= 2 ? xmm0 : ST(0)
1017   // rbx:  index
1018   // rdx:  array
1019   index_check(rdx, rbx); // prefer index in rbx
1020   __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1021 }
1022 
1023 void TemplateTable::aastore() {
1024   Label is_null, ok_is_subtype, done;
1025   transition(vtos, vtos);
1026   // stack: ..., array, index, value
1027   __ movptr(rax, at_tos());    // value
1028   __ movl(rcx, at_tos_p1()); // index
1029   __ movptr(rdx, at_tos_p2()); // array
1030 
1031   Address element_address(rdx, rcx,
1032                           UseCompressedOops? Address::times_4 : Address::times_ptr,
1033                           arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1034 
1035   index_check_without_pop(rdx, rcx);     // kills rbx
1036   __ testptr(rax, rax);
1037   __ jcc(Assembler::zero, is_null);
1038 
1039   // Move subklass into rbx
1040   __ load_klass(rbx, rax);
1041   // Move superklass into rax
1042   __ load_klass(rax, rdx);
1043   __ movptr(rax, Address(rax,
1044                          ObjArrayKlass::element_klass_offset()));
1045   // Compress array + index*oopSize + 12 into a single register.  Frees rcx.
1046   __ lea(rdx, element_address);
1047 
1048   // Generate subtype check.  Blows rcx, rdi
1049   // Superklass in rax.  Subklass in rbx.
1050   __ gen_subtype_check(rbx, ok_is_subtype);
1051 
1052   // Come here on failure
1053   // object is at TOS
1054   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1055 
1056   // Come here on success
1057   __ bind(ok_is_subtype);
1058 
1059   // Get the value we will store
1060   __ movptr(rax, at_tos());
1061   // Now store using the appropriate barrier
1062   do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1063   __ jmp(done);
1064 
1065   // Have a NULL in rax, rdx=array, ecx=index.  Store NULL at ary[idx]
1066   __ bind(is_null);
1067   __ profile_null_seen(rbx);
1068 
1069   // Store a NULL
1070   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1071 
1072   // Pop stack arguments
1073   __ bind(done);
1074   __ addptr(rsp, 3 * Interpreter::stackElementSize);
1075 }
1076 
1077 void TemplateTable::bastore() {
1078   transition(itos, vtos);
1079   __ pop_i(rbx);
1080   // rax: value
1081   // rbx: index
1082   // rdx: array
1083   index_check(rdx, rbx); // prefer index in rbx
1084   // Need to check whether array is boolean or byte
1085   // since both types share the bastore bytecode.
1086   __ load_klass(rcx, rdx);
1087   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1088   int diffbit = Klass::layout_helper_boolean_diffbit();
1089   __ testl(rcx, diffbit);
1090   Label L_skip;
1091   __ jccb(Assembler::zero, L_skip);
1092   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1093   __ bind(L_skip);
1094   __ movb(Address(rdx, rbx,
1095                   Address::times_1,
1096                   arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1097           rax);
1098 }
1099 
1100 void TemplateTable::castore() {
1101   transition(itos, vtos);
1102   __ pop_i(rbx);
1103   // rax: value
1104   // rbx: index
1105   // rdx: array
1106   index_check(rdx, rbx);  // prefer index in rbx
1107   __ movw(Address(rdx, rbx,
1108                   Address::times_2,
1109                   arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1110           rax);
1111 }
1112 
1113 
1114 void TemplateTable::sastore() {
1115   castore();
1116 }
1117 
1118 void TemplateTable::istore(int n) {
1119   transition(itos, vtos);
1120   __ movl(iaddress(n), rax);
1121 }
1122 
1123 void TemplateTable::lstore(int n) {
1124   transition(ltos, vtos);
1125   __ movptr(laddress(n), rax);
1126   NOT_LP64(__ movptr(haddress(n), rdx));
1127 }
1128 
1129 void TemplateTable::fstore(int n) {
1130   transition(ftos, vtos);
1131   __ store_float(faddress(n));
1132 }
1133 
1134 void TemplateTable::dstore(int n) {
1135   transition(dtos, vtos);
1136   __ store_double(daddress(n));
1137 }
1138 
1139 
1140 void TemplateTable::astore(int n) {
1141   transition(vtos, vtos);
1142   __ pop_ptr(rax);
1143   __ movptr(aaddress(n), rax);
1144 }
1145 
1146 void TemplateTable::pop() {
1147   transition(vtos, vtos);
1148   __ addptr(rsp, Interpreter::stackElementSize);
1149 }
1150 
1151 void TemplateTable::pop2() {
1152   transition(vtos, vtos);
1153   __ addptr(rsp, 2 * Interpreter::stackElementSize);
1154 }
1155 
1156 
1157 void TemplateTable::dup() {
1158   transition(vtos, vtos);
1159   __ load_ptr(0, rax);
1160   __ push_ptr(rax);
1161   // stack: ..., a, a
1162 }
1163 
1164 void TemplateTable::dup_x1() {
1165   transition(vtos, vtos);
1166   // stack: ..., a, b
1167   __ load_ptr( 0, rax);  // load b
1168   __ load_ptr( 1, rcx);  // load a
1169   __ store_ptr(1, rax);  // store b
1170   __ store_ptr(0, rcx);  // store a
1171   __ push_ptr(rax);      // push b
1172   // stack: ..., b, a, b
1173 }
1174 
1175 void TemplateTable::dup_x2() {
1176   transition(vtos, vtos);
1177   // stack: ..., a, b, c
1178   __ load_ptr( 0, rax);  // load c
1179   __ load_ptr( 2, rcx);  // load a
1180   __ store_ptr(2, rax);  // store c in a
1181   __ push_ptr(rax);      // push c
1182   // stack: ..., c, b, c, c
1183   __ load_ptr( 2, rax);  // load b
1184   __ store_ptr(2, rcx);  // store a in b
1185   // stack: ..., c, a, c, c
1186   __ store_ptr(1, rax);  // store b in c
1187   // stack: ..., c, a, b, c
1188 }
1189 
1190 void TemplateTable::dup2() {
1191   transition(vtos, vtos);
1192   // stack: ..., a, b
1193   __ load_ptr(1, rax);  // load a
1194   __ push_ptr(rax);     // push a
1195   __ load_ptr(1, rax);  // load b
1196   __ push_ptr(rax);     // push b
1197   // stack: ..., a, b, a, b
1198 }
1199 
1200 
1201 void TemplateTable::dup2_x1() {
1202   transition(vtos, vtos);
1203   // stack: ..., a, b, c
1204   __ load_ptr( 0, rcx);  // load c
1205   __ load_ptr( 1, rax);  // load b
1206   __ push_ptr(rax);      // push b
1207   __ push_ptr(rcx);      // push c
1208   // stack: ..., a, b, c, b, c
1209   __ store_ptr(3, rcx);  // store c in b
1210   // stack: ..., a, c, c, b, c
1211   __ load_ptr( 4, rcx);  // load a
1212   __ store_ptr(2, rcx);  // store a in 2nd c
1213   // stack: ..., a, c, a, b, c
1214   __ store_ptr(4, rax);  // store b in a
1215   // stack: ..., b, c, a, b, c
1216 }
1217 
1218 void TemplateTable::dup2_x2() {
1219   transition(vtos, vtos);
1220   // stack: ..., a, b, c, d
1221   __ load_ptr( 0, rcx);  // load d
1222   __ load_ptr( 1, rax);  // load c
1223   __ push_ptr(rax);      // push c
1224   __ push_ptr(rcx);      // push d
1225   // stack: ..., a, b, c, d, c, d
1226   __ load_ptr( 4, rax);  // load b
1227   __ store_ptr(2, rax);  // store b in d
1228   __ store_ptr(4, rcx);  // store d in b
1229   // stack: ..., a, d, c, b, c, d
1230   __ load_ptr( 5, rcx);  // load a
1231   __ load_ptr( 3, rax);  // load c
1232   __ store_ptr(3, rcx);  // store a in c
1233   __ store_ptr(5, rax);  // store c in a
1234   // stack: ..., c, d, a, b, c, d
1235 }
1236 
1237 void TemplateTable::swap() {
1238   transition(vtos, vtos);
1239   // stack: ..., a, b
1240   __ load_ptr( 1, rcx);  // load a
1241   __ load_ptr( 0, rax);  // load b
1242   __ store_ptr(0, rcx);  // store a in b
1243   __ store_ptr(1, rax);  // store b in a
1244   // stack: ..., b, a
1245 }
1246 
1247 void TemplateTable::iop2(Operation op) {
1248   transition(itos, itos);
1249   switch (op) {
1250   case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1251   case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1252   case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1253   case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1254   case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1255   case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1256   case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break;
1257   case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break;
1258   case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break;
1259   default   : ShouldNotReachHere();
1260   }
1261 }
1262 
1263 void TemplateTable::lop2(Operation op) {
1264   transition(ltos, ltos);
1265 #ifdef _LP64
1266   switch (op) {
1267   case add  :                    __ pop_l(rdx); __ addptr(rax, rdx); break;
1268   case sub  : __ mov(rdx, rax);  __ pop_l(rax); __ subptr(rax, rdx); break;
1269   case _and :                    __ pop_l(rdx); __ andptr(rax, rdx); break;
1270   case _or  :                    __ pop_l(rdx); __ orptr (rax, rdx); break;
1271   case _xor :                    __ pop_l(rdx); __ xorptr(rax, rdx); break;
1272   default   : ShouldNotReachHere();
1273   }
1274 #else
1275   __ pop_l(rbx, rcx);
1276   switch (op) {
1277     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1278     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1279                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1280     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1281     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1282     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1283     default   : ShouldNotReachHere();
1284   }
1285 #endif
1286 }
1287 
1288 void TemplateTable::idiv() {
1289   transition(itos, itos);
1290   __ movl(rcx, rax);
1291   __ pop_i(rax);
1292   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1293   //       they are not equal, one could do a normal division (no correction
1294   //       needed), which may speed up this implementation for the common case.
1295   //       (see also JVM spec., p.243 & p.271)
1296   __ corrected_idivl(rcx);
1297 }
1298 
1299 void TemplateTable::irem() {
1300   transition(itos, itos);
1301   __ movl(rcx, rax);
1302   __ pop_i(rax);
1303   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1304   //       they are not equal, one could do a normal division (no correction
1305   //       needed), which may speed up this implementation for the common case.
1306   //       (see also JVM spec., p.243 & p.271)
1307   __ corrected_idivl(rcx);
1308   __ movl(rax, rdx);
1309 }
1310 
1311 void TemplateTable::lmul() {
1312   transition(ltos, ltos);
1313 #ifdef _LP64
1314   __ pop_l(rdx);
1315   __ imulq(rax, rdx);
1316 #else
1317   __ pop_l(rbx, rcx);
1318   __ push(rcx); __ push(rbx);
1319   __ push(rdx); __ push(rax);
1320   __ lmul(2 * wordSize, 0);
1321   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1322 #endif
1323 }
1324 
1325 void TemplateTable::ldiv() {
1326   transition(ltos, ltos);
1327 #ifdef _LP64
1328   __ mov(rcx, rax);
1329   __ pop_l(rax);
1330   // generate explicit div0 check
1331   __ testq(rcx, rcx);
1332   __ jump_cc(Assembler::zero,
1333              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1334   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1335   //       they are not equal, one could do a normal division (no correction
1336   //       needed), which may speed up this implementation for the common case.
1337   //       (see also JVM spec., p.243 & p.271)
1338   __ corrected_idivq(rcx); // kills rbx
1339 #else
1340   __ pop_l(rbx, rcx);
1341   __ push(rcx); __ push(rbx);
1342   __ push(rdx); __ push(rax);
1343   // check if y = 0
1344   __ orl(rax, rdx);
1345   __ jump_cc(Assembler::zero,
1346              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1347   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1348   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1349 #endif
1350 }
1351 
1352 void TemplateTable::lrem() {
1353   transition(ltos, ltos);
1354 #ifdef _LP64
1355   __ mov(rcx, rax);
1356   __ pop_l(rax);
1357   __ testq(rcx, rcx);
1358   __ jump_cc(Assembler::zero,
1359              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1360   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1361   //       they are not equal, one could do a normal division (no correction
1362   //       needed), which may speed up this implementation for the common case.
1363   //       (see also JVM spec., p.243 & p.271)
1364   __ corrected_idivq(rcx); // kills rbx
1365   __ mov(rax, rdx);
1366 #else
1367   __ pop_l(rbx, rcx);
1368   __ push(rcx); __ push(rbx);
1369   __ push(rdx); __ push(rax);
1370   // check if y = 0
1371   __ orl(rax, rdx);
1372   __ jump_cc(Assembler::zero,
1373              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1374   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1375   __ addptr(rsp, 4 * wordSize);
1376 #endif
1377 }
1378 
1379 void TemplateTable::lshl() {
1380   transition(itos, ltos);
1381   __ movl(rcx, rax);                             // get shift count
1382   #ifdef _LP64
1383   __ pop_l(rax);                                 // get shift value
1384   __ shlq(rax);
1385 #else
1386   __ pop_l(rax, rdx);                            // get shift value
1387   __ lshl(rdx, rax);
1388 #endif
1389 }
1390 
1391 void TemplateTable::lshr() {
1392 #ifdef _LP64
1393   transition(itos, ltos);
1394   __ movl(rcx, rax);                             // get shift count
1395   __ pop_l(rax);                                 // get shift value
1396   __ sarq(rax);
1397 #else
1398   transition(itos, ltos);
1399   __ mov(rcx, rax);                              // get shift count
1400   __ pop_l(rax, rdx);                            // get shift value
1401   __ lshr(rdx, rax, true);
1402 #endif
1403 }
1404 
1405 void TemplateTable::lushr() {
1406   transition(itos, ltos);
1407 #ifdef _LP64
1408   __ movl(rcx, rax);                             // get shift count
1409   __ pop_l(rax);                                 // get shift value
1410   __ shrq(rax);
1411 #else
1412   __ mov(rcx, rax);                              // get shift count
1413   __ pop_l(rax, rdx);                            // get shift value
1414   __ lshr(rdx, rax);
1415 #endif
1416 }
1417 
1418 void TemplateTable::fop2(Operation op) {
1419   transition(ftos, ftos);
1420 
1421   if (UseSSE >= 1) {
1422     switch (op) {
1423     case add:
1424       __ addss(xmm0, at_rsp());
1425       __ addptr(rsp, Interpreter::stackElementSize);
1426       break;
1427     case sub:
1428       __ movflt(xmm1, xmm0);
1429       __ pop_f(xmm0);
1430       __ subss(xmm0, xmm1);
1431       break;
1432     case mul:
1433       __ mulss(xmm0, at_rsp());
1434       __ addptr(rsp, Interpreter::stackElementSize);
1435       break;
1436     case div:
1437       __ movflt(xmm1, xmm0);
1438       __ pop_f(xmm0);
1439       __ divss(xmm0, xmm1);
1440       break;
1441     case rem:
1442       // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1443       // modulo operation. The frem method calls the function
1444       // double fmod(double x, double y) in math.h. The documentation of fmod states:
1445       // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1446       // (signalling or quiet) is returned.
1447       //
1448       // On x86_32 platforms the FPU is used to perform the modulo operation. The
1449       // reason is that on 32-bit Windows the sign of modulo operations diverges from
1450       // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1451       // The fprem instruction used on x86_32 is functionally equivalent to
1452       // SharedRuntime::frem in that it returns a NaN.
1453 #ifdef _LP64
1454       __ movflt(xmm1, xmm0);
1455       __ pop_f(xmm0);
1456       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1457 #else
1458       __ push_f(xmm0);
1459       __ pop_f();
1460       __ fld_s(at_rsp());
1461       __ fremr(rax);
1462       __ f2ieee();
1463       __ pop(rax);  // pop second operand off the stack
1464       __ push_f();
1465       __ pop_f(xmm0);
1466 #endif
1467       break;
1468     default:
1469       ShouldNotReachHere();
1470       break;
1471     }
1472   } else {
1473 #ifdef _LP64
1474     ShouldNotReachHere();
1475 #else
1476     switch (op) {
1477     case add: __ fadd_s (at_rsp());                break;
1478     case sub: __ fsubr_s(at_rsp());                break;
1479     case mul: __ fmul_s (at_rsp());                break;
1480     case div: __ fdivr_s(at_rsp());                break;
1481     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1482     default : ShouldNotReachHere();
1483     }
1484     __ f2ieee();
1485     __ pop(rax);  // pop second operand off the stack
1486 #endif // _LP64
1487   }
1488 }
1489 
1490 void TemplateTable::dop2(Operation op) {
1491   transition(dtos, dtos);
1492   if (UseSSE >= 2) {
1493     switch (op) {
1494     case add:
1495       __ addsd(xmm0, at_rsp());
1496       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1497       break;
1498     case sub:
1499       __ movdbl(xmm1, xmm0);
1500       __ pop_d(xmm0);
1501       __ subsd(xmm0, xmm1);
1502       break;
1503     case mul:
1504       __ mulsd(xmm0, at_rsp());
1505       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1506       break;
1507     case div:
1508       __ movdbl(xmm1, xmm0);
1509       __ pop_d(xmm0);
1510       __ divsd(xmm0, xmm1);
1511       break;
1512     case rem:
1513       // Similar to fop2(), the modulo operation is performed using the
1514       // SharedRuntime::drem method (on x86_64 platforms) or using the
1515       // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1516 #ifdef _LP64
1517       __ movdbl(xmm1, xmm0);
1518       __ pop_d(xmm0);
1519       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1520 #else
1521       __ push_d(xmm0);
1522       __ pop_d();
1523       __ fld_d(at_rsp());
1524       __ fremr(rax);
1525       __ d2ieee();
1526       __ pop(rax);
1527       __ pop(rdx);
1528       __ push_d();
1529       __ pop_d(xmm0);
1530 #endif
1531       break;
1532     default:
1533       ShouldNotReachHere();
1534       break;
1535     }
1536   } else {
1537 #ifdef _LP64
1538     ShouldNotReachHere();
1539 #else
1540     switch (op) {
1541     case add: __ fadd_d (at_rsp());                break;
1542     case sub: __ fsubr_d(at_rsp());                break;
1543     case mul: {
1544       Label L_strict;
1545       Label L_join;
1546       const Address access_flags      (rcx, Method::access_flags_offset());
1547       __ get_method(rcx);
1548       __ movl(rcx, access_flags);
1549       __ testl(rcx, JVM_ACC_STRICT);
1550       __ jccb(Assembler::notZero, L_strict);
1551       __ fmul_d (at_rsp());
1552       __ jmpb(L_join);
1553       __ bind(L_strict);
1554       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1555       __ fmulp();
1556       __ fmul_d (at_rsp());
1557       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1558       __ fmulp();
1559       __ bind(L_join);
1560       break;
1561     }
1562     case div: {
1563       Label L_strict;
1564       Label L_join;
1565       const Address access_flags      (rcx, Method::access_flags_offset());
1566       __ get_method(rcx);
1567       __ movl(rcx, access_flags);
1568       __ testl(rcx, JVM_ACC_STRICT);
1569       __ jccb(Assembler::notZero, L_strict);
1570       __ fdivr_d(at_rsp());
1571       __ jmp(L_join);
1572       __ bind(L_strict);
1573       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1574       __ fmul_d (at_rsp());
1575       __ fdivrp();
1576       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1577       __ fmulp();
1578       __ bind(L_join);
1579       break;
1580     }
1581     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1582     default : ShouldNotReachHere();
1583     }
1584     __ d2ieee();
1585     // Pop double precision number from rsp.
1586     __ pop(rax);
1587     __ pop(rdx);
1588 #endif
1589   }
1590 }
1591 
1592 void TemplateTable::ineg() {
1593   transition(itos, itos);
1594   __ negl(rax);
1595 }
1596 
1597 void TemplateTable::lneg() {
1598   transition(ltos, ltos);
1599   LP64_ONLY(__ negq(rax));
1600   NOT_LP64(__ lneg(rdx, rax));
1601 }
1602 
1603 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1604 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1605   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1606   // of 128-bits operands for SSE instructions.
1607   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1608   // Store the value to a 128-bits operand.
1609   operand[0] = lo;
1610   operand[1] = hi;
1611   return operand;
1612 }
1613 
1614 // Buffer for 128-bits masks used by SSE instructions.
1615 static jlong float_signflip_pool[2*2];
1616 static jlong double_signflip_pool[2*2];
1617 
1618 void TemplateTable::fneg() {
1619   transition(ftos, ftos);
1620   if (UseSSE >= 1) {
1621     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1622     __ xorps(xmm0, ExternalAddress((address) float_signflip));
1623   } else {
1624     LP64_ONLY(ShouldNotReachHere());
1625     NOT_LP64(__ fchs());
1626   }
1627 }
1628 
1629 void TemplateTable::dneg() {
1630   transition(dtos, dtos);
1631   if (UseSSE >= 2) {
1632     static jlong *double_signflip =
1633       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1634     __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1635   } else {
1636 #ifdef _LP64
1637     ShouldNotReachHere();
1638 #else
1639     __ fchs();
1640 #endif
1641   }
1642 }
1643 
1644 void TemplateTable::iinc() {
1645   transition(vtos, vtos);
1646   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1647   locals_index(rbx);
1648   __ addl(iaddress(rbx), rdx);
1649 }
1650 
1651 void TemplateTable::wide_iinc() {
1652   transition(vtos, vtos);
1653   __ movl(rdx, at_bcp(4)); // get constant
1654   locals_index_wide(rbx);
1655   __ bswapl(rdx); // swap bytes & sign-extend constant
1656   __ sarl(rdx, 16);
1657   __ addl(iaddress(rbx), rdx);
1658   // Note: should probably use only one movl to get both
1659   //       the index and the constant -> fix this
1660 }
1661 
1662 void TemplateTable::convert() {
1663 #ifdef _LP64
1664   // Checking
1665 #ifdef ASSERT
1666   {
1667     TosState tos_in  = ilgl;
1668     TosState tos_out = ilgl;
1669     switch (bytecode()) {
1670     case Bytecodes::_i2l: // fall through
1671     case Bytecodes::_i2f: // fall through
1672     case Bytecodes::_i2d: // fall through
1673     case Bytecodes::_i2b: // fall through
1674     case Bytecodes::_i2c: // fall through
1675     case Bytecodes::_i2s: tos_in = itos; break;
1676     case Bytecodes::_l2i: // fall through
1677     case Bytecodes::_l2f: // fall through
1678     case Bytecodes::_l2d: tos_in = ltos; break;
1679     case Bytecodes::_f2i: // fall through
1680     case Bytecodes::_f2l: // fall through
1681     case Bytecodes::_f2d: tos_in = ftos; break;
1682     case Bytecodes::_d2i: // fall through
1683     case Bytecodes::_d2l: // fall through
1684     case Bytecodes::_d2f: tos_in = dtos; break;
1685     default             : ShouldNotReachHere();
1686     }
1687     switch (bytecode()) {
1688     case Bytecodes::_l2i: // fall through
1689     case Bytecodes::_f2i: // fall through
1690     case Bytecodes::_d2i: // fall through
1691     case Bytecodes::_i2b: // fall through
1692     case Bytecodes::_i2c: // fall through
1693     case Bytecodes::_i2s: tos_out = itos; break;
1694     case Bytecodes::_i2l: // fall through
1695     case Bytecodes::_f2l: // fall through
1696     case Bytecodes::_d2l: tos_out = ltos; break;
1697     case Bytecodes::_i2f: // fall through
1698     case Bytecodes::_l2f: // fall through
1699     case Bytecodes::_d2f: tos_out = ftos; break;
1700     case Bytecodes::_i2d: // fall through
1701     case Bytecodes::_l2d: // fall through
1702     case Bytecodes::_f2d: tos_out = dtos; break;
1703     default             : ShouldNotReachHere();
1704     }
1705     transition(tos_in, tos_out);
1706   }
1707 #endif // ASSERT
1708 
1709   static const int64_t is_nan = 0x8000000000000000L;
1710 
1711   // Conversion
1712   switch (bytecode()) {
1713   case Bytecodes::_i2l:
1714     __ movslq(rax, rax);
1715     break;
1716   case Bytecodes::_i2f:
1717     __ cvtsi2ssl(xmm0, rax);
1718     break;
1719   case Bytecodes::_i2d:
1720     __ cvtsi2sdl(xmm0, rax);
1721     break;
1722   case Bytecodes::_i2b:
1723     __ movsbl(rax, rax);
1724     break;
1725   case Bytecodes::_i2c:
1726     __ movzwl(rax, rax);
1727     break;
1728   case Bytecodes::_i2s:
1729     __ movswl(rax, rax);
1730     break;
1731   case Bytecodes::_l2i:
1732     __ movl(rax, rax);
1733     break;
1734   case Bytecodes::_l2f:
1735     __ cvtsi2ssq(xmm0, rax);
1736     break;
1737   case Bytecodes::_l2d:
1738     __ cvtsi2sdq(xmm0, rax);
1739     break;
1740   case Bytecodes::_f2i:
1741   {
1742     Label L;
1743     __ cvttss2sil(rax, xmm0);
1744     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1745     __ jcc(Assembler::notEqual, L);
1746     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1747     __ bind(L);
1748   }
1749     break;
1750   case Bytecodes::_f2l:
1751   {
1752     Label L;
1753     __ cvttss2siq(rax, xmm0);
1754     // NaN or overflow/underflow?
1755     __ cmp64(rax, ExternalAddress((address) &is_nan));
1756     __ jcc(Assembler::notEqual, L);
1757     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1758     __ bind(L);
1759   }
1760     break;
1761   case Bytecodes::_f2d:
1762     __ cvtss2sd(xmm0, xmm0);
1763     break;
1764   case Bytecodes::_d2i:
1765   {
1766     Label L;
1767     __ cvttsd2sil(rax, xmm0);
1768     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1769     __ jcc(Assembler::notEqual, L);
1770     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1771     __ bind(L);
1772   }
1773     break;
1774   case Bytecodes::_d2l:
1775   {
1776     Label L;
1777     __ cvttsd2siq(rax, xmm0);
1778     // NaN or overflow/underflow?
1779     __ cmp64(rax, ExternalAddress((address) &is_nan));
1780     __ jcc(Assembler::notEqual, L);
1781     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1782     __ bind(L);
1783   }
1784     break;
1785   case Bytecodes::_d2f:
1786     __ cvtsd2ss(xmm0, xmm0);
1787     break;
1788   default:
1789     ShouldNotReachHere();
1790   }
1791 #else
1792   // Checking
1793 #ifdef ASSERT
1794   { TosState tos_in  = ilgl;
1795     TosState tos_out = ilgl;
1796     switch (bytecode()) {
1797       case Bytecodes::_i2l: // fall through
1798       case Bytecodes::_i2f: // fall through
1799       case Bytecodes::_i2d: // fall through
1800       case Bytecodes::_i2b: // fall through
1801       case Bytecodes::_i2c: // fall through
1802       case Bytecodes::_i2s: tos_in = itos; break;
1803       case Bytecodes::_l2i: // fall through
1804       case Bytecodes::_l2f: // fall through
1805       case Bytecodes::_l2d: tos_in = ltos; break;
1806       case Bytecodes::_f2i: // fall through
1807       case Bytecodes::_f2l: // fall through
1808       case Bytecodes::_f2d: tos_in = ftos; break;
1809       case Bytecodes::_d2i: // fall through
1810       case Bytecodes::_d2l: // fall through
1811       case Bytecodes::_d2f: tos_in = dtos; break;
1812       default             : ShouldNotReachHere();
1813     }
1814     switch (bytecode()) {
1815       case Bytecodes::_l2i: // fall through
1816       case Bytecodes::_f2i: // fall through
1817       case Bytecodes::_d2i: // fall through
1818       case Bytecodes::_i2b: // fall through
1819       case Bytecodes::_i2c: // fall through
1820       case Bytecodes::_i2s: tos_out = itos; break;
1821       case Bytecodes::_i2l: // fall through
1822       case Bytecodes::_f2l: // fall through
1823       case Bytecodes::_d2l: tos_out = ltos; break;
1824       case Bytecodes::_i2f: // fall through
1825       case Bytecodes::_l2f: // fall through
1826       case Bytecodes::_d2f: tos_out = ftos; break;
1827       case Bytecodes::_i2d: // fall through
1828       case Bytecodes::_l2d: // fall through
1829       case Bytecodes::_f2d: tos_out = dtos; break;
1830       default             : ShouldNotReachHere();
1831     }
1832     transition(tos_in, tos_out);
1833   }
1834 #endif // ASSERT
1835 
1836   // Conversion
1837   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1838   switch (bytecode()) {
1839     case Bytecodes::_i2l:
1840       __ extend_sign(rdx, rax);
1841       break;
1842     case Bytecodes::_i2f:
1843       if (UseSSE >= 1) {
1844         __ cvtsi2ssl(xmm0, rax);
1845       } else {
1846         __ push(rax);          // store int on tos
1847         __ fild_s(at_rsp());   // load int to ST0
1848         __ f2ieee();           // truncate to float size
1849         __ pop(rcx);           // adjust rsp
1850       }
1851       break;
1852     case Bytecodes::_i2d:
1853       if (UseSSE >= 2) {
1854         __ cvtsi2sdl(xmm0, rax);
1855       } else {
1856       __ push(rax);          // add one slot for d2ieee()
1857       __ push(rax);          // store int on tos
1858       __ fild_s(at_rsp());   // load int to ST0
1859       __ d2ieee();           // truncate to double size
1860       __ pop(rcx);           // adjust rsp
1861       __ pop(rcx);
1862       }
1863       break;
1864     case Bytecodes::_i2b:
1865       __ shll(rax, 24);      // truncate upper 24 bits
1866       __ sarl(rax, 24);      // and sign-extend byte
1867       LP64_ONLY(__ movsbl(rax, rax));
1868       break;
1869     case Bytecodes::_i2c:
1870       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1871       LP64_ONLY(__ movzwl(rax, rax));
1872       break;
1873     case Bytecodes::_i2s:
1874       __ shll(rax, 16);      // truncate upper 16 bits
1875       __ sarl(rax, 16);      // and sign-extend short
1876       LP64_ONLY(__ movswl(rax, rax));
1877       break;
1878     case Bytecodes::_l2i:
1879       /* nothing to do */
1880       break;
1881     case Bytecodes::_l2f:
1882       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1883       // 64-bit long values to floats. On 32-bit platforms it is not possible
1884       // to use that instruction with 64-bit operands, therefore the FPU is
1885       // used to perform the conversion.
1886       __ push(rdx);          // store long on tos
1887       __ push(rax);
1888       __ fild_d(at_rsp());   // load long to ST0
1889       __ f2ieee();           // truncate to float size
1890       __ pop(rcx);           // adjust rsp
1891       __ pop(rcx);
1892       if (UseSSE >= 1) {
1893         __ push_f();
1894         __ pop_f(xmm0);
1895       }
1896       break;
1897     case Bytecodes::_l2d:
1898       // On 32-bit platforms the FPU is used for conversion because on
1899       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1900       // instruction with 64-bit operands.
1901       __ push(rdx);          // store long on tos
1902       __ push(rax);
1903       __ fild_d(at_rsp());   // load long to ST0
1904       __ d2ieee();           // truncate to double size
1905       __ pop(rcx);           // adjust rsp
1906       __ pop(rcx);
1907       if (UseSSE >= 2) {
1908         __ push_d();
1909         __ pop_d(xmm0);
1910       }
1911       break;
1912     case Bytecodes::_f2i:
1913       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1914       // as it returns 0 for any NaN.
1915       if (UseSSE >= 1) {
1916         __ push_f(xmm0);
1917       } else {
1918         __ push(rcx);          // reserve space for argument
1919         __ fstp_s(at_rsp());   // pass float argument on stack
1920       }
1921       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1922       break;
1923     case Bytecodes::_f2l:
1924       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1925       // as it returns 0 for any NaN.
1926       if (UseSSE >= 1) {
1927        __ push_f(xmm0);
1928       } else {
1929         __ push(rcx);          // reserve space for argument
1930         __ fstp_s(at_rsp());   // pass float argument on stack
1931       }
1932       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1933       break;
1934     case Bytecodes::_f2d:
1935       if (UseSSE < 1) {
1936         /* nothing to do */
1937       } else if (UseSSE == 1) {
1938         __ push_f(xmm0);
1939         __ pop_f();
1940       } else { // UseSSE >= 2
1941         __ cvtss2sd(xmm0, xmm0);
1942       }
1943       break;
1944     case Bytecodes::_d2i:
1945       if (UseSSE >= 2) {
1946         __ push_d(xmm0);
1947       } else {
1948         __ push(rcx);          // reserve space for argument
1949         __ push(rcx);
1950         __ fstp_d(at_rsp());   // pass double argument on stack
1951       }
1952       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1953       break;
1954     case Bytecodes::_d2l:
1955       if (UseSSE >= 2) {
1956         __ push_d(xmm0);
1957       } else {
1958         __ push(rcx);          // reserve space for argument
1959         __ push(rcx);
1960         __ fstp_d(at_rsp());   // pass double argument on stack
1961       }
1962       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1963       break;
1964     case Bytecodes::_d2f:
1965       if (UseSSE <= 1) {
1966         __ push(rcx);          // reserve space for f2ieee()
1967         __ f2ieee();           // truncate to float size
1968         __ pop(rcx);           // adjust rsp
1969         if (UseSSE == 1) {
1970           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
1971           // the conversion is performed using the FPU in this case.
1972           __ push_f();
1973           __ pop_f(xmm0);
1974         }
1975       } else { // UseSSE >= 2
1976         __ cvtsd2ss(xmm0, xmm0);
1977       }
1978       break;
1979     default             :
1980       ShouldNotReachHere();
1981   }
1982 #endif
1983 }
1984 
1985 void TemplateTable::lcmp() {
1986   transition(ltos, itos);
1987 #ifdef _LP64
1988   Label done;
1989   __ pop_l(rdx);
1990   __ cmpq(rdx, rax);
1991   __ movl(rax, -1);
1992   __ jccb(Assembler::less, done);
1993   __ setb(Assembler::notEqual, rax);
1994   __ movzbl(rax, rax);
1995   __ bind(done);
1996 #else
1997 
1998   // y = rdx:rax
1999   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2000   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2001   __ mov(rax, rcx);
2002 #endif
2003 }
2004 
2005 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2006   if ((is_float && UseSSE >= 1) ||
2007       (!is_float && UseSSE >= 2)) {
2008     Label done;
2009     if (is_float) {
2010       // XXX get rid of pop here, use ... reg, mem32
2011       __ pop_f(xmm1);
2012       __ ucomiss(xmm1, xmm0);
2013     } else {
2014       // XXX get rid of pop here, use ... reg, mem64
2015       __ pop_d(xmm1);
2016       __ ucomisd(xmm1, xmm0);
2017     }
2018     if (unordered_result < 0) {
2019       __ movl(rax, -1);
2020       __ jccb(Assembler::parity, done);
2021       __ jccb(Assembler::below, done);
2022       __ setb(Assembler::notEqual, rdx);
2023       __ movzbl(rax, rdx);
2024     } else {
2025       __ movl(rax, 1);
2026       __ jccb(Assembler::parity, done);
2027       __ jccb(Assembler::above, done);
2028       __ movl(rax, 0);
2029       __ jccb(Assembler::equal, done);
2030       __ decrementl(rax);
2031     }
2032     __ bind(done);
2033   } else {
2034 #ifdef _LP64
2035     ShouldNotReachHere();
2036 #else
2037     if (is_float) {
2038       __ fld_s(at_rsp());
2039     } else {
2040       __ fld_d(at_rsp());
2041       __ pop(rdx);
2042     }
2043     __ pop(rcx);
2044     __ fcmp2int(rax, unordered_result < 0);
2045 #endif // _LP64
2046   }
2047 }
2048 
2049 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2050   __ get_method(rcx); // rcx holds method
2051   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2052                                      // holds bumped taken count
2053 
2054   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2055                              InvocationCounter::counter_offset();
2056   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2057                               InvocationCounter::counter_offset();
2058 
2059   // Load up edx with the branch displacement
2060   if (is_wide) {
2061     __ movl(rdx, at_bcp(1));
2062   } else {
2063     __ load_signed_short(rdx, at_bcp(1));
2064   }
2065   __ bswapl(rdx);
2066 
2067   if (!is_wide) {
2068     __ sarl(rdx, 16);
2069   }
2070   LP64_ONLY(__ movl2ptr(rdx, rdx));
2071 
2072   // Handle all the JSR stuff here, then exit.
2073   // It's much shorter and cleaner than intermingling with the non-JSR
2074   // normal-branch stuff occurring below.
2075   if (is_jsr) {
2076     // Pre-load the next target bytecode into rbx
2077     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2078 
2079     // compute return address as bci in rax
2080     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2081                         in_bytes(ConstMethod::codes_offset())));
2082     __ subptr(rax, Address(rcx, Method::const_offset()));
2083     // Adjust the bcp in r13 by the displacement in rdx
2084     __ addptr(rbcp, rdx);
2085     // jsr returns atos that is not an oop
2086     __ push_i(rax);
2087     __ dispatch_only(vtos);
2088     return;
2089   }
2090 
2091   // Normal (non-jsr) branch handling
2092 
2093   // Adjust the bcp in r13 by the displacement in rdx
2094   __ addptr(rbcp, rdx);
2095 
2096   assert(UseLoopCounter || !UseOnStackReplacement,
2097          "on-stack-replacement requires loop counters");
2098   Label backedge_counter_overflow;
2099   Label profile_method;
2100   Label dispatch;
2101   if (UseLoopCounter) {
2102     // increment backedge counter for backward branches
2103     // rax: MDO
2104     // rbx: MDO bumped taken-count
2105     // rcx: method
2106     // rdx: target offset
2107     // r13: target bcp
2108     // r14: locals pointer
2109     __ testl(rdx, rdx);             // check if forward or backward branch
2110     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2111 
2112     // check if MethodCounters exists
2113     Label has_counters;
2114     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2115     __ testptr(rax, rax);
2116     __ jcc(Assembler::notZero, has_counters);
2117     __ push(rdx);
2118     __ push(rcx);
2119     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2120                rcx);
2121     __ pop(rcx);
2122     __ pop(rdx);
2123     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2124     __ testptr(rax, rax);
2125     __ jcc(Assembler::zero, dispatch);
2126     __ bind(has_counters);
2127 
2128     if (TieredCompilation) {
2129       Label no_mdo;
2130       int increment = InvocationCounter::count_increment;
2131       if (ProfileInterpreter) {
2132         // Are we profiling?
2133         __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2134         __ testptr(rbx, rbx);
2135         __ jccb(Assembler::zero, no_mdo);
2136         // Increment the MDO backedge counter
2137         const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2138                                            in_bytes(InvocationCounter::counter_offset()));
2139         const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2140         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2141                                    rax, false, Assembler::zero, &backedge_counter_overflow);
2142         __ jmp(dispatch);
2143       }
2144       __ bind(no_mdo);
2145       // Increment backedge counter in MethodCounters*
2146       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2147       const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2148       __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2149                                  rax, false, Assembler::zero, &backedge_counter_overflow);
2150     } else { // not TieredCompilation
2151       // increment counter
2152       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2153       __ movl(rax, Address(rcx, be_offset));        // load backedge counter
2154       __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2155       __ movl(Address(rcx, be_offset), rax);        // store counter
2156 
2157       __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
2158 
2159       __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2160       __ addl(rax, Address(rcx, be_offset));        // add both counters
2161 
2162       if (ProfileInterpreter) {
2163         // Test to see if we should create a method data oop
2164         __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2165         __ jcc(Assembler::less, dispatch);
2166 
2167         // if no method data exists, go to profile method
2168         __ test_method_data_pointer(rax, profile_method);
2169 
2170         if (UseOnStackReplacement) {
2171           // check for overflow against rbx which is the MDO taken count
2172           __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2173           __ jcc(Assembler::below, dispatch);
2174 
2175           // When ProfileInterpreter is on, the backedge_count comes
2176           // from the MethodData*, which value does not get reset on
2177           // the call to frequency_counter_overflow().  To avoid
2178           // excessive calls to the overflow routine while the method is
2179           // being compiled, add a second test to make sure the overflow
2180           // function is called only once every overflow_frequency.
2181           const int overflow_frequency = 1024;
2182           __ andl(rbx, overflow_frequency - 1);
2183           __ jcc(Assembler::zero, backedge_counter_overflow);
2184 
2185         }
2186       } else {
2187         if (UseOnStackReplacement) {
2188           // check for overflow against rax, which is the sum of the
2189           // counters
2190           __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2191           __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2192 
2193         }
2194       }
2195     }
2196     __ bind(dispatch);
2197   }
2198 
2199   // Pre-load the next target bytecode into rbx
2200   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2201 
2202   // continue with the bytecode @ target
2203   // rax: return bci for jsr's, unused otherwise
2204   // rbx: target bytecode
2205   // r13: target bcp
2206   __ dispatch_only(vtos);
2207 
2208   if (UseLoopCounter) {
2209     if (ProfileInterpreter) {
2210       // Out-of-line code to allocate method data oop.
2211       __ bind(profile_method);
2212       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2213       __ set_method_data_pointer_for_bcp();
2214       __ jmp(dispatch);
2215     }
2216 
2217     if (UseOnStackReplacement) {
2218       // invocation counter overflow
2219       __ bind(backedge_counter_overflow);
2220       __ negptr(rdx);
2221       __ addptr(rdx, rbcp); // branch bcp
2222       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2223       __ call_VM(noreg,
2224                  CAST_FROM_FN_PTR(address,
2225                                   InterpreterRuntime::frequency_counter_overflow),
2226                  rdx);
2227 
2228       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2229       // rdx: scratch
2230       // r14: locals pointer
2231       // r13: bcp
2232       __ testptr(rax, rax);                        // test result
2233       __ jcc(Assembler::zero, dispatch);         // no osr if null
2234       // nmethod may have been invalidated (VM may block upon call_VM return)
2235       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2236       __ jcc(Assembler::notEqual, dispatch);
2237 
2238       // We have the address of an on stack replacement routine in rax.
2239       // In preparation of invoking it, first we must migrate the locals
2240       // and monitors from off the interpreter frame on the stack.
2241       // Ensure to save the osr nmethod over the migration call,
2242       // it will be preserved in rbx.
2243       __ mov(rbx, rax);
2244 
2245       NOT_LP64(__ get_thread(rcx));
2246 
2247       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2248 
2249       // rax is OSR buffer, move it to expected parameter location
2250       LP64_ONLY(__ mov(j_rarg0, rax));
2251       NOT_LP64(__ mov(rcx, rax));
2252       // We use j_rarg definitions here so that registers don't conflict as parameter
2253       // registers change across platforms as we are in the midst of a calling
2254       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2255 
2256       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2257       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2258 
2259       // pop the interpreter frame
2260       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2261       __ leave();                                // remove frame anchor
2262       __ pop(retaddr);                           // get return address
2263       __ mov(rsp, sender_sp);                   // set sp to sender sp
2264       // Ensure compiled code always sees stack at proper alignment
2265       __ andptr(rsp, -(StackAlignmentInBytes));
2266 
2267       // unlike x86 we need no specialized return from compiled code
2268       // to the interpreter or the call stub.
2269 
2270       // push the return address
2271       __ push(retaddr);
2272 
2273       // and begin the OSR nmethod
2274       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2275     }
2276   }
2277 }
2278 
2279 void TemplateTable::if_0cmp(Condition cc) {
2280   transition(itos, vtos);
2281   // assume branch is more often taken than not (loops use backward branches)
2282   Label not_taken;
2283   __ testl(rax, rax);
2284   __ jcc(j_not(cc), not_taken);
2285   branch(false, false);
2286   __ bind(not_taken);
2287   __ profile_not_taken_branch(rax);
2288 }
2289 
2290 void TemplateTable::if_icmp(Condition cc) {
2291   transition(itos, vtos);
2292   // assume branch is more often taken than not (loops use backward branches)
2293   Label not_taken;
2294   __ pop_i(rdx);
2295   __ cmpl(rdx, rax);
2296   __ jcc(j_not(cc), not_taken);
2297   branch(false, false);
2298   __ bind(not_taken);
2299   __ profile_not_taken_branch(rax);
2300 }
2301 
2302 void TemplateTable::if_nullcmp(Condition cc) {
2303   transition(atos, vtos);
2304   // assume branch is more often taken than not (loops use backward branches)
2305   Label not_taken;
2306   __ testptr(rax, rax);
2307   __ jcc(j_not(cc), not_taken);
2308   branch(false, false);
2309   __ bind(not_taken);
2310   __ profile_not_taken_branch(rax);
2311 }
2312 
2313 void TemplateTable::if_acmp(Condition cc) {
2314   transition(atos, vtos);
2315   // assume branch is more often taken than not (loops use backward branches)
2316   Label not_taken;
2317   __ pop_ptr(rdx);
2318   __ cmpptr(rdx, rax);
2319   __ jcc(j_not(cc), not_taken);
2320   branch(false, false);
2321   __ bind(not_taken);
2322   __ profile_not_taken_branch(rax);
2323 }
2324 
2325 void TemplateTable::ret() {
2326   transition(vtos, vtos);
2327   locals_index(rbx);
2328   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2329   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2330   __ profile_ret(rbx, rcx);
2331   __ get_method(rax);
2332   __ movptr(rbcp, Address(rax, Method::const_offset()));
2333   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2334                       ConstMethod::codes_offset()));
2335   __ dispatch_next(vtos);
2336 }
2337 
2338 void TemplateTable::wide_ret() {
2339   transition(vtos, vtos);
2340   locals_index_wide(rbx);
2341   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2342   __ profile_ret(rbx, rcx);
2343   __ get_method(rax);
2344   __ movptr(rbcp, Address(rax, Method::const_offset()));
2345   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2346   __ dispatch_next(vtos);
2347 }
2348 
2349 void TemplateTable::tableswitch() {
2350   Label default_case, continue_execution;
2351   transition(itos, vtos);
2352 
2353   // align r13/rsi
2354   __ lea(rbx, at_bcp(BytesPerInt));
2355   __ andptr(rbx, -BytesPerInt);
2356   // load lo & hi
2357   __ movl(rcx, Address(rbx, BytesPerInt));
2358   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2359   __ bswapl(rcx);
2360   __ bswapl(rdx);
2361   // check against lo & hi
2362   __ cmpl(rax, rcx);
2363   __ jcc(Assembler::less, default_case);
2364   __ cmpl(rax, rdx);
2365   __ jcc(Assembler::greater, default_case);
2366   // lookup dispatch offset
2367   __ subl(rax, rcx);
2368   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2369   __ profile_switch_case(rax, rbx, rcx);
2370   // continue execution
2371   __ bind(continue_execution);
2372   __ bswapl(rdx);
2373   LP64_ONLY(__ movl2ptr(rdx, rdx));
2374   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2375   __ addptr(rbcp, rdx);
2376   __ dispatch_only(vtos);
2377   // handle default
2378   __ bind(default_case);
2379   __ profile_switch_default(rax);
2380   __ movl(rdx, Address(rbx, 0));
2381   __ jmp(continue_execution);
2382 }
2383 
2384 void TemplateTable::lookupswitch() {
2385   transition(itos, itos);
2386   __ stop("lookupswitch bytecode should have been rewritten");
2387 }
2388 
2389 void TemplateTable::fast_linearswitch() {
2390   transition(itos, vtos);
2391   Label loop_entry, loop, found, continue_execution;
2392   // bswap rax so we can avoid bswapping the table entries
2393   __ bswapl(rax);
2394   // align r13
2395   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2396                                     // this instruction (change offsets
2397                                     // below)
2398   __ andptr(rbx, -BytesPerInt);
2399   // set counter
2400   __ movl(rcx, Address(rbx, BytesPerInt));
2401   __ bswapl(rcx);
2402   __ jmpb(loop_entry);
2403   // table search
2404   __ bind(loop);
2405   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2406   __ jcc(Assembler::equal, found);
2407   __ bind(loop_entry);
2408   __ decrementl(rcx);
2409   __ jcc(Assembler::greaterEqual, loop);
2410   // default case
2411   __ profile_switch_default(rax);
2412   __ movl(rdx, Address(rbx, 0));
2413   __ jmp(continue_execution);
2414   // entry found -> get offset
2415   __ bind(found);
2416   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2417   __ profile_switch_case(rcx, rax, rbx);
2418   // continue execution
2419   __ bind(continue_execution);
2420   __ bswapl(rdx);
2421   __ movl2ptr(rdx, rdx);
2422   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2423   __ addptr(rbcp, rdx);
2424   __ dispatch_only(vtos);
2425 }
2426 
2427 void TemplateTable::fast_binaryswitch() {
2428   transition(itos, vtos);
2429   // Implementation using the following core algorithm:
2430   //
2431   // int binary_search(int key, LookupswitchPair* array, int n) {
2432   //   // Binary search according to "Methodik des Programmierens" by
2433   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2434   //   int i = 0;
2435   //   int j = n;
2436   //   while (i+1 < j) {
2437   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2438   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2439   //     // where a stands for the array and assuming that the (inexisting)
2440   //     // element a[n] is infinitely big.
2441   //     int h = (i + j) >> 1;
2442   //     // i < h < j
2443   //     if (key < array[h].fast_match()) {
2444   //       j = h;
2445   //     } else {
2446   //       i = h;
2447   //     }
2448   //   }
2449   //   // R: a[i] <= key < a[i+1] or Q
2450   //   // (i.e., if key is within array, i is the correct index)
2451   //   return i;
2452   // }
2453 
2454   // Register allocation
2455   const Register key   = rax; // already set (tosca)
2456   const Register array = rbx;
2457   const Register i     = rcx;
2458   const Register j     = rdx;
2459   const Register h     = rdi;
2460   const Register temp  = rsi;
2461 
2462   // Find array start
2463   NOT_LP64(__ save_bcp());
2464 
2465   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2466                                           // get rid of this
2467                                           // instruction (change
2468                                           // offsets below)
2469   __ andptr(array, -BytesPerInt);
2470 
2471   // Initialize i & j
2472   __ xorl(i, i);                            // i = 0;
2473   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2474 
2475   // Convert j into native byteordering
2476   __ bswapl(j);
2477 
2478   // And start
2479   Label entry;
2480   __ jmp(entry);
2481 
2482   // binary search loop
2483   {
2484     Label loop;
2485     __ bind(loop);
2486     // int h = (i + j) >> 1;
2487     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2488     __ sarl(h, 1);                               // h = (i + j) >> 1;
2489     // if (key < array[h].fast_match()) {
2490     //   j = h;
2491     // } else {
2492     //   i = h;
2493     // }
2494     // Convert array[h].match to native byte-ordering before compare
2495     __ movl(temp, Address(array, h, Address::times_8));
2496     __ bswapl(temp);
2497     __ cmpl(key, temp);
2498     // j = h if (key <  array[h].fast_match())
2499     __ cmov32(Assembler::less, j, h);
2500     // i = h if (key >= array[h].fast_match())
2501     __ cmov32(Assembler::greaterEqual, i, h);
2502     // while (i+1 < j)
2503     __ bind(entry);
2504     __ leal(h, Address(i, 1)); // i+1
2505     __ cmpl(h, j);             // i+1 < j
2506     __ jcc(Assembler::less, loop);
2507   }
2508 
2509   // end of binary search, result index is i (must check again!)
2510   Label default_case;
2511   // Convert array[i].match to native byte-ordering before compare
2512   __ movl(temp, Address(array, i, Address::times_8));
2513   __ bswapl(temp);
2514   __ cmpl(key, temp);
2515   __ jcc(Assembler::notEqual, default_case);
2516 
2517   // entry found -> j = offset
2518   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2519   __ profile_switch_case(i, key, array);
2520   __ bswapl(j);
2521   LP64_ONLY(__ movslq(j, j));
2522 
2523   NOT_LP64(__ restore_bcp());
2524   NOT_LP64(__ restore_locals());                           // restore rdi
2525 
2526   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2527   __ addptr(rbcp, j);
2528   __ dispatch_only(vtos);
2529 
2530   // default case -> j = default offset
2531   __ bind(default_case);
2532   __ profile_switch_default(i);
2533   __ movl(j, Address(array, -2 * BytesPerInt));
2534   __ bswapl(j);
2535   LP64_ONLY(__ movslq(j, j));
2536 
2537   NOT_LP64(__ restore_bcp());
2538   NOT_LP64(__ restore_locals());
2539 
2540   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2541   __ addptr(rbcp, j);
2542   __ dispatch_only(vtos);
2543 }
2544 
2545 void TemplateTable::_return(TosState state) {
2546   transition(state, state);
2547 
2548   assert(_desc->calls_vm(),
2549          "inconsistent calls_vm information"); // call in remove_activation
2550 
2551   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2552     assert(state == vtos, "only valid state");
2553     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2554     __ movptr(robj, aaddress(0));
2555     __ load_klass(rdi, robj);
2556     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2557     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2558     Label skip_register_finalizer;
2559     __ jcc(Assembler::zero, skip_register_finalizer);
2560 
2561     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2562 
2563     __ bind(skip_register_finalizer);
2564   }
2565 
2566   // Explicitly reset last_sp, for handling special case in TemplateInterpreter::deopt_reexecute_entry
2567   if (state == vtos) {
2568     __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
2569   }
2570 
2571   // Narrow result if state is itos but result type is smaller.
2572   // Need to narrow in the return bytecode rather than in generate_return_entry
2573   // since compiled code callers expect the result to already be narrowed.
2574   if (state == itos) {
2575     __ narrow(rax);
2576   }
2577   __ remove_activation(state, rbcp);
2578 
2579   __ jmp(rbcp);
2580 }
2581 
2582 // ----------------------------------------------------------------------------
2583 // Volatile variables demand their effects be made known to all CPU's
2584 // in order.  Store buffers on most chips allow reads & writes to
2585 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2586 // without some kind of memory barrier (i.e., it's not sufficient that
2587 // the interpreter does not reorder volatile references, the hardware
2588 // also must not reorder them).
2589 //
2590 // According to the new Java Memory Model (JMM):
2591 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2592 //     writes act as aquire & release, so:
2593 // (2) A read cannot let unrelated NON-volatile memory refs that
2594 //     happen after the read float up to before the read.  It's OK for
2595 //     non-volatile memory refs that happen before the volatile read to
2596 //     float down below it.
2597 // (3) Similar a volatile write cannot let unrelated NON-volatile
2598 //     memory refs that happen BEFORE the write float down to after the
2599 //     write.  It's OK for non-volatile memory refs that happen after the
2600 //     volatile write to float up before it.
2601 //
2602 // We only put in barriers around volatile refs (they are expensive),
2603 // not _between_ memory refs (that would require us to track the
2604 // flavor of the previous memory refs).  Requirements (2) and (3)
2605 // require some barriers before volatile stores and after volatile
2606 // loads.  These nearly cover requirement (1) but miss the
2607 // volatile-store-volatile-load case.  This final case is placed after
2608 // volatile-stores although it could just as well go before
2609 // volatile-loads.
2610 
2611 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2612   // Helper function to insert a is-volatile test and memory barrier
2613   if(!os::is_MP()) return;    // Not needed on single CPU
2614   __ membar(order_constraint);
2615 }
2616 
2617 void TemplateTable::resolve_cache_and_index(int byte_no,
2618                                             Register Rcache,
2619                                             Register index,
2620                                             size_t index_size) {
2621   const Register temp = rbx;
2622   assert_different_registers(Rcache, index, temp);
2623 
2624   Label resolved;
2625 
2626   Bytecodes::Code code = bytecode();
2627   switch (code) {
2628   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2629   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2630   default: break;
2631   }
2632 
2633   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2634   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2635   __ cmpl(temp, code);  // have we resolved this bytecode?
2636   __ jcc(Assembler::equal, resolved);
2637 
2638   // resolve first time through
2639   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2640   __ movl(temp, code);
2641   __ call_VM(noreg, entry, temp);
2642   // Update registers with resolved info
2643   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2644   __ bind(resolved);
2645 }
2646 
2647 // The cache and index registers must be set before call
2648 void TemplateTable::load_field_cp_cache_entry(Register obj,
2649                                               Register cache,
2650                                               Register index,
2651                                               Register off,
2652                                               Register flags,
2653                                               bool is_static = false) {
2654   assert_different_registers(cache, index, flags, off);
2655 
2656   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2657   // Field offset
2658   __ movptr(off, Address(cache, index, Address::times_ptr,
2659                          in_bytes(cp_base_offset +
2660                                   ConstantPoolCacheEntry::f2_offset())));
2661   // Flags
2662   __ movl(flags, Address(cache, index, Address::times_ptr,
2663                          in_bytes(cp_base_offset +
2664                                   ConstantPoolCacheEntry::flags_offset())));
2665 
2666   // klass overwrite register
2667   if (is_static) {
2668     __ movptr(obj, Address(cache, index, Address::times_ptr,
2669                            in_bytes(cp_base_offset +
2670                                     ConstantPoolCacheEntry::f1_offset())));
2671     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2672     __ movptr(obj, Address(obj, mirror_offset));
2673   }
2674 }
2675 
2676 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2677                                                Register method,
2678                                                Register itable_index,
2679                                                Register flags,
2680                                                bool is_invokevirtual,
2681                                                bool is_invokevfinal, /*unused*/
2682                                                bool is_invokedynamic) {
2683   // setup registers
2684   const Register cache = rcx;
2685   const Register index = rdx;
2686   assert_different_registers(method, flags);
2687   assert_different_registers(method, cache, index);
2688   assert_different_registers(itable_index, flags);
2689   assert_different_registers(itable_index, cache, index);
2690   // determine constant pool cache field offsets
2691   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2692   const int method_offset = in_bytes(
2693     ConstantPoolCache::base_offset() +
2694       ((byte_no == f2_byte)
2695        ? ConstantPoolCacheEntry::f2_offset()
2696        : ConstantPoolCacheEntry::f1_offset()));
2697   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2698                                     ConstantPoolCacheEntry::flags_offset());
2699   // access constant pool cache fields
2700   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2701                                     ConstantPoolCacheEntry::f2_offset());
2702 
2703   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2704   resolve_cache_and_index(byte_no, cache, index, index_size);
2705     __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2706 
2707   if (itable_index != noreg) {
2708     // pick up itable or appendix index from f2 also:
2709     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2710   }
2711   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2712 }
2713 
2714 // The registers cache and index expected to be set before call.
2715 // Correct values of the cache and index registers are preserved.
2716 void TemplateTable::jvmti_post_field_access(Register cache,
2717                                             Register index,
2718                                             bool is_static,
2719                                             bool has_tos) {
2720   if (JvmtiExport::can_post_field_access()) {
2721     // Check to see if a field access watch has been set before we take
2722     // the time to call into the VM.
2723     Label L1;
2724     assert_different_registers(cache, index, rax);
2725     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2726     __ testl(rax,rax);
2727     __ jcc(Assembler::zero, L1);
2728 
2729     // cache entry pointer
2730     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2731     __ shll(index, LogBytesPerWord);
2732     __ addptr(cache, index);
2733     if (is_static) {
2734       __ xorptr(rax, rax);      // NULL object reference
2735     } else {
2736       __ pop(atos);         // Get the object
2737       __ verify_oop(rax);
2738       __ push(atos);        // Restore stack state
2739     }
2740     // rax,:   object pointer or NULL
2741     // cache: cache entry pointer
2742     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2743                rax, cache);
2744     __ get_cache_and_index_at_bcp(cache, index, 1);
2745     __ bind(L1);
2746   }
2747 }
2748 
2749 void TemplateTable::pop_and_check_object(Register r) {
2750   __ pop_ptr(r);
2751   __ null_check(r);  // for field access must check obj.
2752   __ verify_oop(r);
2753 }
2754 
2755 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2756   transition(vtos, vtos);
2757 
2758   const Register cache = rcx;
2759   const Register index = rdx;
2760   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2761   const Register off   = rbx;
2762   const Register flags = rax;
2763   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2764 
2765   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2766   jvmti_post_field_access(cache, index, is_static, false);
2767   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2768 
2769   if (!is_static) pop_and_check_object(obj);
2770 
2771   const Address field(obj, off, Address::times_1, 0*wordSize);
2772   NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2773 
2774   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2775 
2776   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2777   // Make sure we don't need to mask edx after the above shift
2778   assert(btos == 0, "change code, btos != 0");
2779 
2780   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2781 
2782   __ jcc(Assembler::notZero, notByte);
2783   // btos
2784   __ load_signed_byte(rax, field);
2785   __ push(btos);
2786   // Rewrite bytecode to be faster
2787   if (!is_static && rc == may_rewrite) {
2788     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2789   }
2790   __ jmp(Done);
2791 
2792   __ bind(notByte);
2793   __ cmpl(flags, ztos);
2794   __ jcc(Assembler::notEqual, notBool);
2795 
2796   // ztos (same code as btos)
2797   __ load_signed_byte(rax, field);
2798   __ push(ztos);
2799   // Rewrite bytecode to be faster
2800   if (!is_static && rc == may_rewrite) {
2801     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2802     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2803   }
2804   __ jmp(Done);
2805 
2806   __ bind(notBool);
2807   __ cmpl(flags, atos);
2808   __ jcc(Assembler::notEqual, notObj);
2809   // atos
2810   __ load_heap_oop(rax, field);
2811   __ push(atos);
2812   if (!is_static && rc == may_rewrite) {
2813     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2814   }
2815   __ jmp(Done);
2816 
2817   __ bind(notObj);
2818   __ cmpl(flags, itos);
2819   __ jcc(Assembler::notEqual, notInt);
2820   // itos
2821   __ movl(rax, field);
2822   __ push(itos);
2823   // Rewrite bytecode to be faster
2824   if (!is_static && rc == may_rewrite) {
2825     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2826   }
2827   __ jmp(Done);
2828 
2829   __ bind(notInt);
2830   __ cmpl(flags, ctos);
2831   __ jcc(Assembler::notEqual, notChar);
2832   // ctos
2833   __ load_unsigned_short(rax, field);
2834   __ push(ctos);
2835   // Rewrite bytecode to be faster
2836   if (!is_static && rc == may_rewrite) {
2837     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2838   }
2839   __ jmp(Done);
2840 
2841   __ bind(notChar);
2842   __ cmpl(flags, stos);
2843   __ jcc(Assembler::notEqual, notShort);
2844   // stos
2845   __ load_signed_short(rax, field);
2846   __ push(stos);
2847   // Rewrite bytecode to be faster
2848   if (!is_static && rc == may_rewrite) {
2849     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2850   }
2851   __ jmp(Done);
2852 
2853   __ bind(notShort);
2854   __ cmpl(flags, ltos);
2855   __ jcc(Assembler::notEqual, notLong);
2856   // ltos
2857 
2858 #ifndef _LP64
2859   // Generate code as if volatile.  There just aren't enough registers to
2860   // save that information and this code is faster than the test.
2861   __ fild_d(field);                // Must load atomically
2862   __ subptr(rsp,2*wordSize);    // Make space for store
2863   __ fistp_d(Address(rsp,0));
2864   __ pop(rax);
2865   __ pop(rdx);
2866 #else
2867   __ movq(rax, field);
2868 #endif
2869 
2870   __ push(ltos);
2871   // Rewrite bytecode to be faster
2872   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2873   __ jmp(Done);
2874 
2875   __ bind(notLong);
2876   __ cmpl(flags, ftos);
2877   __ jcc(Assembler::notEqual, notFloat);
2878   // ftos
2879 
2880   __ load_float(field);
2881   __ push(ftos);
2882   // Rewrite bytecode to be faster
2883   if (!is_static && rc == may_rewrite) {
2884     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2885   }
2886   __ jmp(Done);
2887 
2888   __ bind(notFloat);
2889 #ifdef ASSERT
2890   __ cmpl(flags, dtos);
2891   __ jcc(Assembler::notEqual, notDouble);
2892 #endif
2893   // dtos
2894   __ load_double(field);
2895   __ push(dtos);
2896   // Rewrite bytecode to be faster
2897   if (!is_static && rc == may_rewrite) {
2898     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2899   }
2900 #ifdef ASSERT
2901   __ jmp(Done);
2902 
2903 
2904   __ bind(notDouble);
2905   __ stop("Bad state");
2906 #endif
2907 
2908   __ bind(Done);
2909   // [jk] not needed currently
2910   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2911   //                                              Assembler::LoadStore));
2912 }
2913 
2914 void TemplateTable::getfield(int byte_no) {
2915   getfield_or_static(byte_no, false);
2916 }
2917 
2918 void TemplateTable::nofast_getfield(int byte_no) {
2919   getfield_or_static(byte_no, false, may_not_rewrite);
2920 }
2921 
2922 void TemplateTable::getstatic(int byte_no) {
2923   getfield_or_static(byte_no, true);
2924 }
2925 
2926 
2927 // The registers cache and index expected to be set before call.
2928 // The function may destroy various registers, just not the cache and index registers.
2929 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2930 
2931   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
2932   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
2933   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
2934   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2935 
2936   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2937 
2938   if (JvmtiExport::can_post_field_modification()) {
2939     // Check to see if a field modification watch has been set before
2940     // we take the time to call into the VM.
2941     Label L1;
2942     assert_different_registers(cache, index, rax);
2943     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2944     __ testl(rax, rax);
2945     __ jcc(Assembler::zero, L1);
2946 
2947     __ get_cache_and_index_at_bcp(robj, RDX, 1);
2948 
2949 
2950     if (is_static) {
2951       // Life is simple.  Null out the object pointer.
2952       __ xorl(RBX, RBX);
2953 
2954     } else {
2955       // Life is harder. The stack holds the value on top, followed by
2956       // the object.  We don't know the size of the value, though; it
2957       // could be one or two words depending on its type. As a result,
2958       // we must find the type to determine where the object is.
2959 #ifndef _LP64
2960       Label two_word, valsize_known;
2961 #endif
2962       __ movl(RCX, Address(robj, RDX,
2963                            Address::times_ptr,
2964                            in_bytes(cp_base_offset +
2965                                      ConstantPoolCacheEntry::flags_offset())));
2966       NOT_LP64(__ mov(rbx, rsp));
2967       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2968 
2969       // Make sure we don't need to mask rcx after the above shift
2970       ConstantPoolCacheEntry::verify_tos_state_shift();
2971 #ifdef _LP64
2972       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2973       __ cmpl(c_rarg3, ltos);
2974       __ cmovptr(Assembler::equal,
2975                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2976       __ cmpl(c_rarg3, dtos);
2977       __ cmovptr(Assembler::equal,
2978                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2979 #else
2980       __ cmpl(rcx, ltos);
2981       __ jccb(Assembler::equal, two_word);
2982       __ cmpl(rcx, dtos);
2983       __ jccb(Assembler::equal, two_word);
2984       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2985       __ jmpb(valsize_known);
2986 
2987       __ bind(two_word);
2988       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2989 
2990       __ bind(valsize_known);
2991       // setup object pointer
2992       __ movptr(rbx, Address(rbx, 0));
2993 #endif
2994     }
2995     // cache entry pointer
2996     __ addptr(robj, in_bytes(cp_base_offset));
2997     __ shll(RDX, LogBytesPerWord);
2998     __ addptr(robj, RDX);
2999     // object (tos)
3000     __ mov(RCX, rsp);
3001     // c_rarg1: object pointer set up above (NULL if static)
3002     // c_rarg2: cache entry pointer
3003     // c_rarg3: jvalue object on the stack
3004     __ call_VM(noreg,
3005                CAST_FROM_FN_PTR(address,
3006                                 InterpreterRuntime::post_field_modification),
3007                RBX, robj, RCX);
3008     __ get_cache_and_index_at_bcp(cache, index, 1);
3009     __ bind(L1);
3010   }
3011 }
3012 
3013 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3014   transition(vtos, vtos);
3015 
3016   const Register cache = rcx;
3017   const Register index = rdx;
3018   const Register obj   = rcx;
3019   const Register off   = rbx;
3020   const Register flags = rax;
3021   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3022 
3023   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3024   jvmti_post_field_mod(cache, index, is_static);
3025   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3026 
3027   // [jk] not needed currently
3028   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3029   //                                              Assembler::StoreStore));
3030 
3031   Label notVolatile, Done;
3032   __ movl(rdx, flags);
3033   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3034   __ andl(rdx, 0x1);
3035 
3036   // field addresses
3037   const Address field(obj, off, Address::times_1, 0*wordSize);
3038   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3039 
3040   Label notByte, notBool, notInt, notShort, notChar,
3041         notLong, notFloat, notObj, notDouble;
3042 
3043   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3044 
3045   assert(btos == 0, "change code, btos != 0");
3046   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3047   __ jcc(Assembler::notZero, notByte);
3048 
3049   // btos
3050   {
3051     __ pop(btos);
3052     if (!is_static) pop_and_check_object(obj);
3053     __ movb(field, rax);
3054     if (!is_static && rc == may_rewrite) {
3055       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3056     }
3057     __ jmp(Done);
3058   }
3059 
3060   __ bind(notByte);
3061   __ cmpl(flags, ztos);
3062   __ jcc(Assembler::notEqual, notBool);
3063 
3064   // ztos
3065   {
3066     __ pop(ztos);
3067     if (!is_static) pop_and_check_object(obj);
3068     __ andl(rax, 0x1);
3069     __ movb(field, rax);
3070     if (!is_static && rc == may_rewrite) {
3071       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3072     }
3073     __ jmp(Done);
3074   }
3075 
3076   __ bind(notBool);
3077   __ cmpl(flags, atos);
3078   __ jcc(Assembler::notEqual, notObj);
3079 
3080   // atos
3081   {
3082     __ pop(atos);
3083     if (!is_static) pop_and_check_object(obj);
3084     // Store into the field
3085     do_oop_store(_masm, field, rax, _bs->kind(), false);
3086     if (!is_static && rc == may_rewrite) {
3087       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3088     }
3089     __ jmp(Done);
3090   }
3091 
3092   __ bind(notObj);
3093   __ cmpl(flags, itos);
3094   __ jcc(Assembler::notEqual, notInt);
3095 
3096   // itos
3097   {
3098     __ pop(itos);
3099     if (!is_static) pop_and_check_object(obj);
3100     __ movl(field, rax);
3101     if (!is_static && rc == may_rewrite) {
3102       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3103     }
3104     __ jmp(Done);
3105   }
3106 
3107   __ bind(notInt);
3108   __ cmpl(flags, ctos);
3109   __ jcc(Assembler::notEqual, notChar);
3110 
3111   // ctos
3112   {
3113     __ pop(ctos);
3114     if (!is_static) pop_and_check_object(obj);
3115     __ movw(field, rax);
3116     if (!is_static && rc == may_rewrite) {
3117       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3118     }
3119     __ jmp(Done);
3120   }
3121 
3122   __ bind(notChar);
3123   __ cmpl(flags, stos);
3124   __ jcc(Assembler::notEqual, notShort);
3125 
3126   // stos
3127   {
3128     __ pop(stos);
3129     if (!is_static) pop_and_check_object(obj);
3130     __ movw(field, rax);
3131     if (!is_static && rc == may_rewrite) {
3132       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3133     }
3134     __ jmp(Done);
3135   }
3136 
3137   __ bind(notShort);
3138   __ cmpl(flags, ltos);
3139   __ jcc(Assembler::notEqual, notLong);
3140 
3141   // ltos
3142 #ifdef _LP64
3143   {
3144     __ pop(ltos);
3145     if (!is_static) pop_and_check_object(obj);
3146     __ movq(field, rax);
3147     if (!is_static && rc == may_rewrite) {
3148       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3149     }
3150     __ jmp(Done);
3151   }
3152 #else
3153   {
3154     Label notVolatileLong;
3155     __ testl(rdx, rdx);
3156     __ jcc(Assembler::zero, notVolatileLong);
3157 
3158     __ pop(ltos);  // overwrites rdx, do this after testing volatile.
3159     if (!is_static) pop_and_check_object(obj);
3160 
3161     // Replace with real volatile test
3162     __ push(rdx);
3163     __ push(rax);                 // Must update atomically with FIST
3164     __ fild_d(Address(rsp,0));    // So load into FPU register
3165     __ fistp_d(field);            // and put into memory atomically
3166     __ addptr(rsp, 2*wordSize);
3167     // volatile_barrier();
3168     volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3169                                                  Assembler::StoreStore));
3170     // Don't rewrite volatile version
3171     __ jmp(notVolatile);
3172 
3173     __ bind(notVolatileLong);
3174 
3175     __ pop(ltos);  // overwrites rdx
3176     if (!is_static) pop_and_check_object(obj);
3177     __ movptr(hi, rdx);
3178     __ movptr(field, rax);
3179     // Don't rewrite to _fast_lputfield for potential volatile case.
3180     __ jmp(notVolatile);
3181   }
3182 #endif // _LP64
3183 
3184   __ bind(notLong);
3185   __ cmpl(flags, ftos);
3186   __ jcc(Assembler::notEqual, notFloat);
3187 
3188   // ftos
3189   {
3190     __ pop(ftos);
3191     if (!is_static) pop_and_check_object(obj);
3192     __ store_float(field);
3193     if (!is_static && rc == may_rewrite) {
3194       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3195     }
3196     __ jmp(Done);
3197   }
3198 
3199   __ bind(notFloat);
3200 #ifdef ASSERT
3201   __ cmpl(flags, dtos);
3202   __ jcc(Assembler::notEqual, notDouble);
3203 #endif
3204 
3205   // dtos
3206   {
3207     __ pop(dtos);
3208     if (!is_static) pop_and_check_object(obj);
3209     __ store_double(field);
3210     if (!is_static && rc == may_rewrite) {
3211       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3212     }
3213   }
3214 
3215 #ifdef ASSERT
3216   __ jmp(Done);
3217 
3218   __ bind(notDouble);
3219   __ stop("Bad state");
3220 #endif
3221 
3222   __ bind(Done);
3223 
3224   // Check for volatile store
3225   __ testl(rdx, rdx);
3226   __ jcc(Assembler::zero, notVolatile);
3227   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3228                                                Assembler::StoreStore));
3229   __ bind(notVolatile);
3230 }
3231 
3232 void TemplateTable::putfield(int byte_no) {
3233   putfield_or_static(byte_no, false);
3234 }
3235 
3236 void TemplateTable::nofast_putfield(int byte_no) {
3237   putfield_or_static(byte_no, false, may_not_rewrite);
3238 }
3239 
3240 void TemplateTable::putstatic(int byte_no) {
3241   putfield_or_static(byte_no, true);
3242 }
3243 
3244 void TemplateTable::jvmti_post_fast_field_mod() {
3245 
3246   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3247 
3248   if (JvmtiExport::can_post_field_modification()) {
3249     // Check to see if a field modification watch has been set before
3250     // we take the time to call into the VM.
3251     Label L2;
3252     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3253     __ testl(scratch, scratch);
3254     __ jcc(Assembler::zero, L2);
3255     __ pop_ptr(rbx);                  // copy the object pointer from tos
3256     __ verify_oop(rbx);
3257     __ push_ptr(rbx);                 // put the object pointer back on tos
3258     // Save tos values before call_VM() clobbers them. Since we have
3259     // to do it for every data type, we use the saved values as the
3260     // jvalue object.
3261     switch (bytecode()) {          // load values into the jvalue object
3262     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3263     case Bytecodes::_fast_bputfield: // fall through
3264     case Bytecodes::_fast_zputfield: // fall through
3265     case Bytecodes::_fast_sputfield: // fall through
3266     case Bytecodes::_fast_cputfield: // fall through
3267     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3268     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3269     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3270     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3271 
3272     default:
3273       ShouldNotReachHere();
3274     }
3275     __ mov(scratch, rsp);             // points to jvalue on the stack
3276     // access constant pool cache entry
3277     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3278     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3279     __ verify_oop(rbx);
3280     // rbx: object pointer copied above
3281     // c_rarg2: cache entry pointer
3282     // c_rarg3: jvalue object on the stack
3283     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3284     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3285 
3286     switch (bytecode()) {             // restore tos values
3287     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3288     case Bytecodes::_fast_bputfield: // fall through
3289     case Bytecodes::_fast_zputfield: // fall through
3290     case Bytecodes::_fast_sputfield: // fall through
3291     case Bytecodes::_fast_cputfield: // fall through
3292     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3293     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3294     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3295     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3296     default: break;
3297     }
3298     __ bind(L2);
3299   }
3300 }
3301 
3302 void TemplateTable::fast_storefield(TosState state) {
3303   transition(state, vtos);
3304 
3305   ByteSize base = ConstantPoolCache::base_offset();
3306 
3307   jvmti_post_fast_field_mod();
3308 
3309   // access constant pool cache
3310   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3311 
3312   // test for volatile with rdx but rdx is tos register for lputfield.
3313   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3314                        in_bytes(base +
3315                                 ConstantPoolCacheEntry::flags_offset())));
3316 
3317   // replace index with field offset from cache entry
3318   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3319                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3320 
3321   // [jk] not needed currently
3322   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3323   //                                              Assembler::StoreStore));
3324 
3325   Label notVolatile;
3326   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3327   __ andl(rdx, 0x1);
3328 
3329   // Get object from stack
3330   pop_and_check_object(rcx);
3331 
3332   // field address
3333   const Address field(rcx, rbx, Address::times_1);
3334 
3335   // access field
3336   switch (bytecode()) {
3337   case Bytecodes::_fast_aputfield:
3338     do_oop_store(_masm, field, rax, _bs->kind(), false);
3339     break;
3340   case Bytecodes::_fast_lputfield:
3341 #ifdef _LP64
3342   __ movq(field, rax);
3343 #else
3344   __ stop("should not be rewritten");
3345 #endif
3346     break;
3347   case Bytecodes::_fast_iputfield:
3348     __ movl(field, rax);
3349     break;
3350   case Bytecodes::_fast_zputfield:
3351     __ andl(rax, 0x1);  // boolean is true if LSB is 1
3352     // fall through to bputfield
3353   case Bytecodes::_fast_bputfield:
3354     __ movb(field, rax);
3355     break;
3356   case Bytecodes::_fast_sputfield:
3357     // fall through
3358   case Bytecodes::_fast_cputfield:
3359     __ movw(field, rax);
3360     break;
3361   case Bytecodes::_fast_fputfield:
3362     __ store_float(field);
3363     break;
3364   case Bytecodes::_fast_dputfield:
3365     __ store_double(field);
3366     break;
3367   default:
3368     ShouldNotReachHere();
3369   }
3370 
3371   // Check for volatile store
3372   __ testl(rdx, rdx);
3373   __ jcc(Assembler::zero, notVolatile);
3374   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3375                                                Assembler::StoreStore));
3376   __ bind(notVolatile);
3377 }
3378 
3379 void TemplateTable::fast_accessfield(TosState state) {
3380   transition(atos, state);
3381 
3382   // Do the JVMTI work here to avoid disturbing the register state below
3383   if (JvmtiExport::can_post_field_access()) {
3384     // Check to see if a field access watch has been set before we
3385     // take the time to call into the VM.
3386     Label L1;
3387     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3388     __ testl(rcx, rcx);
3389     __ jcc(Assembler::zero, L1);
3390     // access constant pool cache entry
3391     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3392     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3393     __ verify_oop(rax);
3394     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3395     LP64_ONLY(__ mov(c_rarg1, rax));
3396     // c_rarg1: object pointer copied above
3397     // c_rarg2: cache entry pointer
3398     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3399     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3400     __ pop_ptr(rax); // restore object pointer
3401     __ bind(L1);
3402   }
3403 
3404   // access constant pool cache
3405   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3406   // replace index with field offset from cache entry
3407   // [jk] not needed currently
3408   // if (os::is_MP()) {
3409   //   __ movl(rdx, Address(rcx, rbx, Address::times_8,
3410   //                        in_bytes(ConstantPoolCache::base_offset() +
3411   //                                 ConstantPoolCacheEntry::flags_offset())));
3412   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3413   //   __ andl(rdx, 0x1);
3414   // }
3415   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3416                          in_bytes(ConstantPoolCache::base_offset() +
3417                                   ConstantPoolCacheEntry::f2_offset())));
3418 
3419   // rax: object
3420   __ verify_oop(rax);
3421   __ null_check(rax);
3422   Address field(rax, rbx, Address::times_1);
3423 
3424   // access field
3425   switch (bytecode()) {
3426   case Bytecodes::_fast_agetfield:
3427     __ load_heap_oop(rax, field);
3428     __ verify_oop(rax);
3429     break;
3430   case Bytecodes::_fast_lgetfield:
3431 #ifdef _LP64
3432   __ movq(rax, field);
3433 #else
3434   __ stop("should not be rewritten");
3435 #endif
3436     break;
3437   case Bytecodes::_fast_igetfield:
3438     __ movl(rax, field);
3439     break;
3440   case Bytecodes::_fast_bgetfield:
3441     __ movsbl(rax, field);
3442     break;
3443   case Bytecodes::_fast_sgetfield:
3444     __ load_signed_short(rax, field);
3445     break;
3446   case Bytecodes::_fast_cgetfield:
3447     __ load_unsigned_short(rax, field);
3448     break;
3449   case Bytecodes::_fast_fgetfield:
3450     __ load_float(field);
3451     break;
3452   case Bytecodes::_fast_dgetfield:
3453     __ load_double(field);
3454     break;
3455   default:
3456     ShouldNotReachHere();
3457   }
3458   // [jk] not needed currently
3459   // if (os::is_MP()) {
3460   //   Label notVolatile;
3461   //   __ testl(rdx, rdx);
3462   //   __ jcc(Assembler::zero, notVolatile);
3463   //   __ membar(Assembler::LoadLoad);
3464   //   __ bind(notVolatile);
3465   //};
3466 }
3467 
3468 void TemplateTable::fast_xaccess(TosState state) {
3469   transition(vtos, state);
3470 
3471   // get receiver
3472   __ movptr(rax, aaddress(0));
3473   // access constant pool cache
3474   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3475   __ movptr(rbx,
3476             Address(rcx, rdx, Address::times_ptr,
3477                     in_bytes(ConstantPoolCache::base_offset() +
3478                              ConstantPoolCacheEntry::f2_offset())));
3479   // make sure exception is reported in correct bcp range (getfield is
3480   // next instruction)
3481   __ increment(rbcp);
3482   __ null_check(rax);
3483   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3484   switch (state) {
3485   case itos:
3486     __ movl(rax, field);
3487     break;
3488   case atos:
3489     __ load_heap_oop(rax, field);
3490     __ verify_oop(rax);
3491     break;
3492   case ftos:
3493     __ load_float(field);
3494     break;
3495   default:
3496     ShouldNotReachHere();
3497   }
3498 
3499   // [jk] not needed currently
3500   // if (os::is_MP()) {
3501   //   Label notVolatile;
3502   //   __ movl(rdx, Address(rcx, rdx, Address::times_8,
3503   //                        in_bytes(ConstantPoolCache::base_offset() +
3504   //                                 ConstantPoolCacheEntry::flags_offset())));
3505   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3506   //   __ testl(rdx, 0x1);
3507   //   __ jcc(Assembler::zero, notVolatile);
3508   //   __ membar(Assembler::LoadLoad);
3509   //   __ bind(notVolatile);
3510   // }
3511 
3512   __ decrement(rbcp);
3513 }
3514 
3515 //-----------------------------------------------------------------------------
3516 // Calls
3517 
3518 void TemplateTable::count_calls(Register method, Register temp) {
3519   // implemented elsewhere
3520   ShouldNotReachHere();
3521 }
3522 
3523 void TemplateTable::prepare_invoke(int byte_no,
3524                                    Register method,  // linked method (or i-klass)
3525                                    Register index,   // itable index, MethodType, etc.
3526                                    Register recv,    // if caller wants to see it
3527                                    Register flags    // if caller wants to test it
3528                                    ) {
3529   // determine flags
3530   const Bytecodes::Code code = bytecode();
3531   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3532   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3533   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3534   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3535   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3536   const bool load_receiver       = (recv  != noreg);
3537   const bool save_flags          = (flags != noreg);
3538   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3539   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3540   assert(flags == noreg || flags == rdx, "");
3541   assert(recv  == noreg || recv  == rcx, "");
3542 
3543   // setup registers & access constant pool cache
3544   if (recv  == noreg)  recv  = rcx;
3545   if (flags == noreg)  flags = rdx;
3546   assert_different_registers(method, index, recv, flags);
3547 
3548   // save 'interpreter return address'
3549   __ save_bcp();
3550 
3551   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3552 
3553   // maybe push appendix to arguments (just before return address)
3554   if (is_invokedynamic || is_invokehandle) {
3555     Label L_no_push;
3556     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3557     __ jcc(Assembler::zero, L_no_push);
3558     // Push the appendix as a trailing parameter.
3559     // This must be done before we get the receiver,
3560     // since the parameter_size includes it.
3561     __ push(rbx);
3562     __ mov(rbx, index);
3563     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3564     __ load_resolved_reference_at_index(index, rbx);
3565     __ pop(rbx);
3566     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3567     __ bind(L_no_push);
3568   }
3569 
3570   // load receiver if needed (after appendix is pushed so parameter size is correct)
3571   // Note: no return address pushed yet
3572   if (load_receiver) {
3573     __ movl(recv, flags);
3574     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3575     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3576     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3577     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3578     __ movptr(recv, recv_addr);
3579     __ verify_oop(recv);
3580   }
3581 
3582   if (save_flags) {
3583     __ movl(rbcp, flags);
3584   }
3585 
3586   // compute return type
3587   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3588   // Make sure we don't need to mask flags after the above shift
3589   ConstantPoolCacheEntry::verify_tos_state_shift();
3590   // load return address
3591   {
3592     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3593     ExternalAddress table(table_addr);
3594     LP64_ONLY(__ lea(rscratch1, table));
3595     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3596     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3597   }
3598 
3599   // push return address
3600   __ push(flags);
3601 
3602   // Restore flags value from the constant pool cache, and restore rsi
3603   // for later null checks.  r13 is the bytecode pointer
3604   if (save_flags) {
3605     __ movl(flags, rbcp);
3606     __ restore_bcp();
3607   }
3608 }
3609 
3610 void TemplateTable::invokevirtual_helper(Register index,
3611                                          Register recv,
3612                                          Register flags) {
3613   // Uses temporary registers rax, rdx
3614   assert_different_registers(index, recv, rax, rdx);
3615   assert(index == rbx, "");
3616   assert(recv  == rcx, "");
3617 
3618   // Test for an invoke of a final method
3619   Label notFinal;
3620   __ movl(rax, flags);
3621   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3622   __ jcc(Assembler::zero, notFinal);
3623 
3624   const Register method = index;  // method must be rbx
3625   assert(method == rbx,
3626          "Method* must be rbx for interpreter calling convention");
3627 
3628   // do the call - the index is actually the method to call
3629   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3630 
3631   // It's final, need a null check here!
3632   __ null_check(recv);
3633 
3634   // profile this call
3635   __ profile_final_call(rax);
3636   __ profile_arguments_type(rax, method, rbcp, true);
3637 
3638   __ jump_from_interpreted(method, rax);
3639 
3640   __ bind(notFinal);
3641 
3642   // get receiver klass
3643   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3644   __ load_klass(rax, recv);
3645 
3646   // profile this call
3647   __ profile_virtual_call(rax, rlocals, rdx);
3648   // get target Method* & entry point
3649   __ lookup_virtual_method(rax, index, method);
3650   __ profile_called_method(method, rdx, rbcp);
3651 
3652   __ profile_arguments_type(rdx, method, rbcp, true);
3653   __ jump_from_interpreted(method, rdx);
3654 }
3655 
3656 void TemplateTable::invokevirtual(int byte_no) {
3657   transition(vtos, vtos);
3658   assert(byte_no == f2_byte, "use this argument");
3659   prepare_invoke(byte_no,
3660                  rbx,    // method or vtable index
3661                  noreg,  // unused itable index
3662                  rcx, rdx); // recv, flags
3663 
3664   // rbx: index
3665   // rcx: receiver
3666   // rdx: flags
3667 
3668   invokevirtual_helper(rbx, rcx, rdx);
3669 }
3670 
3671 void TemplateTable::invokespecial(int byte_no) {
3672   transition(vtos, vtos);
3673   assert(byte_no == f1_byte, "use this argument");
3674   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3675                  rcx);  // get receiver also for null check
3676   __ verify_oop(rcx);
3677   __ null_check(rcx);
3678   // do the call
3679   __ profile_call(rax);
3680   __ profile_arguments_type(rax, rbx, rbcp, false);
3681   __ jump_from_interpreted(rbx, rax);
3682 }
3683 
3684 void TemplateTable::invokestatic(int byte_no) {
3685   transition(vtos, vtos);
3686   assert(byte_no == f1_byte, "use this argument");
3687   prepare_invoke(byte_no, rbx);  // get f1 Method*
3688   // do the call
3689   __ profile_call(rax);
3690   __ profile_arguments_type(rax, rbx, rbcp, false);
3691   __ jump_from_interpreted(rbx, rax);
3692 }
3693 
3694 
3695 void TemplateTable::fast_invokevfinal(int byte_no) {
3696   transition(vtos, vtos);
3697   assert(byte_no == f2_byte, "use this argument");
3698   __ stop("fast_invokevfinal not used on x86");
3699 }
3700 
3701 
3702 void TemplateTable::invokeinterface(int byte_no) {
3703   transition(vtos, vtos);
3704   assert(byte_no == f1_byte, "use this argument");
3705   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 itable index
3706                  rcx, rdx); // recv, flags
3707 
3708   // rax: interface klass (from f1)
3709   // rbx: itable index (from f2)
3710   // rcx: receiver
3711   // rdx: flags
3712 
3713   // Special case of invokeinterface called for virtual method of
3714   // java.lang.Object.  See cpCacheOop.cpp for details.
3715   // This code isn't produced by javac, but could be produced by
3716   // another compliant java compiler.
3717   Label notMethod;
3718   __ movl(rlocals, rdx);
3719   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3720 
3721   __ jcc(Assembler::zero, notMethod);
3722 
3723   invokevirtual_helper(rbx, rcx, rdx);
3724   __ bind(notMethod);
3725 
3726   // Get receiver klass into rdx - also a null check
3727   __ restore_locals();  // restore r14
3728   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3729   __ load_klass(rdx, rcx);
3730 
3731   // profile this call
3732   __ profile_virtual_call(rdx, rbcp, rlocals);
3733 
3734   Label no_such_interface, no_such_method;
3735 
3736   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3737                              rdx, rax, rbx,
3738                              // outputs: method, scan temp. reg
3739                              rbx, rbcp,
3740                              no_such_interface);
3741 
3742   // rbx: Method* to call
3743   // rcx: receiver
3744   // Check for abstract method error
3745   // Note: This should be done more efficiently via a throw_abstract_method_error
3746   //       interpreter entry point and a conditional jump to it in case of a null
3747   //       method.
3748   __ testptr(rbx, rbx);
3749   __ jcc(Assembler::zero, no_such_method);
3750 
3751   __ profile_called_method(rbx, rbcp, rdx);
3752   __ profile_arguments_type(rdx, rbx, rbcp, true);
3753 
3754   // do the call
3755   // rcx: receiver
3756   // rbx,: Method*
3757   __ jump_from_interpreted(rbx, rdx);
3758   __ should_not_reach_here();
3759 
3760   // exception handling code follows...
3761   // note: must restore interpreter registers to canonical
3762   //       state for exception handling to work correctly!
3763 
3764   __ bind(no_such_method);
3765   // throw exception
3766   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3767   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3768   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3769   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3770   // the call_VM checks for exception, so we should never return here.
3771   __ should_not_reach_here();
3772 
3773   __ bind(no_such_interface);
3774   // throw exception
3775   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3776   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3777   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3778   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3779                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3780   // the call_VM checks for exception, so we should never return here.
3781   __ should_not_reach_here();
3782 }
3783 
3784 void TemplateTable::invokehandle(int byte_no) {
3785   transition(vtos, vtos);
3786   assert(byte_no == f1_byte, "use this argument");
3787   const Register rbx_method = rbx;
3788   const Register rax_mtype  = rax;
3789   const Register rcx_recv   = rcx;
3790   const Register rdx_flags  = rdx;
3791 
3792   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3793   __ verify_method_ptr(rbx_method);
3794   __ verify_oop(rcx_recv);
3795   __ null_check(rcx_recv);
3796 
3797   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3798   // rbx: MH.invokeExact_MT method (from f2)
3799 
3800   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3801 
3802   // FIXME: profile the LambdaForm also
3803   __ profile_final_call(rax);
3804   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3805 
3806   __ jump_from_interpreted(rbx_method, rdx);
3807 }
3808 
3809 void TemplateTable::invokedynamic(int byte_no) {
3810   transition(vtos, vtos);
3811   assert(byte_no == f1_byte, "use this argument");
3812 
3813   const Register rbx_method   = rbx;
3814   const Register rax_callsite = rax;
3815 
3816   prepare_invoke(byte_no, rbx_method, rax_callsite);
3817 
3818   // rax: CallSite object (from cpool->resolved_references[f1])
3819   // rbx: MH.linkToCallSite method (from f2)
3820 
3821   // Note:  rax_callsite is already pushed by prepare_invoke
3822 
3823   // %%% should make a type profile for any invokedynamic that takes a ref argument
3824   // profile this call
3825   __ profile_call(rbcp);
3826   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3827 
3828   __ verify_oop(rax_callsite);
3829 
3830   __ jump_from_interpreted(rbx_method, rdx);
3831 }
3832 
3833 //-----------------------------------------------------------------------------
3834 // Allocation
3835 
3836 void TemplateTable::_new() {
3837   transition(vtos, atos);
3838   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3839   Label slow_case;
3840   Label slow_case_no_pop;
3841   Label done;
3842   Label initialize_header;
3843   Label initialize_object;  // including clearing the fields
3844   Label allocate_shared;
3845 
3846   __ get_cpool_and_tags(rcx, rax);
3847 
3848   // Make sure the class we're about to instantiate has been resolved.
3849   // This is done before loading InstanceKlass to be consistent with the order
3850   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3851   const int tags_offset = Array<u1>::base_offset_in_bytes();
3852   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3853   __ jcc(Assembler::notEqual, slow_case_no_pop);
3854 
3855   // get InstanceKlass
3856   __ load_resolved_klass_at_index(rcx, rdx, rcx);
3857   __ push(rcx);  // save the contexts of klass for initializing the header
3858 
3859   // make sure klass is initialized & doesn't have finalizer
3860   // make sure klass is fully initialized
3861   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3862   __ jcc(Assembler::notEqual, slow_case);
3863 
3864   // get instance_size in InstanceKlass (scaled to a count of bytes)
3865   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3866   // test to see if it has a finalizer or is malformed in some way
3867   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3868   __ jcc(Assembler::notZero, slow_case);
3869 
3870   //
3871   // Allocate the instance
3872   // 1) Try to allocate in the TLAB
3873   // 2) if fail and the object is large allocate in the shared Eden
3874   // 3) if the above fails (or is not applicable), go to a slow case
3875   // (creates a new TLAB, etc.)
3876 
3877   const bool allow_shared_alloc =
3878     Universe::heap()->supports_inline_contig_alloc();
3879 
3880   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3881 #ifndef _LP64
3882   if (UseTLAB || allow_shared_alloc) {
3883     __ get_thread(thread);
3884   }
3885 #endif // _LP64
3886 
3887   if (UseTLAB) {
3888     __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3889     __ lea(rbx, Address(rax, rdx, Address::times_1));
3890     __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3891     __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3892     __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3893     if (ZeroTLAB) {
3894       // the fields have been already cleared
3895       __ jmp(initialize_header);
3896     } else {
3897       // initialize both the header and fields
3898       __ jmp(initialize_object);
3899     }
3900   }
3901 
3902   // Allocation in the shared Eden, if allowed.
3903   //
3904   // rdx: instance size in bytes
3905   if (allow_shared_alloc) {
3906     __ bind(allocate_shared);
3907 
3908     ExternalAddress heap_top((address)Universe::heap()->top_addr());
3909     ExternalAddress heap_end((address)Universe::heap()->end_addr());
3910 
3911     Label retry;
3912     __ bind(retry);
3913     __ movptr(rax, heap_top);
3914     __ lea(rbx, Address(rax, rdx, Address::times_1));
3915     __ cmpptr(rbx, heap_end);
3916     __ jcc(Assembler::above, slow_case);
3917 
3918     // Compare rax, with the top addr, and if still equal, store the new
3919     // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3920     // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3921     //
3922     // rax,: object begin
3923     // rbx,: object end
3924     // rdx: instance size in bytes
3925     __ locked_cmpxchgptr(rbx, heap_top);
3926 
3927     // if someone beat us on the allocation, try again, otherwise continue
3928     __ jcc(Assembler::notEqual, retry);
3929 
3930     __ incr_allocated_bytes(thread, rdx, 0);
3931   }
3932 
3933   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3934     // The object is initialized before the header.  If the object size is
3935     // zero, go directly to the header initialization.
3936     __ bind(initialize_object);
3937     __ decrement(rdx, sizeof(oopDesc));
3938     __ jcc(Assembler::zero, initialize_header);
3939 
3940     // Initialize topmost object field, divide rdx by 8, check if odd and
3941     // test if zero.
3942     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
3943     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3944 
3945     // rdx must have been multiple of 8
3946 #ifdef ASSERT
3947     // make sure rdx was multiple of 8
3948     Label L;
3949     // Ignore partial flag stall after shrl() since it is debug VM
3950     __ jccb(Assembler::carryClear, L);
3951     __ stop("object size is not multiple of 2 - adjust this code");
3952     __ bind(L);
3953     // rdx must be > 0, no extra check needed here
3954 #endif
3955 
3956     // initialize remaining object fields: rdx was a multiple of 8
3957     { Label loop;
3958     __ bind(loop);
3959     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3960     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3961     __ decrement(rdx);
3962     __ jcc(Assembler::notZero, loop);
3963     }
3964 
3965     // initialize object header only.
3966     __ bind(initialize_header);
3967     if (UseBiasedLocking) {
3968       __ pop(rcx);   // get saved klass back in the register.
3969       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3970       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3971     } else {
3972       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3973                 (intptr_t)markOopDesc::prototype()); // header
3974       __ pop(rcx);   // get saved klass back in the register.
3975     }
3976 #ifdef _LP64
3977     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3978     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
3979 #endif
3980     __ store_klass(rax, rcx);  // klass
3981 
3982     {
3983       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3984       // Trigger dtrace event for fastpath
3985       __ push(atos);
3986       __ call_VM_leaf(
3987            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3988       __ pop(atos);
3989     }
3990 
3991     __ jmp(done);
3992   }
3993 
3994   // slow case
3995   __ bind(slow_case);
3996   __ pop(rcx);   // restore stack pointer to what it was when we came in.
3997   __ bind(slow_case_no_pop);
3998 
3999   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4000   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4001 
4002   __ get_constant_pool(rarg1);
4003   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4004   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4005    __ verify_oop(rax);
4006 
4007   // continue
4008   __ bind(done);
4009 }
4010 
4011 void TemplateTable::newarray() {
4012   transition(itos, atos);
4013   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4014   __ load_unsigned_byte(rarg1, at_bcp(1));
4015   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4016           rarg1, rax);
4017 }
4018 
4019 void TemplateTable::anewarray() {
4020   transition(itos, atos);
4021 
4022   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4023   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4024 
4025   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4026   __ get_constant_pool(rarg1);
4027   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4028           rarg1, rarg2, rax);
4029 }
4030 
4031 void TemplateTable::arraylength() {
4032   transition(atos, itos);
4033   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4034   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4035 }
4036 
4037 void TemplateTable::checkcast() {
4038   transition(atos, atos);
4039   Label done, is_null, ok_is_subtype, quicked, resolved;
4040   __ testptr(rax, rax); // object is in rax
4041   __ jcc(Assembler::zero, is_null);
4042 
4043   // Get cpool & tags index
4044   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4045   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4046   // See if bytecode has already been quicked
4047   __ cmpb(Address(rdx, rbx,
4048                   Address::times_1,
4049                   Array<u1>::base_offset_in_bytes()),
4050           JVM_CONSTANT_Class);
4051   __ jcc(Assembler::equal, quicked);
4052   __ push(atos); // save receiver for result, and for GC
4053   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4054 
4055   // vm_result_2 has metadata result
4056 #ifndef _LP64
4057   // borrow rdi from locals
4058   __ get_thread(rdi);
4059   __ get_vm_result_2(rax, rdi);
4060   __ restore_locals();
4061 #else
4062   __ get_vm_result_2(rax, r15_thread);
4063 #endif
4064 
4065   __ pop_ptr(rdx); // restore receiver
4066   __ jmpb(resolved);
4067 
4068   // Get superklass in rax and subklass in rbx
4069   __ bind(quicked);
4070   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4071   __ load_resolved_klass_at_index(rcx, rbx, rax);
4072 
4073   __ bind(resolved);
4074   __ load_klass(rbx, rdx);
4075 
4076   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4077   // Superklass in rax.  Subklass in rbx.
4078   __ gen_subtype_check(rbx, ok_is_subtype);
4079 
4080   // Come here on failure
4081   __ push_ptr(rdx);
4082   // object is at TOS
4083   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4084 
4085   // Come here on success
4086   __ bind(ok_is_subtype);
4087   __ mov(rax, rdx); // Restore object in rdx
4088 
4089   // Collect counts on whether this check-cast sees NULLs a lot or not.
4090   if (ProfileInterpreter) {
4091     __ jmp(done);
4092     __ bind(is_null);
4093     __ profile_null_seen(rcx);
4094   } else {
4095     __ bind(is_null);   // same as 'done'
4096   }
4097   __ bind(done);
4098 }
4099 
4100 void TemplateTable::instanceof() {
4101   transition(atos, itos);
4102   Label done, is_null, ok_is_subtype, quicked, resolved;
4103   __ testptr(rax, rax);
4104   __ jcc(Assembler::zero, is_null);
4105 
4106   // Get cpool & tags index
4107   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4108   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4109   // See if bytecode has already been quicked
4110   __ cmpb(Address(rdx, rbx,
4111                   Address::times_1,
4112                   Array<u1>::base_offset_in_bytes()),
4113           JVM_CONSTANT_Class);
4114   __ jcc(Assembler::equal, quicked);
4115 
4116   __ push(atos); // save receiver for result, and for GC
4117   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4118   // vm_result_2 has metadata result
4119 
4120 #ifndef _LP64
4121   // borrow rdi from locals
4122   __ get_thread(rdi);
4123   __ get_vm_result_2(rax, rdi);
4124   __ restore_locals();
4125 #else
4126   __ get_vm_result_2(rax, r15_thread);
4127 #endif
4128 
4129   __ pop_ptr(rdx); // restore receiver
4130   __ verify_oop(rdx);
4131   __ load_klass(rdx, rdx);
4132   __ jmpb(resolved);
4133 
4134   // Get superklass in rax and subklass in rdx
4135   __ bind(quicked);
4136   __ load_klass(rdx, rax);
4137   __ load_resolved_klass_at_index(rcx, rbx, rax);
4138 
4139   __ bind(resolved);
4140 
4141   // Generate subtype check.  Blows rcx, rdi
4142   // Superklass in rax.  Subklass in rdx.
4143   __ gen_subtype_check(rdx, ok_is_subtype);
4144 
4145   // Come here on failure
4146   __ xorl(rax, rax);
4147   __ jmpb(done);
4148   // Come here on success
4149   __ bind(ok_is_subtype);
4150   __ movl(rax, 1);
4151 
4152   // Collect counts on whether this test sees NULLs a lot or not.
4153   if (ProfileInterpreter) {
4154     __ jmp(done);
4155     __ bind(is_null);
4156     __ profile_null_seen(rcx);
4157   } else {
4158     __ bind(is_null);   // same as 'done'
4159   }
4160   __ bind(done);
4161   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4162   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4163 }
4164 
4165 
4166 //----------------------------------------------------------------------------------------------------
4167 // Breakpoints
4168 void TemplateTable::_breakpoint() {
4169   // Note: We get here even if we are single stepping..
4170   // jbug insists on setting breakpoints at every bytecode
4171   // even if we are in single step mode.
4172 
4173   transition(vtos, vtos);
4174 
4175   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4176 
4177   // get the unpatched byte code
4178   __ get_method(rarg);
4179   __ call_VM(noreg,
4180              CAST_FROM_FN_PTR(address,
4181                               InterpreterRuntime::get_original_bytecode_at),
4182              rarg, rbcp);
4183   __ mov(rbx, rax);  // why?
4184 
4185   // post the breakpoint event
4186   __ get_method(rarg);
4187   __ call_VM(noreg,
4188              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4189              rarg, rbcp);
4190 
4191   // complete the execution of original bytecode
4192   __ dispatch_only_normal(vtos);
4193 }
4194 
4195 //-----------------------------------------------------------------------------
4196 // Exceptions
4197 
4198 void TemplateTable::athrow() {
4199   transition(atos, vtos);
4200   __ null_check(rax);
4201   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4202 }
4203 
4204 //-----------------------------------------------------------------------------
4205 // Synchronization
4206 //
4207 // Note: monitorenter & exit are symmetric routines; which is reflected
4208 //       in the assembly code structure as well
4209 //
4210 // Stack layout:
4211 //
4212 // [expressions  ] <--- rsp               = expression stack top
4213 // ..
4214 // [expressions  ]
4215 // [monitor entry] <--- monitor block top = expression stack bot
4216 // ..
4217 // [monitor entry]
4218 // [frame data   ] <--- monitor block bot
4219 // ...
4220 // [saved rbp    ] <--- rbp
4221 void TemplateTable::monitorenter() {
4222   transition(atos, vtos);
4223 
4224   // check for NULL object
4225   __ null_check(rax);
4226 
4227   const Address monitor_block_top(
4228         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4229   const Address monitor_block_bot(
4230         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4231   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4232 
4233   Label allocated;
4234 
4235   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4236   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4237   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4238 
4239   // initialize entry pointer
4240   __ xorl(rmon, rmon); // points to free slot or NULL
4241 
4242   // find a free slot in the monitor block (result in rmon)
4243   {
4244     Label entry, loop, exit;
4245     __ movptr(rtop, monitor_block_top); // points to current entry,
4246                                         // starting with top-most entry
4247     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4248                                         // of monitor block
4249     __ jmpb(entry);
4250 
4251     __ bind(loop);
4252     // check if current entry is used
4253     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4254     // if not used then remember entry in rmon
4255     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4256     // check if current entry is for same object
4257     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4258     // if same object then stop searching
4259     __ jccb(Assembler::equal, exit);
4260     // otherwise advance to next entry
4261     __ addptr(rtop, entry_size);
4262     __ bind(entry);
4263     // check if bottom reached
4264     __ cmpptr(rtop, rbot);
4265     // if not at bottom then check this entry
4266     __ jcc(Assembler::notEqual, loop);
4267     __ bind(exit);
4268   }
4269 
4270   __ testptr(rmon, rmon); // check if a slot has been found
4271   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4272 
4273   // allocate one if there's no free slot
4274   {
4275     Label entry, loop;
4276     // 1. compute new pointers          // rsp: old expression stack top
4277     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4278     __ subptr(rsp, entry_size);         // move expression stack top
4279     __ subptr(rmon, entry_size);        // move expression stack bottom
4280     __ mov(rtop, rsp);                  // set start value for copy loop
4281     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4282     __ jmp(entry);
4283     // 2. move expression stack contents
4284     __ bind(loop);
4285     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4286                                                 // word from old location
4287     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4288     __ addptr(rtop, wordSize);                  // advance to next word
4289     __ bind(entry);
4290     __ cmpptr(rtop, rmon);                      // check if bottom reached
4291     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4292                                                 // copy next word
4293   }
4294 
4295   // call run-time routine
4296   // rmon: points to monitor entry
4297   __ bind(allocated);
4298 
4299   // Increment bcp to point to the next bytecode, so exception
4300   // handling for async. exceptions work correctly.
4301   // The object has already been poped from the stack, so the
4302   // expression stack looks correct.
4303   __ increment(rbcp);
4304 
4305   // store object
4306   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4307   __ lock_object(rmon);
4308 
4309   // check to make sure this monitor doesn't cause stack overflow after locking
4310   __ save_bcp();  // in case of exception
4311   __ generate_stack_overflow_check(0);
4312 
4313   // The bcp has already been incremented. Just need to dispatch to
4314   // next instruction.
4315   __ dispatch_next(vtos);
4316 }
4317 
4318 void TemplateTable::monitorexit() {
4319   transition(atos, vtos);
4320 
4321   // check for NULL object
4322   __ null_check(rax);
4323 
4324   const Address monitor_block_top(
4325         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4326   const Address monitor_block_bot(
4327         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4328   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4329 
4330   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4331   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4332 
4333   Label found;
4334 
4335   // find matching slot
4336   {
4337     Label entry, loop;
4338     __ movptr(rtop, monitor_block_top); // points to current entry,
4339                                         // starting with top-most entry
4340     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4341                                         // of monitor block
4342     __ jmpb(entry);
4343 
4344     __ bind(loop);
4345     // check if current entry is for same object
4346     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4347     // if same object then stop searching
4348     __ jcc(Assembler::equal, found);
4349     // otherwise advance to next entry
4350     __ addptr(rtop, entry_size);
4351     __ bind(entry);
4352     // check if bottom reached
4353     __ cmpptr(rtop, rbot);
4354     // if not at bottom then check this entry
4355     __ jcc(Assembler::notEqual, loop);
4356   }
4357 
4358   // error handling. Unlocking was not block-structured
4359   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4360                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4361   __ should_not_reach_here();
4362 
4363   // call run-time routine
4364   __ bind(found);
4365   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4366   __ unlock_object(rtop);
4367   __ pop_ptr(rax); // discard object
4368 }
4369 
4370 // Wide instructions
4371 void TemplateTable::wide() {
4372   transition(vtos, vtos);
4373   __ load_unsigned_byte(rbx, at_bcp(1));
4374   ExternalAddress wtable((address)Interpreter::_wentry_point);
4375   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4376   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4377 }
4378 
4379 // Multi arrays
4380 void TemplateTable::multianewarray() {
4381   transition(vtos, atos);
4382 
4383   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4384   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4385   // last dim is on top of stack; we want address of first one:
4386   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4387   // the latter wordSize to point to the beginning of the array.
4388   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4389   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4390   __ load_unsigned_byte(rbx, at_bcp(3));
4391   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4392 }