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_qputfield:
 245     __ jmp(L_patch_done); // don't patch yet
 246     break;
 247   case Bytecodes::_fast_aputfield:
 248   case Bytecodes::_fast_bputfield:
 249   case Bytecodes::_fast_zputfield:
 250   case Bytecodes::_fast_cputfield:
 251   case Bytecodes::_fast_dputfield:
 252   case Bytecodes::_fast_fputfield:
 253   case Bytecodes::_fast_iputfield:
 254   case Bytecodes::_fast_lputfield:
 255   case Bytecodes::_fast_sputfield:
 256     {
 257       // We skip bytecode quickening for putfield instructions when
 258       // the put_code written to the constant pool cache is zero.
 259       // This is required so that every execution of this instruction
 260       // calls out to InterpreterRuntime::resolve_get_put to do
 261       // additional, required work.
 262       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 263       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 264       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 265       __ movl(bc_reg, bc);
 266       __ cmpl(temp_reg, (int) 0);
 267       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 268     }
 269     break;
 270   default:
 271     assert(byte_no == -1, "sanity");
 272     // the pair bytecodes have already done the load.
 273     if (load_bc_into_bc_reg) {
 274       __ movl(bc_reg, bc);
 275     }
 276   }
 277 
 278   if (JvmtiExport::can_post_breakpoint()) {
 279     Label L_fast_patch;
 280     // if a breakpoint is present we can't rewrite the stream directly
 281     __ movzbl(temp_reg, at_bcp(0));
 282     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 283     __ jcc(Assembler::notEqual, L_fast_patch);
 284     __ get_method(temp_reg);
 285     // Let breakpoint table handling rewrite to quicker bytecode
 286     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 287 #ifndef ASSERT
 288     __ jmpb(L_patch_done);
 289 #else
 290     __ jmp(L_patch_done);
 291 #endif
 292     __ bind(L_fast_patch);
 293   }
 294 
 295 #ifdef ASSERT
 296   Label L_okay;
 297   __ load_unsigned_byte(temp_reg, at_bcp(0));
 298   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 299   __ jcc(Assembler::equal, L_okay);
 300   __ cmpl(temp_reg, bc_reg);
 301   __ jcc(Assembler::equal, L_okay);
 302   __ stop("patching the wrong bytecode");
 303   __ bind(L_okay);
 304 #endif
 305 
 306   // patch bytecode
 307   __ movb(at_bcp(0), bc_reg);
 308   __ bind(L_patch_done);
 309 }
 310 // Individual instructions
 311 
 312 
 313 void TemplateTable::nop() {
 314   transition(vtos, vtos);
 315   // nothing to do
 316 }
 317 
 318 void TemplateTable::shouldnotreachhere() {
 319   transition(vtos, vtos);
 320   __ stop("shouldnotreachhere bytecode");
 321 }
 322 
 323 void TemplateTable::aconst_null() {
 324   transition(vtos, atos);
 325   __ xorl(rax, rax);
 326 }
 327 
 328 void TemplateTable::iconst(int value) {
 329   transition(vtos, itos);
 330   if (value == 0) {
 331     __ xorl(rax, rax);
 332   } else {
 333     __ movl(rax, value);
 334   }
 335 }
 336 
 337 void TemplateTable::lconst(int value) {
 338   transition(vtos, ltos);
 339   if (value == 0) {
 340     __ xorl(rax, rax);
 341   } else {
 342     __ movl(rax, value);
 343   }
 344 #ifndef _LP64
 345   assert(value >= 0, "check this code");
 346   __ xorptr(rdx, rdx);
 347 #endif
 348 }
 349 
 350 
 351 
 352 void TemplateTable::fconst(int value) {
 353   transition(vtos, ftos);
 354   if (UseSSE >= 1) {
 355     static float one = 1.0f, two = 2.0f;
 356     switch (value) {
 357     case 0:
 358       __ xorps(xmm0, xmm0);
 359       break;
 360     case 1:
 361       __ movflt(xmm0, ExternalAddress((address) &one));
 362       break;
 363     case 2:
 364       __ movflt(xmm0, ExternalAddress((address) &two));
 365       break;
 366     default:
 367       ShouldNotReachHere();
 368       break;
 369     }
 370   } else {
 371 #ifdef _LP64
 372     ShouldNotReachHere();
 373 #else
 374            if (value == 0) { __ fldz();
 375     } else if (value == 1) { __ fld1();
 376     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 377     } else                 { ShouldNotReachHere();
 378     }
 379 #endif // _LP64
 380   }
 381 }
 382 
 383 void TemplateTable::dconst(int value) {
 384   transition(vtos, dtos);
 385   if (UseSSE >= 2) {
 386     static double one = 1.0;
 387     switch (value) {
 388     case 0:
 389       __ xorpd(xmm0, xmm0);
 390       break;
 391     case 1:
 392       __ movdbl(xmm0, ExternalAddress((address) &one));
 393       break;
 394     default:
 395       ShouldNotReachHere();
 396       break;
 397     }
 398   } else {
 399 #ifdef _LP64
 400     ShouldNotReachHere();
 401 #else
 402            if (value == 0) { __ fldz();
 403     } else if (value == 1) { __ fld1();
 404     } else                 { ShouldNotReachHere();
 405     }
 406 #endif
 407   }
 408 }
 409 
 410 void TemplateTable::bipush() {
 411   transition(vtos, itos);
 412   __ load_signed_byte(rax, at_bcp(1));
 413 }
 414 
 415 void TemplateTable::sipush() {
 416   transition(vtos, itos);
 417   __ load_unsigned_short(rax, at_bcp(1));
 418   __ bswapl(rax);
 419   __ sarl(rax, 16);
 420 }
 421 
 422 void TemplateTable::ldc(bool wide) {
 423   transition(vtos, vtos);
 424   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 425   Label call_ldc, notFloat, notClass, Done;
 426 
 427   if (wide) {
 428     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 429   } else {
 430     __ load_unsigned_byte(rbx, at_bcp(1));
 431   }
 432 
 433   __ get_cpool_and_tags(rcx, rax);
 434   const int base_offset = ConstantPool::header_size() * wordSize;
 435   const int tags_offset = Array<u1>::base_offset_in_bytes();
 436 
 437   // get type
 438   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 439 
 440   // unresolved class - get the resolved class
 441   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 442   __ jccb(Assembler::equal, call_ldc);
 443 
 444   // unresolved class in error state - call into runtime to throw the error
 445   // from the first resolution attempt
 446   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 447   __ jccb(Assembler::equal, call_ldc);
 448 
 449   // resolved class - need to call vm to get java mirror of the class
 450   __ cmpl(rdx, JVM_CONSTANT_Class);
 451   __ jcc(Assembler::equal, call_ldc);
 452 
 453   // unresolved value type - get the resolved class
 454   __ cmpl(rdx, JVM_CONSTANT_UnresolvedValue);
 455   __ jccb(Assembler::equal, call_ldc);
 456 
 457   // unresolved value type in error state - call into runtime to throw the error
 458   // from the first resolution attempt
 459   __ cmpl(rdx, JVM_CONSTANT_UnresolvedValueInError);
 460   __ jccb(Assembler::equal, call_ldc);
 461 
 462   // resolved value type - need to call vm to get java mirror
 463   __ cmpl(rdx, JVM_CONSTANT_Value);
 464   __ jcc(Assembler::notEqual, notClass);
 465 
 466   __ bind(call_ldc);
 467 
 468   __ movl(rarg, wide);
 469   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 470 
 471   __ push(atos);
 472   __ jmp(Done);
 473 
 474   __ bind(notClass);
 475   __ cmpl(rdx, JVM_CONSTANT_Float);
 476   __ jccb(Assembler::notEqual, notFloat);
 477 
 478   // ftos
 479   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 480   __ push(ftos);
 481   __ jmp(Done);
 482 
 483   __ bind(notFloat);
 484 #ifdef ASSERT
 485   {
 486     Label L;
 487     __ cmpl(rdx, JVM_CONSTANT_Integer);
 488     __ jcc(Assembler::equal, L);
 489     // String and Object are rewritten to fast_aldc
 490     __ stop("unexpected tag type in ldc");
 491     __ bind(L);
 492   }
 493 #endif
 494   // itos JVM_CONSTANT_Integer only
 495   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 496   __ push(itos);
 497   __ bind(Done);
 498 }
 499 
 500 // Fast path for caching oop constants.
 501 void TemplateTable::fast_aldc(bool wide) {
 502   transition(vtos, atos);
 503 
 504   Register result = rax;
 505   Register tmp = rdx;
 506   int index_size = wide ? sizeof(u2) : sizeof(u1);
 507 
 508   Label resolved;
 509 
 510   // We are resolved if the resolved reference cache entry contains a
 511   // non-null object (String, MethodType, etc.)
 512   assert_different_registers(result, tmp);
 513   __ get_cache_index_at_bcp(tmp, 1, index_size);
 514   __ load_resolved_reference_at_index(result, tmp);
 515   __ testl(result, result);
 516   __ jcc(Assembler::notZero, resolved);
 517 
 518   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 519 
 520   // first time invocation - must resolve first
 521   __ movl(tmp, (int)bytecode());
 522   __ call_VM(result, entry, tmp);
 523 
 524   __ bind(resolved);
 525 
 526   if (VerifyOops) {
 527     __ verify_oop(result);
 528   }
 529 }
 530 
 531 void TemplateTable::ldc2_w() {
 532   transition(vtos, vtos);
 533   Label Long, Done;
 534   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 535 
 536   __ get_cpool_and_tags(rcx, rax);
 537   const int base_offset = ConstantPool::header_size() * wordSize;
 538   const int tags_offset = Array<u1>::base_offset_in_bytes();
 539 
 540   // get type
 541   __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
 542           JVM_CONSTANT_Double);
 543   __ jccb(Assembler::notEqual, Long);
 544 
 545   // dtos
 546   __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
 547   __ push(dtos);
 548 
 549   __ jmpb(Done);
 550   __ bind(Long);
 551 
 552   // ltos
 553   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 554   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 555   __ push(ltos);
 556 
 557   __ bind(Done);
 558 }
 559 
 560 void TemplateTable::locals_index(Register reg, int offset) {
 561   __ load_unsigned_byte(reg, at_bcp(offset));
 562   __ negptr(reg);
 563 }
 564 
 565 void TemplateTable::iload() {
 566   iload_internal();
 567 }
 568 
 569 void TemplateTable::nofast_iload() {
 570   iload_internal(may_not_rewrite);
 571 }
 572 
 573 void TemplateTable::iload_internal(RewriteControl rc) {
 574   transition(vtos, itos);
 575   if (RewriteFrequentPairs && rc == may_rewrite) {
 576     Label rewrite, done;
 577     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 578     LP64_ONLY(assert(rbx != bc, "register damaged"));
 579 
 580     // get next byte
 581     __ load_unsigned_byte(rbx,
 582                           at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 583     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 584     // last two iloads in a pair.  Comparing against fast_iload means that
 585     // the next bytecode is neither an iload or a caload, and therefore
 586     // an iload pair.
 587     __ cmpl(rbx, Bytecodes::_iload);
 588     __ jcc(Assembler::equal, done);
 589 
 590     __ cmpl(rbx, Bytecodes::_fast_iload);
 591     __ movl(bc, Bytecodes::_fast_iload2);
 592 
 593     __ jccb(Assembler::equal, rewrite);
 594 
 595     // if _caload, rewrite to fast_icaload
 596     __ cmpl(rbx, Bytecodes::_caload);
 597     __ movl(bc, Bytecodes::_fast_icaload);
 598     __ jccb(Assembler::equal, rewrite);
 599 
 600     // rewrite so iload doesn't check again.
 601     __ movl(bc, Bytecodes::_fast_iload);
 602 
 603     // rewrite
 604     // bc: fast bytecode
 605     __ bind(rewrite);
 606     patch_bytecode(Bytecodes::_iload, bc, rbx, false);
 607     __ bind(done);
 608   }
 609 
 610   // Get the local value into tos
 611   locals_index(rbx);
 612   __ movl(rax, iaddress(rbx));
 613 }
 614 
 615 void TemplateTable::fast_iload2() {
 616   transition(vtos, itos);
 617   locals_index(rbx);
 618   __ movl(rax, iaddress(rbx));
 619   __ push(itos);
 620   locals_index(rbx, 3);
 621   __ movl(rax, iaddress(rbx));
 622 }
 623 
 624 void TemplateTable::fast_iload() {
 625   transition(vtos, itos);
 626   locals_index(rbx);
 627   __ movl(rax, iaddress(rbx));
 628 }
 629 
 630 void TemplateTable::lload() {
 631   transition(vtos, ltos);
 632   locals_index(rbx);
 633   __ movptr(rax, laddress(rbx));
 634   NOT_LP64(__ movl(rdx, haddress(rbx)));
 635 }
 636 
 637 void TemplateTable::fload() {
 638   transition(vtos, ftos);
 639   locals_index(rbx);
 640   __ load_float(faddress(rbx));
 641 }
 642 
 643 void TemplateTable::dload() {
 644   transition(vtos, dtos);
 645   locals_index(rbx);
 646   __ load_double(daddress(rbx));
 647 }
 648 
 649 void TemplateTable::aload() {
 650   transition(vtos, atos);
 651   locals_index(rbx);
 652   __ movptr(rax, aaddress(rbx));
 653 }
 654 
 655 void TemplateTable::vload() {
 656   transition(vtos, qtos);
 657   locals_index(rbx);
 658   __ movptr(rax, aaddress(rbx));
 659 }
 660 
 661 void TemplateTable::locals_index_wide(Register reg) {
 662   __ load_unsigned_short(reg, at_bcp(2));
 663   __ bswapl(reg);
 664   __ shrl(reg, 16);
 665   __ negptr(reg);
 666 }
 667 
 668 void TemplateTable::wide_iload() {
 669   transition(vtos, itos);
 670   locals_index_wide(rbx);
 671   __ movl(rax, iaddress(rbx));
 672 }
 673 
 674 void TemplateTable::wide_lload() {
 675   transition(vtos, ltos);
 676   locals_index_wide(rbx);
 677   __ movptr(rax, laddress(rbx));
 678   NOT_LP64(__ movl(rdx, haddress(rbx)));
 679 }
 680 
 681 void TemplateTable::wide_fload() {
 682   transition(vtos, ftos);
 683   locals_index_wide(rbx);
 684   __ load_float(faddress(rbx));
 685 }
 686 
 687 void TemplateTable::wide_dload() {
 688   transition(vtos, dtos);
 689   locals_index_wide(rbx);
 690   __ load_double(daddress(rbx));
 691 }
 692 
 693 void TemplateTable::wide_aload() {
 694   transition(vtos, atos);
 695   locals_index_wide(rbx);
 696   __ movptr(rax, aaddress(rbx));
 697 }
 698 
 699 void TemplateTable::wide_vload() {
 700   transition(vtos, qtos);
 701   locals_index_wide(rbx);
 702   __ movptr(rax, aaddress(rbx));
 703 }
 704 
 705 void TemplateTable::index_check(Register array, Register index) {
 706   // Pop ptr into array
 707   __ pop_ptr(array);
 708   index_check_without_pop(array, index);
 709 }
 710 
 711 void TemplateTable::index_check_without_pop(Register array, Register index) {
 712   // destroys rbx
 713   // check array
 714   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 715   // sign extend index for use by indexed load
 716   __ movl2ptr(index, index);
 717   // check index
 718   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 719   if (index != rbx) {
 720     // ??? convention: move aberrant index into rbx for exception message
 721     assert(rbx != array, "different registers");
 722     __ movl(rbx, index);
 723   }
 724   __ jump_cc(Assembler::aboveEqual,
 725              ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 726 }
 727 
 728 
 729 void TemplateTable::iaload() {
 730   transition(itos, itos);
 731   // rax: index
 732   // rdx: array
 733   index_check(rdx, rax); // kills rbx
 734   __ movl(rax, Address(rdx, rax,
 735                        Address::times_4,
 736                        arrayOopDesc::base_offset_in_bytes(T_INT)));
 737 }
 738 
 739 void TemplateTable::laload() {
 740   transition(itos, ltos);
 741   // rax: index
 742   // rdx: array
 743   index_check(rdx, rax); // kills rbx
 744   NOT_LP64(__ mov(rbx, rax));
 745   // rbx,: index
 746   __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
 747   NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
 748 }
 749 
 750 
 751 
 752 void TemplateTable::faload() {
 753   transition(itos, ftos);
 754   // rax: index
 755   // rdx: array
 756   index_check(rdx, rax); // kills rbx
 757   __ load_float(Address(rdx, rax,
 758                         Address::times_4,
 759                         arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 760 }
 761 
 762 void TemplateTable::daload() {
 763   transition(itos, dtos);
 764   // rax: index
 765   // rdx: array
 766   index_check(rdx, rax); // kills rbx
 767   __ load_double(Address(rdx, rax,
 768                          Address::times_8,
 769                          arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 770 }
 771 
 772 void TemplateTable::aaload() {
 773   transition(itos, atos);
 774   // rax: index
 775   // rdx: array
 776   index_check(rdx, rax); // kills rbx
 777   __ load_heap_oop(rax, Address(rdx, rax,
 778                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 779                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 780 }
 781 
 782 void TemplateTable::vaload() {
 783   transition(itos, qtos);
 784 
 785   Register array = rcx;
 786   Register index = rax;
 787 
 788   index_check(array, index); // kills rbx, pops array
 789 
 790   __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load) , array, index);
 791 }
 792 
 793 void TemplateTable::baload() {
 794   transition(itos, itos);
 795   // rax: index
 796   // rdx: array
 797   index_check(rdx, rax); // kills rbx
 798   __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 799 }
 800 
 801 void TemplateTable::caload() {
 802   transition(itos, itos);
 803   // rax: index
 804   // rdx: array
 805   index_check(rdx, rax); // kills rbx
 806   __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 807 }
 808 
 809 // iload followed by caload frequent pair
 810 void TemplateTable::fast_icaload() {
 811   transition(vtos, itos);
 812   // load index out of locals
 813   locals_index(rbx);
 814   __ movl(rax, iaddress(rbx));
 815 
 816   // rax: index
 817   // rdx: array
 818   index_check(rdx, rax); // kills rbx
 819   __ load_unsigned_short(rax,
 820                          Address(rdx, rax,
 821                                  Address::times_2,
 822                                  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 823 }
 824 
 825 
 826 void TemplateTable::saload() {
 827   transition(itos, itos);
 828   // rax: index
 829   // rdx: array
 830   index_check(rdx, rax); // kills rbx
 831   __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 832 }
 833 
 834 void TemplateTable::iload(int n) {
 835   transition(vtos, itos);
 836   __ movl(rax, iaddress(n));
 837 }
 838 
 839 void TemplateTable::lload(int n) {
 840   transition(vtos, ltos);
 841   __ movptr(rax, laddress(n));
 842   NOT_LP64(__ movptr(rdx, haddress(n)));
 843 }
 844 
 845 void TemplateTable::fload(int n) {
 846   transition(vtos, ftos);
 847   __ load_float(faddress(n));
 848 }
 849 
 850 void TemplateTable::dload(int n) {
 851   transition(vtos, dtos);
 852   __ load_double(daddress(n));
 853 }
 854 
 855 void TemplateTable::aload(int n) {
 856   transition(vtos, atos);
 857   __ movptr(rax, aaddress(n));
 858 }
 859 
 860 void TemplateTable::aload_0() {
 861   aload_0_internal();
 862 }
 863 
 864 void TemplateTable::nofast_aload_0() {
 865   aload_0_internal(may_not_rewrite);
 866 }
 867 
 868 void TemplateTable::aload_0_internal(RewriteControl rc) {
 869   transition(vtos, atos);
 870   // According to bytecode histograms, the pairs:
 871   //
 872   // _aload_0, _fast_igetfield
 873   // _aload_0, _fast_agetfield
 874   // _aload_0, _fast_fgetfield
 875   //
 876   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 877   // _aload_0 bytecode checks if the next bytecode is either
 878   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 879   // rewrites the current bytecode into a pair bytecode; otherwise it
 880   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 881   // the pair check anymore.
 882   //
 883   // Note: If the next bytecode is _getfield, the rewrite must be
 884   //       delayed, otherwise we may miss an opportunity for a pair.
 885   //
 886   // Also rewrite frequent pairs
 887   //   aload_0, aload_1
 888   //   aload_0, iload_1
 889   // These bytecodes with a small amount of code are most profitable
 890   // to rewrite
 891   if (RewriteFrequentPairs && rc == may_rewrite) {
 892     Label rewrite, done;
 893 
 894     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 895     LP64_ONLY(assert(rbx != bc, "register damaged"));
 896 
 897     // get next byte
 898     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 899 
 900     // if _getfield then wait with rewrite
 901     __ cmpl(rbx, Bytecodes::_getfield);
 902     __ jcc(Assembler::equal, done);
 903 
 904     // if _igetfield then rewrite to _fast_iaccess_0
 905     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 906     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 907     __ movl(bc, Bytecodes::_fast_iaccess_0);
 908     __ jccb(Assembler::equal, rewrite);
 909 
 910     // if _agetfield then rewrite to _fast_aaccess_0
 911     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 912     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 913     __ movl(bc, Bytecodes::_fast_aaccess_0);
 914     __ jccb(Assembler::equal, rewrite);
 915 
 916     // if _fgetfield then rewrite to _fast_faccess_0
 917     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 918     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 919     __ movl(bc, Bytecodes::_fast_faccess_0);
 920     __ jccb(Assembler::equal, rewrite);
 921 
 922     // else rewrite to _fast_aload0
 923     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 924     __ movl(bc, Bytecodes::_fast_aload_0);
 925 
 926     // rewrite
 927     // bc: fast bytecode
 928     __ bind(rewrite);
 929     patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
 930 
 931     __ bind(done);
 932   }
 933 
 934   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 935   aload(0);
 936 }
 937 
 938 void TemplateTable::istore() {
 939   transition(itos, vtos);
 940   locals_index(rbx);
 941   __ movl(iaddress(rbx), rax);
 942 }
 943 
 944 
 945 void TemplateTable::lstore() {
 946   transition(ltos, vtos);
 947   locals_index(rbx);
 948   __ movptr(laddress(rbx), rax);
 949   NOT_LP64(__ movptr(haddress(rbx), rdx));
 950 }
 951 
 952 void TemplateTable::fstore() {
 953   transition(ftos, vtos);
 954   locals_index(rbx);
 955   __ store_float(faddress(rbx));
 956 }
 957 
 958 void TemplateTable::dstore() {
 959   transition(dtos, vtos);
 960   locals_index(rbx);
 961   __ store_double(daddress(rbx));
 962 }
 963 
 964 void TemplateTable::astore() {
 965   transition(vtos, vtos);
 966   __ pop_ptr(rax);
 967   locals_index(rbx);
 968   __ movptr(aaddress(rbx), rax);
 969 }
 970 
 971 void TemplateTable::vstore() {
 972   transition(vtos, vtos);
 973   __ pop_ptr(rax);
 974   locals_index(rbx);
 975   __ movptr(aaddress(rbx), rax);
 976 }
 977 
 978 void TemplateTable::wide_istore() {
 979   transition(vtos, vtos);
 980   __ pop_i();
 981   locals_index_wide(rbx);
 982   __ movl(iaddress(rbx), rax);
 983 }
 984 
 985 void TemplateTable::wide_lstore() {
 986   transition(vtos, vtos);
 987   NOT_LP64(__ pop_l(rax, rdx));
 988   LP64_ONLY(__ pop_l());
 989   locals_index_wide(rbx);
 990   __ movptr(laddress(rbx), rax);
 991   NOT_LP64(__ movl(haddress(rbx), rdx));
 992 }
 993 
 994 void TemplateTable::wide_fstore() {
 995 #ifdef _LP64
 996   transition(vtos, vtos);
 997   __ pop_f(xmm0);
 998   locals_index_wide(rbx);
 999   __ movflt(faddress(rbx), xmm0);
1000 #else
1001   wide_istore();
1002 #endif
1003 }
1004 
1005 void TemplateTable::wide_dstore() {
1006 #ifdef _LP64
1007   transition(vtos, vtos);
1008   __ pop_d(xmm0);
1009   locals_index_wide(rbx);
1010   __ movdbl(daddress(rbx), xmm0);
1011 #else
1012   wide_lstore();
1013 #endif
1014 }
1015 
1016 void TemplateTable::wide_astore() {
1017   transition(vtos, vtos);
1018   __ pop_ptr(rax);
1019   locals_index_wide(rbx);
1020   __ movptr(aaddress(rbx), rax);
1021 }
1022 
1023 void TemplateTable::wide_vstore() {
1024   transition(vtos, vtos);
1025   __ pop_ptr(rax);
1026   locals_index_wide(rbx);
1027   __ movptr(aaddress(rbx), rax);
1028 }
1029 
1030 void TemplateTable::iastore() {
1031   transition(itos, vtos);
1032   __ pop_i(rbx);
1033   // rax: value
1034   // rbx: index
1035   // rdx: array
1036   index_check(rdx, rbx); // prefer index in rbx
1037   __ movl(Address(rdx, rbx,
1038                   Address::times_4,
1039                   arrayOopDesc::base_offset_in_bytes(T_INT)),
1040           rax);
1041 }
1042 
1043 void TemplateTable::lastore() {
1044   transition(ltos, vtos);
1045   __ pop_i(rbx);
1046   // rax,: low(value)
1047   // rcx: array
1048   // rdx: high(value)
1049   index_check(rcx, rbx);  // prefer index in rbx,
1050   // rbx,: index
1051   __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1052   NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1053 }
1054 
1055 
1056 void TemplateTable::fastore() {
1057   transition(ftos, vtos);
1058   __ pop_i(rbx);
1059   // value is in UseSSE >= 1 ? xmm0 : ST(0)
1060   // rbx:  index
1061   // rdx:  array
1062   index_check(rdx, rbx); // prefer index in rbx
1063   __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1064 }
1065 
1066 void TemplateTable::dastore() {
1067   transition(dtos, vtos);
1068   __ pop_i(rbx);
1069   // value is in UseSSE >= 2 ? xmm0 : ST(0)
1070   // rbx:  index
1071   // rdx:  array
1072   index_check(rdx, rbx); // prefer index in rbx
1073   __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1074 }
1075 
1076 void TemplateTable::aastore() {
1077   Label is_null, ok_is_subtype, done;
1078   transition(vtos, vtos);
1079   // stack: ..., array, index, value
1080   __ movptr(rax, at_tos());    // value
1081   __ movl(rcx, at_tos_p1()); // index
1082   __ movptr(rdx, at_tos_p2()); // array
1083 
1084   Address element_address(rdx, rcx,
1085                           UseCompressedOops? Address::times_4 : Address::times_ptr,
1086                           arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1087 
1088   index_check_without_pop(rdx, rcx);     // kills rbx
1089   __ testptr(rax, rax);
1090   __ jcc(Assembler::zero, is_null);
1091 
1092   // Move subklass into rbx
1093   __ load_klass(rbx, rax);
1094   // Move superklass into rax
1095   __ load_klass(rax, rdx);
1096   __ movptr(rax, Address(rax,
1097                          ObjArrayKlass::element_klass_offset()));
1098   // Compress array + index*oopSize + 12 into a single register.  Frees rcx.
1099   __ lea(rdx, element_address);
1100 
1101   // Generate subtype check.  Blows rcx, rdi
1102   // Superklass in rax.  Subklass in rbx.
1103   __ gen_subtype_check(rbx, ok_is_subtype);
1104 
1105   // Come here on failure
1106   // object is at TOS
1107   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1108 
1109   // Come here on success
1110   __ bind(ok_is_subtype);
1111 
1112   // Get the value we will store
1113   __ movptr(rax, at_tos());
1114   // Now store using the appropriate barrier
1115   do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1116   __ jmp(done);
1117 
1118   // Have a NULL in rax, rdx=array, ecx=index.  Store NULL at ary[idx]
1119   __ bind(is_null);
1120   __ profile_null_seen(rbx);
1121 
1122   // Store a NULL
1123   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1124 
1125   // Pop stack arguments
1126   __ bind(done);
1127   __ addptr(rsp, 3 * Interpreter::stackElementSize);
1128 }
1129 
1130 void TemplateTable::vastore() {
1131   transition(vtos, vtos);
1132 
1133   Register value = rcx;
1134   Register index = rbx;
1135   Register array = rax;
1136 
1137   // stack: ..., array, index, value
1138   __ pop_ptr(value);
1139   __ pop_i(index);
1140   __ pop_ptr(array);
1141 
1142   index_check_without_pop(array, index);
1143 
1144   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_store), array, index, value);
1145 }
1146 
1147 void TemplateTable::bastore() {
1148   transition(itos, vtos);
1149   __ pop_i(rbx);
1150   // rax: value
1151   // rbx: index
1152   // rdx: array
1153   index_check(rdx, rbx); // prefer index in rbx
1154   // Need to check whether array is boolean or byte
1155   // since both types share the bastore bytecode.
1156   __ load_klass(rcx, rdx);
1157   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1158   int diffbit = Klass::layout_helper_boolean_diffbit();
1159   __ testl(rcx, diffbit);
1160   Label L_skip;
1161   __ jccb(Assembler::zero, L_skip);
1162   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1163   __ bind(L_skip);
1164   __ movb(Address(rdx, rbx,
1165                   Address::times_1,
1166                   arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1167           rax);
1168 }
1169 
1170 void TemplateTable::castore() {
1171   transition(itos, vtos);
1172   __ pop_i(rbx);
1173   // rax: value
1174   // rbx: index
1175   // rdx: array
1176   index_check(rdx, rbx);  // prefer index in rbx
1177   __ movw(Address(rdx, rbx,
1178                   Address::times_2,
1179                   arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1180           rax);
1181 }
1182 
1183 
1184 void TemplateTable::sastore() {
1185   castore();
1186 }
1187 
1188 void TemplateTable::istore(int n) {
1189   transition(itos, vtos);
1190   __ movl(iaddress(n), rax);
1191 }
1192 
1193 void TemplateTable::lstore(int n) {
1194   transition(ltos, vtos);
1195   __ movptr(laddress(n), rax);
1196   NOT_LP64(__ movptr(haddress(n), rdx));
1197 }
1198 
1199 void TemplateTable::fstore(int n) {
1200   transition(ftos, vtos);
1201   __ store_float(faddress(n));
1202 }
1203 
1204 void TemplateTable::dstore(int n) {
1205   transition(dtos, vtos);
1206   __ store_double(daddress(n));
1207 }
1208 
1209 
1210 void TemplateTable::astore(int n) {
1211   transition(vtos, vtos);
1212   __ pop_ptr(rax);
1213   __ movptr(aaddress(n), rax);
1214 }
1215 
1216 void TemplateTable::pop() {
1217   transition(vtos, vtos);
1218   __ addptr(rsp, Interpreter::stackElementSize);
1219 }
1220 
1221 void TemplateTable::pop2() {
1222   transition(vtos, vtos);
1223   __ addptr(rsp, 2 * Interpreter::stackElementSize);
1224 }
1225 
1226 
1227 void TemplateTable::dup() {
1228   transition(vtos, vtos);
1229   __ load_ptr(0, rax);
1230   __ push_ptr(rax);
1231   // stack: ..., a, a
1232 }
1233 
1234 void TemplateTable::dup_x1() {
1235   transition(vtos, vtos);
1236   // stack: ..., a, b
1237   __ load_ptr( 0, rax);  // load b
1238   __ load_ptr( 1, rcx);  // load a
1239   __ store_ptr(1, rax);  // store b
1240   __ store_ptr(0, rcx);  // store a
1241   __ push_ptr(rax);      // push b
1242   // stack: ..., b, a, b
1243 }
1244 
1245 void TemplateTable::dup_x2() {
1246   transition(vtos, vtos);
1247   // stack: ..., a, b, c
1248   __ load_ptr( 0, rax);  // load c
1249   __ load_ptr( 2, rcx);  // load a
1250   __ store_ptr(2, rax);  // store c in a
1251   __ push_ptr(rax);      // push c
1252   // stack: ..., c, b, c, c
1253   __ load_ptr( 2, rax);  // load b
1254   __ store_ptr(2, rcx);  // store a in b
1255   // stack: ..., c, a, c, c
1256   __ store_ptr(1, rax);  // store b in c
1257   // stack: ..., c, a, b, c
1258 }
1259 
1260 void TemplateTable::dup2() {
1261   transition(vtos, vtos);
1262   // stack: ..., a, b
1263   __ load_ptr(1, rax);  // load a
1264   __ push_ptr(rax);     // push a
1265   __ load_ptr(1, rax);  // load b
1266   __ push_ptr(rax);     // push b
1267   // stack: ..., a, b, a, b
1268 }
1269 
1270 
1271 void TemplateTable::dup2_x1() {
1272   transition(vtos, vtos);
1273   // stack: ..., a, b, c
1274   __ load_ptr( 0, rcx);  // load c
1275   __ load_ptr( 1, rax);  // load b
1276   __ push_ptr(rax);      // push b
1277   __ push_ptr(rcx);      // push c
1278   // stack: ..., a, b, c, b, c
1279   __ store_ptr(3, rcx);  // store c in b
1280   // stack: ..., a, c, c, b, c
1281   __ load_ptr( 4, rcx);  // load a
1282   __ store_ptr(2, rcx);  // store a in 2nd c
1283   // stack: ..., a, c, a, b, c
1284   __ store_ptr(4, rax);  // store b in a
1285   // stack: ..., b, c, a, b, c
1286 }
1287 
1288 void TemplateTable::dup2_x2() {
1289   transition(vtos, vtos);
1290   // stack: ..., a, b, c, d
1291   __ load_ptr( 0, rcx);  // load d
1292   __ load_ptr( 1, rax);  // load c
1293   __ push_ptr(rax);      // push c
1294   __ push_ptr(rcx);      // push d
1295   // stack: ..., a, b, c, d, c, d
1296   __ load_ptr( 4, rax);  // load b
1297   __ store_ptr(2, rax);  // store b in d
1298   __ store_ptr(4, rcx);  // store d in b
1299   // stack: ..., a, d, c, b, c, d
1300   __ load_ptr( 5, rcx);  // load a
1301   __ load_ptr( 3, rax);  // load c
1302   __ store_ptr(3, rcx);  // store a in c
1303   __ store_ptr(5, rax);  // store c in a
1304   // stack: ..., c, d, a, b, c, d
1305 }
1306 
1307 void TemplateTable::swap() {
1308   transition(vtos, vtos);
1309   // stack: ..., a, b
1310   __ load_ptr( 1, rcx);  // load a
1311   __ load_ptr( 0, rax);  // load b
1312   __ store_ptr(0, rcx);  // store a in b
1313   __ store_ptr(1, rax);  // store b in a
1314   // stack: ..., b, a
1315 }
1316 
1317 void TemplateTable::iop2(Operation op) {
1318   transition(itos, itos);
1319   switch (op) {
1320   case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1321   case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1322   case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1323   case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1324   case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1325   case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1326   case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break;
1327   case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break;
1328   case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break;
1329   default   : ShouldNotReachHere();
1330   }
1331 }
1332 
1333 void TemplateTable::lop2(Operation op) {
1334   transition(ltos, ltos);
1335 #ifdef _LP64
1336   switch (op) {
1337   case add  :                    __ pop_l(rdx); __ addptr(rax, rdx); break;
1338   case sub  : __ mov(rdx, rax);  __ pop_l(rax); __ subptr(rax, rdx); break;
1339   case _and :                    __ pop_l(rdx); __ andptr(rax, rdx); break;
1340   case _or  :                    __ pop_l(rdx); __ orptr (rax, rdx); break;
1341   case _xor :                    __ pop_l(rdx); __ xorptr(rax, rdx); break;
1342   default   : ShouldNotReachHere();
1343   }
1344 #else
1345   __ pop_l(rbx, rcx);
1346   switch (op) {
1347     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1348     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1349                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1350     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1351     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1352     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1353     default   : ShouldNotReachHere();
1354   }
1355 #endif
1356 }
1357 
1358 void TemplateTable::idiv() {
1359   transition(itos, itos);
1360   __ movl(rcx, rax);
1361   __ pop_i(rax);
1362   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1363   //       they are not equal, one could do a normal division (no correction
1364   //       needed), which may speed up this implementation for the common case.
1365   //       (see also JVM spec., p.243 & p.271)
1366   __ corrected_idivl(rcx);
1367 }
1368 
1369 void TemplateTable::irem() {
1370   transition(itos, itos);
1371   __ movl(rcx, rax);
1372   __ pop_i(rax);
1373   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1374   //       they are not equal, one could do a normal division (no correction
1375   //       needed), which may speed up this implementation for the common case.
1376   //       (see also JVM spec., p.243 & p.271)
1377   __ corrected_idivl(rcx);
1378   __ movl(rax, rdx);
1379 }
1380 
1381 void TemplateTable::lmul() {
1382   transition(ltos, ltos);
1383 #ifdef _LP64
1384   __ pop_l(rdx);
1385   __ imulq(rax, rdx);
1386 #else
1387   __ pop_l(rbx, rcx);
1388   __ push(rcx); __ push(rbx);
1389   __ push(rdx); __ push(rax);
1390   __ lmul(2 * wordSize, 0);
1391   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1392 #endif
1393 }
1394 
1395 void TemplateTable::ldiv() {
1396   transition(ltos, ltos);
1397 #ifdef _LP64
1398   __ mov(rcx, rax);
1399   __ pop_l(rax);
1400   // generate explicit div0 check
1401   __ testq(rcx, rcx);
1402   __ jump_cc(Assembler::zero,
1403              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1404   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1405   //       they are not equal, one could do a normal division (no correction
1406   //       needed), which may speed up this implementation for the common case.
1407   //       (see also JVM spec., p.243 & p.271)
1408   __ corrected_idivq(rcx); // kills rbx
1409 #else
1410   __ pop_l(rbx, rcx);
1411   __ push(rcx); __ push(rbx);
1412   __ push(rdx); __ push(rax);
1413   // check if y = 0
1414   __ orl(rax, rdx);
1415   __ jump_cc(Assembler::zero,
1416              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1417   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1418   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1419 #endif
1420 }
1421 
1422 void TemplateTable::lrem() {
1423   transition(ltos, ltos);
1424 #ifdef _LP64
1425   __ mov(rcx, rax);
1426   __ pop_l(rax);
1427   __ testq(rcx, rcx);
1428   __ jump_cc(Assembler::zero,
1429              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1430   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1431   //       they are not equal, one could do a normal division (no correction
1432   //       needed), which may speed up this implementation for the common case.
1433   //       (see also JVM spec., p.243 & p.271)
1434   __ corrected_idivq(rcx); // kills rbx
1435   __ mov(rax, rdx);
1436 #else
1437   __ pop_l(rbx, rcx);
1438   __ push(rcx); __ push(rbx);
1439   __ push(rdx); __ push(rax);
1440   // check if y = 0
1441   __ orl(rax, rdx);
1442   __ jump_cc(Assembler::zero,
1443              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1444   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1445   __ addptr(rsp, 4 * wordSize);
1446 #endif
1447 }
1448 
1449 void TemplateTable::lshl() {
1450   transition(itos, ltos);
1451   __ movl(rcx, rax);                             // get shift count
1452   #ifdef _LP64
1453   __ pop_l(rax);                                 // get shift value
1454   __ shlq(rax);
1455 #else
1456   __ pop_l(rax, rdx);                            // get shift value
1457   __ lshl(rdx, rax);
1458 #endif
1459 }
1460 
1461 void TemplateTable::lshr() {
1462 #ifdef _LP64
1463   transition(itos, ltos);
1464   __ movl(rcx, rax);                             // get shift count
1465   __ pop_l(rax);                                 // get shift value
1466   __ sarq(rax);
1467 #else
1468   transition(itos, ltos);
1469   __ mov(rcx, rax);                              // get shift count
1470   __ pop_l(rax, rdx);                            // get shift value
1471   __ lshr(rdx, rax, true);
1472 #endif
1473 }
1474 
1475 void TemplateTable::lushr() {
1476   transition(itos, ltos);
1477 #ifdef _LP64
1478   __ movl(rcx, rax);                             // get shift count
1479   __ pop_l(rax);                                 // get shift value
1480   __ shrq(rax);
1481 #else
1482   __ mov(rcx, rax);                              // get shift count
1483   __ pop_l(rax, rdx);                            // get shift value
1484   __ lshr(rdx, rax);
1485 #endif
1486 }
1487 
1488 void TemplateTable::fop2(Operation op) {
1489   transition(ftos, ftos);
1490 
1491   if (UseSSE >= 1) {
1492     switch (op) {
1493     case add:
1494       __ addss(xmm0, at_rsp());
1495       __ addptr(rsp, Interpreter::stackElementSize);
1496       break;
1497     case sub:
1498       __ movflt(xmm1, xmm0);
1499       __ pop_f(xmm0);
1500       __ subss(xmm0, xmm1);
1501       break;
1502     case mul:
1503       __ mulss(xmm0, at_rsp());
1504       __ addptr(rsp, Interpreter::stackElementSize);
1505       break;
1506     case div:
1507       __ movflt(xmm1, xmm0);
1508       __ pop_f(xmm0);
1509       __ divss(xmm0, xmm1);
1510       break;
1511     case rem:
1512       // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1513       // modulo operation. The frem method calls the function
1514       // double fmod(double x, double y) in math.h. The documentation of fmod states:
1515       // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1516       // (signalling or quiet) is returned.
1517       //
1518       // On x86_32 platforms the FPU is used to perform the modulo operation. The
1519       // reason is that on 32-bit Windows the sign of modulo operations diverges from
1520       // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1521       // The fprem instruction used on x86_32 is functionally equivalent to
1522       // SharedRuntime::frem in that it returns a NaN.
1523 #ifdef _LP64
1524       __ movflt(xmm1, xmm0);
1525       __ pop_f(xmm0);
1526       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1527 #else
1528       __ push_f(xmm0);
1529       __ pop_f();
1530       __ fld_s(at_rsp());
1531       __ fremr(rax);
1532       __ f2ieee();
1533       __ pop(rax);  // pop second operand off the stack
1534       __ push_f();
1535       __ pop_f(xmm0);
1536 #endif
1537       break;
1538     default:
1539       ShouldNotReachHere();
1540       break;
1541     }
1542   } else {
1543 #ifdef _LP64
1544     ShouldNotReachHere();
1545 #else
1546     switch (op) {
1547     case add: __ fadd_s (at_rsp());                break;
1548     case sub: __ fsubr_s(at_rsp());                break;
1549     case mul: __ fmul_s (at_rsp());                break;
1550     case div: __ fdivr_s(at_rsp());                break;
1551     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1552     default : ShouldNotReachHere();
1553     }
1554     __ f2ieee();
1555     __ pop(rax);  // pop second operand off the stack
1556 #endif // _LP64
1557   }
1558 }
1559 
1560 void TemplateTable::dop2(Operation op) {
1561   transition(dtos, dtos);
1562   if (UseSSE >= 2) {
1563     switch (op) {
1564     case add:
1565       __ addsd(xmm0, at_rsp());
1566       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1567       break;
1568     case sub:
1569       __ movdbl(xmm1, xmm0);
1570       __ pop_d(xmm0);
1571       __ subsd(xmm0, xmm1);
1572       break;
1573     case mul:
1574       __ mulsd(xmm0, at_rsp());
1575       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1576       break;
1577     case div:
1578       __ movdbl(xmm1, xmm0);
1579       __ pop_d(xmm0);
1580       __ divsd(xmm0, xmm1);
1581       break;
1582     case rem:
1583       // Similar to fop2(), the modulo operation is performed using the
1584       // SharedRuntime::drem method (on x86_64 platforms) or using the
1585       // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1586 #ifdef _LP64
1587       __ movdbl(xmm1, xmm0);
1588       __ pop_d(xmm0);
1589       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1590 #else
1591       __ push_d(xmm0);
1592       __ pop_d();
1593       __ fld_d(at_rsp());
1594       __ fremr(rax);
1595       __ d2ieee();
1596       __ pop(rax);
1597       __ pop(rdx);
1598       __ push_d();
1599       __ pop_d(xmm0);
1600 #endif
1601       break;
1602     default:
1603       ShouldNotReachHere();
1604       break;
1605     }
1606   } else {
1607 #ifdef _LP64
1608     ShouldNotReachHere();
1609 #else
1610     switch (op) {
1611     case add: __ fadd_d (at_rsp());                break;
1612     case sub: __ fsubr_d(at_rsp());                break;
1613     case mul: {
1614       Label L_strict;
1615       Label L_join;
1616       const Address access_flags      (rcx, Method::access_flags_offset());
1617       __ get_method(rcx);
1618       __ movl(rcx, access_flags);
1619       __ testl(rcx, JVM_ACC_STRICT);
1620       __ jccb(Assembler::notZero, L_strict);
1621       __ fmul_d (at_rsp());
1622       __ jmpb(L_join);
1623       __ bind(L_strict);
1624       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1625       __ fmulp();
1626       __ fmul_d (at_rsp());
1627       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1628       __ fmulp();
1629       __ bind(L_join);
1630       break;
1631     }
1632     case div: {
1633       Label L_strict;
1634       Label L_join;
1635       const Address access_flags      (rcx, Method::access_flags_offset());
1636       __ get_method(rcx);
1637       __ movl(rcx, access_flags);
1638       __ testl(rcx, JVM_ACC_STRICT);
1639       __ jccb(Assembler::notZero, L_strict);
1640       __ fdivr_d(at_rsp());
1641       __ jmp(L_join);
1642       __ bind(L_strict);
1643       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1644       __ fmul_d (at_rsp());
1645       __ fdivrp();
1646       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1647       __ fmulp();
1648       __ bind(L_join);
1649       break;
1650     }
1651     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1652     default : ShouldNotReachHere();
1653     }
1654     __ d2ieee();
1655     // Pop double precision number from rsp.
1656     __ pop(rax);
1657     __ pop(rdx);
1658 #endif
1659   }
1660 }
1661 
1662 void TemplateTable::ineg() {
1663   transition(itos, itos);
1664   __ negl(rax);
1665 }
1666 
1667 void TemplateTable::lneg() {
1668   transition(ltos, ltos);
1669   LP64_ONLY(__ negq(rax));
1670   NOT_LP64(__ lneg(rdx, rax));
1671 }
1672 
1673 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1674 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1675   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1676   // of 128-bits operands for SSE instructions.
1677   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1678   // Store the value to a 128-bits operand.
1679   operand[0] = lo;
1680   operand[1] = hi;
1681   return operand;
1682 }
1683 
1684 // Buffer for 128-bits masks used by SSE instructions.
1685 static jlong float_signflip_pool[2*2];
1686 static jlong double_signflip_pool[2*2];
1687 
1688 void TemplateTable::fneg() {
1689   transition(ftos, ftos);
1690   if (UseSSE >= 1) {
1691     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1692     __ xorps(xmm0, ExternalAddress((address) float_signflip));
1693   } else {
1694     LP64_ONLY(ShouldNotReachHere());
1695     NOT_LP64(__ fchs());
1696   }
1697 }
1698 
1699 void TemplateTable::dneg() {
1700   transition(dtos, dtos);
1701   if (UseSSE >= 2) {
1702     static jlong *double_signflip =
1703       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1704     __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1705   } else {
1706 #ifdef _LP64
1707     ShouldNotReachHere();
1708 #else
1709     __ fchs();
1710 #endif
1711   }
1712 }
1713 
1714 void TemplateTable::iinc() {
1715   transition(vtos, vtos);
1716   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1717   locals_index(rbx);
1718   __ addl(iaddress(rbx), rdx);
1719 }
1720 
1721 void TemplateTable::wide_iinc() {
1722   transition(vtos, vtos);
1723   __ movl(rdx, at_bcp(4)); // get constant
1724   locals_index_wide(rbx);
1725   __ bswapl(rdx); // swap bytes & sign-extend constant
1726   __ sarl(rdx, 16);
1727   __ addl(iaddress(rbx), rdx);
1728   // Note: should probably use only one movl to get both
1729   //       the index and the constant -> fix this
1730 }
1731 
1732 void TemplateTable::convert() {
1733 #ifdef _LP64
1734   // Checking
1735 #ifdef ASSERT
1736   {
1737     TosState tos_in  = ilgl;
1738     TosState tos_out = ilgl;
1739     switch (bytecode()) {
1740     case Bytecodes::_i2l: // fall through
1741     case Bytecodes::_i2f: // fall through
1742     case Bytecodes::_i2d: // fall through
1743     case Bytecodes::_i2b: // fall through
1744     case Bytecodes::_i2c: // fall through
1745     case Bytecodes::_i2s: tos_in = itos; break;
1746     case Bytecodes::_l2i: // fall through
1747     case Bytecodes::_l2f: // fall through
1748     case Bytecodes::_l2d: tos_in = ltos; break;
1749     case Bytecodes::_f2i: // fall through
1750     case Bytecodes::_f2l: // fall through
1751     case Bytecodes::_f2d: tos_in = ftos; break;
1752     case Bytecodes::_d2i: // fall through
1753     case Bytecodes::_d2l: // fall through
1754     case Bytecodes::_d2f: tos_in = dtos; break;
1755     default             : ShouldNotReachHere();
1756     }
1757     switch (bytecode()) {
1758     case Bytecodes::_l2i: // fall through
1759     case Bytecodes::_f2i: // fall through
1760     case Bytecodes::_d2i: // fall through
1761     case Bytecodes::_i2b: // fall through
1762     case Bytecodes::_i2c: // fall through
1763     case Bytecodes::_i2s: tos_out = itos; break;
1764     case Bytecodes::_i2l: // fall through
1765     case Bytecodes::_f2l: // fall through
1766     case Bytecodes::_d2l: tos_out = ltos; break;
1767     case Bytecodes::_i2f: // fall through
1768     case Bytecodes::_l2f: // fall through
1769     case Bytecodes::_d2f: tos_out = ftos; break;
1770     case Bytecodes::_i2d: // fall through
1771     case Bytecodes::_l2d: // fall through
1772     case Bytecodes::_f2d: tos_out = dtos; break;
1773     default             : ShouldNotReachHere();
1774     }
1775     transition(tos_in, tos_out);
1776   }
1777 #endif // ASSERT
1778 
1779   static const int64_t is_nan = 0x8000000000000000L;
1780 
1781   // Conversion
1782   switch (bytecode()) {
1783   case Bytecodes::_i2l:
1784     __ movslq(rax, rax);
1785     break;
1786   case Bytecodes::_i2f:
1787     __ cvtsi2ssl(xmm0, rax);
1788     break;
1789   case Bytecodes::_i2d:
1790     __ cvtsi2sdl(xmm0, rax);
1791     break;
1792   case Bytecodes::_i2b:
1793     __ movsbl(rax, rax);
1794     break;
1795   case Bytecodes::_i2c:
1796     __ movzwl(rax, rax);
1797     break;
1798   case Bytecodes::_i2s:
1799     __ movswl(rax, rax);
1800     break;
1801   case Bytecodes::_l2i:
1802     __ movl(rax, rax);
1803     break;
1804   case Bytecodes::_l2f:
1805     __ cvtsi2ssq(xmm0, rax);
1806     break;
1807   case Bytecodes::_l2d:
1808     __ cvtsi2sdq(xmm0, rax);
1809     break;
1810   case Bytecodes::_f2i:
1811   {
1812     Label L;
1813     __ cvttss2sil(rax, xmm0);
1814     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1815     __ jcc(Assembler::notEqual, L);
1816     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1817     __ bind(L);
1818   }
1819     break;
1820   case Bytecodes::_f2l:
1821   {
1822     Label L;
1823     __ cvttss2siq(rax, xmm0);
1824     // NaN or overflow/underflow?
1825     __ cmp64(rax, ExternalAddress((address) &is_nan));
1826     __ jcc(Assembler::notEqual, L);
1827     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1828     __ bind(L);
1829   }
1830     break;
1831   case Bytecodes::_f2d:
1832     __ cvtss2sd(xmm0, xmm0);
1833     break;
1834   case Bytecodes::_d2i:
1835   {
1836     Label L;
1837     __ cvttsd2sil(rax, xmm0);
1838     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1839     __ jcc(Assembler::notEqual, L);
1840     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1841     __ bind(L);
1842   }
1843     break;
1844   case Bytecodes::_d2l:
1845   {
1846     Label L;
1847     __ cvttsd2siq(rax, xmm0);
1848     // NaN or overflow/underflow?
1849     __ cmp64(rax, ExternalAddress((address) &is_nan));
1850     __ jcc(Assembler::notEqual, L);
1851     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1852     __ bind(L);
1853   }
1854     break;
1855   case Bytecodes::_d2f:
1856     __ cvtsd2ss(xmm0, xmm0);
1857     break;
1858   default:
1859     ShouldNotReachHere();
1860   }
1861 #else
1862   // Checking
1863 #ifdef ASSERT
1864   { TosState tos_in  = ilgl;
1865     TosState tos_out = ilgl;
1866     switch (bytecode()) {
1867       case Bytecodes::_i2l: // fall through
1868       case Bytecodes::_i2f: // fall through
1869       case Bytecodes::_i2d: // fall through
1870       case Bytecodes::_i2b: // fall through
1871       case Bytecodes::_i2c: // fall through
1872       case Bytecodes::_i2s: tos_in = itos; break;
1873       case Bytecodes::_l2i: // fall through
1874       case Bytecodes::_l2f: // fall through
1875       case Bytecodes::_l2d: tos_in = ltos; break;
1876       case Bytecodes::_f2i: // fall through
1877       case Bytecodes::_f2l: // fall through
1878       case Bytecodes::_f2d: tos_in = ftos; break;
1879       case Bytecodes::_d2i: // fall through
1880       case Bytecodes::_d2l: // fall through
1881       case Bytecodes::_d2f: tos_in = dtos; break;
1882       default             : ShouldNotReachHere();
1883     }
1884     switch (bytecode()) {
1885       case Bytecodes::_l2i: // fall through
1886       case Bytecodes::_f2i: // fall through
1887       case Bytecodes::_d2i: // fall through
1888       case Bytecodes::_i2b: // fall through
1889       case Bytecodes::_i2c: // fall through
1890       case Bytecodes::_i2s: tos_out = itos; break;
1891       case Bytecodes::_i2l: // fall through
1892       case Bytecodes::_f2l: // fall through
1893       case Bytecodes::_d2l: tos_out = ltos; break;
1894       case Bytecodes::_i2f: // fall through
1895       case Bytecodes::_l2f: // fall through
1896       case Bytecodes::_d2f: tos_out = ftos; break;
1897       case Bytecodes::_i2d: // fall through
1898       case Bytecodes::_l2d: // fall through
1899       case Bytecodes::_f2d: tos_out = dtos; break;
1900       default             : ShouldNotReachHere();
1901     }
1902     transition(tos_in, tos_out);
1903   }
1904 #endif // ASSERT
1905 
1906   // Conversion
1907   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1908   switch (bytecode()) {
1909     case Bytecodes::_i2l:
1910       __ extend_sign(rdx, rax);
1911       break;
1912     case Bytecodes::_i2f:
1913       if (UseSSE >= 1) {
1914         __ cvtsi2ssl(xmm0, rax);
1915       } else {
1916         __ push(rax);          // store int on tos
1917         __ fild_s(at_rsp());   // load int to ST0
1918         __ f2ieee();           // truncate to float size
1919         __ pop(rcx);           // adjust rsp
1920       }
1921       break;
1922     case Bytecodes::_i2d:
1923       if (UseSSE >= 2) {
1924         __ cvtsi2sdl(xmm0, rax);
1925       } else {
1926       __ push(rax);          // add one slot for d2ieee()
1927       __ push(rax);          // store int on tos
1928       __ fild_s(at_rsp());   // load int to ST0
1929       __ d2ieee();           // truncate to double size
1930       __ pop(rcx);           // adjust rsp
1931       __ pop(rcx);
1932       }
1933       break;
1934     case Bytecodes::_i2b:
1935       __ shll(rax, 24);      // truncate upper 24 bits
1936       __ sarl(rax, 24);      // and sign-extend byte
1937       LP64_ONLY(__ movsbl(rax, rax));
1938       break;
1939     case Bytecodes::_i2c:
1940       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1941       LP64_ONLY(__ movzwl(rax, rax));
1942       break;
1943     case Bytecodes::_i2s:
1944       __ shll(rax, 16);      // truncate upper 16 bits
1945       __ sarl(rax, 16);      // and sign-extend short
1946       LP64_ONLY(__ movswl(rax, rax));
1947       break;
1948     case Bytecodes::_l2i:
1949       /* nothing to do */
1950       break;
1951     case Bytecodes::_l2f:
1952       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1953       // 64-bit long values to floats. On 32-bit platforms it is not possible
1954       // to use that instruction with 64-bit operands, therefore the FPU is
1955       // used to perform the conversion.
1956       __ push(rdx);          // store long on tos
1957       __ push(rax);
1958       __ fild_d(at_rsp());   // load long to ST0
1959       __ f2ieee();           // truncate to float size
1960       __ pop(rcx);           // adjust rsp
1961       __ pop(rcx);
1962       if (UseSSE >= 1) {
1963         __ push_f();
1964         __ pop_f(xmm0);
1965       }
1966       break;
1967     case Bytecodes::_l2d:
1968       // On 32-bit platforms the FPU is used for conversion because on
1969       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1970       // instruction with 64-bit operands.
1971       __ push(rdx);          // store long on tos
1972       __ push(rax);
1973       __ fild_d(at_rsp());   // load long to ST0
1974       __ d2ieee();           // truncate to double size
1975       __ pop(rcx);           // adjust rsp
1976       __ pop(rcx);
1977       if (UseSSE >= 2) {
1978         __ push_d();
1979         __ pop_d(xmm0);
1980       }
1981       break;
1982     case Bytecodes::_f2i:
1983       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1984       // as it returns 0 for any NaN.
1985       if (UseSSE >= 1) {
1986         __ push_f(xmm0);
1987       } else {
1988         __ push(rcx);          // reserve space for argument
1989         __ fstp_s(at_rsp());   // pass float argument on stack
1990       }
1991       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1992       break;
1993     case Bytecodes::_f2l:
1994       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1995       // as it returns 0 for any NaN.
1996       if (UseSSE >= 1) {
1997        __ push_f(xmm0);
1998       } else {
1999         __ push(rcx);          // reserve space for argument
2000         __ fstp_s(at_rsp());   // pass float argument on stack
2001       }
2002       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2003       break;
2004     case Bytecodes::_f2d:
2005       if (UseSSE < 1) {
2006         /* nothing to do */
2007       } else if (UseSSE == 1) {
2008         __ push_f(xmm0);
2009         __ pop_f();
2010       } else { // UseSSE >= 2
2011         __ cvtss2sd(xmm0, xmm0);
2012       }
2013       break;
2014     case Bytecodes::_d2i:
2015       if (UseSSE >= 2) {
2016         __ push_d(xmm0);
2017       } else {
2018         __ push(rcx);          // reserve space for argument
2019         __ push(rcx);
2020         __ fstp_d(at_rsp());   // pass double argument on stack
2021       }
2022       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2023       break;
2024     case Bytecodes::_d2l:
2025       if (UseSSE >= 2) {
2026         __ push_d(xmm0);
2027       } else {
2028         __ push(rcx);          // reserve space for argument
2029         __ push(rcx);
2030         __ fstp_d(at_rsp());   // pass double argument on stack
2031       }
2032       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2033       break;
2034     case Bytecodes::_d2f:
2035       if (UseSSE <= 1) {
2036         __ push(rcx);          // reserve space for f2ieee()
2037         __ f2ieee();           // truncate to float size
2038         __ pop(rcx);           // adjust rsp
2039         if (UseSSE == 1) {
2040           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2041           // the conversion is performed using the FPU in this case.
2042           __ push_f();
2043           __ pop_f(xmm0);
2044         }
2045       } else { // UseSSE >= 2
2046         __ cvtsd2ss(xmm0, xmm0);
2047       }
2048       break;
2049     default             :
2050       ShouldNotReachHere();
2051   }
2052 #endif
2053 }
2054 
2055 void TemplateTable::lcmp() {
2056   transition(ltos, itos);
2057 #ifdef _LP64
2058   Label done;
2059   __ pop_l(rdx);
2060   __ cmpq(rdx, rax);
2061   __ movl(rax, -1);
2062   __ jccb(Assembler::less, done);
2063   __ setb(Assembler::notEqual, rax);
2064   __ movzbl(rax, rax);
2065   __ bind(done);
2066 #else
2067 
2068   // y = rdx:rax
2069   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2070   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2071   __ mov(rax, rcx);
2072 #endif
2073 }
2074 
2075 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2076   if ((is_float && UseSSE >= 1) ||
2077       (!is_float && UseSSE >= 2)) {
2078     Label done;
2079     if (is_float) {
2080       // XXX get rid of pop here, use ... reg, mem32
2081       __ pop_f(xmm1);
2082       __ ucomiss(xmm1, xmm0);
2083     } else {
2084       // XXX get rid of pop here, use ... reg, mem64
2085       __ pop_d(xmm1);
2086       __ ucomisd(xmm1, xmm0);
2087     }
2088     if (unordered_result < 0) {
2089       __ movl(rax, -1);
2090       __ jccb(Assembler::parity, done);
2091       __ jccb(Assembler::below, done);
2092       __ setb(Assembler::notEqual, rdx);
2093       __ movzbl(rax, rdx);
2094     } else {
2095       __ movl(rax, 1);
2096       __ jccb(Assembler::parity, done);
2097       __ jccb(Assembler::above, done);
2098       __ movl(rax, 0);
2099       __ jccb(Assembler::equal, done);
2100       __ decrementl(rax);
2101     }
2102     __ bind(done);
2103   } else {
2104 #ifdef _LP64
2105     ShouldNotReachHere();
2106 #else
2107     if (is_float) {
2108       __ fld_s(at_rsp());
2109     } else {
2110       __ fld_d(at_rsp());
2111       __ pop(rdx);
2112     }
2113     __ pop(rcx);
2114     __ fcmp2int(rax, unordered_result < 0);
2115 #endif // _LP64
2116   }
2117 }
2118 
2119 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2120   if (ValueTypesThreadLocalRecycling) {
2121     Label no_vt_recycling, no_fixing_required;
2122     const Register thread1 = NOT_LP64(rbx) LP64_ONLY(r15_thread);
2123     NOT_LP64(__ get_thread(thread1));
2124     __ movptr(rbx, Address(thread1, in_bytes(JavaThread::vt_alloc_ptr_offset())));
2125     __ testptr(rbx, rbx);
2126     __ jcc(Assembler::zero, no_vt_recycling);
2127     __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2128     __ testptr(rcx, rcx);
2129     __ jcc(Assembler::notZero, no_fixing_required);
2130     // vt_alloc_ptr in JavaThread is non-null but frame vt_alloc_ptr is null
2131     // which means frame vt_alloc_ptr needs to be initialized
2132     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::fix_frame_vt_alloc_ptr));
2133     __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2134     __ bind(no_fixing_required);
2135     __ testptr(rcx, rbx);
2136     __ jcc(Assembler::equal, no_vt_recycling);
2137     __ andptr(rcx, VTBufferChunk::chunk_mask());
2138     __ movl(rcx, Address(rcx, VTBufferChunk::index_offset()));
2139     __ andptr(rbx, VTBufferChunk::chunk_mask());
2140     __ movl(rbx, Address(rbx, VTBufferChunk::index_offset()));
2141     __ subl(rbx, rcx);
2142     __ get_method(rcx);
2143     __ movl(rcx, Address(rcx, Method::max_vt_buffer_offset()));
2144     __ cmpl(rbx, rcx);
2145     __ jcc(Assembler::lessEqual, no_vt_recycling);
2146     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::recycle_buffered_values));
2147     __ bind(no_vt_recycling);
2148   }
2149 
2150   __ get_method(rcx); // rcx holds method
2151   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2152                                      // holds bumped taken count
2153 
2154   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2155                              InvocationCounter::counter_offset();
2156   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2157                               InvocationCounter::counter_offset();
2158 
2159   // Load up edx with the branch displacement
2160   if (is_wide) {
2161     __ movl(rdx, at_bcp(1));
2162   } else {
2163     __ load_signed_short(rdx, at_bcp(1));
2164   }
2165   __ bswapl(rdx);
2166 
2167   if (!is_wide) {
2168     __ sarl(rdx, 16);
2169   }
2170   LP64_ONLY(__ movl2ptr(rdx, rdx));
2171 
2172   // Handle all the JSR stuff here, then exit.
2173   // It's much shorter and cleaner than intermingling with the non-JSR
2174   // normal-branch stuff occurring below.
2175   if (is_jsr) {
2176     // Pre-load the next target bytecode into rbx
2177     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2178 
2179     // compute return address as bci in rax
2180     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2181                         in_bytes(ConstMethod::codes_offset())));
2182     __ subptr(rax, Address(rcx, Method::const_offset()));
2183     // Adjust the bcp in r13 by the displacement in rdx
2184     __ addptr(rbcp, rdx);
2185     // jsr returns atos that is not an oop
2186     __ push_i(rax);
2187     __ dispatch_only(vtos, true);
2188     return;
2189   }
2190 
2191   // Normal (non-jsr) branch handling
2192 
2193   // Adjust the bcp in r13 by the displacement in rdx
2194   __ addptr(rbcp, rdx);
2195 
2196   assert(UseLoopCounter || !UseOnStackReplacement,
2197          "on-stack-replacement requires loop counters");
2198   Label backedge_counter_overflow;
2199   Label profile_method;
2200   Label dispatch;
2201   if (UseLoopCounter) {
2202     // increment backedge counter for backward branches
2203     // rax: MDO
2204     // rbx: MDO bumped taken-count
2205     // rcx: method
2206     // rdx: target offset
2207     // r13: target bcp
2208     // r14: locals pointer
2209     __ testl(rdx, rdx);             // check if forward or backward branch
2210     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2211 
2212     // check if MethodCounters exists
2213     Label has_counters;
2214     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2215     __ testptr(rax, rax);
2216     __ jcc(Assembler::notZero, has_counters);
2217     __ push(rdx);
2218     __ push(rcx);
2219     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2220                rcx);
2221     __ pop(rcx);
2222     __ pop(rdx);
2223     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2224     __ testptr(rax, rax);
2225     __ jcc(Assembler::zero, dispatch);
2226     __ bind(has_counters);
2227 
2228     if (TieredCompilation) {
2229       Label no_mdo;
2230       int increment = InvocationCounter::count_increment;
2231       if (ProfileInterpreter) {
2232         // Are we profiling?
2233         __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2234         __ testptr(rbx, rbx);
2235         __ jccb(Assembler::zero, no_mdo);
2236         // Increment the MDO backedge counter
2237         const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2238                                            in_bytes(InvocationCounter::counter_offset()));
2239         const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2240         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2241                                    rax, false, Assembler::zero, &backedge_counter_overflow);
2242         __ jmp(dispatch);
2243       }
2244       __ bind(no_mdo);
2245       // Increment backedge counter in MethodCounters*
2246       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2247       const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2248       __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2249                                  rax, false, Assembler::zero, &backedge_counter_overflow);
2250     } else { // not TieredCompilation
2251       // increment counter
2252       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2253       __ movl(rax, Address(rcx, be_offset));        // load backedge counter
2254       __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2255       __ movl(Address(rcx, be_offset), rax);        // store counter
2256 
2257       __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
2258 
2259       __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2260       __ addl(rax, Address(rcx, be_offset));        // add both counters
2261 
2262       if (ProfileInterpreter) {
2263         // Test to see if we should create a method data oop
2264         __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2265         __ jcc(Assembler::less, dispatch);
2266 
2267         // if no method data exists, go to profile method
2268         __ test_method_data_pointer(rax, profile_method);
2269 
2270         if (UseOnStackReplacement) {
2271           // check for overflow against rbx which is the MDO taken count
2272           __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2273           __ jcc(Assembler::below, dispatch);
2274 
2275           // When ProfileInterpreter is on, the backedge_count comes
2276           // from the MethodData*, which value does not get reset on
2277           // the call to frequency_counter_overflow().  To avoid
2278           // excessive calls to the overflow routine while the method is
2279           // being compiled, add a second test to make sure the overflow
2280           // function is called only once every overflow_frequency.
2281           const int overflow_frequency = 1024;
2282           __ andl(rbx, overflow_frequency - 1);
2283           __ jcc(Assembler::zero, backedge_counter_overflow);
2284 
2285         }
2286       } else {
2287         if (UseOnStackReplacement) {
2288           // check for overflow against rax, which is the sum of the
2289           // counters
2290           __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2291           __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2292 
2293         }
2294       }
2295     }
2296     __ bind(dispatch);
2297   }
2298 
2299   // Pre-load the next target bytecode into rbx
2300   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2301 
2302   // continue with the bytecode @ target
2303   // rax: return bci for jsr's, unused otherwise
2304   // rbx: target bytecode
2305   // r13: target bcp
2306   __ dispatch_only(vtos, true);
2307 
2308   if (UseLoopCounter) {
2309     if (ProfileInterpreter) {
2310       // Out-of-line code to allocate method data oop.
2311       __ bind(profile_method);
2312       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2313       __ set_method_data_pointer_for_bcp();
2314       __ jmp(dispatch);
2315     }
2316 
2317     if (UseOnStackReplacement) {
2318       // invocation counter overflow
2319       __ bind(backedge_counter_overflow);
2320       __ negptr(rdx);
2321       __ addptr(rdx, rbcp); // branch bcp
2322       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2323       __ call_VM(noreg,
2324                  CAST_FROM_FN_PTR(address,
2325                                   InterpreterRuntime::frequency_counter_overflow),
2326                  rdx);
2327 
2328       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2329       // rdx: scratch
2330       // r14: locals pointer
2331       // r13: bcp
2332       __ testptr(rax, rax);                        // test result
2333       __ jcc(Assembler::zero, dispatch);         // no osr if null
2334       // nmethod may have been invalidated (VM may block upon call_VM return)
2335       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2336       __ jcc(Assembler::notEqual, dispatch);
2337 
2338       // We have the address of an on stack replacement routine in rax.
2339       // In preparation of invoking it, first we must migrate the locals
2340       // and monitors from off the interpreter frame on the stack.
2341       // Ensure to save the osr nmethod over the migration call,
2342       // it will be preserved in rbx.
2343       __ mov(rbx, rax);
2344 
2345       NOT_LP64(__ get_thread(rcx));
2346 
2347       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2348 
2349       // rax is OSR buffer, move it to expected parameter location
2350       LP64_ONLY(__ mov(j_rarg0, rax));
2351       NOT_LP64(__ mov(rcx, rax));
2352       // We use j_rarg definitions here so that registers don't conflict as parameter
2353       // registers change across platforms as we are in the midst of a calling
2354       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2355 
2356       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2357       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2358 
2359       // pop the interpreter frame
2360       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2361       __ leave();                                // remove frame anchor
2362       __ pop(retaddr);                           // get return address
2363       __ mov(rsp, sender_sp);                   // set sp to sender sp
2364       // Ensure compiled code always sees stack at proper alignment
2365       __ andptr(rsp, -(StackAlignmentInBytes));
2366 
2367       // unlike x86 we need no specialized return from compiled code
2368       // to the interpreter or the call stub.
2369 
2370       // push the return address
2371       __ push(retaddr);
2372 
2373       // and begin the OSR nmethod
2374       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2375     }
2376   }
2377 }
2378 
2379 void TemplateTable::if_0cmp(Condition cc) {
2380   transition(itos, vtos);
2381   // assume branch is more often taken than not (loops use backward branches)
2382   Label not_taken;
2383   __ testl(rax, rax);
2384   __ jcc(j_not(cc), not_taken);
2385   branch(false, false);
2386   __ bind(not_taken);
2387   __ profile_not_taken_branch(rax);
2388 }
2389 
2390 void TemplateTable::if_icmp(Condition cc) {
2391   transition(itos, vtos);
2392   // assume branch is more often taken than not (loops use backward branches)
2393   Label not_taken;
2394   __ pop_i(rdx);
2395   __ cmpl(rdx, rax);
2396   __ jcc(j_not(cc), not_taken);
2397   branch(false, false);
2398   __ bind(not_taken);
2399   __ profile_not_taken_branch(rax);
2400 }
2401 
2402 void TemplateTable::if_nullcmp(Condition cc) {
2403   transition(atos, vtos);
2404   // assume branch is more often taken than not (loops use backward branches)
2405   Label not_taken;
2406   __ testptr(rax, rax);
2407   __ jcc(j_not(cc), not_taken);
2408   branch(false, false);
2409   __ bind(not_taken);
2410   __ profile_not_taken_branch(rax);
2411 }
2412 
2413 void TemplateTable::if_acmp(Condition cc) {
2414   transition(atos, vtos);
2415   // assume branch is more often taken than not (loops use backward branches)
2416   Label not_taken;
2417   __ pop_ptr(rdx);
2418   __ cmpoop(rdx, rax);
2419   __ jcc(j_not(cc), not_taken);
2420   branch(false, false);
2421   __ bind(not_taken);
2422   __ profile_not_taken_branch(rax);
2423 }
2424 
2425 void TemplateTable::ret() {
2426   transition(vtos, vtos);
2427   locals_index(rbx);
2428   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2429   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2430   __ profile_ret(rbx, rcx);
2431   __ get_method(rax);
2432   __ movptr(rbcp, Address(rax, Method::const_offset()));
2433   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2434                       ConstMethod::codes_offset()));
2435   __ dispatch_next(vtos, 0, true);
2436 }
2437 
2438 void TemplateTable::wide_ret() {
2439   transition(vtos, vtos);
2440   locals_index_wide(rbx);
2441   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2442   __ profile_ret(rbx, rcx);
2443   __ get_method(rax);
2444   __ movptr(rbcp, Address(rax, Method::const_offset()));
2445   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2446   __ dispatch_next(vtos, 0, true);
2447 }
2448 
2449 void TemplateTable::tableswitch() {
2450   Label default_case, continue_execution;
2451   transition(itos, vtos);
2452 
2453   // align r13/rsi
2454   __ lea(rbx, at_bcp(BytesPerInt));
2455   __ andptr(rbx, -BytesPerInt);
2456   // load lo & hi
2457   __ movl(rcx, Address(rbx, BytesPerInt));
2458   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2459   __ bswapl(rcx);
2460   __ bswapl(rdx);
2461   // check against lo & hi
2462   __ cmpl(rax, rcx);
2463   __ jcc(Assembler::less, default_case);
2464   __ cmpl(rax, rdx);
2465   __ jcc(Assembler::greater, default_case);
2466   // lookup dispatch offset
2467   __ subl(rax, rcx);
2468   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2469   __ profile_switch_case(rax, rbx, rcx);
2470   // continue execution
2471   __ bind(continue_execution);
2472   __ bswapl(rdx);
2473   LP64_ONLY(__ movl2ptr(rdx, rdx));
2474   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2475   __ addptr(rbcp, rdx);
2476   __ dispatch_only(vtos, true);
2477   // handle default
2478   __ bind(default_case);
2479   __ profile_switch_default(rax);
2480   __ movl(rdx, Address(rbx, 0));
2481   __ jmp(continue_execution);
2482 }
2483 
2484 void TemplateTable::lookupswitch() {
2485   transition(itos, itos);
2486   __ stop("lookupswitch bytecode should have been rewritten");
2487 }
2488 
2489 void TemplateTable::fast_linearswitch() {
2490   transition(itos, vtos);
2491   Label loop_entry, loop, found, continue_execution;
2492   // bswap rax so we can avoid bswapping the table entries
2493   __ bswapl(rax);
2494   // align r13
2495   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2496                                     // this instruction (change offsets
2497                                     // below)
2498   __ andptr(rbx, -BytesPerInt);
2499   // set counter
2500   __ movl(rcx, Address(rbx, BytesPerInt));
2501   __ bswapl(rcx);
2502   __ jmpb(loop_entry);
2503   // table search
2504   __ bind(loop);
2505   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2506   __ jcc(Assembler::equal, found);
2507   __ bind(loop_entry);
2508   __ decrementl(rcx);
2509   __ jcc(Assembler::greaterEqual, loop);
2510   // default case
2511   __ profile_switch_default(rax);
2512   __ movl(rdx, Address(rbx, 0));
2513   __ jmp(continue_execution);
2514   // entry found -> get offset
2515   __ bind(found);
2516   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2517   __ profile_switch_case(rcx, rax, rbx);
2518   // continue execution
2519   __ bind(continue_execution);
2520   __ bswapl(rdx);
2521   __ movl2ptr(rdx, rdx);
2522   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2523   __ addptr(rbcp, rdx);
2524   __ dispatch_only(vtos, true);
2525 }
2526 
2527 void TemplateTable::fast_binaryswitch() {
2528   transition(itos, vtos);
2529   // Implementation using the following core algorithm:
2530   //
2531   // int binary_search(int key, LookupswitchPair* array, int n) {
2532   //   // Binary search according to "Methodik des Programmierens" by
2533   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2534   //   int i = 0;
2535   //   int j = n;
2536   //   while (i+1 < j) {
2537   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2538   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2539   //     // where a stands for the array and assuming that the (inexisting)
2540   //     // element a[n] is infinitely big.
2541   //     int h = (i + j) >> 1;
2542   //     // i < h < j
2543   //     if (key < array[h].fast_match()) {
2544   //       j = h;
2545   //     } else {
2546   //       i = h;
2547   //     }
2548   //   }
2549   //   // R: a[i] <= key < a[i+1] or Q
2550   //   // (i.e., if key is within array, i is the correct index)
2551   //   return i;
2552   // }
2553 
2554   // Register allocation
2555   const Register key   = rax; // already set (tosca)
2556   const Register array = rbx;
2557   const Register i     = rcx;
2558   const Register j     = rdx;
2559   const Register h     = rdi;
2560   const Register temp  = rsi;
2561 
2562   // Find array start
2563   NOT_LP64(__ save_bcp());
2564 
2565   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2566                                           // get rid of this
2567                                           // instruction (change
2568                                           // offsets below)
2569   __ andptr(array, -BytesPerInt);
2570 
2571   // Initialize i & j
2572   __ xorl(i, i);                            // i = 0;
2573   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2574 
2575   // Convert j into native byteordering
2576   __ bswapl(j);
2577 
2578   // And start
2579   Label entry;
2580   __ jmp(entry);
2581 
2582   // binary search loop
2583   {
2584     Label loop;
2585     __ bind(loop);
2586     // int h = (i + j) >> 1;
2587     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2588     __ sarl(h, 1);                               // h = (i + j) >> 1;
2589     // if (key < array[h].fast_match()) {
2590     //   j = h;
2591     // } else {
2592     //   i = h;
2593     // }
2594     // Convert array[h].match to native byte-ordering before compare
2595     __ movl(temp, Address(array, h, Address::times_8));
2596     __ bswapl(temp);
2597     __ cmpl(key, temp);
2598     // j = h if (key <  array[h].fast_match())
2599     __ cmov32(Assembler::less, j, h);
2600     // i = h if (key >= array[h].fast_match())
2601     __ cmov32(Assembler::greaterEqual, i, h);
2602     // while (i+1 < j)
2603     __ bind(entry);
2604     __ leal(h, Address(i, 1)); // i+1
2605     __ cmpl(h, j);             // i+1 < j
2606     __ jcc(Assembler::less, loop);
2607   }
2608 
2609   // end of binary search, result index is i (must check again!)
2610   Label default_case;
2611   // Convert array[i].match to native byte-ordering before compare
2612   __ movl(temp, Address(array, i, Address::times_8));
2613   __ bswapl(temp);
2614   __ cmpl(key, temp);
2615   __ jcc(Assembler::notEqual, default_case);
2616 
2617   // entry found -> j = offset
2618   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2619   __ profile_switch_case(i, key, array);
2620   __ bswapl(j);
2621   LP64_ONLY(__ movslq(j, j));
2622 
2623   NOT_LP64(__ restore_bcp());
2624   NOT_LP64(__ restore_locals());                           // restore rdi
2625 
2626   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2627   __ addptr(rbcp, j);
2628   __ dispatch_only(vtos, true);
2629 
2630   // default case -> j = default offset
2631   __ bind(default_case);
2632   __ profile_switch_default(i);
2633   __ movl(j, Address(array, -2 * BytesPerInt));
2634   __ bswapl(j);
2635   LP64_ONLY(__ movslq(j, j));
2636 
2637   NOT_LP64(__ restore_bcp());
2638   NOT_LP64(__ restore_locals());
2639 
2640   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2641   __ addptr(rbcp, j);
2642   __ dispatch_only(vtos, true);
2643 }
2644 
2645 void TemplateTable::_return(TosState state) {
2646   transition(state, state);
2647 
2648   assert(_desc->calls_vm(),
2649          "inconsistent calls_vm information"); // call in remove_activation
2650 
2651   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2652     assert(state == vtos, "only valid state");
2653     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2654     __ movptr(robj, aaddress(0));
2655     __ load_klass(rdi, robj);
2656     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2657     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2658     Label skip_register_finalizer;
2659     __ jcc(Assembler::zero, skip_register_finalizer);
2660 
2661     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2662 
2663     __ bind(skip_register_finalizer);
2664   }
2665 
2666 #ifdef _LP64
2667   if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2668     Label no_safepoint;
2669     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2670     __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2671     __ jcc(Assembler::zero, no_safepoint);
2672     __ push(state);
2673     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2674                                     InterpreterRuntime::at_safepoint));
2675     __ pop(state);
2676     __ bind(no_safepoint);
2677   }
2678 #endif
2679   if (state == qtos) {
2680     const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread);
2681     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::return_value), rax);
2682     NOT_LP64(__ get_thread(thread1));
2683     __ get_vm_result(rax, thread1);
2684   }
2685 
2686   // Narrow result if state is itos but result type is smaller.
2687   // Need to narrow in the return bytecode rather than in generate_return_entry
2688   // since compiled code callers expect the result to already be narrowed.
2689   if (state == itos) {
2690     __ narrow(rax);
2691   }
2692 
2693 #ifdef ASSERT
2694   if (EnableMVT || EnableValhalla) {
2695     if (state == atos) {
2696       const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread);
2697       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::check_areturn), rax);
2698       NOT_LP64(__ get_thread(thread1));
2699       __ get_vm_result(rax, thread1);
2700     }
2701   }
2702 #endif // ASSERT
2703 
2704   __ remove_activation(state, rbcp, true, true, true, state == qtos && ValueTypeReturnedAsFields);
2705 
2706   __ jmp(rbcp);
2707 }
2708 
2709 // ----------------------------------------------------------------------------
2710 // Volatile variables demand their effects be made known to all CPU's
2711 // in order.  Store buffers on most chips allow reads & writes to
2712 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2713 // without some kind of memory barrier (i.e., it's not sufficient that
2714 // the interpreter does not reorder volatile references, the hardware
2715 // also must not reorder them).
2716 //
2717 // According to the new Java Memory Model (JMM):
2718 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2719 //     writes act as aquire & release, so:
2720 // (2) A read cannot let unrelated NON-volatile memory refs that
2721 //     happen after the read float up to before the read.  It's OK for
2722 //     non-volatile memory refs that happen before the volatile read to
2723 //     float down below it.
2724 // (3) Similar a volatile write cannot let unrelated NON-volatile
2725 //     memory refs that happen BEFORE the write float down to after the
2726 //     write.  It's OK for non-volatile memory refs that happen after the
2727 //     volatile write to float up before it.
2728 //
2729 // We only put in barriers around volatile refs (they are expensive),
2730 // not _between_ memory refs (that would require us to track the
2731 // flavor of the previous memory refs).  Requirements (2) and (3)
2732 // require some barriers before volatile stores and after volatile
2733 // loads.  These nearly cover requirement (1) but miss the
2734 // volatile-store-volatile-load case.  This final case is placed after
2735 // volatile-stores although it could just as well go before
2736 // volatile-loads.
2737 
2738 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2739   // Helper function to insert a is-volatile test and memory barrier
2740   if(!os::is_MP()) return;    // Not needed on single CPU
2741   __ membar(order_constraint);
2742 }
2743 
2744 void TemplateTable::resolve_cache_and_index(int byte_no,
2745                                             Register Rcache,
2746                                             Register index,
2747                                             size_t index_size) {
2748   const Register temp = rbx;
2749   assert_different_registers(Rcache, index, temp);
2750 
2751   Label resolved;
2752 
2753   Bytecodes::Code code = bytecode();
2754   switch (code) {
2755   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2756   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2757   default: break;
2758   }
2759 
2760   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2761   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2762   __ cmpl(temp, code);  // have we resolved this bytecode?
2763   __ jcc(Assembler::equal, resolved);
2764 
2765   // resolve first time through
2766   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2767   __ movl(temp, code);
2768   __ call_VM(noreg, entry, temp);
2769   // Update registers with resolved info
2770   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2771   __ bind(resolved);
2772 }
2773 
2774 // The cache and index registers must be set before call
2775 void TemplateTable::load_field_cp_cache_entry(Register obj,
2776                                               Register cache,
2777                                               Register index,
2778                                               Register off,
2779                                               Register flags,
2780                                               bool is_static = false) {
2781   assert_different_registers(cache, index, flags, off);
2782 
2783   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2784   // Field offset
2785   __ movptr(off, Address(cache, index, Address::times_ptr,
2786                          in_bytes(cp_base_offset +
2787                                   ConstantPoolCacheEntry::f2_offset())));
2788   // Flags
2789   __ movl(flags, Address(cache, index, Address::times_ptr,
2790                          in_bytes(cp_base_offset +
2791                                   ConstantPoolCacheEntry::flags_offset())));
2792 
2793   // klass overwrite register
2794   if (is_static) {
2795     __ movptr(obj, Address(cache, index, Address::times_ptr,
2796                            in_bytes(cp_base_offset +
2797                                     ConstantPoolCacheEntry::f1_offset())));
2798     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2799     __ movptr(obj, Address(obj, mirror_offset));
2800     __ resolve_oop_handle(obj);
2801   }
2802 }
2803 
2804 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2805                                                Register method,
2806                                                Register itable_index,
2807                                                Register flags,
2808                                                bool is_invokevirtual,
2809                                                bool is_invokevfinal, /*unused*/
2810                                                bool is_invokedynamic) {
2811   // setup registers
2812   const Register cache = rcx;
2813   const Register index = rdx;
2814   assert_different_registers(method, flags);
2815   assert_different_registers(method, cache, index);
2816   assert_different_registers(itable_index, flags);
2817   assert_different_registers(itable_index, cache, index);
2818   // determine constant pool cache field offsets
2819   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2820   const int method_offset = in_bytes(
2821     ConstantPoolCache::base_offset() +
2822       ((byte_no == f2_byte)
2823        ? ConstantPoolCacheEntry::f2_offset()
2824        : ConstantPoolCacheEntry::f1_offset()));
2825   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2826                                     ConstantPoolCacheEntry::flags_offset());
2827   // access constant pool cache fields
2828   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2829                                     ConstantPoolCacheEntry::f2_offset());
2830 
2831   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2832   resolve_cache_and_index(byte_no, cache, index, index_size);
2833     __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2834 
2835   if (itable_index != noreg) {
2836     // pick up itable or appendix index from f2 also:
2837     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2838   }
2839   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2840 }
2841 
2842 // The registers cache and index expected to be set before call.
2843 // Correct values of the cache and index registers are preserved.
2844 void TemplateTable::jvmti_post_field_access(Register cache,
2845                                             Register index,
2846                                             bool is_static,
2847                                             bool has_tos) {
2848   if (JvmtiExport::can_post_field_access()) {
2849     // Check to see if a field access watch has been set before we take
2850     // the time to call into the VM.
2851     Label L1;
2852     assert_different_registers(cache, index, rax);
2853     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2854     __ testl(rax,rax);
2855     __ jcc(Assembler::zero, L1);
2856 
2857     // cache entry pointer
2858     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2859     __ shll(index, LogBytesPerWord);
2860     __ addptr(cache, index);
2861     if (is_static) {
2862       __ xorptr(rax, rax);      // NULL object reference
2863     } else {
2864       __ pop(atos);         // Get the object
2865       __ verify_oop(rax);
2866       __ push(atos);        // Restore stack state
2867     }
2868     // rax,:   object pointer or NULL
2869     // cache: cache entry pointer
2870     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2871                rax, cache);
2872     __ get_cache_and_index_at_bcp(cache, index, 1);
2873     __ bind(L1);
2874   }
2875 }
2876 
2877 void TemplateTable::pop_and_check_object(Register r) {
2878   __ pop_ptr(r);
2879   __ null_check(r);  // for field access must check obj.
2880   __ verify_oop(r);
2881 }
2882 
2883 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2884   transition(vtos, vtos);
2885 
2886   const Register cache = rcx;
2887   const Register index = rdx;
2888   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2889   const Register off   = rbx;
2890   const Register flags = rax;
2891   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2892   const Register flags2 = rdx;
2893 
2894   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2895   jvmti_post_field_access(cache, index, is_static, false);
2896   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2897 
2898   const Address field(obj, off, Address::times_1, 0*wordSize);
2899   NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2900 
2901   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notValueType, notDouble;
2902 
2903   if (!is_static) {
2904     __ movptr(rcx, Address(cache, index, Address::times_ptr,
2905                            in_bytes(ConstantPoolCache::base_offset() +
2906                                     ConstantPoolCacheEntry::f1_offset())));
2907   }
2908 
2909   __ movl(flags2, flags);
2910 
2911   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2912   // Make sure we don't need to mask edx after the above shift
2913   assert(btos == 0, "change code, btos != 0");
2914 
2915   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2916 
2917   __ jcc(Assembler::notZero, notByte);
2918   // btos
2919   if (!is_static) pop_and_check_object(obj);
2920   __ load_signed_byte(rax, field);
2921   __ push(btos);
2922   // Rewrite bytecode to be faster
2923   if (!is_static && rc == may_rewrite) {
2924     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2925   }
2926   __ jmp(Done);
2927 
2928   __ bind(notByte);
2929 
2930   __ cmpl(flags, qtos);
2931   __ jcc(Assembler::notEqual, notValueType);
2932   // qtos
2933   if (is_static) {
2934     Label initialized;
2935     // Issue below if the static field has not been initialized yet
2936     __ load_heap_oop(rax, field);
2937     __ testptr(rax, rax);
2938     __ jcc(Assembler::notZero, initialized);
2939     __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
2940     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_value_field),
2941          obj, flags2);
2942     __ verify_oop(rax);
2943     __ bind(initialized);
2944     __ push(qtos);
2945   } else {
2946     __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
2947     pop_and_check_object(rbx);
2948     call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::qgetfield),
2949             rbx, flags2, rcx);
2950     __ verify_oop(rax);
2951     __ push(qtos);
2952     // Bytecode rewrite?
2953   }
2954   __ jmp(Done);
2955 
2956   __ bind(notValueType);
2957 
2958   if (!is_static) pop_and_check_object(obj);
2959 
2960   __ cmpl(flags, ztos);
2961   __ jcc(Assembler::notEqual, notBool);
2962 
2963   // ztos (same code as btos)
2964   __ load_signed_byte(rax, field);
2965   __ push(ztos);
2966   // Rewrite bytecode to be faster
2967   if (!is_static && rc == may_rewrite) {
2968     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2969     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2970   }
2971   __ jmp(Done);
2972 
2973   __ bind(notBool);
2974   __ cmpl(flags, atos);
2975   __ jcc(Assembler::notEqual, notObj);
2976   // atos
2977   __ load_heap_oop(rax, field);
2978   __ push(atos);
2979   if (!is_static && rc == may_rewrite) {
2980     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2981   }
2982   __ jmp(Done);
2983 
2984   __ bind(notObj);
2985   __ cmpl(flags, itos);
2986   __ jcc(Assembler::notEqual, notInt);
2987   // itos
2988   __ movl(rax, field);
2989   __ push(itos);
2990   // Rewrite bytecode to be faster
2991   if (!is_static && rc == may_rewrite) {
2992     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2993   }
2994   __ jmp(Done);
2995 
2996   __ bind(notInt);
2997   __ cmpl(flags, ctos);
2998   __ jcc(Assembler::notEqual, notChar);
2999   // ctos
3000   __ load_unsigned_short(rax, field);
3001   __ push(ctos);
3002   // Rewrite bytecode to be faster
3003   if (!is_static && rc == may_rewrite) {
3004     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3005   }
3006   __ jmp(Done);
3007 
3008   __ bind(notChar);
3009   __ cmpl(flags, stos);
3010   __ jcc(Assembler::notEqual, notShort);
3011   // stos
3012   __ load_signed_short(rax, field);
3013   __ push(stos);
3014   // Rewrite bytecode to be faster
3015   if (!is_static && rc == may_rewrite) {
3016     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3017   }
3018   __ jmp(Done);
3019 
3020   __ bind(notShort);
3021   __ cmpl(flags, ltos);
3022   __ jcc(Assembler::notEqual, notLong);
3023   // ltos
3024 
3025 #ifndef _LP64
3026   // Generate code as if volatile.  There just aren't enough registers to
3027   // save that information and this code is faster than the test.
3028   __ fild_d(field);                // Must load atomically
3029   __ subptr(rsp,2*wordSize);    // Make space for store
3030   __ fistp_d(Address(rsp,0));
3031   __ pop(rax);
3032   __ pop(rdx);
3033 #else
3034   __ movq(rax, field);
3035 #endif
3036 
3037   __ push(ltos);
3038   // Rewrite bytecode to be faster
3039   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3040   __ jmp(Done);
3041 
3042   __ bind(notLong);
3043   __ cmpl(flags, ftos);
3044   __ jcc(Assembler::notEqual, notFloat);
3045   // ftos
3046 
3047   __ load_float(field);
3048   __ push(ftos);
3049   // Rewrite bytecode to be faster
3050   if (!is_static && rc == may_rewrite) {
3051     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3052   }
3053   __ jmp(Done);
3054 
3055   __ bind(notFloat);
3056 #ifdef ASSERT
3057   __ cmpl(flags, dtos);
3058   __ jcc(Assembler::notEqual, notDouble);
3059 #endif
3060   // dtos
3061   __ load_double(field);
3062   __ push(dtos);
3063   // Rewrite bytecode to be faster
3064   if (!is_static && rc == may_rewrite) {
3065     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3066   }
3067 #ifdef ASSERT
3068   __ jmp(Done);
3069 
3070 
3071   __ bind(notDouble);
3072   __ stop("Bad state");
3073 #endif
3074 
3075   __ bind(Done);
3076   // [jk] not needed currently
3077   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3078   //                                              Assembler::LoadStore));
3079 }
3080 
3081 void TemplateTable::getfield(int byte_no) {
3082   getfield_or_static(byte_no, false);
3083 }
3084 
3085 void TemplateTable::nofast_getfield(int byte_no) {
3086   getfield_or_static(byte_no, false, may_not_rewrite);
3087 }
3088 
3089 void TemplateTable::getstatic(int byte_no) {
3090   getfield_or_static(byte_no, true);
3091 }
3092 
3093 void TemplateTable::vwithfield() {
3094   transition(vtos, qtos);
3095 
3096   Register cache = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3097   Register index = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3098 
3099   resolve_cache_and_index(f2_byte, cache, index, sizeof(u2));
3100 
3101   call_VM(rbx, CAST_FROM_FN_PTR(address, InterpreterRuntime::vwithfield), cache);
3102   // new value type is returned in rbx
3103   // stack adjustement is returned in rax
3104   __ verify_oop(rbx);
3105   __ addptr(rsp, rax);
3106   __ movptr(rax, rbx);
3107 }
3108 
3109 // The registers cache and index expected to be set before call.
3110 // The function may destroy various registers, just not the cache and index registers.
3111 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3112 
3113   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
3114   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
3115   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
3116   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3117 
3118   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3119 
3120   if (JvmtiExport::can_post_field_modification()) {
3121     // Check to see if a field modification watch has been set before
3122     // we take the time to call into the VM.
3123     Label L1;
3124     assert_different_registers(cache, index, rax);
3125     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3126     __ testl(rax, rax);
3127     __ jcc(Assembler::zero, L1);
3128 
3129     __ get_cache_and_index_at_bcp(robj, RDX, 1);
3130 
3131 
3132     if (is_static) {
3133       // Life is simple.  Null out the object pointer.
3134       __ xorl(RBX, RBX);
3135 
3136     } else {
3137       // Life is harder. The stack holds the value on top, followed by
3138       // the object.  We don't know the size of the value, though; it
3139       // could be one or two words depending on its type. As a result,
3140       // we must find the type to determine where the object is.
3141 #ifndef _LP64
3142       Label two_word, valsize_known;
3143 #endif
3144       __ movl(RCX, Address(robj, RDX,
3145                            Address::times_ptr,
3146                            in_bytes(cp_base_offset +
3147                                      ConstantPoolCacheEntry::flags_offset())));
3148       NOT_LP64(__ mov(rbx, rsp));
3149       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3150 
3151       // Make sure we don't need to mask rcx after the above shift
3152       ConstantPoolCacheEntry::verify_tos_state_shift();
3153 #ifdef _LP64
3154       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3155       __ cmpl(c_rarg3, ltos);
3156       __ cmovptr(Assembler::equal,
3157                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3158       __ cmpl(c_rarg3, dtos);
3159       __ cmovptr(Assembler::equal,
3160                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3161 #else
3162       __ cmpl(rcx, ltos);
3163       __ jccb(Assembler::equal, two_word);
3164       __ cmpl(rcx, dtos);
3165       __ jccb(Assembler::equal, two_word);
3166       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3167       __ jmpb(valsize_known);
3168 
3169       __ bind(two_word);
3170       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3171 
3172       __ bind(valsize_known);
3173       // setup object pointer
3174       __ movptr(rbx, Address(rbx, 0));
3175 #endif
3176     }
3177     // cache entry pointer
3178     __ addptr(robj, in_bytes(cp_base_offset));
3179     __ shll(RDX, LogBytesPerWord);
3180     __ addptr(robj, RDX);
3181     // object (tos)
3182     __ mov(RCX, rsp);
3183     // c_rarg1: object pointer set up above (NULL if static)
3184     // c_rarg2: cache entry pointer
3185     // c_rarg3: jvalue object on the stack
3186     __ call_VM(noreg,
3187                CAST_FROM_FN_PTR(address,
3188                                 InterpreterRuntime::post_field_modification),
3189                RBX, robj, RCX);
3190     __ get_cache_and_index_at_bcp(cache, index, 1);
3191     __ bind(L1);
3192   }
3193 }
3194 
3195 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3196   transition(vtos, vtos);
3197 
3198   const Register cache = rcx;
3199   const Register index = rdx;
3200   const Register obj   = rcx;
3201   const Register off   = rbx;
3202   const Register flags = rax;
3203   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3204   const Register flags2 = rdx;
3205 
3206   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3207   jvmti_post_field_mod(cache, index, is_static);
3208   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3209 
3210   // [jk] not needed currently
3211   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3212   //                                              Assembler::StoreStore));
3213 
3214   Label notVolatile, Done;
3215 
3216   __ movl(flags2, flags);
3217 
3218   // field addresses
3219   const Address field(obj, off, Address::times_1, 0*wordSize);
3220   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3221 
3222   Label notByte, notBool, notInt, notShort, notChar,
3223         notLong, notFloat, notObj, notValueType, notDouble;
3224 
3225   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3226 
3227   assert(btos == 0, "change code, btos != 0");
3228   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3229   __ jcc(Assembler::notZero, notByte);
3230 
3231   // btos
3232   {
3233     __ pop(btos);
3234     if (!is_static) pop_and_check_object(obj);
3235     __ movb(field, rax);
3236     if (!is_static && rc == may_rewrite) {
3237       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3238     }
3239     __ jmp(Done);
3240   }
3241 
3242   __ bind(notByte);
3243   __ cmpl(flags, ztos);
3244   __ jcc(Assembler::notEqual, notBool);
3245 
3246   // ztos
3247   {
3248     __ pop(ztos);
3249     if (!is_static) pop_and_check_object(obj);
3250     __ andl(rax, 0x1);
3251     __ movb(field, rax);
3252     if (!is_static && rc == may_rewrite) {
3253       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3254     }
3255     __ jmp(Done);
3256   }
3257 
3258   __ bind(notBool);
3259   __ cmpl(flags, atos);
3260   __ jcc(Assembler::notEqual, notObj);
3261 
3262   // atos
3263   {
3264     __ pop(atos);
3265     if (!is_static) pop_and_check_object(obj);
3266     // Store into the field
3267     do_oop_store(_masm, field, rax, _bs->kind(), false);
3268     if (!is_static && rc == may_rewrite) {
3269       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3270     }
3271     __ jmp(Done);
3272   }
3273 
3274   __ bind(notObj);
3275   __ cmpl(flags, qtos);
3276   __ jcc(Assembler::notEqual, notValueType);
3277 
3278   // qtos
3279   {
3280     __ pop(qtos); // => rax == value
3281     if (!is_static) {
3282       // value types in non-static fields are embedded
3283       pop_and_check_object(rbx);
3284       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::qputfield),
3285           rbx, rax, rcx);
3286       __ jmp(notVolatile); // value types are never volatile
3287     } else {
3288       // Store into the static field
3289       // Value types in static fields are currently handled with indirection
3290       // but a copy to the Java heap might be required if the value is currently
3291       // stored in a thread local buffer
3292       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::qputstatic), rax, off, obj);
3293     }
3294     __ jmp(Done);
3295   }
3296 
3297   __ bind(notValueType);
3298   __ cmpl(flags, itos);
3299   __ jcc(Assembler::notEqual, notInt);
3300 
3301   // itos
3302   {
3303     __ pop(itos);
3304     if (!is_static) pop_and_check_object(obj);
3305     __ movl(field, rax);
3306     if (!is_static && rc == may_rewrite) {
3307       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3308     }
3309     __ jmp(Done);
3310   }
3311 
3312   __ bind(notInt);
3313   __ cmpl(flags, ctos);
3314   __ jcc(Assembler::notEqual, notChar);
3315 
3316   // ctos
3317   {
3318     __ pop(ctos);
3319     if (!is_static) pop_and_check_object(obj);
3320     __ movw(field, rax);
3321     if (!is_static && rc == may_rewrite) {
3322       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3323     }
3324     __ jmp(Done);
3325   }
3326 
3327   __ bind(notChar);
3328   __ cmpl(flags, stos);
3329   __ jcc(Assembler::notEqual, notShort);
3330 
3331   // stos
3332   {
3333     __ pop(stos);
3334     if (!is_static) pop_and_check_object(obj);
3335     __ movw(field, rax);
3336     if (!is_static && rc == may_rewrite) {
3337       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3338     }
3339     __ jmp(Done);
3340   }
3341 
3342   __ bind(notShort);
3343   __ cmpl(flags, ltos);
3344   __ jcc(Assembler::notEqual, notLong);
3345 
3346   // ltos
3347 #ifdef _LP64
3348   {
3349     __ pop(ltos);
3350     if (!is_static) pop_and_check_object(obj);
3351     __ movq(field, rax);
3352     if (!is_static && rc == may_rewrite) {
3353       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3354     }
3355     __ jmp(Done);
3356   }
3357 #else
3358   {
3359     Label notVolatileLong;
3360     __ testl(rdx, rdx);
3361     __ jcc(Assembler::zero, notVolatileLong);
3362 
3363     __ pop(ltos);  // overwrites rdx, do this after testing volatile.
3364     if (!is_static) pop_and_check_object(obj);
3365 
3366     // Replace with real volatile test
3367     __ push(rdx);
3368     __ push(rax);                 // Must update atomically with FIST
3369     __ fild_d(Address(rsp,0));    // So load into FPU register
3370     __ fistp_d(field);            // and put into memory atomically
3371     __ addptr(rsp, 2*wordSize);
3372     // volatile_barrier();
3373     volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3374                                                  Assembler::StoreStore));
3375     // Don't rewrite volatile version
3376     __ jmp(notVolatile);
3377 
3378     __ bind(notVolatileLong);
3379 
3380     __ pop(ltos);  // overwrites rdx
3381     if (!is_static) pop_and_check_object(obj);
3382     __ movptr(hi, rdx);
3383     __ movptr(field, rax);
3384     // Don't rewrite to _fast_lputfield for potential volatile case.
3385     __ jmp(notVolatile);
3386   }
3387 #endif // _LP64
3388 
3389   __ bind(notLong);
3390   __ cmpl(flags, ftos);
3391   __ jcc(Assembler::notEqual, notFloat);
3392 
3393   // ftos
3394   {
3395     __ pop(ftos);
3396     if (!is_static) pop_and_check_object(obj);
3397     __ store_float(field);
3398     if (!is_static && rc == may_rewrite) {
3399       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3400     }
3401     __ jmp(Done);
3402   }
3403 
3404   __ bind(notFloat);
3405 #ifdef ASSERT
3406   __ cmpl(flags, dtos);
3407   __ jcc(Assembler::notEqual, notDouble);
3408 #endif
3409 
3410   // dtos
3411   {
3412     __ pop(dtos);
3413     if (!is_static) pop_and_check_object(obj);
3414     __ store_double(field);
3415     if (!is_static && rc == may_rewrite) {
3416       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3417     }
3418   }
3419 
3420 #ifdef ASSERT
3421   __ jmp(Done);
3422 
3423   __ bind(notDouble);
3424   __ stop("Bad state");
3425 #endif
3426 
3427   __ bind(Done);
3428 
3429   __ shrl(flags2, ConstantPoolCacheEntry::is_volatile_shift);
3430   __ andl(flags2, 0x1);
3431 
3432   // Check for volatile store
3433   __ testl(flags2, flags2);
3434   __ jcc(Assembler::zero, notVolatile);
3435   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3436                                                Assembler::StoreStore));
3437   __ bind(notVolatile);
3438 }
3439 
3440 void TemplateTable::putfield(int byte_no) {
3441   putfield_or_static(byte_no, false);
3442 }
3443 
3444 void TemplateTable::nofast_putfield(int byte_no) {
3445   putfield_or_static(byte_no, false, may_not_rewrite);
3446 }
3447 
3448 void TemplateTable::putstatic(int byte_no) {
3449   putfield_or_static(byte_no, true);
3450 }
3451 
3452 void TemplateTable::jvmti_post_fast_field_mod() {
3453 
3454   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3455 
3456   if (JvmtiExport::can_post_field_modification()) {
3457     // Check to see if a field modification watch has been set before
3458     // we take the time to call into the VM.
3459     Label L2;
3460     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3461     __ testl(scratch, scratch);
3462     __ jcc(Assembler::zero, L2);
3463     __ pop_ptr(rbx);                  // copy the object pointer from tos
3464     __ verify_oop(rbx);
3465     __ push_ptr(rbx);                 // put the object pointer back on tos
3466     // Save tos values before call_VM() clobbers them. Since we have
3467     // to do it for every data type, we use the saved values as the
3468     // jvalue object.
3469     switch (bytecode()) {          // load values into the jvalue object
3470     case Bytecodes::_fast_qputfield: //fall through
3471     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3472     case Bytecodes::_fast_bputfield: // fall through
3473     case Bytecodes::_fast_zputfield: // fall through
3474     case Bytecodes::_fast_sputfield: // fall through
3475     case Bytecodes::_fast_cputfield: // fall through
3476     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3477     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3478     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3479     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3480 
3481     default:
3482       ShouldNotReachHere();
3483     }
3484     __ mov(scratch, rsp);             // points to jvalue on the stack
3485     // access constant pool cache entry
3486     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3487     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3488     __ verify_oop(rbx);
3489     // rbx: object pointer copied above
3490     // c_rarg2: cache entry pointer
3491     // c_rarg3: jvalue object on the stack
3492     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3493     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3494 
3495     switch (bytecode()) {             // restore tos values
3496     case Bytecodes::_fast_qputfield: // fall through
3497     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3498     case Bytecodes::_fast_bputfield: // fall through
3499     case Bytecodes::_fast_zputfield: // fall through
3500     case Bytecodes::_fast_sputfield: // fall through
3501     case Bytecodes::_fast_cputfield: // fall through
3502     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3503     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3504     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3505     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3506     default: break;
3507     }
3508     __ bind(L2);
3509   }
3510 }
3511 
3512 void TemplateTable::fast_storefield(TosState state) {
3513   transition(state, vtos);
3514 
3515   ByteSize base = ConstantPoolCache::base_offset();
3516 
3517   jvmti_post_fast_field_mod();
3518 
3519   // access constant pool cache
3520   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3521 
3522   // test for volatile with rdx but rdx is tos register for lputfield.
3523   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3524                        in_bytes(base +
3525                                 ConstantPoolCacheEntry::flags_offset())));
3526 
3527   // replace index with field offset from cache entry
3528   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3529                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3530 
3531   // [jk] not needed currently
3532   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3533   //                                              Assembler::StoreStore));
3534 
3535   Label notVolatile;
3536   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3537   __ andl(rdx, 0x1);
3538 
3539   // Get object from stack
3540   pop_and_check_object(rcx);
3541 
3542   // field address
3543   const Address field(rcx, rbx, Address::times_1);
3544 
3545   // access field
3546   switch (bytecode()) {
3547   case Bytecodes::_fast_qputfield:
3548     __ stop("should not be rewritten yet");
3549     break;
3550   case Bytecodes::_fast_aputfield:
3551     do_oop_store(_masm, field, rax, _bs->kind(), false);
3552     break;
3553   case Bytecodes::_fast_lputfield:
3554 #ifdef _LP64
3555   __ movq(field, rax);
3556 #else
3557   __ stop("should not be rewritten");
3558 #endif
3559     break;
3560   case Bytecodes::_fast_iputfield:
3561     __ movl(field, rax);
3562     break;
3563   case Bytecodes::_fast_zputfield:
3564     __ andl(rax, 0x1);  // boolean is true if LSB is 1
3565     // fall through to bputfield
3566   case Bytecodes::_fast_bputfield:
3567     __ movb(field, rax);
3568     break;
3569   case Bytecodes::_fast_sputfield:
3570     // fall through
3571   case Bytecodes::_fast_cputfield:
3572     __ movw(field, rax);
3573     break;
3574   case Bytecodes::_fast_fputfield:
3575     __ store_float(field);
3576     break;
3577   case Bytecodes::_fast_dputfield:
3578     __ store_double(field);
3579     break;
3580   default:
3581     ShouldNotReachHere();
3582   }
3583 
3584   // Check for volatile store
3585   __ testl(rdx, rdx);
3586   __ jcc(Assembler::zero, notVolatile);
3587   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3588                                                Assembler::StoreStore));
3589   __ bind(notVolatile);
3590 }
3591 
3592 void TemplateTable::fast_accessfield(TosState state) {
3593   transition(atos, state);
3594 
3595   // Do the JVMTI work here to avoid disturbing the register state below
3596   if (JvmtiExport::can_post_field_access()) {
3597     // Check to see if a field access watch has been set before we
3598     // take the time to call into the VM.
3599     Label L1;
3600     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3601     __ testl(rcx, rcx);
3602     __ jcc(Assembler::zero, L1);
3603     // access constant pool cache entry
3604     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3605     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3606     __ verify_oop(rax);
3607     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3608     LP64_ONLY(__ mov(c_rarg1, rax));
3609     // c_rarg1: object pointer copied above
3610     // c_rarg2: cache entry pointer
3611     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3612     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3613     __ pop_ptr(rax); // restore object pointer
3614     __ bind(L1);
3615   }
3616 
3617   // access constant pool cache
3618   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3619   // replace index with field offset from cache entry
3620   // [jk] not needed currently
3621   // if (os::is_MP()) {
3622   //   __ movl(rdx, Address(rcx, rbx, Address::times_8,
3623   //                        in_bytes(ConstantPoolCache::base_offset() +
3624   //                                 ConstantPoolCacheEntry::flags_offset())));
3625   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3626   //   __ andl(rdx, 0x1);
3627   // }
3628   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3629                          in_bytes(ConstantPoolCache::base_offset() +
3630                                   ConstantPoolCacheEntry::f2_offset())));
3631 
3632   // rax: object
3633   __ verify_oop(rax);
3634   __ null_check(rax);
3635   Address field(rax, rbx, Address::times_1);
3636 
3637   // access field
3638   switch (bytecode()) {
3639   case Bytecodes::_fast_qgetfield:
3640     __ stop("should not be rewritten yet");
3641     break;
3642   case Bytecodes::_fast_agetfield:
3643     __ load_heap_oop(rax, field);
3644     __ verify_oop(rax);
3645     break;
3646   case Bytecodes::_fast_lgetfield:
3647 #ifdef _LP64
3648   __ movq(rax, field);
3649 #else
3650   __ stop("should not be rewritten");
3651 #endif
3652     break;
3653   case Bytecodes::_fast_igetfield:
3654     __ movl(rax, field);
3655     break;
3656   case Bytecodes::_fast_bgetfield:
3657     __ movsbl(rax, field);
3658     break;
3659   case Bytecodes::_fast_sgetfield:
3660     __ load_signed_short(rax, field);
3661     break;
3662   case Bytecodes::_fast_cgetfield:
3663     __ load_unsigned_short(rax, field);
3664     break;
3665   case Bytecodes::_fast_fgetfield:
3666     __ load_float(field);
3667     break;
3668   case Bytecodes::_fast_dgetfield:
3669     __ load_double(field);
3670     break;
3671   default:
3672     ShouldNotReachHere();
3673   }
3674   // [jk] not needed currently
3675   // if (os::is_MP()) {
3676   //   Label notVolatile;
3677   //   __ testl(rdx, rdx);
3678   //   __ jcc(Assembler::zero, notVolatile);
3679   //   __ membar(Assembler::LoadLoad);
3680   //   __ bind(notVolatile);
3681   //};
3682 }
3683 
3684 void TemplateTable::fast_xaccess(TosState state) {
3685   transition(vtos, state);
3686 
3687   // get receiver
3688   __ movptr(rax, aaddress(0));
3689   // access constant pool cache
3690   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3691   __ movptr(rbx,
3692             Address(rcx, rdx, Address::times_ptr,
3693                     in_bytes(ConstantPoolCache::base_offset() +
3694                              ConstantPoolCacheEntry::f2_offset())));
3695   // make sure exception is reported in correct bcp range (getfield is
3696   // next instruction)
3697   __ increment(rbcp);
3698   __ null_check(rax);
3699   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3700   switch (state) {
3701   case itos:
3702     __ movl(rax, field);
3703     break;
3704   case atos:
3705     __ load_heap_oop(rax, field);
3706     __ verify_oop(rax);
3707     break;
3708   case ftos:
3709     __ load_float(field);
3710     break;
3711   default:
3712     ShouldNotReachHere();
3713   }
3714 
3715   // [jk] not needed currently
3716   // if (os::is_MP()) {
3717   //   Label notVolatile;
3718   //   __ movl(rdx, Address(rcx, rdx, Address::times_8,
3719   //                        in_bytes(ConstantPoolCache::base_offset() +
3720   //                                 ConstantPoolCacheEntry::flags_offset())));
3721   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3722   //   __ testl(rdx, 0x1);
3723   //   __ jcc(Assembler::zero, notVolatile);
3724   //   __ membar(Assembler::LoadLoad);
3725   //   __ bind(notVolatile);
3726   // }
3727 
3728   __ decrement(rbcp);
3729 }
3730 
3731 //-----------------------------------------------------------------------------
3732 // Calls
3733 
3734 void TemplateTable::count_calls(Register method, Register temp) {
3735   // implemented elsewhere
3736   ShouldNotReachHere();
3737 }
3738 
3739 void TemplateTable::prepare_invoke(int byte_no,
3740                                    Register method,  // linked method (or i-klass)
3741                                    Register index,   // itable index, MethodType, etc.
3742                                    Register recv,    // if caller wants to see it
3743                                    Register flags    // if caller wants to test it
3744                                    ) {
3745   // determine flags
3746   const Bytecodes::Code code = bytecode();
3747   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3748   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3749   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3750   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3751   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3752   const bool load_receiver       = (recv  != noreg);
3753   const bool save_flags          = (flags != noreg);
3754   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3755   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3756   assert(flags == noreg || flags == rdx, "");
3757   assert(recv  == noreg || recv  == rcx, "");
3758 
3759   // setup registers & access constant pool cache
3760   if (recv  == noreg)  recv  = rcx;
3761   if (flags == noreg)  flags = rdx;
3762   assert_different_registers(method, index, recv, flags);
3763 
3764   // save 'interpreter return address'
3765   __ save_bcp();
3766 
3767   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3768 
3769   // maybe push appendix to arguments (just before return address)
3770   if (is_invokedynamic || is_invokehandle) {
3771     Label L_no_push;
3772     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3773     __ jcc(Assembler::zero, L_no_push);
3774     // Push the appendix as a trailing parameter.
3775     // This must be done before we get the receiver,
3776     // since the parameter_size includes it.
3777     __ push(rbx);
3778     __ mov(rbx, index);
3779     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3780     __ load_resolved_reference_at_index(index, rbx);
3781     __ pop(rbx);
3782     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3783     __ bind(L_no_push);
3784   }
3785 
3786   // load receiver if needed (after appendix is pushed so parameter size is correct)
3787   // Note: no return address pushed yet
3788   if (load_receiver) {
3789     __ movl(recv, flags);
3790     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3791     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3792     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3793     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3794     __ movptr(recv, recv_addr);
3795     __ verify_oop(recv);
3796   }
3797 
3798   if (save_flags) {
3799     __ movl(rbcp, flags);
3800   }
3801 
3802   // compute return type
3803   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3804   // Make sure we don't need to mask flags after the above shift
3805   ConstantPoolCacheEntry::verify_tos_state_shift();
3806   // load return address
3807   {
3808     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3809     ExternalAddress table(table_addr);
3810     LP64_ONLY(__ lea(rscratch1, table));
3811     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3812     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3813   }
3814 
3815   // push return address
3816   __ push(flags);
3817 
3818   // Restore flags value from the constant pool cache, and restore rsi
3819   // for later null checks.  r13 is the bytecode pointer
3820   if (save_flags) {
3821     __ movl(flags, rbcp);
3822     __ restore_bcp();
3823   }
3824 }
3825 
3826 void TemplateTable::invokevirtual_helper(Register index,
3827                                          Register recv,
3828                                          Register flags) {
3829   // Uses temporary registers rax, rdx
3830   assert_different_registers(index, recv, rax, rdx);
3831   assert(index == rbx, "");
3832   assert(recv  == rcx, "");
3833 
3834   // Test for an invoke of a final method
3835   Label notFinal;
3836   __ movl(rax, flags);
3837   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3838   __ jcc(Assembler::zero, notFinal);
3839 
3840   const Register method = index;  // method must be rbx
3841   assert(method == rbx,
3842          "Method* must be rbx for interpreter calling convention");
3843 
3844   // do the call - the index is actually the method to call
3845   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3846 
3847   // It's final, need a null check here!
3848   __ null_check(recv);
3849 
3850   // profile this call
3851   __ profile_final_call(rax);
3852   __ profile_arguments_type(rax, method, rbcp, true);
3853 
3854   __ jump_from_interpreted(method, rax);
3855 
3856   __ bind(notFinal);
3857 
3858   // get receiver klass
3859   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3860   __ load_klass(rax, recv);
3861 
3862   // profile this call
3863   __ profile_virtual_call(rax, rlocals, rdx);
3864   // get target Method* & entry point
3865   __ lookup_virtual_method(rax, index, method);
3866   __ profile_called_method(method, rdx, rbcp);
3867 
3868   __ profile_arguments_type(rdx, method, rbcp, true);
3869   __ jump_from_interpreted(method, rdx);
3870 }
3871 
3872 void TemplateTable::invokevirtual(int byte_no) {
3873   transition(vtos, vtos);
3874   assert(byte_no == f2_byte, "use this argument");
3875   prepare_invoke(byte_no,
3876                  rbx,    // method or vtable index
3877                  noreg,  // unused itable index
3878                  rcx, rdx); // recv, flags
3879 
3880   // rbx: index
3881   // rcx: receiver
3882   // rdx: flags
3883 
3884   invokevirtual_helper(rbx, rcx, rdx);
3885 }
3886 
3887 void TemplateTable::invokespecial(int byte_no) {
3888   transition(vtos, vtos);
3889   assert(byte_no == f1_byte, "use this argument");
3890   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3891                  rcx);  // get receiver also for null check
3892   __ verify_oop(rcx);
3893   __ null_check(rcx);
3894   // do the call
3895   __ profile_call(rax);
3896   __ profile_arguments_type(rax, rbx, rbcp, false);
3897   __ jump_from_interpreted(rbx, rax);
3898 }
3899 
3900 void TemplateTable::invokestatic(int byte_no) {
3901   transition(vtos, vtos);
3902   assert(byte_no == f1_byte, "use this argument");
3903   prepare_invoke(byte_no, rbx);  // get f1 Method*
3904   // do the call
3905   __ profile_call(rax);
3906   __ profile_arguments_type(rax, rbx, rbcp, false);
3907   __ jump_from_interpreted(rbx, rax);
3908 }
3909 
3910 
3911 void TemplateTable::fast_invokevfinal(int byte_no) {
3912   transition(vtos, vtos);
3913   assert(byte_no == f2_byte, "use this argument");
3914   __ stop("fast_invokevfinal not used on x86");
3915 }
3916 
3917 
3918 void TemplateTable::invokeinterface(int byte_no) {
3919   transition(vtos, vtos);
3920   assert(byte_no == f1_byte, "use this argument");
3921   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 itable index
3922                  rcx, rdx); // recv, flags
3923 
3924   // rax: interface klass (from f1)
3925   // rbx: itable index (from f2)
3926   // rcx: receiver
3927   // rdx: flags
3928 
3929   // Special case of invokeinterface called for virtual method of
3930   // java.lang.Object.  See cpCacheOop.cpp for details.
3931   // This code isn't produced by javac, but could be produced by
3932   // another compliant java compiler.
3933   Label notMethod;
3934   __ movl(rlocals, rdx);
3935   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3936 
3937   __ jcc(Assembler::zero, notMethod);
3938 
3939   invokevirtual_helper(rbx, rcx, rdx);
3940   __ bind(notMethod);
3941 
3942   // Get receiver klass into rdx - also a null check
3943   __ restore_locals();  // restore r14
3944   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3945   __ load_klass(rdx, rcx);
3946 
3947   // profile this call
3948   __ profile_virtual_call(rdx, rbcp, rlocals);
3949 
3950   Label no_such_interface, no_such_method;
3951 
3952   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3953                              rdx, rax, rbx,
3954                              // outputs: method, scan temp. reg
3955                              rbx, rbcp,
3956                              no_such_interface);
3957 
3958   // rbx: Method* to call
3959   // rcx: receiver
3960   // Check for abstract method error
3961   // Note: This should be done more efficiently via a throw_abstract_method_error
3962   //       interpreter entry point and a conditional jump to it in case of a null
3963   //       method.
3964   __ testptr(rbx, rbx);
3965   __ jcc(Assembler::zero, no_such_method);
3966 
3967   __ profile_called_method(rbx, rbcp, rdx);
3968   __ profile_arguments_type(rdx, rbx, rbcp, true);
3969 
3970   // do the call
3971   // rcx: receiver
3972   // rbx,: Method*
3973   __ jump_from_interpreted(rbx, rdx);
3974   __ should_not_reach_here();
3975 
3976   // exception handling code follows...
3977   // note: must restore interpreter registers to canonical
3978   //       state for exception handling to work correctly!
3979 
3980   __ bind(no_such_method);
3981   // throw exception
3982   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3983   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3984   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3985   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3986   // the call_VM checks for exception, so we should never return here.
3987   __ should_not_reach_here();
3988 
3989   __ bind(no_such_interface);
3990   // throw exception
3991   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3992   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3993   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3994   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3995                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3996   // the call_VM checks for exception, so we should never return here.
3997   __ should_not_reach_here();
3998 }
3999 
4000 void TemplateTable::invokehandle(int byte_no) {
4001   transition(vtos, vtos);
4002   assert(byte_no == f1_byte, "use this argument");
4003   const Register rbx_method = rbx;
4004   const Register rax_mtype  = rax;
4005   const Register rcx_recv   = rcx;
4006   const Register rdx_flags  = rdx;
4007 
4008   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
4009   __ verify_method_ptr(rbx_method);
4010   __ verify_oop(rcx_recv);
4011   __ null_check(rcx_recv);
4012 
4013   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4014   // rbx: MH.invokeExact_MT method (from f2)
4015 
4016   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
4017 
4018   // FIXME: profile the LambdaForm also
4019   __ profile_final_call(rax);
4020   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4021 
4022   __ jump_from_interpreted(rbx_method, rdx);
4023 }
4024 
4025 void TemplateTable::invokedynamic(int byte_no) {
4026   transition(vtos, vtos);
4027   assert(byte_no == f1_byte, "use this argument");
4028 
4029   const Register rbx_method   = rbx;
4030   const Register rax_callsite = rax;
4031 
4032   prepare_invoke(byte_no, rbx_method, rax_callsite);
4033 
4034   // rax: CallSite object (from cpool->resolved_references[f1])
4035   // rbx: MH.linkToCallSite method (from f2)
4036 
4037   // Note:  rax_callsite is already pushed by prepare_invoke
4038 
4039   // %%% should make a type profile for any invokedynamic that takes a ref argument
4040   // profile this call
4041   __ profile_call(rbcp);
4042   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4043 
4044   __ verify_oop(rax_callsite);
4045 
4046   __ jump_from_interpreted(rbx_method, rdx);
4047 }
4048 
4049 //-----------------------------------------------------------------------------
4050 // Allocation
4051 
4052 void TemplateTable::_new() {
4053   transition(vtos, atos);
4054   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4055   Label slow_case;
4056   Label slow_case_no_pop;
4057   Label done;
4058   Label initialize_header;
4059   Label initialize_object;  // including clearing the fields
4060   Label allocate_shared;
4061 
4062   __ get_cpool_and_tags(rcx, rax);
4063 
4064   // Make sure the class we're about to instantiate has been resolved.
4065   // This is done before loading InstanceKlass to be consistent with the order
4066   // how Constant Pool is updated (see ConstantPool::klass_at_put)
4067   const int tags_offset = Array<u1>::base_offset_in_bytes();
4068   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4069   __ jcc(Assembler::notEqual, slow_case_no_pop);
4070 
4071   // get InstanceKlass
4072   __ load_resolved_klass_at_index(rcx, rdx, rcx);
4073   __ push(rcx);  // save the contexts of klass for initializing the header
4074 
4075   // make sure klass is initialized & doesn't have finalizer
4076   // make sure klass is fully initialized
4077   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4078   __ jcc(Assembler::notEqual, slow_case);
4079 
4080   // get instance_size in InstanceKlass (scaled to a count of bytes)
4081   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4082   // test to see if it has a finalizer or is malformed in some way
4083   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4084   __ jcc(Assembler::notZero, slow_case);
4085 
4086   //
4087   // Allocate the instance
4088   // 1) Try to allocate in the TLAB
4089   // 2) if fail and the object is large allocate in the shared Eden
4090   // 3) if the above fails (or is not applicable), go to a slow case
4091   // (creates a new TLAB, etc.)
4092 
4093   const bool allow_shared_alloc =
4094     Universe::heap()->supports_inline_contig_alloc();
4095 
4096   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4097 #ifndef _LP64
4098   if (UseTLAB || allow_shared_alloc) {
4099     __ get_thread(thread);
4100   }
4101 #endif // _LP64
4102 
4103   if (UseTLAB) {
4104     __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4105     __ lea(rbx, Address(rax, rdx, Address::times_1));
4106     __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4107     __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
4108     __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4109     if (ZeroTLAB) {
4110       // the fields have been already cleared
4111       __ jmp(initialize_header);
4112     } else {
4113       // initialize both the header and fields
4114       __ jmp(initialize_object);
4115     }
4116   }
4117 
4118   // Allocation in the shared Eden, if allowed.
4119   //
4120   // rdx: instance size in bytes
4121   if (allow_shared_alloc) {
4122     __ bind(allocate_shared);
4123 
4124     ExternalAddress heap_top((address)Universe::heap()->top_addr());
4125     ExternalAddress heap_end((address)Universe::heap()->end_addr());
4126 
4127     Label retry;
4128     __ bind(retry);
4129     __ movptr(rax, heap_top);
4130     __ lea(rbx, Address(rax, rdx, Address::times_1));
4131     __ cmpptr(rbx, heap_end);
4132     __ jcc(Assembler::above, slow_case);
4133 
4134     // Compare rax, with the top addr, and if still equal, store the new
4135     // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4136     // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4137     //
4138     // rax,: object begin
4139     // rbx,: object end
4140     // rdx: instance size in bytes
4141     __ locked_cmpxchgptr(rbx, heap_top);
4142 
4143     // if someone beat us on the allocation, try again, otherwise continue
4144     __ jcc(Assembler::notEqual, retry);
4145 
4146     __ incr_allocated_bytes(thread, rdx, 0);
4147   }
4148 
4149   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
4150     // The object is initialized before the header.  If the object size is
4151     // zero, go directly to the header initialization.
4152     __ bind(initialize_object);
4153     __ decrement(rdx, sizeof(oopDesc));
4154     __ jcc(Assembler::zero, initialize_header);
4155 
4156     // Initialize topmost object field, divide rdx by 8, check if odd and
4157     // test if zero.
4158     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
4159     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4160 
4161     // rdx must have been multiple of 8
4162 #ifdef ASSERT
4163     // make sure rdx was multiple of 8
4164     Label L;
4165     // Ignore partial flag stall after shrl() since it is debug VM
4166     __ jccb(Assembler::carryClear, L);
4167     __ stop("object size is not multiple of 2 - adjust this code");
4168     __ bind(L);
4169     // rdx must be > 0, no extra check needed here
4170 #endif
4171 
4172     // initialize remaining object fields: rdx was a multiple of 8
4173     { Label loop;
4174     __ bind(loop);
4175     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4176     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4177     __ decrement(rdx);
4178     __ jcc(Assembler::notZero, loop);
4179     }
4180 
4181     // initialize object header only.
4182     __ bind(initialize_header);
4183     if (UseBiasedLocking) {
4184       __ pop(rcx);   // get saved klass back in the register.
4185       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4186       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4187     } else {
4188       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4189                 (intptr_t)markOopDesc::prototype()); // header
4190       __ pop(rcx);   // get saved klass back in the register.
4191     }
4192 #ifdef _LP64
4193     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4194     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4195 #endif
4196     __ store_klass(rax, rcx);  // klass
4197 
4198     {
4199       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4200       // Trigger dtrace event for fastpath
4201       __ push(atos);
4202       __ call_VM_leaf(
4203            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4204       __ pop(atos);
4205     }
4206 
4207     __ jmp(done);
4208   }
4209 
4210   // slow case
4211   __ bind(slow_case);
4212   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4213   __ bind(slow_case_no_pop);
4214 
4215   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4216   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4217 
4218   __ get_constant_pool(rarg1);
4219   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4220   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4221    __ verify_oop(rax);
4222 
4223   // continue
4224   __ bind(done);
4225 }
4226 
4227 void TemplateTable::vdefault() {
4228   transition(vtos, qtos);
4229 
4230   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4231   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4232 
4233   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4234   __ get_constant_pool(rarg1);
4235 
4236   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vdefault),
4237       rarg1, rarg2);
4238   __ verify_oop(rax);
4239 }
4240 
4241 void TemplateTable::newarray() {
4242   transition(itos, atos);
4243   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4244   __ load_unsigned_byte(rarg1, at_bcp(1));
4245   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4246           rarg1, rax);
4247 }
4248 
4249 void TemplateTable::anewarray() {
4250   transition(itos, atos);
4251 
4252   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4253   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4254 
4255   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4256   __ get_constant_pool(rarg1);
4257   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4258           rarg1, rarg2, rax);
4259 }
4260 
4261 void TemplateTable::arraylength() {
4262   transition(atos, itos);
4263   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4264   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4265 }
4266 
4267 void TemplateTable::checkcast() {
4268   transition(atos, atos);
4269   Label done, is_null, ok_is_subtype, quicked, resolved;
4270   __ testptr(rax, rax); // object is in rax
4271   __ jcc(Assembler::zero, is_null);
4272 
4273   // Get cpool & tags index
4274   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4275   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4276   // See if bytecode has already been quicked
4277   __ cmpb(Address(rdx, rbx,
4278                   Address::times_1,
4279                   Array<u1>::base_offset_in_bytes()),
4280           JVM_CONSTANT_Class);
4281   __ jcc(Assembler::equal, quicked);
4282   __ push(atos); // save receiver for result, and for GC
4283   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4284 
4285   // vm_result_2 has metadata result
4286 #ifndef _LP64
4287   // borrow rdi from locals
4288   __ get_thread(rdi);
4289   __ get_vm_result_2(rax, rdi);
4290   __ restore_locals();
4291 #else
4292   __ get_vm_result_2(rax, r15_thread);
4293 #endif
4294 
4295   __ pop_ptr(rdx); // restore receiver
4296   __ jmpb(resolved);
4297 
4298   // Get superklass in rax and subklass in rbx
4299   __ bind(quicked);
4300   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4301   __ load_resolved_klass_at_index(rcx, rbx, rax);
4302 
4303   __ bind(resolved);
4304   __ load_klass(rbx, rdx);
4305 
4306   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4307   // Superklass in rax.  Subklass in rbx.
4308   __ gen_subtype_check(rbx, ok_is_subtype);
4309 
4310   // Come here on failure
4311   __ push_ptr(rdx);
4312   // object is at TOS
4313   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4314 
4315   // Come here on success
4316   __ bind(ok_is_subtype);
4317   __ mov(rax, rdx); // Restore object in rdx
4318 
4319   // Collect counts on whether this check-cast sees NULLs a lot or not.
4320   if (ProfileInterpreter) {
4321     __ jmp(done);
4322     __ bind(is_null);
4323     __ profile_null_seen(rcx);
4324   } else {
4325     __ bind(is_null);   // same as 'done'
4326   }
4327   __ bind(done);
4328 }
4329 
4330 void TemplateTable::instanceof() {
4331   transition(atos, itos);
4332   Label done, is_null, ok_is_subtype, quicked, resolved;
4333   __ testptr(rax, rax);
4334   __ jcc(Assembler::zero, is_null);
4335 
4336   // Get cpool & tags index
4337   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4338   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4339   // See if bytecode has already been quicked
4340   __ cmpb(Address(rdx, rbx,
4341                   Address::times_1,
4342                   Array<u1>::base_offset_in_bytes()),
4343           JVM_CONSTANT_Class);
4344   __ jcc(Assembler::equal, quicked);
4345 
4346   __ push(atos); // save receiver for result, and for GC
4347   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4348   // vm_result_2 has metadata result
4349 
4350 #ifndef _LP64
4351   // borrow rdi from locals
4352   __ get_thread(rdi);
4353   __ get_vm_result_2(rax, rdi);
4354   __ restore_locals();
4355 #else
4356   __ get_vm_result_2(rax, r15_thread);
4357 #endif
4358 
4359   __ pop_ptr(rdx); // restore receiver
4360   __ verify_oop(rdx);
4361   __ load_klass(rdx, rdx);
4362   __ jmpb(resolved);
4363 
4364   // Get superklass in rax and subklass in rdx
4365   __ bind(quicked);
4366   __ load_klass(rdx, rax);
4367   __ load_resolved_klass_at_index(rcx, rbx, rax);
4368 
4369   __ bind(resolved);
4370 
4371   // Generate subtype check.  Blows rcx, rdi
4372   // Superklass in rax.  Subklass in rdx.
4373   __ gen_subtype_check(rdx, ok_is_subtype);
4374 
4375   // Come here on failure
4376   __ xorl(rax, rax);
4377   __ jmpb(done);
4378   // Come here on success
4379   __ bind(ok_is_subtype);
4380   __ movl(rax, 1);
4381 
4382   // Collect counts on whether this test sees NULLs a lot or not.
4383   if (ProfileInterpreter) {
4384     __ jmp(done);
4385     __ bind(is_null);
4386     __ profile_null_seen(rcx);
4387   } else {
4388     __ bind(is_null);   // same as 'done'
4389   }
4390   __ bind(done);
4391   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4392   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4393 }
4394 
4395 void TemplateTable::_vbox() {
4396   transition(qtos, atos);
4397 
4398   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4399   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4400 
4401   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4402   __ get_constant_pool(rarg1);
4403   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vbox),
4404       rarg1, rarg2, rax);
4405 }
4406 
4407 void TemplateTable::_vunbox() {
4408   transition(atos, qtos);
4409 
4410   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4411   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4412 
4413   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4414   __ get_constant_pool(rarg1);
4415   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vunbox),
4416       rarg1, rarg2, rax);
4417 }
4418 
4419 //----------------------------------------------------------------------------------------------------
4420 // Breakpoints
4421 void TemplateTable::_breakpoint() {
4422   // Note: We get here even if we are single stepping..
4423   // jbug insists on setting breakpoints at every bytecode
4424   // even if we are in single step mode.
4425 
4426   transition(vtos, vtos);
4427 
4428   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4429 
4430   // get the unpatched byte code
4431   __ get_method(rarg);
4432   __ call_VM(noreg,
4433              CAST_FROM_FN_PTR(address,
4434                               InterpreterRuntime::get_original_bytecode_at),
4435              rarg, rbcp);
4436   __ mov(rbx, rax);  // why?
4437 
4438   // post the breakpoint event
4439   __ get_method(rarg);
4440   __ call_VM(noreg,
4441              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4442              rarg, rbcp);
4443 
4444   // complete the execution of original bytecode
4445   __ dispatch_only_normal(vtos);
4446 }
4447 
4448 //-----------------------------------------------------------------------------
4449 // Exceptions
4450 
4451 void TemplateTable::athrow() {
4452   transition(atos, vtos);
4453   __ null_check(rax);
4454   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4455 }
4456 
4457 //-----------------------------------------------------------------------------
4458 // Synchronization
4459 //
4460 // Note: monitorenter & exit are symmetric routines; which is reflected
4461 //       in the assembly code structure as well
4462 //
4463 // Stack layout:
4464 //
4465 // [expressions  ] <--- rsp               = expression stack top
4466 // ..
4467 // [expressions  ]
4468 // [monitor entry] <--- monitor block top = expression stack bot
4469 // ..
4470 // [monitor entry]
4471 // [frame data   ] <--- monitor block bot
4472 // ...
4473 // [saved rbp    ] <--- rbp
4474 void TemplateTable::monitorenter() {
4475   transition(atos, vtos);
4476 
4477   // check for NULL object
4478   __ null_check(rax);
4479 
4480   const Address monitor_block_top(
4481         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4482   const Address monitor_block_bot(
4483         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4484   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4485 
4486   Label allocated;
4487 
4488   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4489   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4490   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4491 
4492   // initialize entry pointer
4493   __ xorl(rmon, rmon); // points to free slot or NULL
4494 
4495   // find a free slot in the monitor block (result in rmon)
4496   {
4497     Label entry, loop, exit;
4498     __ movptr(rtop, monitor_block_top); // points to current entry,
4499                                         // starting with top-most entry
4500     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4501                                         // of monitor block
4502     __ jmpb(entry);
4503 
4504     __ bind(loop);
4505     // check if current entry is used
4506     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4507     // if not used then remember entry in rmon
4508     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4509     // check if current entry is for same object
4510     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4511     // if same object then stop searching
4512     __ jccb(Assembler::equal, exit);
4513     // otherwise advance to next entry
4514     __ addptr(rtop, entry_size);
4515     __ bind(entry);
4516     // check if bottom reached
4517     __ cmpptr(rtop, rbot);
4518     // if not at bottom then check this entry
4519     __ jcc(Assembler::notEqual, loop);
4520     __ bind(exit);
4521   }
4522 
4523   __ testptr(rmon, rmon); // check if a slot has been found
4524   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4525 
4526   // allocate one if there's no free slot
4527   {
4528     Label entry, loop;
4529     // 1. compute new pointers          // rsp: old expression stack top
4530     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4531     __ subptr(rsp, entry_size);         // move expression stack top
4532     __ subptr(rmon, entry_size);        // move expression stack bottom
4533     __ mov(rtop, rsp);                  // set start value for copy loop
4534     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4535     __ jmp(entry);
4536     // 2. move expression stack contents
4537     __ bind(loop);
4538     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4539                                                 // word from old location
4540     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4541     __ addptr(rtop, wordSize);                  // advance to next word
4542     __ bind(entry);
4543     __ cmpptr(rtop, rmon);                      // check if bottom reached
4544     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4545                                                 // copy next word
4546   }
4547 
4548   // call run-time routine
4549   // rmon: points to monitor entry
4550   __ bind(allocated);
4551 
4552   // Increment bcp to point to the next bytecode, so exception
4553   // handling for async. exceptions work correctly.
4554   // The object has already been poped from the stack, so the
4555   // expression stack looks correct.
4556   __ increment(rbcp);
4557 
4558   // store object
4559   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4560   __ lock_object(rmon);
4561 
4562   // check to make sure this monitor doesn't cause stack overflow after locking
4563   __ save_bcp();  // in case of exception
4564   __ generate_stack_overflow_check(0);
4565 
4566   // The bcp has already been incremented. Just need to dispatch to
4567   // next instruction.
4568   __ dispatch_next(vtos);
4569 }
4570 
4571 void TemplateTable::monitorexit() {
4572   transition(atos, vtos);
4573 
4574   // check for NULL object
4575   __ null_check(rax);
4576 
4577   const Address monitor_block_top(
4578         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4579   const Address monitor_block_bot(
4580         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4581   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4582 
4583   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4584   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4585 
4586   Label found;
4587 
4588   // find matching slot
4589   {
4590     Label entry, loop;
4591     __ movptr(rtop, monitor_block_top); // points to current entry,
4592                                         // starting with top-most entry
4593     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4594                                         // of monitor block
4595     __ jmpb(entry);
4596 
4597     __ bind(loop);
4598     // check if current entry is for same object
4599     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4600     // if same object then stop searching
4601     __ jcc(Assembler::equal, found);
4602     // otherwise advance to next entry
4603     __ addptr(rtop, entry_size);
4604     __ bind(entry);
4605     // check if bottom reached
4606     __ cmpptr(rtop, rbot);
4607     // if not at bottom then check this entry
4608     __ jcc(Assembler::notEqual, loop);
4609   }
4610 
4611   // error handling. Unlocking was not block-structured
4612   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4613                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4614   __ should_not_reach_here();
4615 
4616   // call run-time routine
4617   __ bind(found);
4618   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4619   __ unlock_object(rtop);
4620   __ pop_ptr(rax); // discard object
4621 }
4622 
4623 // Wide instructions
4624 void TemplateTable::wide() {
4625   transition(vtos, vtos);
4626   __ load_unsigned_byte(rbx, at_bcp(1));
4627   ExternalAddress wtable((address)Interpreter::_wentry_point);
4628   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4629   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4630 }
4631 
4632 // Multi arrays
4633 void TemplateTable::multianewarray() {
4634   transition(vtos, atos);
4635 
4636   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4637   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4638   // last dim is on top of stack; we want address of first one:
4639   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4640   // the latter wordSize to point to the beginning of the array.
4641   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4642   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4643   __ load_unsigned_byte(rbx, at_bcp(3));
4644   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4645 }