1 /*
   2  * Copyright (c) 1997, 2019, 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 "compiler/disassembler.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterRuntime.hpp"
  30 #include "interpreter/interp_masm.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "memory/universe.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/objArrayKlass.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/methodHandles.hpp"
  37 #include "runtime/frame.inline.hpp"
  38 #include "runtime/safepointMechanism.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/stubRoutines.hpp"
  41 #include "runtime/synchronizer.hpp"
  42 #include "utilities/macros.hpp"
  43 
  44 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  45 
  46 // Global Register Names
  47 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  48 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  49 
  50 // Platform-dependent initialization
  51 void TemplateTable::pd_initialize() {
  52   // No x86 specific initialization
  53 }
  54 
  55 // Address Computation: local variables
  56 static inline Address iaddress(int n) {
  57   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  58 }
  59 
  60 static inline Address laddress(int n) {
  61   return iaddress(n + 1);
  62 }
  63 
  64 #ifndef _LP64
  65 static inline Address haddress(int n) {
  66   return iaddress(n + 0);
  67 }
  68 #endif
  69 
  70 static inline Address faddress(int n) {
  71   return iaddress(n);
  72 }
  73 
  74 static inline Address daddress(int n) {
  75   return laddress(n);
  76 }
  77 
  78 static inline Address aaddress(int n) {
  79   return iaddress(n);
  80 }
  81 
  82 static inline Address iaddress(Register r) {
  83   return Address(rlocals, r, Address::times_ptr);
  84 }
  85 
  86 static inline Address laddress(Register r) {
  87   return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
  88 }
  89 
  90 #ifndef _LP64
  91 static inline Address haddress(Register r)       {
  92   return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  93 }
  94 #endif
  95 
  96 static inline Address faddress(Register r) {
  97   return iaddress(r);
  98 }
  99 
 100 static inline Address daddress(Register r) {
 101   return laddress(r);
 102 }
 103 
 104 static inline Address aaddress(Register r) {
 105   return iaddress(r);
 106 }
 107 
 108 
 109 // expression stack
 110 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
 111 // data beyond the rsp which is potentially unsafe in an MT environment;
 112 // an interrupt may overwrite that data.)
 113 static inline Address at_rsp   () {
 114   return Address(rsp, 0);
 115 }
 116 
 117 // At top of Java expression stack which may be different than esp().  It
 118 // isn't for category 1 objects.
 119 static inline Address at_tos   () {
 120   return Address(rsp,  Interpreter::expr_offset_in_bytes(0));
 121 }
 122 
 123 static inline Address at_tos_p1() {
 124   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
 125 }
 126 
 127 static inline Address at_tos_p2() {
 128   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
 129 }
 130 
 131 // Condition conversion
 132 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 133   switch (cc) {
 134   case TemplateTable::equal        : return Assembler::notEqual;
 135   case TemplateTable::not_equal    : return Assembler::equal;
 136   case TemplateTable::less         : return Assembler::greaterEqual;
 137   case TemplateTable::less_equal   : return Assembler::greater;
 138   case TemplateTable::greater      : return Assembler::lessEqual;
 139   case TemplateTable::greater_equal: return Assembler::less;
 140   }
 141   ShouldNotReachHere();
 142   return Assembler::zero;
 143 }
 144 
 145 
 146 
 147 // Miscelaneous helper routines
 148 // Store an oop (or NULL) at the address described by obj.
 149 // If val == noreg this means store a NULL
 150 
 151 
 152 static void do_oop_store(InterpreterMacroAssembler* _masm,
 153                          Address dst,
 154                          Register val,
 155                          DecoratorSet decorators = 0) {
 156   assert(val == noreg || val == rax, "parameter is just for looks");
 157   __ store_heap_oop(dst, val, rdx, rbx, decorators);
 158 }
 159 
 160 static void do_oop_load(InterpreterMacroAssembler* _masm,
 161                         Address src,
 162                         Register dst,
 163                         DecoratorSet decorators = 0) {
 164   __ load_heap_oop(dst, src, rdx, rbx, decorators);
 165 }
 166 
 167 Address TemplateTable::at_bcp(int offset) {
 168   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 169   return Address(rbcp, offset);
 170 }
 171 
 172 
 173 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 174                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 175                                    int byte_no) {
 176   if (!RewriteBytecodes)  return;
 177   Label L_patch_done;
 178 
 179   switch (bc) {
 180   case Bytecodes::_fast_aputfield:
 181   case Bytecodes::_fast_bputfield:
 182   case Bytecodes::_fast_zputfield:
 183   case Bytecodes::_fast_cputfield:
 184   case Bytecodes::_fast_dputfield:
 185   case Bytecodes::_fast_fputfield:
 186   case Bytecodes::_fast_iputfield:
 187   case Bytecodes::_fast_lputfield:
 188   case Bytecodes::_fast_sputfield:
 189     {
 190       // We skip bytecode quickening for putfield instructions when
 191       // the put_code written to the constant pool cache is zero.
 192       // This is required so that every execution of this instruction
 193       // calls out to InterpreterRuntime::resolve_get_put to do
 194       // additional, required work.
 195       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 196       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 197       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 198       __ movl(bc_reg, bc);
 199       __ cmpl(temp_reg, (int) 0);
 200       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 201     }
 202     break;
 203   default:
 204     assert(byte_no == -1, "sanity");
 205     // the pair bytecodes have already done the load.
 206     if (load_bc_into_bc_reg) {
 207       __ movl(bc_reg, bc);
 208     }
 209   }
 210 
 211   if (JvmtiExport::can_post_breakpoint()) {
 212     Label L_fast_patch;
 213     // if a breakpoint is present we can't rewrite the stream directly
 214     __ movzbl(temp_reg, at_bcp(0));
 215     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 216     __ jcc(Assembler::notEqual, L_fast_patch);
 217     __ get_method(temp_reg);
 218     // Let breakpoint table handling rewrite to quicker bytecode
 219     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 220 #ifndef ASSERT
 221     __ jmpb(L_patch_done);
 222 #else
 223     __ jmp(L_patch_done);
 224 #endif
 225     __ bind(L_fast_patch);
 226   }
 227 
 228 #ifdef ASSERT
 229   Label L_okay;
 230   __ load_unsigned_byte(temp_reg, at_bcp(0));
 231   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 232   __ jcc(Assembler::equal, L_okay);
 233   __ cmpl(temp_reg, bc_reg);
 234   __ jcc(Assembler::equal, L_okay);
 235   __ stop("patching the wrong bytecode");
 236   __ bind(L_okay);
 237 #endif
 238 
 239   // patch bytecode
 240   __ movb(at_bcp(0), bc_reg);
 241   __ bind(L_patch_done);
 242 }
 243 // Individual instructions
 244 
 245 
 246 void TemplateTable::nop() {
 247   transition(vtos, vtos);
 248   // nothing to do
 249 }
 250 
 251 void TemplateTable::shouldnotreachhere() {
 252   transition(vtos, vtos);
 253   __ stop("shouldnotreachhere bytecode");
 254 }
 255 
 256 void TemplateTable::aconst_null() {
 257   transition(vtos, atos);
 258   __ xorl(rax, rax);
 259 }
 260 
 261 void TemplateTable::iconst(int value) {
 262   transition(vtos, itos);
 263   if (value == 0) {
 264     __ xorl(rax, rax);
 265   } else {
 266     __ movl(rax, value);
 267   }
 268 }
 269 
 270 void TemplateTable::lconst(int value) {
 271   transition(vtos, ltos);
 272   if (value == 0) {
 273     __ xorl(rax, rax);
 274   } else {
 275     __ movl(rax, value);
 276   }
 277 #ifndef _LP64
 278   assert(value >= 0, "check this code");
 279   __ xorptr(rdx, rdx);
 280 #endif
 281 }
 282 
 283 
 284 
 285 void TemplateTable::fconst(int value) {
 286   transition(vtos, ftos);
 287   if (UseSSE >= 1) {
 288     static float one = 1.0f, two = 2.0f;
 289     switch (value) {
 290     case 0:
 291       __ xorps(xmm0, xmm0);
 292       break;
 293     case 1:
 294       __ movflt(xmm0, ExternalAddress((address) &one));
 295       break;
 296     case 2:
 297       __ movflt(xmm0, ExternalAddress((address) &two));
 298       break;
 299     default:
 300       ShouldNotReachHere();
 301       break;
 302     }
 303   } else {
 304 #ifdef _LP64
 305     ShouldNotReachHere();
 306 #else
 307            if (value == 0) { __ fldz();
 308     } else if (value == 1) { __ fld1();
 309     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 310     } else                 { ShouldNotReachHere();
 311     }
 312 #endif // _LP64
 313   }
 314 }
 315 
 316 void TemplateTable::dconst(int value) {
 317   transition(vtos, dtos);
 318   if (UseSSE >= 2) {
 319     static double one = 1.0;
 320     switch (value) {
 321     case 0:
 322       __ xorpd(xmm0, xmm0);
 323       break;
 324     case 1:
 325       __ movdbl(xmm0, ExternalAddress((address) &one));
 326       break;
 327     default:
 328       ShouldNotReachHere();
 329       break;
 330     }
 331   } else {
 332 #ifdef _LP64
 333     ShouldNotReachHere();
 334 #else
 335            if (value == 0) { __ fldz();
 336     } else if (value == 1) { __ fld1();
 337     } else                 { ShouldNotReachHere();
 338     }
 339 #endif
 340   }
 341 }
 342 
 343 void TemplateTable::bipush() {
 344   transition(vtos, itos);
 345   __ load_signed_byte(rax, at_bcp(1));
 346 }
 347 
 348 void TemplateTable::sipush() {
 349   transition(vtos, itos);
 350   __ load_unsigned_short(rax, at_bcp(1));
 351   __ bswapl(rax);
 352   __ sarl(rax, 16);
 353 }
 354 
 355 void TemplateTable::ldc(bool wide) {
 356   transition(vtos, vtos);
 357   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 358   Label call_ldc, notFloat, notClass, notInt, Done;
 359 
 360   if (wide) {
 361     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 362   } else {
 363     __ load_unsigned_byte(rbx, at_bcp(1));
 364   }
 365 
 366   __ get_cpool_and_tags(rcx, rax);
 367   const int base_offset = ConstantPool::header_size() * wordSize;
 368   const int tags_offset = Array<u1>::base_offset_in_bytes();
 369 
 370   // get type
 371   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 372 
 373   // unresolved class - get the resolved class
 374   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 375   __ jccb(Assembler::equal, call_ldc);
 376 
 377   // unresolved class in error state - call into runtime to throw the error
 378   // from the first resolution attempt
 379   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 380   __ jccb(Assembler::equal, call_ldc);
 381 
 382   // resolved class - need to call vm to get java mirror of the class
 383   __ cmpl(rdx, JVM_CONSTANT_Class);
 384   __ jcc(Assembler::notEqual, notClass);
 385 
 386   __ bind(call_ldc);
 387 
 388   __ movl(rarg, wide);
 389   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 390 
 391   __ push(atos);
 392   __ jmp(Done);
 393 
 394   __ bind(notClass);
 395   __ cmpl(rdx, JVM_CONSTANT_Float);
 396   __ jccb(Assembler::notEqual, notFloat);
 397 
 398   // ftos
 399   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 400   __ push(ftos);
 401   __ jmp(Done);
 402 
 403   __ bind(notFloat);
 404   __ cmpl(rdx, JVM_CONSTANT_Integer);
 405   __ jccb(Assembler::notEqual, notInt);
 406 
 407   // itos
 408   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 409   __ push(itos);
 410   __ jmp(Done);
 411 
 412   // assume the tag is for condy; if not, the VM runtime will tell us
 413   __ bind(notInt);
 414   condy_helper(Done);
 415 
 416   __ bind(Done);
 417 }
 418 
 419 // Fast path for caching oop constants.
 420 void TemplateTable::fast_aldc(bool wide) {
 421   transition(vtos, atos);
 422 
 423   Register result = rax;
 424   Register tmp = rdx;
 425   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 426   int index_size = wide ? sizeof(u2) : sizeof(u1);
 427 
 428   Label resolved;
 429 
 430   // We are resolved if the resolved reference cache entry contains a
 431   // non-null object (String, MethodType, etc.)
 432   assert_different_registers(result, tmp);
 433   __ get_cache_index_at_bcp(tmp, 1, index_size);
 434   __ load_resolved_reference_at_index(result, tmp);
 435   __ testptr(result, result);
 436   __ jcc(Assembler::notZero, resolved);
 437 
 438   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 439 
 440   // first time invocation - must resolve first
 441   __ movl(rarg, (int)bytecode());
 442   __ call_VM(result, entry, rarg);
 443   __ bind(resolved);
 444 
 445   { // Check for the null sentinel.
 446     // If we just called the VM, it already did the mapping for us,
 447     // but it's harmless to retry.
 448     Label notNull;
 449     ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
 450     __ movptr(tmp, null_sentinel);
 451     __ cmpoop(tmp, result);
 452     __ jccb(Assembler::notEqual, notNull);
 453     __ xorptr(result, result);  // NULL object reference
 454     __ bind(notNull);
 455   }
 456 
 457   if (VerifyOops) {
 458     __ verify_oop(result);
 459   }
 460 }
 461 
 462 void TemplateTable::ldc2_w() {
 463   transition(vtos, vtos);
 464   Label notDouble, notLong, Done;
 465   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 466 
 467   __ get_cpool_and_tags(rcx, rax);
 468   const int base_offset = ConstantPool::header_size() * wordSize;
 469   const int tags_offset = Array<u1>::base_offset_in_bytes();
 470 
 471   // get type
 472   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 473   __ cmpl(rdx, JVM_CONSTANT_Double);
 474   __ jccb(Assembler::notEqual, notDouble);
 475 
 476   // dtos
 477   __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
 478   __ push(dtos);
 479 
 480   __ jmp(Done);
 481   __ bind(notDouble);
 482   __ cmpl(rdx, JVM_CONSTANT_Long);
 483   __ jccb(Assembler::notEqual, notLong);
 484 
 485   // ltos
 486   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 487   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 488   __ push(ltos);
 489   __ jmp(Done);
 490 
 491   __ bind(notLong);
 492   condy_helper(Done);
 493 
 494   __ bind(Done);
 495 }
 496 
 497 void TemplateTable::condy_helper(Label& Done) {
 498   const Register obj = rax;
 499   const Register off = rbx;
 500   const Register flags = rcx;
 501   const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 502   __ movl(rarg, (int)bytecode());
 503   call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
 504 #ifndef _LP64
 505   // borrow rdi from locals
 506   __ get_thread(rdi);
 507   __ get_vm_result_2(flags, rdi);
 508   __ restore_locals();
 509 #else
 510   __ get_vm_result_2(flags, r15_thread);
 511 #endif
 512   // VMr = obj = base address to find primitive value to push
 513   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 514   __ movl(off, flags);
 515   __ andl(off, ConstantPoolCacheEntry::field_index_mask);
 516   const Address field(obj, off, Address::times_1, 0*wordSize);
 517 
 518   // What sort of thing are we loading?
 519   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
 520   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
 521 
 522   switch (bytecode()) {
 523   case Bytecodes::_ldc:
 524   case Bytecodes::_ldc_w:
 525     {
 526       // tos in (itos, ftos, stos, btos, ctos, ztos)
 527       Label notInt, notFloat, notShort, notByte, notChar, notBool;
 528       __ cmpl(flags, itos);
 529       __ jcc(Assembler::notEqual, notInt);
 530       // itos
 531       __ movl(rax, field);
 532       __ push(itos);
 533       __ jmp(Done);
 534 
 535       __ bind(notInt);
 536       __ cmpl(flags, ftos);
 537       __ jcc(Assembler::notEqual, notFloat);
 538       // ftos
 539       __ load_float(field);
 540       __ push(ftos);
 541       __ jmp(Done);
 542 
 543       __ bind(notFloat);
 544       __ cmpl(flags, stos);
 545       __ jcc(Assembler::notEqual, notShort);
 546       // stos
 547       __ load_signed_short(rax, field);
 548       __ push(stos);
 549       __ jmp(Done);
 550 
 551       __ bind(notShort);
 552       __ cmpl(flags, btos);
 553       __ jcc(Assembler::notEqual, notByte);
 554       // btos
 555       __ load_signed_byte(rax, field);
 556       __ push(btos);
 557       __ jmp(Done);
 558 
 559       __ bind(notByte);
 560       __ cmpl(flags, ctos);
 561       __ jcc(Assembler::notEqual, notChar);
 562       // ctos
 563       __ load_unsigned_short(rax, field);
 564       __ push(ctos);
 565       __ jmp(Done);
 566 
 567       __ bind(notChar);
 568       __ cmpl(flags, ztos);
 569       __ jcc(Assembler::notEqual, notBool);
 570       // ztos
 571       __ load_signed_byte(rax, field);
 572       __ push(ztos);
 573       __ jmp(Done);
 574 
 575       __ bind(notBool);
 576       break;
 577     }
 578 
 579   case Bytecodes::_ldc2_w:
 580     {
 581       Label notLong, notDouble;
 582       __ cmpl(flags, ltos);
 583       __ jcc(Assembler::notEqual, notLong);
 584       // ltos
 585       // Loading high word first because movptr clobbers rax
 586       NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
 587       __ movptr(rax, field);
 588       __ push(ltos);
 589       __ jmp(Done);
 590 
 591       __ bind(notLong);
 592       __ cmpl(flags, dtos);
 593       __ jcc(Assembler::notEqual, notDouble);
 594       // dtos
 595       __ load_double(field);
 596       __ push(dtos);
 597       __ jmp(Done);
 598 
 599       __ bind(notDouble);
 600       break;
 601     }
 602 
 603   default:
 604     ShouldNotReachHere();
 605   }
 606 
 607   __ stop("bad ldc/condy");
 608 }
 609 
 610 void TemplateTable::locals_index(Register reg, int offset) {
 611   __ load_unsigned_byte(reg, at_bcp(offset));
 612   __ negptr(reg);
 613 }
 614 
 615 void TemplateTable::iload() {
 616   iload_internal();
 617 }
 618 
 619 void TemplateTable::nofast_iload() {
 620   iload_internal(may_not_rewrite);
 621 }
 622 
 623 void TemplateTable::iload_internal(RewriteControl rc) {
 624   transition(vtos, itos);
 625   if (RewriteFrequentPairs && rc == may_rewrite) {
 626     Label rewrite, done;
 627     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 628     LP64_ONLY(assert(rbx != bc, "register damaged"));
 629 
 630     // get next byte
 631     __ load_unsigned_byte(rbx,
 632                           at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 633     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 634     // last two iloads in a pair.  Comparing against fast_iload means that
 635     // the next bytecode is neither an iload or a caload, and therefore
 636     // an iload pair.
 637     __ cmpl(rbx, Bytecodes::_iload);
 638     __ jcc(Assembler::equal, done);
 639 
 640     __ cmpl(rbx, Bytecodes::_fast_iload);
 641     __ movl(bc, Bytecodes::_fast_iload2);
 642 
 643     __ jccb(Assembler::equal, rewrite);
 644 
 645     // if _caload, rewrite to fast_icaload
 646     __ cmpl(rbx, Bytecodes::_caload);
 647     __ movl(bc, Bytecodes::_fast_icaload);
 648     __ jccb(Assembler::equal, rewrite);
 649 
 650     // rewrite so iload doesn't check again.
 651     __ movl(bc, Bytecodes::_fast_iload);
 652 
 653     // rewrite
 654     // bc: fast bytecode
 655     __ bind(rewrite);
 656     patch_bytecode(Bytecodes::_iload, bc, rbx, false);
 657     __ bind(done);
 658   }
 659 
 660   // Get the local value into tos
 661   locals_index(rbx);
 662   __ movl(rax, iaddress(rbx));
 663 }
 664 
 665 void TemplateTable::fast_iload2() {
 666   transition(vtos, itos);
 667   locals_index(rbx);
 668   __ movl(rax, iaddress(rbx));
 669   __ push(itos);
 670   locals_index(rbx, 3);
 671   __ movl(rax, iaddress(rbx));
 672 }
 673 
 674 void TemplateTable::fast_iload() {
 675   transition(vtos, itos);
 676   locals_index(rbx);
 677   __ movl(rax, iaddress(rbx));
 678 }
 679 
 680 void TemplateTable::lload() {
 681   transition(vtos, ltos);
 682   locals_index(rbx);
 683   __ movptr(rax, laddress(rbx));
 684   NOT_LP64(__ movl(rdx, haddress(rbx)));
 685 }
 686 
 687 void TemplateTable::fload() {
 688   transition(vtos, ftos);
 689   locals_index(rbx);
 690   __ load_float(faddress(rbx));
 691 }
 692 
 693 void TemplateTable::dload() {
 694   transition(vtos, dtos);
 695   locals_index(rbx);
 696   __ load_double(daddress(rbx));
 697 }
 698 
 699 void TemplateTable::aload() {
 700   transition(vtos, atos);
 701   locals_index(rbx);
 702   __ movptr(rax, aaddress(rbx));
 703 }
 704 
 705 void TemplateTable::locals_index_wide(Register reg) {
 706   __ load_unsigned_short(reg, at_bcp(2));
 707   __ bswapl(reg);
 708   __ shrl(reg, 16);
 709   __ negptr(reg);
 710 }
 711 
 712 void TemplateTable::wide_iload() {
 713   transition(vtos, itos);
 714   locals_index_wide(rbx);
 715   __ movl(rax, iaddress(rbx));
 716 }
 717 
 718 void TemplateTable::wide_lload() {
 719   transition(vtos, ltos);
 720   locals_index_wide(rbx);
 721   __ movptr(rax, laddress(rbx));
 722   NOT_LP64(__ movl(rdx, haddress(rbx)));
 723 }
 724 
 725 void TemplateTable::wide_fload() {
 726   transition(vtos, ftos);
 727   locals_index_wide(rbx);
 728   __ load_float(faddress(rbx));
 729 }
 730 
 731 void TemplateTable::wide_dload() {
 732   transition(vtos, dtos);
 733   locals_index_wide(rbx);
 734   __ load_double(daddress(rbx));
 735 }
 736 
 737 void TemplateTable::wide_aload() {
 738   transition(vtos, atos);
 739   locals_index_wide(rbx);
 740   __ movptr(rax, aaddress(rbx));
 741 }
 742 
 743 void TemplateTable::index_check(Register array, Register index) {
 744   // Pop ptr into array
 745   __ pop_ptr(array);
 746   index_check_without_pop(array, index);
 747 }
 748 
 749 void TemplateTable::index_check_without_pop(Register array, Register index) {
 750   // destroys rbx
 751   // check array
 752   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 753   // sign extend index for use by indexed load
 754   __ movl2ptr(index, index);
 755   // check index
 756   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 757   if (index != rbx) {
 758     // ??? convention: move aberrant index into rbx for exception message
 759     assert(rbx != array, "different registers");
 760     __ movl(rbx, index);
 761   }
 762   Label skip;
 763   __ jccb(Assembler::below, skip);
 764   // Pass array to create more detailed exceptions.
 765   __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
 766   __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 767   __ bind(skip);
 768 }
 769 
 770 void TemplateTable::iaload() {
 771   transition(itos, itos);
 772   // rax: index
 773   // rdx: array
 774   index_check(rdx, rax); // kills rbx
 775   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
 776                     Address(rdx, rax, Address::times_4,
 777                             arrayOopDesc::base_offset_in_bytes(T_INT)),
 778                     noreg, noreg);
 779 }
 780 
 781 void TemplateTable::laload() {
 782   transition(itos, ltos);
 783   // rax: index
 784   // rdx: array
 785   index_check(rdx, rax); // kills rbx
 786   NOT_LP64(__ mov(rbx, rax));
 787   // rbx,: index
 788   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
 789                     Address(rdx, rbx, Address::times_8,
 790                             arrayOopDesc::base_offset_in_bytes(T_LONG)),
 791                     noreg, noreg);
 792 }
 793 
 794 
 795 
 796 void TemplateTable::faload() {
 797   transition(itos, ftos);
 798   // rax: index
 799   // rdx: array
 800   index_check(rdx, rax); // kills rbx
 801   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
 802                     Address(rdx, rax,
 803                             Address::times_4,
 804                             arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
 805                     noreg, noreg);
 806 }
 807 
 808 void TemplateTable::daload() {
 809   transition(itos, dtos);
 810   // rax: index
 811   // rdx: array
 812   index_check(rdx, rax); // kills rbx
 813   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
 814                     Address(rdx, rax,
 815                             Address::times_8,
 816                             arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
 817                     noreg, noreg);
 818 }
 819 
 820 void TemplateTable::aaload() {
 821   transition(itos, atos);
 822   // rax: index
 823   // rdx: array
 824   index_check(rdx, rax); // kills rbx
 825   do_oop_load(_masm,
 826               Address(rdx, rax,
 827                       UseCompressedOops ? Address::times_4 : Address::times_ptr,
 828                       arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
 829               rax,
 830               IS_ARRAY);
 831 }
 832 
 833 void TemplateTable::baload() {
 834   transition(itos, itos);
 835   // rax: index
 836   // rdx: array
 837   index_check(rdx, rax); // kills rbx
 838   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
 839                     Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
 840                     noreg, noreg);
 841 }
 842 
 843 void TemplateTable::caload() {
 844   transition(itos, itos);
 845   // rax: index
 846   // rdx: array
 847   index_check(rdx, rax); // kills rbx
 848   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 849                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 850                     noreg, noreg);
 851 }
 852 
 853 // iload followed by caload frequent pair
 854 void TemplateTable::fast_icaload() {
 855   transition(vtos, itos);
 856   // load index out of locals
 857   locals_index(rbx);
 858   __ movl(rax, iaddress(rbx));
 859 
 860   // rax: index
 861   // rdx: array
 862   index_check(rdx, rax); // kills rbx
 863   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 864                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 865                     noreg, noreg);
 866 }
 867 
 868 
 869 void TemplateTable::saload() {
 870   transition(itos, itos);
 871   // rax: index
 872   // rdx: array
 873   index_check(rdx, rax); // kills rbx
 874   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
 875                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
 876                     noreg, noreg);
 877 }
 878 
 879 void TemplateTable::iload(int n) {
 880   transition(vtos, itos);
 881   __ movl(rax, iaddress(n));
 882 }
 883 
 884 void TemplateTable::lload(int n) {
 885   transition(vtos, ltos);
 886   __ movptr(rax, laddress(n));
 887   NOT_LP64(__ movptr(rdx, haddress(n)));
 888 }
 889 
 890 void TemplateTable::fload(int n) {
 891   transition(vtos, ftos);
 892   __ load_float(faddress(n));
 893 }
 894 
 895 void TemplateTable::dload(int n) {
 896   transition(vtos, dtos);
 897   __ load_double(daddress(n));
 898 }
 899 
 900 void TemplateTable::aload(int n) {
 901   transition(vtos, atos);
 902   __ movptr(rax, aaddress(n));
 903 }
 904 
 905 void TemplateTable::aload_0() {
 906   aload_0_internal();
 907 }
 908 
 909 void TemplateTable::nofast_aload_0() {
 910   aload_0_internal(may_not_rewrite);
 911 }
 912 
 913 void TemplateTable::aload_0_internal(RewriteControl rc) {
 914   transition(vtos, atos);
 915   // According to bytecode histograms, the pairs:
 916   //
 917   // _aload_0, _fast_igetfield
 918   // _aload_0, _fast_agetfield
 919   // _aload_0, _fast_fgetfield
 920   //
 921   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 922   // _aload_0 bytecode checks if the next bytecode is either
 923   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 924   // rewrites the current bytecode into a pair bytecode; otherwise it
 925   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 926   // the pair check anymore.
 927   //
 928   // Note: If the next bytecode is _getfield, the rewrite must be
 929   //       delayed, otherwise we may miss an opportunity for a pair.
 930   //
 931   // Also rewrite frequent pairs
 932   //   aload_0, aload_1
 933   //   aload_0, iload_1
 934   // These bytecodes with a small amount of code are most profitable
 935   // to rewrite
 936   if (RewriteFrequentPairs && rc == may_rewrite) {
 937     Label rewrite, done;
 938 
 939     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 940     LP64_ONLY(assert(rbx != bc, "register damaged"));
 941 
 942     // get next byte
 943     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 944 
 945     // if _getfield then wait with rewrite
 946     __ cmpl(rbx, Bytecodes::_getfield);
 947     __ jcc(Assembler::equal, done);
 948 
 949     // if _igetfield then rewrite to _fast_iaccess_0
 950     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 951     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 952     __ movl(bc, Bytecodes::_fast_iaccess_0);
 953     __ jccb(Assembler::equal, rewrite);
 954 
 955     // if _agetfield then rewrite to _fast_aaccess_0
 956     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 957     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 958     __ movl(bc, Bytecodes::_fast_aaccess_0);
 959     __ jccb(Assembler::equal, rewrite);
 960 
 961     // if _fgetfield then rewrite to _fast_faccess_0
 962     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 963     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 964     __ movl(bc, Bytecodes::_fast_faccess_0);
 965     __ jccb(Assembler::equal, rewrite);
 966 
 967     // else rewrite to _fast_aload0
 968     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 969     __ movl(bc, Bytecodes::_fast_aload_0);
 970 
 971     // rewrite
 972     // bc: fast bytecode
 973     __ bind(rewrite);
 974     patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
 975 
 976     __ bind(done);
 977   }
 978 
 979   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 980   aload(0);
 981 }
 982 
 983 void TemplateTable::istore() {
 984   transition(itos, vtos);
 985   locals_index(rbx);
 986   __ movl(iaddress(rbx), rax);
 987 }
 988 
 989 
 990 void TemplateTable::lstore() {
 991   transition(ltos, vtos);
 992   locals_index(rbx);
 993   __ movptr(laddress(rbx), rax);
 994   NOT_LP64(__ movptr(haddress(rbx), rdx));
 995 }
 996 
 997 void TemplateTable::fstore() {
 998   transition(ftos, vtos);
 999   locals_index(rbx);
1000   __ store_float(faddress(rbx));
1001 }
1002 
1003 void TemplateTable::dstore() {
1004   transition(dtos, vtos);
1005   locals_index(rbx);
1006   __ store_double(daddress(rbx));
1007 }
1008 
1009 void TemplateTable::astore() {
1010   transition(vtos, vtos);
1011   __ pop_ptr(rax);
1012   locals_index(rbx);
1013   __ movptr(aaddress(rbx), rax);
1014 }
1015 
1016 void TemplateTable::wide_istore() {
1017   transition(vtos, vtos);
1018   __ pop_i();
1019   locals_index_wide(rbx);
1020   __ movl(iaddress(rbx), rax);
1021 }
1022 
1023 void TemplateTable::wide_lstore() {
1024   transition(vtos, vtos);
1025   NOT_LP64(__ pop_l(rax, rdx));
1026   LP64_ONLY(__ pop_l());
1027   locals_index_wide(rbx);
1028   __ movptr(laddress(rbx), rax);
1029   NOT_LP64(__ movl(haddress(rbx), rdx));
1030 }
1031 
1032 void TemplateTable::wide_fstore() {
1033 #ifdef _LP64
1034   transition(vtos, vtos);
1035   __ pop_f(xmm0);
1036   locals_index_wide(rbx);
1037   __ movflt(faddress(rbx), xmm0);
1038 #else
1039   wide_istore();
1040 #endif
1041 }
1042 
1043 void TemplateTable::wide_dstore() {
1044 #ifdef _LP64
1045   transition(vtos, vtos);
1046   __ pop_d(xmm0);
1047   locals_index_wide(rbx);
1048   __ movdbl(daddress(rbx), xmm0);
1049 #else
1050   wide_lstore();
1051 #endif
1052 }
1053 
1054 void TemplateTable::wide_astore() {
1055   transition(vtos, vtos);
1056   __ pop_ptr(rax);
1057   locals_index_wide(rbx);
1058   __ movptr(aaddress(rbx), rax);
1059 }
1060 
1061 void TemplateTable::iastore() {
1062   transition(itos, vtos);
1063   __ pop_i(rbx);
1064   // rax: value
1065   // rbx: index
1066   // rdx: array
1067   index_check(rdx, rbx); // prefer index in rbx
1068   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1069                      Address(rdx, rbx, Address::times_4,
1070                              arrayOopDesc::base_offset_in_bytes(T_INT)),
1071                      rax, noreg, noreg);
1072 }
1073 
1074 void TemplateTable::lastore() {
1075   transition(ltos, vtos);
1076   __ pop_i(rbx);
1077   // rax,: low(value)
1078   // rcx: array
1079   // rdx: high(value)
1080   index_check(rcx, rbx);  // prefer index in rbx,
1081   // rbx,: index
1082   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1083                      Address(rcx, rbx, Address::times_8,
1084                              arrayOopDesc::base_offset_in_bytes(T_LONG)),
1085                      noreg /* ltos */, noreg, noreg);
1086 }
1087 
1088 
1089 void TemplateTable::fastore() {
1090   transition(ftos, vtos);
1091   __ pop_i(rbx);
1092   // value is in UseSSE >= 1 ? xmm0 : ST(0)
1093   // rbx:  index
1094   // rdx:  array
1095   index_check(rdx, rbx); // prefer index in rbx
1096   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1097                      Address(rdx, rbx, Address::times_4,
1098                              arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1099                      noreg /* ftos */, noreg, noreg);
1100 }
1101 
1102 void TemplateTable::dastore() {
1103   transition(dtos, vtos);
1104   __ pop_i(rbx);
1105   // value is in UseSSE >= 2 ? xmm0 : ST(0)
1106   // rbx:  index
1107   // rdx:  array
1108   index_check(rdx, rbx); // prefer index in rbx
1109   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1110                      Address(rdx, rbx, Address::times_8,
1111                              arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1112                      noreg /* dtos */, noreg, noreg);
1113 }
1114 
1115 void TemplateTable::aastore() {
1116   Label is_null, ok_is_subtype, done;
1117   transition(vtos, vtos);
1118   // stack: ..., array, index, value
1119   __ movptr(rax, at_tos());    // value
1120   __ movl(rcx, at_tos_p1()); // index
1121   __ movptr(rdx, at_tos_p2()); // array
1122 
1123   Address element_address(rdx, rcx,
1124                           UseCompressedOops? Address::times_4 : Address::times_ptr,
1125                           arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1126 
1127   index_check_without_pop(rdx, rcx);     // kills rbx
1128   __ testptr(rax, rax);
1129   __ jcc(Assembler::zero, is_null);
1130 
1131   // Move subklass into rbx
1132   __ load_klass(rbx, rax);
1133   // Move superklass into rax
1134   __ load_klass(rax, rdx);
1135   __ movptr(rax, Address(rax,
1136                          ObjArrayKlass::element_klass_offset()));
1137 
1138   // Generate subtype check.  Blows rcx, rdi
1139   // Superklass in rax.  Subklass in rbx.
1140   __ gen_subtype_check(rbx, ok_is_subtype);
1141 
1142   // Come here on failure
1143   // object is at TOS
1144   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1145 
1146   // Come here on success
1147   __ bind(ok_is_subtype);
1148 
1149   // Get the value we will store
1150   __ movptr(rax, at_tos());
1151   __ movl(rcx, at_tos_p1()); // index
1152   // Now store using the appropriate barrier
1153   do_oop_store(_masm, element_address, rax, IS_ARRAY);
1154   __ jmp(done);
1155 
1156   // Have a NULL in rax, rdx=array, ecx=index.  Store NULL at ary[idx]
1157   __ bind(is_null);
1158   __ profile_null_seen(rbx);
1159 
1160   // Store a NULL
1161   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1162 
1163   // Pop stack arguments
1164   __ bind(done);
1165   __ addptr(rsp, 3 * Interpreter::stackElementSize);
1166 }
1167 
1168 void TemplateTable::bastore() {
1169   transition(itos, vtos);
1170   __ pop_i(rbx);
1171   // rax: value
1172   // rbx: index
1173   // rdx: array
1174   index_check(rdx, rbx); // prefer index in rbx
1175   // Need to check whether array is boolean or byte
1176   // since both types share the bastore bytecode.
1177   __ load_klass(rcx, rdx);
1178   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1179   int diffbit = Klass::layout_helper_boolean_diffbit();
1180   __ testl(rcx, diffbit);
1181   Label L_skip;
1182   __ jccb(Assembler::zero, L_skip);
1183   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1184   __ bind(L_skip);
1185   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1186                      Address(rdx, rbx,Address::times_1,
1187                              arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1188                      rax, noreg, noreg);
1189 }
1190 
1191 void TemplateTable::castore() {
1192   transition(itos, vtos);
1193   __ pop_i(rbx);
1194   // rax: value
1195   // rbx: index
1196   // rdx: array
1197   index_check(rdx, rbx);  // prefer index in rbx
1198   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1199                      Address(rdx, rbx, Address::times_2,
1200                              arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1201                      rax, noreg, noreg);
1202 }
1203 
1204 
1205 void TemplateTable::sastore() {
1206   castore();
1207 }
1208 
1209 void TemplateTable::istore(int n) {
1210   transition(itos, vtos);
1211   __ movl(iaddress(n), rax);
1212 }
1213 
1214 void TemplateTable::lstore(int n) {
1215   transition(ltos, vtos);
1216   __ movptr(laddress(n), rax);
1217   NOT_LP64(__ movptr(haddress(n), rdx));
1218 }
1219 
1220 void TemplateTable::fstore(int n) {
1221   transition(ftos, vtos);
1222   __ store_float(faddress(n));
1223 }
1224 
1225 void TemplateTable::dstore(int n) {
1226   transition(dtos, vtos);
1227   __ store_double(daddress(n));
1228 }
1229 
1230 
1231 void TemplateTable::astore(int n) {
1232   transition(vtos, vtos);
1233   __ pop_ptr(rax);
1234   __ movptr(aaddress(n), rax);
1235 }
1236 
1237 void TemplateTable::pop() {
1238   transition(vtos, vtos);
1239   __ addptr(rsp, Interpreter::stackElementSize);
1240 }
1241 
1242 void TemplateTable::pop2() {
1243   transition(vtos, vtos);
1244   __ addptr(rsp, 2 * Interpreter::stackElementSize);
1245 }
1246 
1247 
1248 void TemplateTable::dup() {
1249   transition(vtos, vtos);
1250   __ load_ptr(0, rax);
1251   __ push_ptr(rax);
1252   // stack: ..., a, a
1253 }
1254 
1255 void TemplateTable::dup_x1() {
1256   transition(vtos, vtos);
1257   // stack: ..., a, b
1258   __ load_ptr( 0, rax);  // load b
1259   __ load_ptr( 1, rcx);  // load a
1260   __ store_ptr(1, rax);  // store b
1261   __ store_ptr(0, rcx);  // store a
1262   __ push_ptr(rax);      // push b
1263   // stack: ..., b, a, b
1264 }
1265 
1266 void TemplateTable::dup_x2() {
1267   transition(vtos, vtos);
1268   // stack: ..., a, b, c
1269   __ load_ptr( 0, rax);  // load c
1270   __ load_ptr( 2, rcx);  // load a
1271   __ store_ptr(2, rax);  // store c in a
1272   __ push_ptr(rax);      // push c
1273   // stack: ..., c, b, c, c
1274   __ load_ptr( 2, rax);  // load b
1275   __ store_ptr(2, rcx);  // store a in b
1276   // stack: ..., c, a, c, c
1277   __ store_ptr(1, rax);  // store b in c
1278   // stack: ..., c, a, b, c
1279 }
1280 
1281 void TemplateTable::dup2() {
1282   transition(vtos, vtos);
1283   // stack: ..., a, b
1284   __ load_ptr(1, rax);  // load a
1285   __ push_ptr(rax);     // push a
1286   __ load_ptr(1, rax);  // load b
1287   __ push_ptr(rax);     // push b
1288   // stack: ..., a, b, a, b
1289 }
1290 
1291 
1292 void TemplateTable::dup2_x1() {
1293   transition(vtos, vtos);
1294   // stack: ..., a, b, c
1295   __ load_ptr( 0, rcx);  // load c
1296   __ load_ptr( 1, rax);  // load b
1297   __ push_ptr(rax);      // push b
1298   __ push_ptr(rcx);      // push c
1299   // stack: ..., a, b, c, b, c
1300   __ store_ptr(3, rcx);  // store c in b
1301   // stack: ..., a, c, c, b, c
1302   __ load_ptr( 4, rcx);  // load a
1303   __ store_ptr(2, rcx);  // store a in 2nd c
1304   // stack: ..., a, c, a, b, c
1305   __ store_ptr(4, rax);  // store b in a
1306   // stack: ..., b, c, a, b, c
1307 }
1308 
1309 void TemplateTable::dup2_x2() {
1310   transition(vtos, vtos);
1311   // stack: ..., a, b, c, d
1312   __ load_ptr( 0, rcx);  // load d
1313   __ load_ptr( 1, rax);  // load c
1314   __ push_ptr(rax);      // push c
1315   __ push_ptr(rcx);      // push d
1316   // stack: ..., a, b, c, d, c, d
1317   __ load_ptr( 4, rax);  // load b
1318   __ store_ptr(2, rax);  // store b in d
1319   __ store_ptr(4, rcx);  // store d in b
1320   // stack: ..., a, d, c, b, c, d
1321   __ load_ptr( 5, rcx);  // load a
1322   __ load_ptr( 3, rax);  // load c
1323   __ store_ptr(3, rcx);  // store a in c
1324   __ store_ptr(5, rax);  // store c in a
1325   // stack: ..., c, d, a, b, c, d
1326 }
1327 
1328 void TemplateTable::swap() {
1329   transition(vtos, vtos);
1330   // stack: ..., a, b
1331   __ load_ptr( 1, rcx);  // load a
1332   __ load_ptr( 0, rax);  // load b
1333   __ store_ptr(0, rcx);  // store a in b
1334   __ store_ptr(1, rax);  // store b in a
1335   // stack: ..., b, a
1336 }
1337 
1338 void TemplateTable::iop2(Operation op) {
1339   transition(itos, itos);
1340   switch (op) {
1341   case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1342   case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1343   case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1344   case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1345   case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1346   case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1347   case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break;
1348   case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break;
1349   case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break;
1350   default   : ShouldNotReachHere();
1351   }
1352 }
1353 
1354 void TemplateTable::lop2(Operation op) {
1355   transition(ltos, ltos);
1356 #ifdef _LP64
1357   switch (op) {
1358   case add  :                    __ pop_l(rdx); __ addptr(rax, rdx); break;
1359   case sub  : __ mov(rdx, rax);  __ pop_l(rax); __ subptr(rax, rdx); break;
1360   case _and :                    __ pop_l(rdx); __ andptr(rax, rdx); break;
1361   case _or  :                    __ pop_l(rdx); __ orptr (rax, rdx); break;
1362   case _xor :                    __ pop_l(rdx); __ xorptr(rax, rdx); break;
1363   default   : ShouldNotReachHere();
1364   }
1365 #else
1366   __ pop_l(rbx, rcx);
1367   switch (op) {
1368     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1369     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1370                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1371     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1372     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1373     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1374     default   : ShouldNotReachHere();
1375   }
1376 #endif
1377 }
1378 
1379 void TemplateTable::idiv() {
1380   transition(itos, itos);
1381   __ movl(rcx, rax);
1382   __ pop_i(rax);
1383   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1384   //       they are not equal, one could do a normal division (no correction
1385   //       needed), which may speed up this implementation for the common case.
1386   //       (see also JVM spec., p.243 & p.271)
1387   __ corrected_idivl(rcx);
1388 }
1389 
1390 void TemplateTable::irem() {
1391   transition(itos, itos);
1392   __ movl(rcx, rax);
1393   __ pop_i(rax);
1394   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1395   //       they are not equal, one could do a normal division (no correction
1396   //       needed), which may speed up this implementation for the common case.
1397   //       (see also JVM spec., p.243 & p.271)
1398   __ corrected_idivl(rcx);
1399   __ movl(rax, rdx);
1400 }
1401 
1402 void TemplateTable::lmul() {
1403   transition(ltos, ltos);
1404 #ifdef _LP64
1405   __ pop_l(rdx);
1406   __ imulq(rax, rdx);
1407 #else
1408   __ pop_l(rbx, rcx);
1409   __ push(rcx); __ push(rbx);
1410   __ push(rdx); __ push(rax);
1411   __ lmul(2 * wordSize, 0);
1412   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1413 #endif
1414 }
1415 
1416 void TemplateTable::ldiv() {
1417   transition(ltos, ltos);
1418 #ifdef _LP64
1419   __ mov(rcx, rax);
1420   __ pop_l(rax);
1421   // generate explicit div0 check
1422   __ testq(rcx, rcx);
1423   __ jump_cc(Assembler::zero,
1424              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1425   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1426   //       they are not equal, one could do a normal division (no correction
1427   //       needed), which may speed up this implementation for the common case.
1428   //       (see also JVM spec., p.243 & p.271)
1429   __ corrected_idivq(rcx); // kills rbx
1430 #else
1431   __ pop_l(rbx, rcx);
1432   __ push(rcx); __ push(rbx);
1433   __ push(rdx); __ push(rax);
1434   // check if y = 0
1435   __ orl(rax, rdx);
1436   __ jump_cc(Assembler::zero,
1437              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1438   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1439   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1440 #endif
1441 }
1442 
1443 void TemplateTable::lrem() {
1444   transition(ltos, ltos);
1445 #ifdef _LP64
1446   __ mov(rcx, rax);
1447   __ pop_l(rax);
1448   __ testq(rcx, rcx);
1449   __ jump_cc(Assembler::zero,
1450              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1451   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1452   //       they are not equal, one could do a normal division (no correction
1453   //       needed), which may speed up this implementation for the common case.
1454   //       (see also JVM spec., p.243 & p.271)
1455   __ corrected_idivq(rcx); // kills rbx
1456   __ mov(rax, rdx);
1457 #else
1458   __ pop_l(rbx, rcx);
1459   __ push(rcx); __ push(rbx);
1460   __ push(rdx); __ push(rax);
1461   // check if y = 0
1462   __ orl(rax, rdx);
1463   __ jump_cc(Assembler::zero,
1464              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1465   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1466   __ addptr(rsp, 4 * wordSize);
1467 #endif
1468 }
1469 
1470 void TemplateTable::lshl() {
1471   transition(itos, ltos);
1472   __ movl(rcx, rax);                             // get shift count
1473   #ifdef _LP64
1474   __ pop_l(rax);                                 // get shift value
1475   __ shlq(rax);
1476 #else
1477   __ pop_l(rax, rdx);                            // get shift value
1478   __ lshl(rdx, rax);
1479 #endif
1480 }
1481 
1482 void TemplateTable::lshr() {
1483 #ifdef _LP64
1484   transition(itos, ltos);
1485   __ movl(rcx, rax);                             // get shift count
1486   __ pop_l(rax);                                 // get shift value
1487   __ sarq(rax);
1488 #else
1489   transition(itos, ltos);
1490   __ mov(rcx, rax);                              // get shift count
1491   __ pop_l(rax, rdx);                            // get shift value
1492   __ lshr(rdx, rax, true);
1493 #endif
1494 }
1495 
1496 void TemplateTable::lushr() {
1497   transition(itos, ltos);
1498 #ifdef _LP64
1499   __ movl(rcx, rax);                             // get shift count
1500   __ pop_l(rax);                                 // get shift value
1501   __ shrq(rax);
1502 #else
1503   __ mov(rcx, rax);                              // get shift count
1504   __ pop_l(rax, rdx);                            // get shift value
1505   __ lshr(rdx, rax);
1506 #endif
1507 }
1508 
1509 void TemplateTable::fop2(Operation op) {
1510   transition(ftos, ftos);
1511 
1512   if (UseSSE >= 1) {
1513     switch (op) {
1514     case add:
1515       __ addss(xmm0, at_rsp());
1516       __ addptr(rsp, Interpreter::stackElementSize);
1517       break;
1518     case sub:
1519       __ movflt(xmm1, xmm0);
1520       __ pop_f(xmm0);
1521       __ subss(xmm0, xmm1);
1522       break;
1523     case mul:
1524       __ mulss(xmm0, at_rsp());
1525       __ addptr(rsp, Interpreter::stackElementSize);
1526       break;
1527     case div:
1528       __ movflt(xmm1, xmm0);
1529       __ pop_f(xmm0);
1530       __ divss(xmm0, xmm1);
1531       break;
1532     case rem:
1533       // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1534       // modulo operation. The frem method calls the function
1535       // double fmod(double x, double y) in math.h. The documentation of fmod states:
1536       // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1537       // (signalling or quiet) is returned.
1538       //
1539       // On x86_32 platforms the FPU is used to perform the modulo operation. The
1540       // reason is that on 32-bit Windows the sign of modulo operations diverges from
1541       // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1542       // The fprem instruction used on x86_32 is functionally equivalent to
1543       // SharedRuntime::frem in that it returns a NaN.
1544 #ifdef _LP64
1545       __ movflt(xmm1, xmm0);
1546       __ pop_f(xmm0);
1547       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1548 #else
1549       __ push_f(xmm0);
1550       __ pop_f();
1551       __ fld_s(at_rsp());
1552       __ fremr(rax);
1553       __ f2ieee();
1554       __ pop(rax);  // pop second operand off the stack
1555       __ push_f();
1556       __ pop_f(xmm0);
1557 #endif
1558       break;
1559     default:
1560       ShouldNotReachHere();
1561       break;
1562     }
1563   } else {
1564 #ifdef _LP64
1565     ShouldNotReachHere();
1566 #else
1567     switch (op) {
1568     case add: __ fadd_s (at_rsp());                break;
1569     case sub: __ fsubr_s(at_rsp());                break;
1570     case mul: __ fmul_s (at_rsp());                break;
1571     case div: __ fdivr_s(at_rsp());                break;
1572     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1573     default : ShouldNotReachHere();
1574     }
1575     __ f2ieee();
1576     __ pop(rax);  // pop second operand off the stack
1577 #endif // _LP64
1578   }
1579 }
1580 
1581 void TemplateTable::dop2(Operation op) {
1582   transition(dtos, dtos);
1583   if (UseSSE >= 2) {
1584     switch (op) {
1585     case add:
1586       __ addsd(xmm0, at_rsp());
1587       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1588       break;
1589     case sub:
1590       __ movdbl(xmm1, xmm0);
1591       __ pop_d(xmm0);
1592       __ subsd(xmm0, xmm1);
1593       break;
1594     case mul:
1595       __ mulsd(xmm0, at_rsp());
1596       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1597       break;
1598     case div:
1599       __ movdbl(xmm1, xmm0);
1600       __ pop_d(xmm0);
1601       __ divsd(xmm0, xmm1);
1602       break;
1603     case rem:
1604       // Similar to fop2(), the modulo operation is performed using the
1605       // SharedRuntime::drem method (on x86_64 platforms) or using the
1606       // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1607 #ifdef _LP64
1608       __ movdbl(xmm1, xmm0);
1609       __ pop_d(xmm0);
1610       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1611 #else
1612       __ push_d(xmm0);
1613       __ pop_d();
1614       __ fld_d(at_rsp());
1615       __ fremr(rax);
1616       __ d2ieee();
1617       __ pop(rax);
1618       __ pop(rdx);
1619       __ push_d();
1620       __ pop_d(xmm0);
1621 #endif
1622       break;
1623     default:
1624       ShouldNotReachHere();
1625       break;
1626     }
1627   } else {
1628 #ifdef _LP64
1629     ShouldNotReachHere();
1630 #else
1631     switch (op) {
1632     case add: __ fadd_d (at_rsp());                break;
1633     case sub: __ fsubr_d(at_rsp());                break;
1634     case mul: {
1635       Label L_strict;
1636       Label L_join;
1637       const Address access_flags      (rcx, Method::access_flags_offset());
1638       __ get_method(rcx);
1639       __ movl(rcx, access_flags);
1640       __ testl(rcx, JVM_ACC_STRICT);
1641       __ jccb(Assembler::notZero, L_strict);
1642       __ fmul_d (at_rsp());
1643       __ jmpb(L_join);
1644       __ bind(L_strict);
1645       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1646       __ fmulp();
1647       __ fmul_d (at_rsp());
1648       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1649       __ fmulp();
1650       __ bind(L_join);
1651       break;
1652     }
1653     case div: {
1654       Label L_strict;
1655       Label L_join;
1656       const Address access_flags      (rcx, Method::access_flags_offset());
1657       __ get_method(rcx);
1658       __ movl(rcx, access_flags);
1659       __ testl(rcx, JVM_ACC_STRICT);
1660       __ jccb(Assembler::notZero, L_strict);
1661       __ fdivr_d(at_rsp());
1662       __ jmp(L_join);
1663       __ bind(L_strict);
1664       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1665       __ fmul_d (at_rsp());
1666       __ fdivrp();
1667       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1668       __ fmulp();
1669       __ bind(L_join);
1670       break;
1671     }
1672     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1673     default : ShouldNotReachHere();
1674     }
1675     __ d2ieee();
1676     // Pop double precision number from rsp.
1677     __ pop(rax);
1678     __ pop(rdx);
1679 #endif
1680   }
1681 }
1682 
1683 void TemplateTable::ineg() {
1684   transition(itos, itos);
1685   __ negl(rax);
1686 }
1687 
1688 void TemplateTable::lneg() {
1689   transition(ltos, ltos);
1690   LP64_ONLY(__ negq(rax));
1691   NOT_LP64(__ lneg(rdx, rax));
1692 }
1693 
1694 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1695 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1696   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1697   // of 128-bits operands for SSE instructions.
1698   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1699   // Store the value to a 128-bits operand.
1700   operand[0] = lo;
1701   operand[1] = hi;
1702   return operand;
1703 }
1704 
1705 // Buffer for 128-bits masks used by SSE instructions.
1706 static jlong float_signflip_pool[2*2];
1707 static jlong double_signflip_pool[2*2];
1708 
1709 void TemplateTable::fneg() {
1710   transition(ftos, ftos);
1711   if (UseSSE >= 1) {
1712     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1713     __ xorps(xmm0, ExternalAddress((address) float_signflip));
1714   } else {
1715     LP64_ONLY(ShouldNotReachHere());
1716     NOT_LP64(__ fchs());
1717   }
1718 }
1719 
1720 void TemplateTable::dneg() {
1721   transition(dtos, dtos);
1722   if (UseSSE >= 2) {
1723     static jlong *double_signflip =
1724       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1725     __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1726   } else {
1727 #ifdef _LP64
1728     ShouldNotReachHere();
1729 #else
1730     __ fchs();
1731 #endif
1732   }
1733 }
1734 
1735 void TemplateTable::iinc() {
1736   transition(vtos, vtos);
1737   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1738   locals_index(rbx);
1739   __ addl(iaddress(rbx), rdx);
1740 }
1741 
1742 void TemplateTable::wide_iinc() {
1743   transition(vtos, vtos);
1744   __ movl(rdx, at_bcp(4)); // get constant
1745   locals_index_wide(rbx);
1746   __ bswapl(rdx); // swap bytes & sign-extend constant
1747   __ sarl(rdx, 16);
1748   __ addl(iaddress(rbx), rdx);
1749   // Note: should probably use only one movl to get both
1750   //       the index and the constant -> fix this
1751 }
1752 
1753 void TemplateTable::convert() {
1754 #ifdef _LP64
1755   // Checking
1756 #ifdef ASSERT
1757   {
1758     TosState tos_in  = ilgl;
1759     TosState tos_out = ilgl;
1760     switch (bytecode()) {
1761     case Bytecodes::_i2l: // fall through
1762     case Bytecodes::_i2f: // fall through
1763     case Bytecodes::_i2d: // fall through
1764     case Bytecodes::_i2b: // fall through
1765     case Bytecodes::_i2c: // fall through
1766     case Bytecodes::_i2s: tos_in = itos; break;
1767     case Bytecodes::_l2i: // fall through
1768     case Bytecodes::_l2f: // fall through
1769     case Bytecodes::_l2d: tos_in = ltos; break;
1770     case Bytecodes::_f2i: // fall through
1771     case Bytecodes::_f2l: // fall through
1772     case Bytecodes::_f2d: tos_in = ftos; break;
1773     case Bytecodes::_d2i: // fall through
1774     case Bytecodes::_d2l: // fall through
1775     case Bytecodes::_d2f: tos_in = dtos; break;
1776     default             : ShouldNotReachHere();
1777     }
1778     switch (bytecode()) {
1779     case Bytecodes::_l2i: // fall through
1780     case Bytecodes::_f2i: // fall through
1781     case Bytecodes::_d2i: // fall through
1782     case Bytecodes::_i2b: // fall through
1783     case Bytecodes::_i2c: // fall through
1784     case Bytecodes::_i2s: tos_out = itos; break;
1785     case Bytecodes::_i2l: // fall through
1786     case Bytecodes::_f2l: // fall through
1787     case Bytecodes::_d2l: tos_out = ltos; break;
1788     case Bytecodes::_i2f: // fall through
1789     case Bytecodes::_l2f: // fall through
1790     case Bytecodes::_d2f: tos_out = ftos; break;
1791     case Bytecodes::_i2d: // fall through
1792     case Bytecodes::_l2d: // fall through
1793     case Bytecodes::_f2d: tos_out = dtos; break;
1794     default             : ShouldNotReachHere();
1795     }
1796     transition(tos_in, tos_out);
1797   }
1798 #endif // ASSERT
1799 
1800   static const int64_t is_nan = 0x8000000000000000L;
1801 
1802   // Conversion
1803   switch (bytecode()) {
1804   case Bytecodes::_i2l:
1805     __ movslq(rax, rax);
1806     break;
1807   case Bytecodes::_i2f:
1808     __ cvtsi2ssl(xmm0, rax);
1809     break;
1810   case Bytecodes::_i2d:
1811     __ cvtsi2sdl(xmm0, rax);
1812     break;
1813   case Bytecodes::_i2b:
1814     __ movsbl(rax, rax);
1815     break;
1816   case Bytecodes::_i2c:
1817     __ movzwl(rax, rax);
1818     break;
1819   case Bytecodes::_i2s:
1820     __ movswl(rax, rax);
1821     break;
1822   case Bytecodes::_l2i:
1823     __ movl(rax, rax);
1824     break;
1825   case Bytecodes::_l2f:
1826     __ cvtsi2ssq(xmm0, rax);
1827     break;
1828   case Bytecodes::_l2d:
1829     __ cvtsi2sdq(xmm0, rax);
1830     break;
1831   case Bytecodes::_f2i:
1832   {
1833     Label L;
1834     __ cvttss2sil(rax, xmm0);
1835     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1836     __ jcc(Assembler::notEqual, L);
1837     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1838     __ bind(L);
1839   }
1840     break;
1841   case Bytecodes::_f2l:
1842   {
1843     Label L;
1844     __ cvttss2siq(rax, xmm0);
1845     // NaN or overflow/underflow?
1846     __ cmp64(rax, ExternalAddress((address) &is_nan));
1847     __ jcc(Assembler::notEqual, L);
1848     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1849     __ bind(L);
1850   }
1851     break;
1852   case Bytecodes::_f2d:
1853     __ cvtss2sd(xmm0, xmm0);
1854     break;
1855   case Bytecodes::_d2i:
1856   {
1857     Label L;
1858     __ cvttsd2sil(rax, xmm0);
1859     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1860     __ jcc(Assembler::notEqual, L);
1861     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1862     __ bind(L);
1863   }
1864     break;
1865   case Bytecodes::_d2l:
1866   {
1867     Label L;
1868     __ cvttsd2siq(rax, xmm0);
1869     // NaN or overflow/underflow?
1870     __ cmp64(rax, ExternalAddress((address) &is_nan));
1871     __ jcc(Assembler::notEqual, L);
1872     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1873     __ bind(L);
1874   }
1875     break;
1876   case Bytecodes::_d2f:
1877     __ cvtsd2ss(xmm0, xmm0);
1878     break;
1879   default:
1880     ShouldNotReachHere();
1881   }
1882 #else
1883   // Checking
1884 #ifdef ASSERT
1885   { TosState tos_in  = ilgl;
1886     TosState tos_out = ilgl;
1887     switch (bytecode()) {
1888       case Bytecodes::_i2l: // fall through
1889       case Bytecodes::_i2f: // fall through
1890       case Bytecodes::_i2d: // fall through
1891       case Bytecodes::_i2b: // fall through
1892       case Bytecodes::_i2c: // fall through
1893       case Bytecodes::_i2s: tos_in = itos; break;
1894       case Bytecodes::_l2i: // fall through
1895       case Bytecodes::_l2f: // fall through
1896       case Bytecodes::_l2d: tos_in = ltos; break;
1897       case Bytecodes::_f2i: // fall through
1898       case Bytecodes::_f2l: // fall through
1899       case Bytecodes::_f2d: tos_in = ftos; break;
1900       case Bytecodes::_d2i: // fall through
1901       case Bytecodes::_d2l: // fall through
1902       case Bytecodes::_d2f: tos_in = dtos; break;
1903       default             : ShouldNotReachHere();
1904     }
1905     switch (bytecode()) {
1906       case Bytecodes::_l2i: // fall through
1907       case Bytecodes::_f2i: // fall through
1908       case Bytecodes::_d2i: // fall through
1909       case Bytecodes::_i2b: // fall through
1910       case Bytecodes::_i2c: // fall through
1911       case Bytecodes::_i2s: tos_out = itos; break;
1912       case Bytecodes::_i2l: // fall through
1913       case Bytecodes::_f2l: // fall through
1914       case Bytecodes::_d2l: tos_out = ltos; break;
1915       case Bytecodes::_i2f: // fall through
1916       case Bytecodes::_l2f: // fall through
1917       case Bytecodes::_d2f: tos_out = ftos; break;
1918       case Bytecodes::_i2d: // fall through
1919       case Bytecodes::_l2d: // fall through
1920       case Bytecodes::_f2d: tos_out = dtos; break;
1921       default             : ShouldNotReachHere();
1922     }
1923     transition(tos_in, tos_out);
1924   }
1925 #endif // ASSERT
1926 
1927   // Conversion
1928   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1929   switch (bytecode()) {
1930     case Bytecodes::_i2l:
1931       __ extend_sign(rdx, rax);
1932       break;
1933     case Bytecodes::_i2f:
1934       if (UseSSE >= 1) {
1935         __ cvtsi2ssl(xmm0, rax);
1936       } else {
1937         __ push(rax);          // store int on tos
1938         __ fild_s(at_rsp());   // load int to ST0
1939         __ f2ieee();           // truncate to float size
1940         __ pop(rcx);           // adjust rsp
1941       }
1942       break;
1943     case Bytecodes::_i2d:
1944       if (UseSSE >= 2) {
1945         __ cvtsi2sdl(xmm0, rax);
1946       } else {
1947       __ push(rax);          // add one slot for d2ieee()
1948       __ push(rax);          // store int on tos
1949       __ fild_s(at_rsp());   // load int to ST0
1950       __ d2ieee();           // truncate to double size
1951       __ pop(rcx);           // adjust rsp
1952       __ pop(rcx);
1953       }
1954       break;
1955     case Bytecodes::_i2b:
1956       __ shll(rax, 24);      // truncate upper 24 bits
1957       __ sarl(rax, 24);      // and sign-extend byte
1958       LP64_ONLY(__ movsbl(rax, rax));
1959       break;
1960     case Bytecodes::_i2c:
1961       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1962       LP64_ONLY(__ movzwl(rax, rax));
1963       break;
1964     case Bytecodes::_i2s:
1965       __ shll(rax, 16);      // truncate upper 16 bits
1966       __ sarl(rax, 16);      // and sign-extend short
1967       LP64_ONLY(__ movswl(rax, rax));
1968       break;
1969     case Bytecodes::_l2i:
1970       /* nothing to do */
1971       break;
1972     case Bytecodes::_l2f:
1973       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1974       // 64-bit long values to floats. On 32-bit platforms it is not possible
1975       // to use that instruction with 64-bit operands, therefore the FPU is
1976       // used to perform the conversion.
1977       __ push(rdx);          // store long on tos
1978       __ push(rax);
1979       __ fild_d(at_rsp());   // load long to ST0
1980       __ f2ieee();           // truncate to float size
1981       __ pop(rcx);           // adjust rsp
1982       __ pop(rcx);
1983       if (UseSSE >= 1) {
1984         __ push_f();
1985         __ pop_f(xmm0);
1986       }
1987       break;
1988     case Bytecodes::_l2d:
1989       // On 32-bit platforms the FPU is used for conversion because on
1990       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1991       // instruction with 64-bit operands.
1992       __ push(rdx);          // store long on tos
1993       __ push(rax);
1994       __ fild_d(at_rsp());   // load long to ST0
1995       __ d2ieee();           // truncate to double size
1996       __ pop(rcx);           // adjust rsp
1997       __ pop(rcx);
1998       if (UseSSE >= 2) {
1999         __ push_d();
2000         __ pop_d(xmm0);
2001       }
2002       break;
2003     case Bytecodes::_f2i:
2004       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2005       // as it returns 0 for any NaN.
2006       if (UseSSE >= 1) {
2007         __ push_f(xmm0);
2008       } else {
2009         __ push(rcx);          // reserve space for argument
2010         __ fstp_s(at_rsp());   // pass float argument on stack
2011       }
2012       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2013       break;
2014     case Bytecodes::_f2l:
2015       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2016       // as it returns 0 for any NaN.
2017       if (UseSSE >= 1) {
2018        __ push_f(xmm0);
2019       } else {
2020         __ push(rcx);          // reserve space for argument
2021         __ fstp_s(at_rsp());   // pass float argument on stack
2022       }
2023       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2024       break;
2025     case Bytecodes::_f2d:
2026       if (UseSSE < 1) {
2027         /* nothing to do */
2028       } else if (UseSSE == 1) {
2029         __ push_f(xmm0);
2030         __ pop_f();
2031       } else { // UseSSE >= 2
2032         __ cvtss2sd(xmm0, xmm0);
2033       }
2034       break;
2035     case Bytecodes::_d2i:
2036       if (UseSSE >= 2) {
2037         __ push_d(xmm0);
2038       } else {
2039         __ push(rcx);          // reserve space for argument
2040         __ push(rcx);
2041         __ fstp_d(at_rsp());   // pass double argument on stack
2042       }
2043       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2044       break;
2045     case Bytecodes::_d2l:
2046       if (UseSSE >= 2) {
2047         __ push_d(xmm0);
2048       } else {
2049         __ push(rcx);          // reserve space for argument
2050         __ push(rcx);
2051         __ fstp_d(at_rsp());   // pass double argument on stack
2052       }
2053       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2054       break;
2055     case Bytecodes::_d2f:
2056       if (UseSSE <= 1) {
2057         __ push(rcx);          // reserve space for f2ieee()
2058         __ f2ieee();           // truncate to float size
2059         __ pop(rcx);           // adjust rsp
2060         if (UseSSE == 1) {
2061           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2062           // the conversion is performed using the FPU in this case.
2063           __ push_f();
2064           __ pop_f(xmm0);
2065         }
2066       } else { // UseSSE >= 2
2067         __ cvtsd2ss(xmm0, xmm0);
2068       }
2069       break;
2070     default             :
2071       ShouldNotReachHere();
2072   }
2073 #endif
2074 }
2075 
2076 void TemplateTable::lcmp() {
2077   transition(ltos, itos);
2078 #ifdef _LP64
2079   Label done;
2080   __ pop_l(rdx);
2081   __ cmpq(rdx, rax);
2082   __ movl(rax, -1);
2083   __ jccb(Assembler::less, done);
2084   __ setb(Assembler::notEqual, rax);
2085   __ movzbl(rax, rax);
2086   __ bind(done);
2087 #else
2088 
2089   // y = rdx:rax
2090   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2091   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2092   __ mov(rax, rcx);
2093 #endif
2094 }
2095 
2096 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2097   if ((is_float && UseSSE >= 1) ||
2098       (!is_float && UseSSE >= 2)) {
2099     Label done;
2100     if (is_float) {
2101       // XXX get rid of pop here, use ... reg, mem32
2102       __ pop_f(xmm1);
2103       __ ucomiss(xmm1, xmm0);
2104     } else {
2105       // XXX get rid of pop here, use ... reg, mem64
2106       __ pop_d(xmm1);
2107       __ ucomisd(xmm1, xmm0);
2108     }
2109     if (unordered_result < 0) {
2110       __ movl(rax, -1);
2111       __ jccb(Assembler::parity, done);
2112       __ jccb(Assembler::below, done);
2113       __ setb(Assembler::notEqual, rdx);
2114       __ movzbl(rax, rdx);
2115     } else {
2116       __ movl(rax, 1);
2117       __ jccb(Assembler::parity, done);
2118       __ jccb(Assembler::above, done);
2119       __ movl(rax, 0);
2120       __ jccb(Assembler::equal, done);
2121       __ decrementl(rax);
2122     }
2123     __ bind(done);
2124   } else {
2125 #ifdef _LP64
2126     ShouldNotReachHere();
2127 #else
2128     if (is_float) {
2129       __ fld_s(at_rsp());
2130     } else {
2131       __ fld_d(at_rsp());
2132       __ pop(rdx);
2133     }
2134     __ pop(rcx);
2135     __ fcmp2int(rax, unordered_result < 0);
2136 #endif // _LP64
2137   }
2138 }
2139 
2140 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2141   __ get_method(rcx); // rcx holds method
2142   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2143                                      // holds bumped taken count
2144 
2145   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2146                              InvocationCounter::counter_offset();
2147   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2148                               InvocationCounter::counter_offset();
2149 
2150   // Load up edx with the branch displacement
2151   if (is_wide) {
2152     __ movl(rdx, at_bcp(1));
2153   } else {
2154     __ load_signed_short(rdx, at_bcp(1));
2155   }
2156   __ bswapl(rdx);
2157 
2158   if (!is_wide) {
2159     __ sarl(rdx, 16);
2160   }
2161   LP64_ONLY(__ movl2ptr(rdx, rdx));
2162 
2163   // Handle all the JSR stuff here, then exit.
2164   // It's much shorter and cleaner than intermingling with the non-JSR
2165   // normal-branch stuff occurring below.
2166   if (is_jsr) {
2167     // Pre-load the next target bytecode into rbx
2168     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2169 
2170     // compute return address as bci in rax
2171     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2172                         in_bytes(ConstMethod::codes_offset())));
2173     __ subptr(rax, Address(rcx, Method::const_offset()));
2174     // Adjust the bcp in r13 by the displacement in rdx
2175     __ addptr(rbcp, rdx);
2176     // jsr returns atos that is not an oop
2177     __ push_i(rax);
2178     __ dispatch_only(vtos, true);
2179     return;
2180   }
2181 
2182   // Normal (non-jsr) branch handling
2183 
2184   // Adjust the bcp in r13 by the displacement in rdx
2185   __ addptr(rbcp, rdx);
2186 
2187   assert(UseLoopCounter || !UseOnStackReplacement,
2188          "on-stack-replacement requires loop counters");
2189   Label backedge_counter_overflow;
2190   Label profile_method;
2191   Label dispatch;
2192   if (UseLoopCounter) {
2193     // increment backedge counter for backward branches
2194     // rax: MDO
2195     // rbx: MDO bumped taken-count
2196     // rcx: method
2197     // rdx: target offset
2198     // r13: target bcp
2199     // r14: locals pointer
2200     __ testl(rdx, rdx);             // check if forward or backward branch
2201     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2202 
2203     // check if MethodCounters exists
2204     Label has_counters;
2205     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2206     __ testptr(rax, rax);
2207     __ jcc(Assembler::notZero, has_counters);
2208     __ push(rdx);
2209     __ push(rcx);
2210     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2211                rcx);
2212     __ pop(rcx);
2213     __ pop(rdx);
2214     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2215     __ testptr(rax, rax);
2216     __ jcc(Assembler::zero, dispatch);
2217     __ bind(has_counters);
2218 
2219     if (TieredCompilation) {
2220       Label no_mdo;
2221       int increment = InvocationCounter::count_increment;
2222       if (ProfileInterpreter) {
2223         // Are we profiling?
2224         __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2225         __ testptr(rbx, rbx);
2226         __ jccb(Assembler::zero, no_mdo);
2227         // Increment the MDO backedge counter
2228         const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2229                                            in_bytes(InvocationCounter::counter_offset()));
2230         const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2231         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, rax, false, Assembler::zero,
2232                                    UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2233         __ jmp(dispatch);
2234       }
2235       __ bind(no_mdo);
2236       // Increment backedge counter in MethodCounters*
2237       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2238       const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2239       __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2240                                  rax, false, Assembler::zero,
2241                                  UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2242     } else { // not TieredCompilation
2243       // increment counter
2244       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2245       __ movl(rax, Address(rcx, be_offset));        // load backedge counter
2246       __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2247       __ movl(Address(rcx, be_offset), rax);        // store counter
2248 
2249       __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
2250 
2251       __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2252       __ addl(rax, Address(rcx, be_offset));        // add both counters
2253 
2254       if (ProfileInterpreter) {
2255         // Test to see if we should create a method data oop
2256         __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2257         __ jcc(Assembler::less, dispatch);
2258 
2259         // if no method data exists, go to profile method
2260         __ test_method_data_pointer(rax, profile_method);
2261 
2262         if (UseOnStackReplacement) {
2263           // check for overflow against rbx which is the MDO taken count
2264           __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2265           __ jcc(Assembler::below, dispatch);
2266 
2267           // When ProfileInterpreter is on, the backedge_count comes
2268           // from the MethodData*, which value does not get reset on
2269           // the call to frequency_counter_overflow().  To avoid
2270           // excessive calls to the overflow routine while the method is
2271           // being compiled, add a second test to make sure the overflow
2272           // function is called only once every overflow_frequency.
2273           const int overflow_frequency = 1024;
2274           __ andl(rbx, overflow_frequency - 1);
2275           __ jcc(Assembler::zero, backedge_counter_overflow);
2276 
2277         }
2278       } else {
2279         if (UseOnStackReplacement) {
2280           // check for overflow against rax, which is the sum of the
2281           // counters
2282           __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2283           __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2284 
2285         }
2286       }
2287     }
2288     __ bind(dispatch);
2289   }
2290 
2291   // Pre-load the next target bytecode into rbx
2292   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2293 
2294   // continue with the bytecode @ target
2295   // rax: return bci for jsr's, unused otherwise
2296   // rbx: target bytecode
2297   // r13: target bcp
2298   __ dispatch_only(vtos, true);
2299 
2300   if (UseLoopCounter) {
2301     if (ProfileInterpreter) {
2302       // Out-of-line code to allocate method data oop.
2303       __ bind(profile_method);
2304       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2305       __ set_method_data_pointer_for_bcp();
2306       __ jmp(dispatch);
2307     }
2308 
2309     if (UseOnStackReplacement) {
2310       // invocation counter overflow
2311       __ bind(backedge_counter_overflow);
2312       __ negptr(rdx);
2313       __ addptr(rdx, rbcp); // branch bcp
2314       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2315       __ call_VM(noreg,
2316                  CAST_FROM_FN_PTR(address,
2317                                   InterpreterRuntime::frequency_counter_overflow),
2318                  rdx);
2319 
2320       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2321       // rdx: scratch
2322       // r14: locals pointer
2323       // r13: bcp
2324       __ testptr(rax, rax);                        // test result
2325       __ jcc(Assembler::zero, dispatch);         // no osr if null
2326       // nmethod may have been invalidated (VM may block upon call_VM return)
2327       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2328       __ jcc(Assembler::notEqual, dispatch);
2329 
2330       // We have the address of an on stack replacement routine in rax.
2331       // In preparation of invoking it, first we must migrate the locals
2332       // and monitors from off the interpreter frame on the stack.
2333       // Ensure to save the osr nmethod over the migration call,
2334       // it will be preserved in rbx.
2335       __ mov(rbx, rax);
2336 
2337       NOT_LP64(__ get_thread(rcx));
2338 
2339       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2340 
2341       // rax is OSR buffer, move it to expected parameter location
2342       LP64_ONLY(__ mov(j_rarg0, rax));
2343       NOT_LP64(__ mov(rcx, rax));
2344       // We use j_rarg definitions here so that registers don't conflict as parameter
2345       // registers change across platforms as we are in the midst of a calling
2346       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2347 
2348       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2349       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2350 
2351       // pop the interpreter frame
2352       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2353       __ leave();                                // remove frame anchor
2354       __ pop(retaddr);                           // get return address
2355       __ mov(rsp, sender_sp);                   // set sp to sender sp
2356       // Ensure compiled code always sees stack at proper alignment
2357       __ andptr(rsp, -(StackAlignmentInBytes));
2358 
2359       // unlike x86 we need no specialized return from compiled code
2360       // to the interpreter or the call stub.
2361 
2362       // push the return address
2363       __ push(retaddr);
2364 
2365       // and begin the OSR nmethod
2366       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2367     }
2368   }
2369 }
2370 
2371 void TemplateTable::if_0cmp(Condition cc) {
2372   transition(itos, vtos);
2373   // assume branch is more often taken than not (loops use backward branches)
2374   Label not_taken;
2375   __ testl(rax, rax);
2376   __ jcc(j_not(cc), not_taken);
2377   branch(false, false);
2378   __ bind(not_taken);
2379   __ profile_not_taken_branch(rax);
2380 }
2381 
2382 void TemplateTable::if_icmp(Condition cc) {
2383   transition(itos, vtos);
2384   // assume branch is more often taken than not (loops use backward branches)
2385   Label not_taken;
2386   __ pop_i(rdx);
2387   __ cmpl(rdx, rax);
2388   __ jcc(j_not(cc), not_taken);
2389   branch(false, false);
2390   __ bind(not_taken);
2391   __ profile_not_taken_branch(rax);
2392 }
2393 
2394 void TemplateTable::if_nullcmp(Condition cc) {
2395   transition(atos, vtos);
2396   // assume branch is more often taken than not (loops use backward branches)
2397   Label not_taken;
2398   __ testptr(rax, rax);
2399   __ jcc(j_not(cc), not_taken);
2400   branch(false, false);
2401   __ bind(not_taken);
2402   __ profile_not_taken_branch(rax);
2403 }
2404 
2405 void TemplateTable::if_acmp(Condition cc) {
2406   transition(atos, vtos);
2407   // assume branch is more often taken than not (loops use backward branches)
2408   Label not_taken;
2409   __ pop_ptr(rdx);
2410   __ cmpoop(rdx, rax);
2411   __ jcc(j_not(cc), not_taken);
2412   branch(false, false);
2413   __ bind(not_taken);
2414   __ profile_not_taken_branch(rax);
2415 }
2416 
2417 void TemplateTable::ret() {
2418   transition(vtos, vtos);
2419   locals_index(rbx);
2420   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2421   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2422   __ profile_ret(rbx, rcx);
2423   __ get_method(rax);
2424   __ movptr(rbcp, Address(rax, Method::const_offset()));
2425   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2426                       ConstMethod::codes_offset()));
2427   __ dispatch_next(vtos, 0, true);
2428 }
2429 
2430 void TemplateTable::wide_ret() {
2431   transition(vtos, vtos);
2432   locals_index_wide(rbx);
2433   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2434   __ profile_ret(rbx, rcx);
2435   __ get_method(rax);
2436   __ movptr(rbcp, Address(rax, Method::const_offset()));
2437   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2438   __ dispatch_next(vtos, 0, true);
2439 }
2440 
2441 void TemplateTable::tableswitch() {
2442   Label default_case, continue_execution;
2443   transition(itos, vtos);
2444 
2445   // align r13/rsi
2446   __ lea(rbx, at_bcp(BytesPerInt));
2447   __ andptr(rbx, -BytesPerInt);
2448   // load lo & hi
2449   __ movl(rcx, Address(rbx, BytesPerInt));
2450   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2451   __ bswapl(rcx);
2452   __ bswapl(rdx);
2453   // check against lo & hi
2454   __ cmpl(rax, rcx);
2455   __ jcc(Assembler::less, default_case);
2456   __ cmpl(rax, rdx);
2457   __ jcc(Assembler::greater, default_case);
2458   // lookup dispatch offset
2459   __ subl(rax, rcx);
2460   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2461   __ profile_switch_case(rax, rbx, rcx);
2462   // continue execution
2463   __ bind(continue_execution);
2464   __ bswapl(rdx);
2465   LP64_ONLY(__ movl2ptr(rdx, rdx));
2466   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2467   __ addptr(rbcp, rdx);
2468   __ dispatch_only(vtos, true);
2469   // handle default
2470   __ bind(default_case);
2471   __ profile_switch_default(rax);
2472   __ movl(rdx, Address(rbx, 0));
2473   __ jmp(continue_execution);
2474 }
2475 
2476 void TemplateTable::lookupswitch() {
2477   transition(itos, itos);
2478   __ stop("lookupswitch bytecode should have been rewritten");
2479 }
2480 
2481 void TemplateTable::fast_linearswitch() {
2482   transition(itos, vtos);
2483   Label loop_entry, loop, found, continue_execution;
2484   // bswap rax so we can avoid bswapping the table entries
2485   __ bswapl(rax);
2486   // align r13
2487   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2488                                     // this instruction (change offsets
2489                                     // below)
2490   __ andptr(rbx, -BytesPerInt);
2491   // set counter
2492   __ movl(rcx, Address(rbx, BytesPerInt));
2493   __ bswapl(rcx);
2494   __ jmpb(loop_entry);
2495   // table search
2496   __ bind(loop);
2497   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2498   __ jcc(Assembler::equal, found);
2499   __ bind(loop_entry);
2500   __ decrementl(rcx);
2501   __ jcc(Assembler::greaterEqual, loop);
2502   // default case
2503   __ profile_switch_default(rax);
2504   __ movl(rdx, Address(rbx, 0));
2505   __ jmp(continue_execution);
2506   // entry found -> get offset
2507   __ bind(found);
2508   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2509   __ profile_switch_case(rcx, rax, rbx);
2510   // continue execution
2511   __ bind(continue_execution);
2512   __ bswapl(rdx);
2513   __ movl2ptr(rdx, rdx);
2514   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2515   __ addptr(rbcp, rdx);
2516   __ dispatch_only(vtos, true);
2517 }
2518 
2519 void TemplateTable::fast_binaryswitch() {
2520   transition(itos, vtos);
2521   // Implementation using the following core algorithm:
2522   //
2523   // int binary_search(int key, LookupswitchPair* array, int n) {
2524   //   // Binary search according to "Methodik des Programmierens" by
2525   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2526   //   int i = 0;
2527   //   int j = n;
2528   //   while (i+1 < j) {
2529   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2530   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2531   //     // where a stands for the array and assuming that the (inexisting)
2532   //     // element a[n] is infinitely big.
2533   //     int h = (i + j) >> 1;
2534   //     // i < h < j
2535   //     if (key < array[h].fast_match()) {
2536   //       j = h;
2537   //     } else {
2538   //       i = h;
2539   //     }
2540   //   }
2541   //   // R: a[i] <= key < a[i+1] or Q
2542   //   // (i.e., if key is within array, i is the correct index)
2543   //   return i;
2544   // }
2545 
2546   // Register allocation
2547   const Register key   = rax; // already set (tosca)
2548   const Register array = rbx;
2549   const Register i     = rcx;
2550   const Register j     = rdx;
2551   const Register h     = rdi;
2552   const Register temp  = rsi;
2553 
2554   // Find array start
2555   NOT_LP64(__ save_bcp());
2556 
2557   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2558                                           // get rid of this
2559                                           // instruction (change
2560                                           // offsets below)
2561   __ andptr(array, -BytesPerInt);
2562 
2563   // Initialize i & j
2564   __ xorl(i, i);                            // i = 0;
2565   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2566 
2567   // Convert j into native byteordering
2568   __ bswapl(j);
2569 
2570   // And start
2571   Label entry;
2572   __ jmp(entry);
2573 
2574   // binary search loop
2575   {
2576     Label loop;
2577     __ bind(loop);
2578     // int h = (i + j) >> 1;
2579     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2580     __ sarl(h, 1);                               // h = (i + j) >> 1;
2581     // if (key < array[h].fast_match()) {
2582     //   j = h;
2583     // } else {
2584     //   i = h;
2585     // }
2586     // Convert array[h].match to native byte-ordering before compare
2587     __ movl(temp, Address(array, h, Address::times_8));
2588     __ bswapl(temp);
2589     __ cmpl(key, temp);
2590     // j = h if (key <  array[h].fast_match())
2591     __ cmov32(Assembler::less, j, h);
2592     // i = h if (key >= array[h].fast_match())
2593     __ cmov32(Assembler::greaterEqual, i, h);
2594     // while (i+1 < j)
2595     __ bind(entry);
2596     __ leal(h, Address(i, 1)); // i+1
2597     __ cmpl(h, j);             // i+1 < j
2598     __ jcc(Assembler::less, loop);
2599   }
2600 
2601   // end of binary search, result index is i (must check again!)
2602   Label default_case;
2603   // Convert array[i].match to native byte-ordering before compare
2604   __ movl(temp, Address(array, i, Address::times_8));
2605   __ bswapl(temp);
2606   __ cmpl(key, temp);
2607   __ jcc(Assembler::notEqual, default_case);
2608 
2609   // entry found -> j = offset
2610   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2611   __ profile_switch_case(i, key, array);
2612   __ bswapl(j);
2613   LP64_ONLY(__ movslq(j, j));
2614 
2615   NOT_LP64(__ restore_bcp());
2616   NOT_LP64(__ restore_locals());                           // restore rdi
2617 
2618   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2619   __ addptr(rbcp, j);
2620   __ dispatch_only(vtos, true);
2621 
2622   // default case -> j = default offset
2623   __ bind(default_case);
2624   __ profile_switch_default(i);
2625   __ movl(j, Address(array, -2 * BytesPerInt));
2626   __ bswapl(j);
2627   LP64_ONLY(__ movslq(j, j));
2628 
2629   NOT_LP64(__ restore_bcp());
2630   NOT_LP64(__ restore_locals());
2631 
2632   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2633   __ addptr(rbcp, j);
2634   __ dispatch_only(vtos, true);
2635 }
2636 
2637 void TemplateTable::_return(TosState state) {
2638   transition(state, state);
2639 
2640   // {
2641   //   Label not_rb;
2642   //   Register aa = rcx, bb = rdi;
2643   //   __ movptr(aa, Address(rsp, 0));
2644   //   __ lea(bb, ExternalAddress(StubRoutines::cont_returnBarrier()));
2645   //   __ cmpq(aa, bb);
2646   //   // __ cmpq(ExternalAddress(StubRoutines::cont_returnBarrier()).addr(), aa);
2647   //   __ jcc(Assembler::notZero, not_rb);
2648   //   __ stop("WQWWQWQW");
2649   //   __ bind(not_rb);
2650   // }
2651   assert(_desc->calls_vm(),
2652          "inconsistent calls_vm information"); // call in remove_activation
2653 
2654   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2655     assert(state == vtos, "only valid state");
2656     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2657     __ movptr(robj, aaddress(0));
2658     __ load_klass(rdi, robj);
2659     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2660     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2661     Label skip_register_finalizer;
2662     __ jcc(Assembler::zero, skip_register_finalizer);
2663 
2664     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2665 
2666     __ bind(skip_register_finalizer);
2667   }
2668 
2669   if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2670     Label no_safepoint;
2671     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2672 #ifdef _LP64
2673     __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2674 #else
2675     const Register thread = rdi;
2676     __ get_thread(thread);
2677     __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2678 #endif
2679     __ jcc(Assembler::zero, no_safepoint);
2680     __ push(state);
2681     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2682                                     InterpreterRuntime::at_safepoint));
2683     __ pop(state);
2684     __ bind(no_safepoint);
2685   }
2686 
2687   // Narrow result if state is itos but result type is smaller.
2688   // Need to narrow in the return bytecode rather than in generate_return_entry
2689   // since compiled code callers expect the result to already be narrowed.
2690   if (state == itos) {
2691     __ narrow(rax);
2692   }
2693   __ remove_activation(state, rbcp);
2694 
2695   __ jmp(rbcp);
2696 }
2697 
2698 // ----------------------------------------------------------------------------
2699 // Volatile variables demand their effects be made known to all CPU's
2700 // in order.  Store buffers on most chips allow reads & writes to
2701 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2702 // without some kind of memory barrier (i.e., it's not sufficient that
2703 // the interpreter does not reorder volatile references, the hardware
2704 // also must not reorder them).
2705 //
2706 // According to the new Java Memory Model (JMM):
2707 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2708 //     writes act as aquire & release, so:
2709 // (2) A read cannot let unrelated NON-volatile memory refs that
2710 //     happen after the read float up to before the read.  It's OK for
2711 //     non-volatile memory refs that happen before the volatile read to
2712 //     float down below it.
2713 // (3) Similar a volatile write cannot let unrelated NON-volatile
2714 //     memory refs that happen BEFORE the write float down to after the
2715 //     write.  It's OK for non-volatile memory refs that happen after the
2716 //     volatile write to float up before it.
2717 //
2718 // We only put in barriers around volatile refs (they are expensive),
2719 // not _between_ memory refs (that would require us to track the
2720 // flavor of the previous memory refs).  Requirements (2) and (3)
2721 // require some barriers before volatile stores and after volatile
2722 // loads.  These nearly cover requirement (1) but miss the
2723 // volatile-store-volatile-load case.  This final case is placed after
2724 // volatile-stores although it could just as well go before
2725 // volatile-loads.
2726 
2727 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2728   // Helper function to insert a is-volatile test and memory barrier
2729   __ membar(order_constraint);
2730 }
2731 
2732 void TemplateTable::resolve_cache_and_index(int byte_no,
2733                                             Register cache,
2734                                             Register index,
2735                                             size_t index_size) {
2736   const Register temp = rbx;
2737   assert_different_registers(cache, index, temp);
2738 
2739   Label L_clinit_barrier_slow;
2740   Label resolved;
2741 
2742   Bytecodes::Code code = bytecode();
2743   switch (code) {
2744   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2745   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2746   default: break;
2747   }
2748 
2749   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2750   __ get_cache_and_index_and_bytecode_at_bcp(cache, index, temp, byte_no, 1, index_size);
2751   __ cmpl(temp, code);  // have we resolved this bytecode?
2752   __ jcc(Assembler::equal, resolved);
2753 
2754   // resolve first time through
2755   // Class initialization barrier slow path lands here as well.
2756   __ bind(L_clinit_barrier_slow);
2757   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2758   __ movl(temp, code);
2759   __ call_VM(noreg, entry, temp);
2760   // Update registers with resolved info
2761   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
2762 
2763   __ bind(resolved);
2764 
2765   // Class initialization barrier for static methods
2766   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2767     const Register method = temp;
2768     const Register klass  = temp;
2769     const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2770     assert(thread != noreg, "x86_32 not supported");
2771 
2772     __ load_resolved_method_at_index(byte_no, method, cache, index);
2773     __ load_method_holder(klass, method);
2774     __ clinit_barrier(klass, thread, NULL /*L_fast_path*/, &L_clinit_barrier_slow);
2775   }
2776 }
2777 
2778 // The cache and index registers must be set before call
2779 void TemplateTable::load_field_cp_cache_entry(Register obj,
2780                                               Register cache,
2781                                               Register index,
2782                                               Register off,
2783                                               Register flags,
2784                                               bool is_static = false) {
2785   assert_different_registers(cache, index, flags, off);
2786 
2787   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2788   // Field offset
2789   __ movptr(off, Address(cache, index, Address::times_ptr,
2790                          in_bytes(cp_base_offset +
2791                                   ConstantPoolCacheEntry::f2_offset())));
2792   // Flags
2793   __ movl(flags, Address(cache, index, Address::times_ptr,
2794                          in_bytes(cp_base_offset +
2795                                   ConstantPoolCacheEntry::flags_offset())));
2796 
2797   // klass overwrite register
2798   if (is_static) {
2799     __ movptr(obj, Address(cache, index, Address::times_ptr,
2800                            in_bytes(cp_base_offset +
2801                                     ConstantPoolCacheEntry::f1_offset())));
2802     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2803     __ movptr(obj, Address(obj, mirror_offset));
2804     __ resolve_oop_handle(obj);
2805   }
2806 }
2807 
2808 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2809                                                Register method,
2810                                                Register itable_index,
2811                                                Register flags,
2812                                                bool is_invokevirtual,
2813                                                bool is_invokevfinal, /*unused*/
2814                                                bool is_invokedynamic) {
2815   // setup registers
2816   const Register cache = rcx;
2817   const Register index = rdx;
2818   assert_different_registers(method, flags);
2819   assert_different_registers(method, cache, index);
2820   assert_different_registers(itable_index, flags);
2821   assert_different_registers(itable_index, cache, index);
2822   // determine constant pool cache field offsets
2823   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2824   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2825                                     ConstantPoolCacheEntry::flags_offset());
2826   // access constant pool cache fields
2827   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2828                                     ConstantPoolCacheEntry::f2_offset());
2829 
2830   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2831   resolve_cache_and_index(byte_no, cache, index, index_size);
2832   __ load_resolved_method_at_index(byte_no, method, cache, index);
2833 
2834   if (itable_index != noreg) {
2835     // pick up itable or appendix index from f2 also:
2836     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2837   }
2838   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2839 }
2840 
2841 // The registers cache and index expected to be set before call.
2842 // Correct values of the cache and index registers are preserved.
2843 void TemplateTable::jvmti_post_field_access(Register cache,
2844                                             Register index,
2845                                             bool is_static,
2846                                             bool has_tos) {
2847   if (JvmtiExport::can_post_field_access()) {
2848     // Check to see if a field access watch has been set before we take
2849     // the time to call into the VM.
2850     Label L1;
2851     assert_different_registers(cache, index, rax);
2852     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2853     __ testl(rax,rax);
2854     __ jcc(Assembler::zero, L1);
2855 
2856     // cache entry pointer
2857     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2858     __ shll(index, LogBytesPerWord);
2859     __ addptr(cache, index);
2860     if (is_static) {
2861       __ xorptr(rax, rax);      // NULL object reference
2862     } else {
2863       __ pop(atos);         // Get the object
2864       __ verify_oop(rax);
2865       __ push(atos);        // Restore stack state
2866     }
2867     // rax,:   object pointer or NULL
2868     // cache: cache entry pointer
2869     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2870                rax, cache);
2871     __ get_cache_and_index_at_bcp(cache, index, 1);
2872     __ bind(L1);
2873   }
2874 }
2875 
2876 void TemplateTable::pop_and_check_object(Register r) {
2877   __ pop_ptr(r);
2878   __ null_check(r);  // for field access must check obj.
2879   __ verify_oop(r);
2880 }
2881 
2882 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2883   transition(vtos, vtos);
2884 
2885   const Register cache = rcx;
2886   const Register index = rdx;
2887   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2888   const Register off   = rbx;
2889   const Register flags = rax;
2890   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2891 
2892   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2893   jvmti_post_field_access(cache, index, is_static, false);
2894   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2895 
2896   if (!is_static) pop_and_check_object(obj);
2897 
2898   const Address field(obj, off, Address::times_1, 0*wordSize);
2899 
2900   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2901 
2902   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2903   // Make sure we don't need to mask edx after the above shift
2904   assert(btos == 0, "change code, btos != 0");
2905 
2906   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2907 
2908   __ jcc(Assembler::notZero, notByte);
2909   // btos
2910   __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2911   __ push(btos);
2912   // Rewrite bytecode to be faster
2913   if (!is_static && rc == may_rewrite) {
2914     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2915   }
2916   __ jmp(Done);
2917 
2918   __ bind(notByte);
2919   __ cmpl(flags, ztos);
2920   __ jcc(Assembler::notEqual, notBool);
2921 
2922   // ztos (same code as btos)
2923   __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2924   __ push(ztos);
2925   // Rewrite bytecode to be faster
2926   if (!is_static && rc == may_rewrite) {
2927     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2928     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2929   }
2930   __ jmp(Done);
2931 
2932   __ bind(notBool);
2933   __ cmpl(flags, atos);
2934   __ jcc(Assembler::notEqual, notObj);
2935   // atos
2936   do_oop_load(_masm, field, rax);
2937   __ push(atos);
2938   if (!is_static && rc == may_rewrite) {
2939     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2940   }
2941   __ jmp(Done);
2942 
2943   __ bind(notObj);
2944   __ cmpl(flags, itos);
2945   __ jcc(Assembler::notEqual, notInt);
2946   // itos
2947   __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2948   __ push(itos);
2949   // Rewrite bytecode to be faster
2950   if (!is_static && rc == may_rewrite) {
2951     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2952   }
2953   __ jmp(Done);
2954 
2955   __ bind(notInt);
2956   __ cmpl(flags, ctos);
2957   __ jcc(Assembler::notEqual, notChar);
2958   // ctos
2959   __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2960   __ push(ctos);
2961   // Rewrite bytecode to be faster
2962   if (!is_static && rc == may_rewrite) {
2963     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2964   }
2965   __ jmp(Done);
2966 
2967   __ bind(notChar);
2968   __ cmpl(flags, stos);
2969   __ jcc(Assembler::notEqual, notShort);
2970   // stos
2971   __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2972   __ push(stos);
2973   // Rewrite bytecode to be faster
2974   if (!is_static && rc == may_rewrite) {
2975     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2976   }
2977   __ jmp(Done);
2978 
2979   __ bind(notShort);
2980   __ cmpl(flags, ltos);
2981   __ jcc(Assembler::notEqual, notLong);
2982   // ltos
2983     // Generate code as if volatile (x86_32).  There just aren't enough registers to
2984     // save that information and this code is faster than the test.
2985   __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2986   __ push(ltos);
2987   // Rewrite bytecode to be faster
2988   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2989   __ jmp(Done);
2990 
2991   __ bind(notLong);
2992   __ cmpl(flags, ftos);
2993   __ jcc(Assembler::notEqual, notFloat);
2994   // ftos
2995 
2996   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2997   __ push(ftos);
2998   // Rewrite bytecode to be faster
2999   if (!is_static && rc == may_rewrite) {
3000     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3001   }
3002   __ jmp(Done);
3003 
3004   __ bind(notFloat);
3005 #ifdef ASSERT
3006   Label notDouble;
3007   __ cmpl(flags, dtos);
3008   __ jcc(Assembler::notEqual, notDouble);
3009 #endif
3010   // dtos
3011   // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3012   __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
3013   __ push(dtos);
3014   // Rewrite bytecode to be faster
3015   if (!is_static && rc == may_rewrite) {
3016     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3017   }
3018 #ifdef ASSERT
3019   __ jmp(Done);
3020 
3021   __ bind(notDouble);
3022   __ stop("Bad state");
3023 #endif
3024 
3025   __ bind(Done);
3026   // [jk] not needed currently
3027   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3028   //                                              Assembler::LoadStore));
3029 }
3030 
3031 void TemplateTable::getfield(int byte_no) {
3032   getfield_or_static(byte_no, false);
3033 }
3034 
3035 void TemplateTable::nofast_getfield(int byte_no) {
3036   getfield_or_static(byte_no, false, may_not_rewrite);
3037 }
3038 
3039 void TemplateTable::getstatic(int byte_no) {
3040   getfield_or_static(byte_no, true);
3041 }
3042 
3043 
3044 // The registers cache and index expected to be set before call.
3045 // The function may destroy various registers, just not the cache and index registers.
3046 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3047 
3048   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
3049   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
3050   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
3051   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3052 
3053   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3054 
3055   if (JvmtiExport::can_post_field_modification()) {
3056     // Check to see if a field modification watch has been set before
3057     // we take the time to call into the VM.
3058     Label L1;
3059     assert_different_registers(cache, index, rax);
3060     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3061     __ testl(rax, rax);
3062     __ jcc(Assembler::zero, L1);
3063 
3064     __ get_cache_and_index_at_bcp(robj, RDX, 1);
3065 
3066 
3067     if (is_static) {
3068       // Life is simple.  Null out the object pointer.
3069       __ xorl(RBX, RBX);
3070 
3071     } else {
3072       // Life is harder. The stack holds the value on top, followed by
3073       // the object.  We don't know the size of the value, though; it
3074       // could be one or two words depending on its type. As a result,
3075       // we must find the type to determine where the object is.
3076 #ifndef _LP64
3077       Label two_word, valsize_known;
3078 #endif
3079       __ movl(RCX, Address(robj, RDX,
3080                            Address::times_ptr,
3081                            in_bytes(cp_base_offset +
3082                                      ConstantPoolCacheEntry::flags_offset())));
3083       NOT_LP64(__ mov(rbx, rsp));
3084       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3085 
3086       // Make sure we don't need to mask rcx after the above shift
3087       ConstantPoolCacheEntry::verify_tos_state_shift();
3088 #ifdef _LP64
3089       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3090       __ cmpl(c_rarg3, ltos);
3091       __ cmovptr(Assembler::equal,
3092                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3093       __ cmpl(c_rarg3, dtos);
3094       __ cmovptr(Assembler::equal,
3095                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3096 #else
3097       __ cmpl(rcx, ltos);
3098       __ jccb(Assembler::equal, two_word);
3099       __ cmpl(rcx, dtos);
3100       __ jccb(Assembler::equal, two_word);
3101       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3102       __ jmpb(valsize_known);
3103 
3104       __ bind(two_word);
3105       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3106 
3107       __ bind(valsize_known);
3108       // setup object pointer
3109       __ movptr(rbx, Address(rbx, 0));
3110 #endif
3111     }
3112     // cache entry pointer
3113     __ addptr(robj, in_bytes(cp_base_offset));
3114     __ shll(RDX, LogBytesPerWord);
3115     __ addptr(robj, RDX);
3116     // object (tos)
3117     __ mov(RCX, rsp);
3118     // c_rarg1: object pointer set up above (NULL if static)
3119     // c_rarg2: cache entry pointer
3120     // c_rarg3: jvalue object on the stack
3121     __ call_VM(noreg,
3122                CAST_FROM_FN_PTR(address,
3123                                 InterpreterRuntime::post_field_modification),
3124                RBX, robj, RCX);
3125     __ get_cache_and_index_at_bcp(cache, index, 1);
3126     __ bind(L1);
3127   }
3128 }
3129 
3130 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3131   transition(vtos, vtos);
3132 
3133   const Register cache = rcx;
3134   const Register index = rdx;
3135   const Register obj   = rcx;
3136   const Register off   = rbx;
3137   const Register flags = rax;
3138 
3139   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3140   jvmti_post_field_mod(cache, index, is_static);
3141   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3142 
3143   // [jk] not needed currently
3144   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3145   //                                              Assembler::StoreStore));
3146 
3147   Label notVolatile, Done;
3148   __ movl(rdx, flags);
3149   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3150   __ andl(rdx, 0x1);
3151 
3152   // Check for volatile store
3153   __ testl(rdx, rdx);
3154   __ jcc(Assembler::zero, notVolatile);
3155 
3156   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3157   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3158                                                Assembler::StoreStore));
3159   __ jmp(Done);
3160   __ bind(notVolatile);
3161 
3162   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3163 
3164   __ bind(Done);
3165 }
3166 
3167 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3168                                               Register obj, Register off, Register flags) {
3169 
3170   // field addresses
3171   const Address field(obj, off, Address::times_1, 0*wordSize);
3172   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3173 
3174   Label notByte, notBool, notInt, notShort, notChar,
3175         notLong, notFloat, notObj;
3176   Label Done;
3177 
3178   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3179 
3180   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3181 
3182   assert(btos == 0, "change code, btos != 0");
3183   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3184   __ jcc(Assembler::notZero, notByte);
3185 
3186   // btos
3187   {
3188     __ pop(btos);
3189     if (!is_static) pop_and_check_object(obj);
3190     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3191     if (!is_static && rc == may_rewrite) {
3192       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3193     }
3194     __ jmp(Done);
3195   }
3196 
3197   __ bind(notByte);
3198   __ cmpl(flags, ztos);
3199   __ jcc(Assembler::notEqual, notBool);
3200 
3201   // ztos
3202   {
3203     __ pop(ztos);
3204     if (!is_static) pop_and_check_object(obj);
3205     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3206     if (!is_static && rc == may_rewrite) {
3207       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3208     }
3209     __ jmp(Done);
3210   }
3211 
3212   __ bind(notBool);
3213   __ cmpl(flags, atos);
3214   __ jcc(Assembler::notEqual, notObj);
3215 
3216   // atos
3217   {
3218     __ pop(atos);
3219     if (!is_static) pop_and_check_object(obj);
3220     // Store into the field
3221     do_oop_store(_masm, field, rax);
3222     if (!is_static && rc == may_rewrite) {
3223       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3224     }
3225     __ jmp(Done);
3226   }
3227 
3228   __ bind(notObj);
3229   __ cmpl(flags, itos);
3230   __ jcc(Assembler::notEqual, notInt);
3231 
3232   // itos
3233   {
3234     __ pop(itos);
3235     if (!is_static) pop_and_check_object(obj);
3236     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3237     if (!is_static && rc == may_rewrite) {
3238       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3239     }
3240     __ jmp(Done);
3241   }
3242 
3243   __ bind(notInt);
3244   __ cmpl(flags, ctos);
3245   __ jcc(Assembler::notEqual, notChar);
3246 
3247   // ctos
3248   {
3249     __ pop(ctos);
3250     if (!is_static) pop_and_check_object(obj);
3251     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3252     if (!is_static && rc == may_rewrite) {
3253       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3254     }
3255     __ jmp(Done);
3256   }
3257 
3258   __ bind(notChar);
3259   __ cmpl(flags, stos);
3260   __ jcc(Assembler::notEqual, notShort);
3261 
3262   // stos
3263   {
3264     __ pop(stos);
3265     if (!is_static) pop_and_check_object(obj);
3266     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3267     if (!is_static && rc == may_rewrite) {
3268       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3269     }
3270     __ jmp(Done);
3271   }
3272 
3273   __ bind(notShort);
3274   __ cmpl(flags, ltos);
3275   __ jcc(Assembler::notEqual, notLong);
3276 
3277   // ltos
3278   {
3279     __ pop(ltos);
3280     if (!is_static) pop_and_check_object(obj);
3281     // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3282     __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg);
3283 #ifdef _LP64
3284     if (!is_static && rc == may_rewrite) {
3285       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3286     }
3287 #endif // _LP64
3288     __ jmp(Done);
3289   }
3290 
3291   __ bind(notLong);
3292   __ cmpl(flags, ftos);
3293   __ jcc(Assembler::notEqual, notFloat);
3294 
3295   // ftos
3296   {
3297     __ pop(ftos);
3298     if (!is_static) pop_and_check_object(obj);
3299     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3300     if (!is_static && rc == may_rewrite) {
3301       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3302     }
3303     __ jmp(Done);
3304   }
3305 
3306   __ bind(notFloat);
3307 #ifdef ASSERT
3308   Label notDouble;
3309   __ cmpl(flags, dtos);
3310   __ jcc(Assembler::notEqual, notDouble);
3311 #endif
3312 
3313   // dtos
3314   {
3315     __ pop(dtos);
3316     if (!is_static) pop_and_check_object(obj);
3317     // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3318     __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg);
3319     if (!is_static && rc == may_rewrite) {
3320       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3321     }
3322   }
3323 
3324 #ifdef ASSERT
3325   __ jmp(Done);
3326 
3327   __ bind(notDouble);
3328   __ stop("Bad state");
3329 #endif
3330 
3331   __ bind(Done);
3332 }
3333 
3334 void TemplateTable::putfield(int byte_no) {
3335   putfield_or_static(byte_no, false);
3336 }
3337 
3338 void TemplateTable::nofast_putfield(int byte_no) {
3339   putfield_or_static(byte_no, false, may_not_rewrite);
3340 }
3341 
3342 void TemplateTable::putstatic(int byte_no) {
3343   putfield_or_static(byte_no, true);
3344 }
3345 
3346 void TemplateTable::jvmti_post_fast_field_mod() {
3347 
3348   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3349 
3350   if (JvmtiExport::can_post_field_modification()) {
3351     // Check to see if a field modification watch has been set before
3352     // we take the time to call into the VM.
3353     Label L2;
3354     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3355     __ testl(scratch, scratch);
3356     __ jcc(Assembler::zero, L2);
3357     __ pop_ptr(rbx);                  // copy the object pointer from tos
3358     __ verify_oop(rbx);
3359     __ push_ptr(rbx);                 // put the object pointer back on tos
3360     // Save tos values before call_VM() clobbers them. Since we have
3361     // to do it for every data type, we use the saved values as the
3362     // jvalue object.
3363     switch (bytecode()) {          // load values into the jvalue object
3364     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3365     case Bytecodes::_fast_bputfield: // fall through
3366     case Bytecodes::_fast_zputfield: // fall through
3367     case Bytecodes::_fast_sputfield: // fall through
3368     case Bytecodes::_fast_cputfield: // fall through
3369     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3370     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3371     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3372     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3373 
3374     default:
3375       ShouldNotReachHere();
3376     }
3377     __ mov(scratch, rsp);             // points to jvalue on the stack
3378     // access constant pool cache entry
3379     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3380     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3381     __ verify_oop(rbx);
3382     // rbx: object pointer copied above
3383     // c_rarg2: cache entry pointer
3384     // c_rarg3: jvalue object on the stack
3385     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3386     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3387 
3388     switch (bytecode()) {             // restore tos values
3389     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3390     case Bytecodes::_fast_bputfield: // fall through
3391     case Bytecodes::_fast_zputfield: // fall through
3392     case Bytecodes::_fast_sputfield: // fall through
3393     case Bytecodes::_fast_cputfield: // fall through
3394     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3395     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3396     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3397     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3398     default: break;
3399     }
3400     __ bind(L2);
3401   }
3402 }
3403 
3404 void TemplateTable::fast_storefield(TosState state) {
3405   transition(state, vtos);
3406 
3407   ByteSize base = ConstantPoolCache::base_offset();
3408 
3409   jvmti_post_fast_field_mod();
3410 
3411   // access constant pool cache
3412   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3413 
3414   // test for volatile with rdx but rdx is tos register for lputfield.
3415   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3416                        in_bytes(base +
3417                                 ConstantPoolCacheEntry::flags_offset())));
3418 
3419   // replace index with field offset from cache entry
3420   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3421                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3422 
3423   // [jk] not needed currently
3424   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3425   //                                              Assembler::StoreStore));
3426 
3427   Label notVolatile, Done;
3428   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3429   __ andl(rdx, 0x1);
3430 
3431   // Get object from stack
3432   pop_and_check_object(rcx);
3433 
3434   // field address
3435   const Address field(rcx, rbx, Address::times_1);
3436 
3437   // Check for volatile store
3438   __ testl(rdx, rdx);
3439   __ jcc(Assembler::zero, notVolatile);
3440 
3441   fast_storefield_helper(field, rax);
3442   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3443                                                Assembler::StoreStore));
3444   __ jmp(Done);
3445   __ bind(notVolatile);
3446 
3447   fast_storefield_helper(field, rax);
3448 
3449   __ bind(Done);
3450 }
3451 
3452 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3453 
3454   // access field
3455   switch (bytecode()) {
3456   case Bytecodes::_fast_aputfield:
3457     do_oop_store(_masm, field, rax);
3458     break;
3459   case Bytecodes::_fast_lputfield:
3460 #ifdef _LP64
3461     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg);
3462 #else
3463   __ stop("should not be rewritten");
3464 #endif
3465     break;
3466   case Bytecodes::_fast_iputfield:
3467     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3468     break;
3469   case Bytecodes::_fast_zputfield:
3470     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3471     break;
3472   case Bytecodes::_fast_bputfield:
3473     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3474     break;
3475   case Bytecodes::_fast_sputfield:
3476     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3477     break;
3478   case Bytecodes::_fast_cputfield:
3479     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3480     break;
3481   case Bytecodes::_fast_fputfield:
3482     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg);
3483     break;
3484   case Bytecodes::_fast_dputfield:
3485     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg);
3486     break;
3487   default:
3488     ShouldNotReachHere();
3489   }
3490 }
3491 
3492 void TemplateTable::fast_accessfield(TosState state) {
3493   transition(atos, state);
3494 
3495   // Do the JVMTI work here to avoid disturbing the register state below
3496   if (JvmtiExport::can_post_field_access()) {
3497     // Check to see if a field access watch has been set before we
3498     // take the time to call into the VM.
3499     Label L1;
3500     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3501     __ testl(rcx, rcx);
3502     __ jcc(Assembler::zero, L1);
3503     // access constant pool cache entry
3504     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3505     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3506     __ verify_oop(rax);
3507     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3508     LP64_ONLY(__ mov(c_rarg1, rax));
3509     // c_rarg1: object pointer copied above
3510     // c_rarg2: cache entry pointer
3511     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3512     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3513     __ pop_ptr(rax); // restore object pointer
3514     __ bind(L1);
3515   }
3516 
3517   // access constant pool cache
3518   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3519   // replace index with field offset from cache entry
3520   // [jk] not needed currently
3521   // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3522   //                      in_bytes(ConstantPoolCache::base_offset() +
3523   //                               ConstantPoolCacheEntry::flags_offset())));
3524   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3525   // __ andl(rdx, 0x1);
3526   //
3527   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3528                          in_bytes(ConstantPoolCache::base_offset() +
3529                                   ConstantPoolCacheEntry::f2_offset())));
3530 
3531   // rax: object
3532   __ verify_oop(rax);
3533   __ null_check(rax);
3534   Address field(rax, rbx, Address::times_1);
3535 
3536   // access field
3537   switch (bytecode()) {
3538   case Bytecodes::_fast_agetfield:
3539     do_oop_load(_masm, field, rax);
3540     __ verify_oop(rax);
3541     break;
3542   case Bytecodes::_fast_lgetfield:
3543 #ifdef _LP64
3544     __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3545 #else
3546   __ stop("should not be rewritten");
3547 #endif
3548     break;
3549   case Bytecodes::_fast_igetfield:
3550     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3551     break;
3552   case Bytecodes::_fast_bgetfield:
3553     __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3554     break;
3555   case Bytecodes::_fast_sgetfield:
3556     __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3557     break;
3558   case Bytecodes::_fast_cgetfield:
3559     __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3560     break;
3561   case Bytecodes::_fast_fgetfield:
3562     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3563     break;
3564   case Bytecodes::_fast_dgetfield:
3565     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3566     break;
3567   default:
3568     ShouldNotReachHere();
3569   }
3570   // [jk] not needed currently
3571   //   Label notVolatile;
3572   //   __ testl(rdx, rdx);
3573   //   __ jcc(Assembler::zero, notVolatile);
3574   //   __ membar(Assembler::LoadLoad);
3575   //   __ bind(notVolatile);
3576 }
3577 
3578 void TemplateTable::fast_xaccess(TosState state) {
3579   transition(vtos, state);
3580 
3581   // get receiver
3582   __ movptr(rax, aaddress(0));
3583   // access constant pool cache
3584   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3585   __ movptr(rbx,
3586             Address(rcx, rdx, Address::times_ptr,
3587                     in_bytes(ConstantPoolCache::base_offset() +
3588                              ConstantPoolCacheEntry::f2_offset())));
3589   // make sure exception is reported in correct bcp range (getfield is
3590   // next instruction)
3591   __ increment(rbcp);
3592   __ null_check(rax);
3593   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3594   switch (state) {
3595   case itos:
3596     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3597     break;
3598   case atos:
3599     do_oop_load(_masm, field, rax);
3600     __ verify_oop(rax);
3601     break;
3602   case ftos:
3603     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3604     break;
3605   default:
3606     ShouldNotReachHere();
3607   }
3608 
3609   // [jk] not needed currently
3610   // Label notVolatile;
3611   // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3612   //                      in_bytes(ConstantPoolCache::base_offset() +
3613   //                               ConstantPoolCacheEntry::flags_offset())));
3614   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3615   // __ testl(rdx, 0x1);
3616   // __ jcc(Assembler::zero, notVolatile);
3617   // __ membar(Assembler::LoadLoad);
3618   // __ bind(notVolatile);
3619 
3620   __ decrement(rbcp);
3621 }
3622 
3623 //-----------------------------------------------------------------------------
3624 // Calls
3625 
3626 void TemplateTable::count_calls(Register method, Register temp) {
3627   // implemented elsewhere
3628   ShouldNotReachHere();
3629 }
3630 
3631 void TemplateTable::prepare_invoke(int byte_no,
3632                                    Register method,  // linked method (or i-klass)
3633                                    Register index,   // itable index, MethodType, etc.
3634                                    Register recv,    // if caller wants to see it
3635                                    Register flags    // if caller wants to test it
3636                                    ) {
3637   // determine flags
3638   const Bytecodes::Code code = bytecode();
3639   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3640   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3641   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3642   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3643   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3644   const bool load_receiver       = (recv  != noreg);
3645   const bool save_flags          = (flags != noreg);
3646   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3647   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3648   assert(flags == noreg || flags == rdx, "");
3649   assert(recv  == noreg || recv  == rcx, "");
3650 
3651   // setup registers & access constant pool cache
3652   if (recv  == noreg)  recv  = rcx;
3653   if (flags == noreg)  flags = rdx;
3654   assert_different_registers(method, index, recv, flags);
3655 
3656   // save 'interpreter return address'
3657   __ save_bcp();
3658 
3659   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3660 
3661   // maybe push appendix to arguments (just before return address)
3662   if (is_invokedynamic || is_invokehandle) {
3663     Label L_no_push;
3664     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3665     __ jcc(Assembler::zero, L_no_push);
3666     // Push the appendix as a trailing parameter.
3667     // This must be done before we get the receiver,
3668     // since the parameter_size includes it.
3669     __ push(rbx);
3670     __ mov(rbx, index);
3671     __ load_resolved_reference_at_index(index, rbx);
3672     __ pop(rbx);
3673     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3674     __ bind(L_no_push);
3675   }
3676 
3677   // load receiver if needed (after appendix is pushed so parameter size is correct)
3678   // Note: no return address pushed yet
3679   if (load_receiver) {
3680     __ movl(recv, flags);
3681     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3682     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3683     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3684     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3685     __ movptr(recv, recv_addr);
3686     __ verify_oop(recv);
3687   }
3688 
3689   if (save_flags) {
3690     __ movl(rbcp, flags);
3691   }
3692 
3693   // compute return type
3694   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3695   // Make sure we don't need to mask flags after the above shift
3696   ConstantPoolCacheEntry::verify_tos_state_shift();
3697   // load return address
3698   {
3699     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3700     ExternalAddress table(table_addr);
3701     LP64_ONLY(__ lea(rscratch1, table));
3702     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3703     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3704   }
3705 
3706   // push return address
3707   __ push(flags);
3708 
3709   // Restore flags value from the constant pool cache, and restore rsi
3710   // for later null checks.  r13 is the bytecode pointer
3711   if (save_flags) {
3712     __ movl(flags, rbcp);
3713     __ restore_bcp();
3714   }
3715 }
3716 
3717 void TemplateTable::invokevirtual_helper(Register index,
3718                                          Register recv,
3719                                          Register flags) {
3720   // Uses temporary registers rax, rdx
3721   assert_different_registers(index, recv, rax, rdx);
3722   assert(index == rbx, "");
3723   assert(recv  == rcx, "");
3724 
3725   // Test for an invoke of a final method
3726   Label notFinal;
3727   __ movl(rax, flags);
3728   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3729   __ jcc(Assembler::zero, notFinal);
3730 
3731   const Register method = index;  // method must be rbx
3732   assert(method == rbx,
3733          "Method* must be rbx for interpreter calling convention");
3734 
3735   // do the call - the index is actually the method to call
3736   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3737 
3738   // It's final, need a null check here!
3739   __ null_check(recv);
3740 
3741   // profile this call
3742   __ profile_final_call(rax);
3743   __ profile_arguments_type(rax, method, rbcp, true);
3744 
3745   __ jump_from_interpreted(method, rax);
3746 
3747   __ bind(notFinal);
3748 
3749   // get receiver klass
3750   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3751   __ load_klass(rax, recv);
3752 
3753   // profile this call
3754   __ profile_virtual_call(rax, rlocals, rdx);
3755   // get target Method* & entry point
3756   __ lookup_virtual_method(rax, index, method);
3757   __ profile_called_method(method, rdx, rbcp);
3758 
3759   __ profile_arguments_type(rdx, method, rbcp, true);
3760   __ jump_from_interpreted(method, rdx);
3761 }
3762 
3763 void TemplateTable::invokevirtual(int byte_no) {
3764   transition(vtos, vtos);
3765   assert(byte_no == f2_byte, "use this argument");
3766   prepare_invoke(byte_no,
3767                  rbx,    // method or vtable index
3768                  noreg,  // unused itable index
3769                  rcx, rdx); // recv, flags
3770 
3771   // rbx: index
3772   // rcx: receiver
3773   // rdx: flags
3774 
3775   invokevirtual_helper(rbx, rcx, rdx);
3776 }
3777 
3778 void TemplateTable::invokespecial(int byte_no) {
3779   transition(vtos, vtos);
3780   assert(byte_no == f1_byte, "use this argument");
3781   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3782                  rcx);  // get receiver also for null check
3783   __ verify_oop(rcx);
3784   __ null_check(rcx);
3785   // do the call
3786   __ profile_call(rax);
3787   __ profile_arguments_type(rax, rbx, rbcp, false);
3788   __ jump_from_interpreted(rbx, rax);
3789 }
3790 
3791 void TemplateTable::invokestatic(int byte_no) {
3792   transition(vtos, vtos);
3793   assert(byte_no == f1_byte, "use this argument");
3794   prepare_invoke(byte_no, rbx);  // get f1 Method*
3795   // do the call
3796   __ profile_call(rax);
3797   __ profile_arguments_type(rax, rbx, rbcp, false);
3798   __ jump_from_interpreted(rbx, rax);
3799 }
3800 
3801 
3802 void TemplateTable::fast_invokevfinal(int byte_no) {
3803   transition(vtos, vtos);
3804   assert(byte_no == f2_byte, "use this argument");
3805   __ stop("fast_invokevfinal not used on x86");
3806 }
3807 
3808 
3809 void TemplateTable::invokeinterface(int byte_no) {
3810   transition(vtos, vtos);
3811   assert(byte_no == f1_byte, "use this argument");
3812   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 Method*
3813                  rcx, rdx); // recv, flags
3814 
3815   // rax: reference klass (from f1) if interface method
3816   // rbx: method (from f2)
3817   // rcx: receiver
3818   // rdx: flags
3819 
3820   // First check for Object case, then private interface method,
3821   // then regular interface method.
3822 
3823   // Special case of invokeinterface called for virtual method of
3824   // java.lang.Object.  See cpCache.cpp for details.
3825   Label notObjectMethod;
3826   __ movl(rlocals, rdx);
3827   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3828   __ jcc(Assembler::zero, notObjectMethod);
3829   invokevirtual_helper(rbx, rcx, rdx);
3830   // no return from above
3831   __ bind(notObjectMethod);
3832 
3833   Label no_such_interface; // for receiver subtype check
3834   Register recvKlass; // used for exception processing
3835 
3836   // Check for private method invocation - indicated by vfinal
3837   Label notVFinal;
3838   __ movl(rlocals, rdx);
3839   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3840   __ jcc(Assembler::zero, notVFinal);
3841 
3842   // Get receiver klass into rlocals - also a null check
3843   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3844   __ load_klass(rlocals, rcx);
3845 
3846   Label subtype;
3847   __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3848   // If we get here the typecheck failed
3849   recvKlass = rdx;
3850   __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3851   __ jmp(no_such_interface);
3852 
3853   __ bind(subtype);
3854 
3855   // do the call - rbx is actually the method to call
3856 
3857   __ profile_final_call(rdx);
3858   __ profile_arguments_type(rdx, rbx, rbcp, true);
3859 
3860   __ jump_from_interpreted(rbx, rdx);
3861   // no return from above
3862   __ bind(notVFinal);
3863 
3864   // Get receiver klass into rdx - also a null check
3865   __ restore_locals();  // restore r14
3866   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3867   __ load_klass(rdx, rcx);
3868 
3869   Label no_such_method;
3870 
3871   // Preserve method for throw_AbstractMethodErrorVerbose.
3872   __ mov(rcx, rbx);
3873   // Receiver subtype check against REFC.
3874   // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3875   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3876                              rdx, rax, noreg,
3877                              // outputs: scan temp. reg, scan temp. reg
3878                              rbcp, rlocals,
3879                              no_such_interface,
3880                              /*return_method=*/false);
3881 
3882   // profile this call
3883   __ restore_bcp(); // rbcp was destroyed by receiver type check
3884   __ profile_virtual_call(rdx, rbcp, rlocals);
3885 
3886   // Get declaring interface class from method, and itable index
3887   __ load_method_holder(rax, rbx);
3888   __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3889   __ subl(rbx, Method::itable_index_max);
3890   __ negl(rbx);
3891 
3892   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3893   __ mov(rlocals, rdx);
3894   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3895                              rlocals, rax, rbx,
3896                              // outputs: method, scan temp. reg
3897                              rbx, rbcp,
3898                              no_such_interface);
3899 
3900   // rbx: Method* to call
3901   // rcx: receiver
3902   // Check for abstract method error
3903   // Note: This should be done more efficiently via a throw_abstract_method_error
3904   //       interpreter entry point and a conditional jump to it in case of a null
3905   //       method.
3906   __ testptr(rbx, rbx);
3907   __ jcc(Assembler::zero, no_such_method);
3908 
3909   __ profile_called_method(rbx, rbcp, rdx);
3910   __ profile_arguments_type(rdx, rbx, rbcp, true);
3911 
3912   // do the call
3913   // rcx: receiver
3914   // rbx,: Method*
3915   __ jump_from_interpreted(rbx, rdx);
3916   __ should_not_reach_here();
3917 
3918   // exception handling code follows...
3919   // note: must restore interpreter registers to canonical
3920   //       state for exception handling to work correctly!
3921 
3922   __ bind(no_such_method);
3923   // throw exception
3924   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3925   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3926   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3927   // Pass arguments for generating a verbose error message.
3928 #ifdef _LP64
3929   recvKlass = c_rarg1;
3930   Register method    = c_rarg2;
3931   if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3932   if (method != rcx)    { __ movq(method, rcx);    }
3933 #else
3934   recvKlass = rdx;
3935   Register method    = rcx;
3936 #endif
3937   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3938              recvKlass, method);
3939   // The call_VM checks for exception, so we should never return here.
3940   __ should_not_reach_here();
3941 
3942   __ bind(no_such_interface);
3943   // throw exception
3944   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3945   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3946   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3947   // Pass arguments for generating a verbose error message.
3948   LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3949   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3950              recvKlass, rax);
3951   // the call_VM checks for exception, so we should never return here.
3952   __ should_not_reach_here();
3953 }
3954 
3955 void TemplateTable::invokehandle(int byte_no) {
3956   transition(vtos, vtos);
3957   assert(byte_no == f1_byte, "use this argument");
3958   const Register rbx_method = rbx;
3959   const Register rax_mtype  = rax;
3960   const Register rcx_recv   = rcx;
3961   const Register rdx_flags  = rdx;
3962 
3963   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3964   __ verify_method_ptr(rbx_method);
3965   __ verify_oop(rcx_recv);
3966   __ null_check(rcx_recv);
3967 
3968   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3969   // rbx: MH.invokeExact_MT method (from f2)
3970 
3971   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3972 
3973   // FIXME: profile the LambdaForm also
3974   __ profile_final_call(rax);
3975   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3976 
3977   __ jump_from_interpreted(rbx_method, rdx);
3978 }
3979 
3980 void TemplateTable::invokedynamic(int byte_no) {
3981   transition(vtos, vtos);
3982   assert(byte_no == f1_byte, "use this argument");
3983 
3984   const Register rbx_method   = rbx;
3985   const Register rax_callsite = rax;
3986 
3987   prepare_invoke(byte_no, rbx_method, rax_callsite);
3988 
3989   // rax: CallSite object (from cpool->resolved_references[f1])
3990   // rbx: MH.linkToCallSite method (from f2)
3991 
3992   // Note:  rax_callsite is already pushed by prepare_invoke
3993 
3994   // %%% should make a type profile for any invokedynamic that takes a ref argument
3995   // profile this call
3996   __ profile_call(rbcp);
3997   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3998 
3999   __ verify_oop(rax_callsite);
4000 
4001   __ jump_from_interpreted(rbx_method, rdx);
4002 }
4003 
4004 //-----------------------------------------------------------------------------
4005 // Allocation
4006 
4007 void TemplateTable::_new() {
4008   transition(vtos, atos);
4009   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4010   Label slow_case;
4011   Label slow_case_no_pop;
4012   Label done;
4013   Label initialize_header;
4014   Label initialize_object;  // including clearing the fields
4015 
4016   __ get_cpool_and_tags(rcx, rax);
4017 
4018   // Make sure the class we're about to instantiate has been resolved.
4019   // This is done before loading InstanceKlass to be consistent with the order
4020   // how Constant Pool is updated (see ConstantPool::klass_at_put)
4021   const int tags_offset = Array<u1>::base_offset_in_bytes();
4022   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4023   __ jcc(Assembler::notEqual, slow_case_no_pop);
4024 
4025   // get InstanceKlass
4026   __ load_resolved_klass_at_index(rcx, rcx, rdx);
4027   __ push(rcx);  // save the contexts of klass for initializing the header
4028 
4029   // make sure klass is initialized & doesn't have finalizer
4030   // make sure klass is fully initialized
4031   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4032   __ jcc(Assembler::notEqual, slow_case);
4033 
4034   // get instance_size in InstanceKlass (scaled to a count of bytes)
4035   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4036   // test to see if it has a finalizer or is malformed in some way
4037   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4038   __ jcc(Assembler::notZero, slow_case);
4039 
4040   // Allocate the instance:
4041   //  If TLAB is enabled:
4042   //    Try to allocate in the TLAB.
4043   //    If fails, go to the slow path.
4044   //  Else If inline contiguous allocations are enabled:
4045   //    Try to allocate in eden.
4046   //    If fails due to heap end, go to slow path.
4047   //
4048   //  If TLAB is enabled OR inline contiguous is enabled:
4049   //    Initialize the allocation.
4050   //    Exit.
4051   //
4052   //  Go to slow path.
4053 
4054   const bool allow_shared_alloc =
4055     Universe::heap()->supports_inline_contig_alloc();
4056 
4057   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4058 #ifndef _LP64
4059   if (UseTLAB || allow_shared_alloc) {
4060     __ get_thread(thread);
4061   }
4062 #endif // _LP64
4063 
4064   if (UseTLAB) {
4065     __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
4066     if (ZeroTLAB) {
4067       // the fields have been already cleared
4068       __ jmp(initialize_header);
4069     } else {
4070       // initialize both the header and fields
4071       __ jmp(initialize_object);
4072     }
4073   } else {
4074     // Allocation in the shared Eden, if allowed.
4075     //
4076     // rdx: instance size in bytes
4077     __ eden_allocate(thread, rax, rdx, 0, rbx, slow_case);
4078   }
4079 
4080   // If UseTLAB or allow_shared_alloc are true, the object is created above and
4081   // there is an initialize need. Otherwise, skip and go to the slow path.
4082   if (UseTLAB || allow_shared_alloc) {
4083     // The object is initialized before the header.  If the object size is
4084     // zero, go directly to the header initialization.
4085     __ bind(initialize_object);
4086     __ decrement(rdx, sizeof(oopDesc));
4087     __ jcc(Assembler::zero, initialize_header);
4088 
4089     // Initialize topmost object field, divide rdx by 8, check if odd and
4090     // test if zero.
4091     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
4092     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4093 
4094     // rdx must have been multiple of 8
4095 #ifdef ASSERT
4096     // make sure rdx was multiple of 8
4097     Label L;
4098     // Ignore partial flag stall after shrl() since it is debug VM
4099     __ jcc(Assembler::carryClear, L);
4100     __ stop("object size is not multiple of 2 - adjust this code");
4101     __ bind(L);
4102     // rdx must be > 0, no extra check needed here
4103 #endif
4104 
4105     // initialize remaining object fields: rdx was a multiple of 8
4106     { Label loop;
4107     __ bind(loop);
4108     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4109     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4110     __ decrement(rdx);
4111     __ jcc(Assembler::notZero, loop);
4112     }
4113 
4114     // initialize object header only.
4115     __ bind(initialize_header);
4116     if (UseBiasedLocking) {
4117       __ pop(rcx);   // get saved klass back in the register.
4118       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4119       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4120     } else {
4121       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4122                 (intptr_t)markOopDesc::prototype()); // header
4123       __ pop(rcx);   // get saved klass back in the register.
4124     }
4125 #ifdef _LP64
4126     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4127     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4128 #endif
4129     __ store_klass(rax, rcx);  // klass
4130 
4131     {
4132       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4133       // Trigger dtrace event for fastpath
4134       __ push(atos);
4135       __ call_VM_leaf(
4136            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4137       __ pop(atos);
4138     }
4139 
4140     __ jmp(done);
4141   }
4142 
4143   // slow case
4144   __ bind(slow_case);
4145   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4146   __ bind(slow_case_no_pop);
4147 
4148   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4149   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4150 
4151   __ get_constant_pool(rarg1);
4152   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4153   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4154    __ verify_oop(rax);
4155 
4156   // continue
4157   __ bind(done);
4158 }
4159 
4160 void TemplateTable::newarray() {
4161   transition(itos, atos);
4162   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4163   __ load_unsigned_byte(rarg1, at_bcp(1));
4164   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4165           rarg1, rax);
4166 }
4167 
4168 void TemplateTable::anewarray() {
4169   transition(itos, atos);
4170 
4171   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4172   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4173 
4174   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4175   __ get_constant_pool(rarg1);
4176   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4177           rarg1, rarg2, rax);
4178 }
4179 
4180 void TemplateTable::arraylength() {
4181   transition(atos, itos);
4182   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4183   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4184 }
4185 
4186 void TemplateTable::checkcast() {
4187   transition(atos, atos);
4188   Label done, is_null, ok_is_subtype, quicked, resolved;
4189   __ testptr(rax, rax); // object is in rax
4190   __ jcc(Assembler::zero, is_null);
4191 
4192   // Get cpool & tags index
4193   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4194   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4195   // See if bytecode has already been quicked
4196   __ cmpb(Address(rdx, rbx,
4197                   Address::times_1,
4198                   Array<u1>::base_offset_in_bytes()),
4199           JVM_CONSTANT_Class);
4200   __ jcc(Assembler::equal, quicked);
4201   __ push(atos); // save receiver for result, and for GC
4202   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4203 
4204   // vm_result_2 has metadata result
4205 #ifndef _LP64
4206   // borrow rdi from locals
4207   __ get_thread(rdi);
4208   __ get_vm_result_2(rax, rdi);
4209   __ restore_locals();
4210 #else
4211   __ get_vm_result_2(rax, r15_thread);
4212 #endif
4213 
4214   __ pop_ptr(rdx); // restore receiver
4215   __ jmpb(resolved);
4216 
4217   // Get superklass in rax and subklass in rbx
4218   __ bind(quicked);
4219   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4220   __ load_resolved_klass_at_index(rax, rcx, rbx);
4221 
4222   __ bind(resolved);
4223   __ load_klass(rbx, rdx);
4224 
4225   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4226   // Superklass in rax.  Subklass in rbx.
4227   __ gen_subtype_check(rbx, ok_is_subtype);
4228 
4229   // Come here on failure
4230   __ push_ptr(rdx);
4231   // object is at TOS
4232   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4233 
4234   // Come here on success
4235   __ bind(ok_is_subtype);
4236   __ mov(rax, rdx); // Restore object in rdx
4237 
4238   // Collect counts on whether this check-cast sees NULLs a lot or not.
4239   if (ProfileInterpreter) {
4240     __ jmp(done);
4241     __ bind(is_null);
4242     __ profile_null_seen(rcx);
4243   } else {
4244     __ bind(is_null);   // same as 'done'
4245   }
4246   __ bind(done);
4247 }
4248 
4249 void TemplateTable::instanceof() {
4250   transition(atos, itos);
4251   Label done, is_null, ok_is_subtype, quicked, resolved;
4252   __ testptr(rax, rax);
4253   __ jcc(Assembler::zero, is_null);
4254 
4255   // Get cpool & tags index
4256   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4257   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4258   // See if bytecode has already been quicked
4259   __ cmpb(Address(rdx, rbx,
4260                   Address::times_1,
4261                   Array<u1>::base_offset_in_bytes()),
4262           JVM_CONSTANT_Class);
4263   __ jcc(Assembler::equal, quicked);
4264 
4265   __ push(atos); // save receiver for result, and for GC
4266   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4267   // vm_result_2 has metadata result
4268 
4269 #ifndef _LP64
4270   // borrow rdi from locals
4271   __ get_thread(rdi);
4272   __ get_vm_result_2(rax, rdi);
4273   __ restore_locals();
4274 #else
4275   __ get_vm_result_2(rax, r15_thread);
4276 #endif
4277 
4278   __ pop_ptr(rdx); // restore receiver
4279   __ verify_oop(rdx);
4280   __ load_klass(rdx, rdx);
4281   __ jmpb(resolved);
4282 
4283   // Get superklass in rax and subklass in rdx
4284   __ bind(quicked);
4285   __ load_klass(rdx, rax);
4286   __ load_resolved_klass_at_index(rax, rcx, rbx);
4287 
4288   __ bind(resolved);
4289 
4290   // Generate subtype check.  Blows rcx, rdi
4291   // Superklass in rax.  Subklass in rdx.
4292   __ gen_subtype_check(rdx, ok_is_subtype);
4293 
4294   // Come here on failure
4295   __ xorl(rax, rax);
4296   __ jmpb(done);
4297   // Come here on success
4298   __ bind(ok_is_subtype);
4299   __ movl(rax, 1);
4300 
4301   // Collect counts on whether this test sees NULLs a lot or not.
4302   if (ProfileInterpreter) {
4303     __ jmp(done);
4304     __ bind(is_null);
4305     __ profile_null_seen(rcx);
4306   } else {
4307     __ bind(is_null);   // same as 'done'
4308   }
4309   __ bind(done);
4310   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4311   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4312 }
4313 
4314 
4315 //----------------------------------------------------------------------------------------------------
4316 // Breakpoints
4317 void TemplateTable::_breakpoint() {
4318   // Note: We get here even if we are single stepping..
4319   // jbug insists on setting breakpoints at every bytecode
4320   // even if we are in single step mode.
4321 
4322   transition(vtos, vtos);
4323 
4324   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4325 
4326   // get the unpatched byte code
4327   __ get_method(rarg);
4328   __ call_VM(noreg,
4329              CAST_FROM_FN_PTR(address,
4330                               InterpreterRuntime::get_original_bytecode_at),
4331              rarg, rbcp);
4332   __ mov(rbx, rax);  // why?
4333 
4334   // post the breakpoint event
4335   __ get_method(rarg);
4336   __ call_VM(noreg,
4337              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4338              rarg, rbcp);
4339 
4340   // complete the execution of original bytecode
4341   __ dispatch_only_normal(vtos);
4342 }
4343 
4344 //-----------------------------------------------------------------------------
4345 // Exceptions
4346 
4347 void TemplateTable::athrow() {
4348   transition(atos, vtos);
4349   __ null_check(rax);
4350   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4351 }
4352 
4353 //-----------------------------------------------------------------------------
4354 // Synchronization
4355 //
4356 // Note: monitorenter & exit are symmetric routines; which is reflected
4357 //       in the assembly code structure as well
4358 //
4359 // Stack layout:
4360 //
4361 // [expressions  ] <--- rsp               = expression stack top
4362 // ..
4363 // [expressions  ]
4364 // [monitor entry] <--- monitor block top = expression stack bot
4365 // ..
4366 // [monitor entry]
4367 // [frame data   ] <--- monitor block bot
4368 // ...
4369 // [saved rbp    ] <--- rbp
4370 void TemplateTable::monitorenter() {
4371   transition(atos, vtos);
4372 
4373   // check for NULL object
4374   __ null_check(rax);
4375 
4376   __ resolve(IS_NOT_NULL, rax);
4377 
4378   const Address monitor_block_top(
4379         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4380   const Address monitor_block_bot(
4381         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4382   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4383 
4384   Label allocated;
4385 
4386   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4387   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4388   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4389 
4390   // initialize entry pointer
4391   __ xorl(rmon, rmon); // points to free slot or NULL
4392 
4393   // find a free slot in the monitor block (result in rmon)
4394   {
4395     Label entry, loop, exit;
4396     __ movptr(rtop, monitor_block_top); // points to current entry,
4397                                         // starting with top-most entry
4398     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4399                                         // of monitor block
4400     __ jmpb(entry);
4401 
4402     __ bind(loop);
4403     // check if current entry is used
4404     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4405     // if not used then remember entry in rmon
4406     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4407     // check if current entry is for same object
4408     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4409     // if same object then stop searching
4410     __ jccb(Assembler::equal, exit);
4411     // otherwise advance to next entry
4412     __ addptr(rtop, entry_size);
4413     __ bind(entry);
4414     // check if bottom reached
4415     __ cmpptr(rtop, rbot);
4416     // if not at bottom then check this entry
4417     __ jcc(Assembler::notEqual, loop);
4418     __ bind(exit);
4419   }
4420 
4421   __ testptr(rmon, rmon); // check if a slot has been found
4422   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4423 
4424   // allocate one if there's no free slot
4425   {
4426     Label entry, loop;
4427     // 1. compute new pointers          // rsp: old expression stack top
4428     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4429     __ subptr(rsp, entry_size);         // move expression stack top
4430     __ subptr(rmon, entry_size);        // move expression stack bottom
4431     __ mov(rtop, rsp);                  // set start value for copy loop
4432     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4433     __ jmp(entry);
4434     // 2. move expression stack contents
4435     __ bind(loop);
4436     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4437                                                 // word from old location
4438     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4439     __ addptr(rtop, wordSize);                  // advance to next word
4440     __ bind(entry);
4441     __ cmpptr(rtop, rmon);                      // check if bottom reached
4442     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4443                                                 // copy next word
4444   }
4445 
4446   // call run-time routine
4447   // rmon: points to monitor entry
4448   __ bind(allocated);
4449 
4450   // Increment bcp to point to the next bytecode, so exception
4451   // handling for async. exceptions work correctly.
4452   // The object has already been poped from the stack, so the
4453   // expression stack looks correct.
4454   __ increment(rbcp);
4455 
4456   // store object
4457   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4458   __ lock_object(rmon);
4459 
4460   // check to make sure this monitor doesn't cause stack overflow after locking
4461   __ save_bcp();  // in case of exception
4462   __ generate_stack_overflow_check(0);
4463 
4464   // The bcp has already been incremented. Just need to dispatch to
4465   // next instruction.
4466   __ dispatch_next(vtos);
4467 }
4468 
4469 void TemplateTable::monitorexit() {
4470   transition(atos, vtos);
4471 
4472   // check for NULL object
4473   __ null_check(rax);
4474 
4475   __ resolve(IS_NOT_NULL, rax);
4476 
4477   const Address monitor_block_top(
4478         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4479   const Address monitor_block_bot(
4480         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4481   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4482 
4483   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4484   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4485 
4486   Label found;
4487 
4488   // find matching slot
4489   {
4490     Label entry, loop;
4491     __ movptr(rtop, monitor_block_top); // points to current entry,
4492                                         // starting with top-most entry
4493     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4494                                         // of monitor block
4495     __ jmpb(entry);
4496 
4497     __ bind(loop);
4498     // check if current entry is for same object
4499     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4500     // if same object then stop searching
4501     __ jcc(Assembler::equal, found);
4502     // otherwise advance to next entry
4503     __ addptr(rtop, entry_size);
4504     __ bind(entry);
4505     // check if bottom reached
4506     __ cmpptr(rtop, rbot);
4507     // if not at bottom then check this entry
4508     __ jcc(Assembler::notEqual, loop);
4509   }
4510 
4511   // error handling. Unlocking was not block-structured
4512   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4513                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4514   __ should_not_reach_here();
4515 
4516   // call run-time routine
4517   __ bind(found);
4518   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4519   __ unlock_object(rtop);
4520   __ pop_ptr(rax); // discard object
4521 }
4522 
4523 // Wide instructions
4524 void TemplateTable::wide() {
4525   transition(vtos, vtos);
4526   __ load_unsigned_byte(rbx, at_bcp(1));
4527   ExternalAddress wtable((address)Interpreter::_wentry_point);
4528   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4529   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4530 }
4531 
4532 // Multi arrays
4533 void TemplateTable::multianewarray() {
4534   transition(vtos, atos);
4535 
4536   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4537   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4538   // last dim is on top of stack; we want address of first one:
4539   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4540   // the latter wordSize to point to the beginning of the array.
4541   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4542   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4543   __ load_unsigned_byte(rbx, at_bcp(3));
4544   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4545 }