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   assert(_desc->calls_vm(),
2641          "inconsistent calls_vm information"); // call in remove_activation
2642 
2643   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2644     assert(state == vtos, "only valid state");
2645     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2646     __ movptr(robj, aaddress(0));
2647     __ load_klass(rdi, robj);
2648     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2649     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2650     Label skip_register_finalizer;
2651     __ jcc(Assembler::zero, skip_register_finalizer);
2652 
2653     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2654 
2655     __ bind(skip_register_finalizer);
2656   }
2657 
2658   if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2659     Label no_safepoint;
2660     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2661 #ifdef _LP64
2662     __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2663 #else
2664     const Register thread = rdi;
2665     __ get_thread(thread);
2666     __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2667 #endif
2668     __ jcc(Assembler::zero, no_safepoint);
2669     __ push(state);
2670     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2671                                     InterpreterRuntime::at_safepoint));
2672     __ pop(state);
2673     __ bind(no_safepoint);
2674   }
2675 
2676   // Narrow result if state is itos but result type is smaller.
2677   // Need to narrow in the return bytecode rather than in generate_return_entry
2678   // since compiled code callers expect the result to already be narrowed.
2679   if (state == itos) {
2680     __ narrow(rax);
2681   }
2682   __ remove_activation(state, rbcp);
2683 
2684   __ jmp(rbcp);
2685 }
2686 
2687 // ----------------------------------------------------------------------------
2688 // Volatile variables demand their effects be made known to all CPU's
2689 // in order.  Store buffers on most chips allow reads & writes to
2690 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2691 // without some kind of memory barrier (i.e., it's not sufficient that
2692 // the interpreter does not reorder volatile references, the hardware
2693 // also must not reorder them).
2694 //
2695 // According to the new Java Memory Model (JMM):
2696 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2697 //     writes act as aquire & release, so:
2698 // (2) A read cannot let unrelated NON-volatile memory refs that
2699 //     happen after the read float up to before the read.  It's OK for
2700 //     non-volatile memory refs that happen before the volatile read to
2701 //     float down below it.
2702 // (3) Similar a volatile write cannot let unrelated NON-volatile
2703 //     memory refs that happen BEFORE the write float down to after the
2704 //     write.  It's OK for non-volatile memory refs that happen after the
2705 //     volatile write to float up before it.
2706 //
2707 // We only put in barriers around volatile refs (they are expensive),
2708 // not _between_ memory refs (that would require us to track the
2709 // flavor of the previous memory refs).  Requirements (2) and (3)
2710 // require some barriers before volatile stores and after volatile
2711 // loads.  These nearly cover requirement (1) but miss the
2712 // volatile-store-volatile-load case.  This final case is placed after
2713 // volatile-stores although it could just as well go before
2714 // volatile-loads.
2715 
2716 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2717   // Helper function to insert a is-volatile test and memory barrier
2718   __ membar(order_constraint);
2719 }
2720 
2721 void TemplateTable::resolve_cache_and_index(int byte_no,
2722                                             Register cache,
2723                                             Register index,
2724                                             size_t index_size) {
2725   const Register temp = rbx;
2726   assert_different_registers(cache, index, temp);
2727 
2728   Label L_clinit_barrier_slow;
2729   Label resolved;
2730 
2731   Bytecodes::Code code = bytecode();
2732   switch (code) {
2733   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2734   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2735   default: break;
2736   }
2737 
2738   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2739   __ get_cache_and_index_and_bytecode_at_bcp(cache, index, temp, byte_no, 1, index_size);
2740   __ cmpl(temp, code);  // have we resolved this bytecode?
2741   __ jcc(Assembler::equal, resolved);
2742 
2743   // resolve first time through
2744   // Class initialization barrier slow path lands here as well.
2745   __ bind(L_clinit_barrier_slow);
2746   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2747   __ movl(temp, code);
2748   __ call_VM(noreg, entry, temp);
2749   // Update registers with resolved info
2750   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
2751 
2752   __ bind(resolved);
2753 
2754   // Class initialization barrier for static methods
2755   if (UseFastClassInitChecks && bytecode() == Bytecodes::_invokestatic) {
2756     const Register method = temp;
2757     const Register klass  = temp;
2758     const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2759     assert(thread != noreg, "x86_32 not supported");
2760 
2761     __ load_resolved_method_at_index(byte_no, cache, index, method);
2762     __ load_method_holder(klass, method);
2763     __ clinit_barrier(klass, thread, NULL /*L_fast_path*/, &L_clinit_barrier_slow);
2764   }
2765 }
2766 
2767 // The cache and index registers must be set before call
2768 void TemplateTable::load_field_cp_cache_entry(Register obj,
2769                                               Register cache,
2770                                               Register index,
2771                                               Register off,
2772                                               Register flags,
2773                                               bool is_static = false) {
2774   assert_different_registers(cache, index, flags, off);
2775 
2776   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2777   // Field offset
2778   __ movptr(off, Address(cache, index, Address::times_ptr,
2779                          in_bytes(cp_base_offset +
2780                                   ConstantPoolCacheEntry::f2_offset())));
2781   // Flags
2782   __ movl(flags, Address(cache, index, Address::times_ptr,
2783                          in_bytes(cp_base_offset +
2784                                   ConstantPoolCacheEntry::flags_offset())));
2785 
2786   // klass overwrite register
2787   if (is_static) {
2788     __ movptr(obj, Address(cache, index, Address::times_ptr,
2789                            in_bytes(cp_base_offset +
2790                                     ConstantPoolCacheEntry::f1_offset())));
2791     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2792     __ movptr(obj, Address(obj, mirror_offset));
2793     __ resolve_oop_handle(obj);
2794   }
2795 }
2796 
2797 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2798                                                Register method,
2799                                                Register itable_index,
2800                                                Register flags,
2801                                                bool is_invokevirtual,
2802                                                bool is_invokevfinal, /*unused*/
2803                                                bool is_invokedynamic) {
2804   // setup registers
2805   const Register cache = rcx;
2806   const Register index = rdx;
2807   assert_different_registers(method, flags);
2808   assert_different_registers(method, cache, index);
2809   assert_different_registers(itable_index, flags);
2810   assert_different_registers(itable_index, cache, index);
2811   // determine constant pool cache field offsets
2812   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2813   const int method_offset = in_bytes(
2814     ConstantPoolCache::base_offset() +
2815       ((byte_no == f2_byte)
2816        ? ConstantPoolCacheEntry::f2_offset()
2817        : ConstantPoolCacheEntry::f1_offset()));
2818   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2819                                     ConstantPoolCacheEntry::flags_offset());
2820   // access constant pool cache fields
2821   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2822                                     ConstantPoolCacheEntry::f2_offset());
2823 
2824   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2825   resolve_cache_and_index(byte_no, cache, index, index_size);
2826   __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2827 
2828   if (itable_index != noreg) {
2829     // pick up itable or appendix index from f2 also:
2830     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2831   }
2832   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2833 }
2834 
2835 // The registers cache and index expected to be set before call.
2836 // Correct values of the cache and index registers are preserved.
2837 void TemplateTable::jvmti_post_field_access(Register cache,
2838                                             Register index,
2839                                             bool is_static,
2840                                             bool has_tos) {
2841   if (JvmtiExport::can_post_field_access()) {
2842     // Check to see if a field access watch has been set before we take
2843     // the time to call into the VM.
2844     Label L1;
2845     assert_different_registers(cache, index, rax);
2846     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2847     __ testl(rax,rax);
2848     __ jcc(Assembler::zero, L1);
2849 
2850     // cache entry pointer
2851     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2852     __ shll(index, LogBytesPerWord);
2853     __ addptr(cache, index);
2854     if (is_static) {
2855       __ xorptr(rax, rax);      // NULL object reference
2856     } else {
2857       __ pop(atos);         // Get the object
2858       __ verify_oop(rax);
2859       __ push(atos);        // Restore stack state
2860     }
2861     // rax,:   object pointer or NULL
2862     // cache: cache entry pointer
2863     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2864                rax, cache);
2865     __ get_cache_and_index_at_bcp(cache, index, 1);
2866     __ bind(L1);
2867   }
2868 }
2869 
2870 void TemplateTable::pop_and_check_object(Register r) {
2871   __ pop_ptr(r);
2872   __ null_check(r);  // for field access must check obj.
2873   __ verify_oop(r);
2874 }
2875 
2876 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2877   transition(vtos, vtos);
2878 
2879   const Register cache = rcx;
2880   const Register index = rdx;
2881   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2882   const Register off   = rbx;
2883   const Register flags = rax;
2884   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2885 
2886   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2887   jvmti_post_field_access(cache, index, is_static, false);
2888   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2889 
2890   if (!is_static) pop_and_check_object(obj);
2891 
2892   const Address field(obj, off, Address::times_1, 0*wordSize);
2893 
2894   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2895 
2896   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2897   // Make sure we don't need to mask edx after the above shift
2898   assert(btos == 0, "change code, btos != 0");
2899 
2900   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2901 
2902   __ jcc(Assembler::notZero, notByte);
2903   // btos
2904   __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2905   __ push(btos);
2906   // Rewrite bytecode to be faster
2907   if (!is_static && rc == may_rewrite) {
2908     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2909   }
2910   __ jmp(Done);
2911 
2912   __ bind(notByte);
2913   __ cmpl(flags, ztos);
2914   __ jcc(Assembler::notEqual, notBool);
2915 
2916   // ztos (same code as btos)
2917   __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2918   __ push(ztos);
2919   // Rewrite bytecode to be faster
2920   if (!is_static && rc == may_rewrite) {
2921     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2922     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2923   }
2924   __ jmp(Done);
2925 
2926   __ bind(notBool);
2927   __ cmpl(flags, atos);
2928   __ jcc(Assembler::notEqual, notObj);
2929   // atos
2930   do_oop_load(_masm, field, rax);
2931   __ push(atos);
2932   if (!is_static && rc == may_rewrite) {
2933     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2934   }
2935   __ jmp(Done);
2936 
2937   __ bind(notObj);
2938   __ cmpl(flags, itos);
2939   __ jcc(Assembler::notEqual, notInt);
2940   // itos
2941   __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2942   __ push(itos);
2943   // Rewrite bytecode to be faster
2944   if (!is_static && rc == may_rewrite) {
2945     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2946   }
2947   __ jmp(Done);
2948 
2949   __ bind(notInt);
2950   __ cmpl(flags, ctos);
2951   __ jcc(Assembler::notEqual, notChar);
2952   // ctos
2953   __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2954   __ push(ctos);
2955   // Rewrite bytecode to be faster
2956   if (!is_static && rc == may_rewrite) {
2957     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2958   }
2959   __ jmp(Done);
2960 
2961   __ bind(notChar);
2962   __ cmpl(flags, stos);
2963   __ jcc(Assembler::notEqual, notShort);
2964   // stos
2965   __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2966   __ push(stos);
2967   // Rewrite bytecode to be faster
2968   if (!is_static && rc == may_rewrite) {
2969     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2970   }
2971   __ jmp(Done);
2972 
2973   __ bind(notShort);
2974   __ cmpl(flags, ltos);
2975   __ jcc(Assembler::notEqual, notLong);
2976   // ltos
2977     // Generate code as if volatile (x86_32).  There just aren't enough registers to
2978     // save that information and this code is faster than the test.
2979   __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2980   __ push(ltos);
2981   // Rewrite bytecode to be faster
2982   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2983   __ jmp(Done);
2984 
2985   __ bind(notLong);
2986   __ cmpl(flags, ftos);
2987   __ jcc(Assembler::notEqual, notFloat);
2988   // ftos
2989 
2990   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2991   __ push(ftos);
2992   // Rewrite bytecode to be faster
2993   if (!is_static && rc == may_rewrite) {
2994     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2995   }
2996   __ jmp(Done);
2997 
2998   __ bind(notFloat);
2999 #ifdef ASSERT
3000   Label notDouble;
3001   __ cmpl(flags, dtos);
3002   __ jcc(Assembler::notEqual, notDouble);
3003 #endif
3004   // dtos
3005   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3006   __ push(dtos);
3007   // Rewrite bytecode to be faster
3008   if (!is_static && rc == may_rewrite) {
3009     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3010   }
3011 #ifdef ASSERT
3012   __ jmp(Done);
3013 
3014   __ bind(notDouble);
3015   __ stop("Bad state");
3016 #endif
3017 
3018   __ bind(Done);
3019   // [jk] not needed currently
3020   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3021   //                                              Assembler::LoadStore));
3022 }
3023 
3024 void TemplateTable::getfield(int byte_no) {
3025   getfield_or_static(byte_no, false);
3026 }
3027 
3028 void TemplateTable::nofast_getfield(int byte_no) {
3029   getfield_or_static(byte_no, false, may_not_rewrite);
3030 }
3031 
3032 void TemplateTable::getstatic(int byte_no) {
3033   getfield_or_static(byte_no, true);
3034 }
3035 
3036 
3037 // The registers cache and index expected to be set before call.
3038 // The function may destroy various registers, just not the cache and index registers.
3039 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3040 
3041   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
3042   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
3043   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
3044   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3045 
3046   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3047 
3048   if (JvmtiExport::can_post_field_modification()) {
3049     // Check to see if a field modification watch has been set before
3050     // we take the time to call into the VM.
3051     Label L1;
3052     assert_different_registers(cache, index, rax);
3053     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3054     __ testl(rax, rax);
3055     __ jcc(Assembler::zero, L1);
3056 
3057     __ get_cache_and_index_at_bcp(robj, RDX, 1);
3058 
3059 
3060     if (is_static) {
3061       // Life is simple.  Null out the object pointer.
3062       __ xorl(RBX, RBX);
3063 
3064     } else {
3065       // Life is harder. The stack holds the value on top, followed by
3066       // the object.  We don't know the size of the value, though; it
3067       // could be one or two words depending on its type. As a result,
3068       // we must find the type to determine where the object is.
3069 #ifndef _LP64
3070       Label two_word, valsize_known;
3071 #endif
3072       __ movl(RCX, Address(robj, RDX,
3073                            Address::times_ptr,
3074                            in_bytes(cp_base_offset +
3075                                      ConstantPoolCacheEntry::flags_offset())));
3076       NOT_LP64(__ mov(rbx, rsp));
3077       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3078 
3079       // Make sure we don't need to mask rcx after the above shift
3080       ConstantPoolCacheEntry::verify_tos_state_shift();
3081 #ifdef _LP64
3082       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3083       __ cmpl(c_rarg3, ltos);
3084       __ cmovptr(Assembler::equal,
3085                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3086       __ cmpl(c_rarg3, dtos);
3087       __ cmovptr(Assembler::equal,
3088                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3089 #else
3090       __ cmpl(rcx, ltos);
3091       __ jccb(Assembler::equal, two_word);
3092       __ cmpl(rcx, dtos);
3093       __ jccb(Assembler::equal, two_word);
3094       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3095       __ jmpb(valsize_known);
3096 
3097       __ bind(two_word);
3098       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3099 
3100       __ bind(valsize_known);
3101       // setup object pointer
3102       __ movptr(rbx, Address(rbx, 0));
3103 #endif
3104     }
3105     // cache entry pointer
3106     __ addptr(robj, in_bytes(cp_base_offset));
3107     __ shll(RDX, LogBytesPerWord);
3108     __ addptr(robj, RDX);
3109     // object (tos)
3110     __ mov(RCX, rsp);
3111     // c_rarg1: object pointer set up above (NULL if static)
3112     // c_rarg2: cache entry pointer
3113     // c_rarg3: jvalue object on the stack
3114     __ call_VM(noreg,
3115                CAST_FROM_FN_PTR(address,
3116                                 InterpreterRuntime::post_field_modification),
3117                RBX, robj, RCX);
3118     __ get_cache_and_index_at_bcp(cache, index, 1);
3119     __ bind(L1);
3120   }
3121 }
3122 
3123 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3124   transition(vtos, vtos);
3125 
3126   const Register cache = rcx;
3127   const Register index = rdx;
3128   const Register obj   = rcx;
3129   const Register off   = rbx;
3130   const Register flags = rax;
3131 
3132   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3133   jvmti_post_field_mod(cache, index, is_static);
3134   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3135 
3136   // [jk] not needed currently
3137   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3138   //                                              Assembler::StoreStore));
3139 
3140   Label notVolatile, Done;
3141   __ movl(rdx, flags);
3142   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3143   __ andl(rdx, 0x1);
3144 
3145   // Check for volatile store
3146   __ testl(rdx, rdx);
3147   __ jcc(Assembler::zero, notVolatile);
3148 
3149   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3150   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3151                                                Assembler::StoreStore));
3152   __ jmp(Done);
3153   __ bind(notVolatile);
3154 
3155   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3156 
3157   __ bind(Done);
3158 }
3159 
3160 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3161                                               Register obj, Register off, Register flags) {
3162 
3163   // field addresses
3164   const Address field(obj, off, Address::times_1, 0*wordSize);
3165   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3166 
3167   Label notByte, notBool, notInt, notShort, notChar,
3168         notLong, notFloat, notObj;
3169   Label Done;
3170 
3171   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3172 
3173   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3174 
3175   assert(btos == 0, "change code, btos != 0");
3176   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3177   __ jcc(Assembler::notZero, notByte);
3178 
3179   // btos
3180   {
3181     __ pop(btos);
3182     if (!is_static) pop_and_check_object(obj);
3183     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3184     if (!is_static && rc == may_rewrite) {
3185       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3186     }
3187     __ jmp(Done);
3188   }
3189 
3190   __ bind(notByte);
3191   __ cmpl(flags, ztos);
3192   __ jcc(Assembler::notEqual, notBool);
3193 
3194   // ztos
3195   {
3196     __ pop(ztos);
3197     if (!is_static) pop_and_check_object(obj);
3198     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3199     if (!is_static && rc == may_rewrite) {
3200       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3201     }
3202     __ jmp(Done);
3203   }
3204 
3205   __ bind(notBool);
3206   __ cmpl(flags, atos);
3207   __ jcc(Assembler::notEqual, notObj);
3208 
3209   // atos
3210   {
3211     __ pop(atos);
3212     if (!is_static) pop_and_check_object(obj);
3213     // Store into the field
3214     do_oop_store(_masm, field, rax);
3215     if (!is_static && rc == may_rewrite) {
3216       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3217     }
3218     __ jmp(Done);
3219   }
3220 
3221   __ bind(notObj);
3222   __ cmpl(flags, itos);
3223   __ jcc(Assembler::notEqual, notInt);
3224 
3225   // itos
3226   {
3227     __ pop(itos);
3228     if (!is_static) pop_and_check_object(obj);
3229     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3230     if (!is_static && rc == may_rewrite) {
3231       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3232     }
3233     __ jmp(Done);
3234   }
3235 
3236   __ bind(notInt);
3237   __ cmpl(flags, ctos);
3238   __ jcc(Assembler::notEqual, notChar);
3239 
3240   // ctos
3241   {
3242     __ pop(ctos);
3243     if (!is_static) pop_and_check_object(obj);
3244     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3245     if (!is_static && rc == may_rewrite) {
3246       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3247     }
3248     __ jmp(Done);
3249   }
3250 
3251   __ bind(notChar);
3252   __ cmpl(flags, stos);
3253   __ jcc(Assembler::notEqual, notShort);
3254 
3255   // stos
3256   {
3257     __ pop(stos);
3258     if (!is_static) pop_and_check_object(obj);
3259     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3260     if (!is_static && rc == may_rewrite) {
3261       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3262     }
3263     __ jmp(Done);
3264   }
3265 
3266   __ bind(notShort);
3267   __ cmpl(flags, ltos);
3268   __ jcc(Assembler::notEqual, notLong);
3269 
3270   // ltos
3271   {
3272     __ pop(ltos);
3273     if (!is_static) pop_and_check_object(obj);
3274     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos*/, noreg, noreg);
3275 #ifdef _LP64
3276     if (!is_static && rc == may_rewrite) {
3277       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3278     }
3279 #endif // _LP64
3280     __ jmp(Done);
3281   }
3282 
3283   __ bind(notLong);
3284   __ cmpl(flags, ftos);
3285   __ jcc(Assembler::notEqual, notFloat);
3286 
3287   // ftos
3288   {
3289     __ pop(ftos);
3290     if (!is_static) pop_and_check_object(obj);
3291     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3292     if (!is_static && rc == may_rewrite) {
3293       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3294     }
3295     __ jmp(Done);
3296   }
3297 
3298   __ bind(notFloat);
3299 #ifdef ASSERT
3300   Label notDouble;
3301   __ cmpl(flags, dtos);
3302   __ jcc(Assembler::notEqual, notDouble);
3303 #endif
3304 
3305   // dtos
3306   {
3307     __ pop(dtos);
3308     if (!is_static) pop_and_check_object(obj);
3309     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3310     if (!is_static && rc == may_rewrite) {
3311       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3312     }
3313   }
3314 
3315 #ifdef ASSERT
3316   __ jmp(Done);
3317 
3318   __ bind(notDouble);
3319   __ stop("Bad state");
3320 #endif
3321 
3322   __ bind(Done);
3323 }
3324 
3325 void TemplateTable::putfield(int byte_no) {
3326   putfield_or_static(byte_no, false);
3327 }
3328 
3329 void TemplateTable::nofast_putfield(int byte_no) {
3330   putfield_or_static(byte_no, false, may_not_rewrite);
3331 }
3332 
3333 void TemplateTable::putstatic(int byte_no) {
3334   putfield_or_static(byte_no, true);
3335 }
3336 
3337 void TemplateTable::jvmti_post_fast_field_mod() {
3338 
3339   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3340 
3341   if (JvmtiExport::can_post_field_modification()) {
3342     // Check to see if a field modification watch has been set before
3343     // we take the time to call into the VM.
3344     Label L2;
3345     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3346     __ testl(scratch, scratch);
3347     __ jcc(Assembler::zero, L2);
3348     __ pop_ptr(rbx);                  // copy the object pointer from tos
3349     __ verify_oop(rbx);
3350     __ push_ptr(rbx);                 // put the object pointer back on tos
3351     // Save tos values before call_VM() clobbers them. Since we have
3352     // to do it for every data type, we use the saved values as the
3353     // jvalue object.
3354     switch (bytecode()) {          // load values into the jvalue object
3355     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3356     case Bytecodes::_fast_bputfield: // fall through
3357     case Bytecodes::_fast_zputfield: // fall through
3358     case Bytecodes::_fast_sputfield: // fall through
3359     case Bytecodes::_fast_cputfield: // fall through
3360     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3361     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3362     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3363     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3364 
3365     default:
3366       ShouldNotReachHere();
3367     }
3368     __ mov(scratch, rsp);             // points to jvalue on the stack
3369     // access constant pool cache entry
3370     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3371     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3372     __ verify_oop(rbx);
3373     // rbx: object pointer copied above
3374     // c_rarg2: cache entry pointer
3375     // c_rarg3: jvalue object on the stack
3376     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3377     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3378 
3379     switch (bytecode()) {             // restore tos values
3380     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3381     case Bytecodes::_fast_bputfield: // fall through
3382     case Bytecodes::_fast_zputfield: // fall through
3383     case Bytecodes::_fast_sputfield: // fall through
3384     case Bytecodes::_fast_cputfield: // fall through
3385     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3386     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3387     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3388     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3389     default: break;
3390     }
3391     __ bind(L2);
3392   }
3393 }
3394 
3395 void TemplateTable::fast_storefield(TosState state) {
3396   transition(state, vtos);
3397 
3398   ByteSize base = ConstantPoolCache::base_offset();
3399 
3400   jvmti_post_fast_field_mod();
3401 
3402   // access constant pool cache
3403   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3404 
3405   // test for volatile with rdx but rdx is tos register for lputfield.
3406   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3407                        in_bytes(base +
3408                                 ConstantPoolCacheEntry::flags_offset())));
3409 
3410   // replace index with field offset from cache entry
3411   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3412                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3413 
3414   // [jk] not needed currently
3415   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3416   //                                              Assembler::StoreStore));
3417 
3418   Label notVolatile, Done;
3419   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3420   __ andl(rdx, 0x1);
3421 
3422   // Get object from stack
3423   pop_and_check_object(rcx);
3424 
3425   // field address
3426   const Address field(rcx, rbx, Address::times_1);
3427 
3428   // Check for volatile store
3429   __ testl(rdx, rdx);
3430   __ jcc(Assembler::zero, notVolatile);
3431 
3432   fast_storefield_helper(field, rax);
3433   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3434                                                Assembler::StoreStore));
3435   __ jmp(Done);
3436   __ bind(notVolatile);
3437 
3438   fast_storefield_helper(field, rax);
3439 
3440   __ bind(Done);
3441 }
3442 
3443 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3444 
3445   // access field
3446   switch (bytecode()) {
3447   case Bytecodes::_fast_aputfield:
3448     do_oop_store(_masm, field, rax);
3449     break;
3450   case Bytecodes::_fast_lputfield:
3451 #ifdef _LP64
3452     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg);
3453 #else
3454   __ stop("should not be rewritten");
3455 #endif
3456     break;
3457   case Bytecodes::_fast_iputfield:
3458     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3459     break;
3460   case Bytecodes::_fast_zputfield:
3461     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3462     break;
3463   case Bytecodes::_fast_bputfield:
3464     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3465     break;
3466   case Bytecodes::_fast_sputfield:
3467     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3468     break;
3469   case Bytecodes::_fast_cputfield:
3470     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3471     break;
3472   case Bytecodes::_fast_fputfield:
3473     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg);
3474     break;
3475   case Bytecodes::_fast_dputfield:
3476     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg);
3477     break;
3478   default:
3479     ShouldNotReachHere();
3480   }
3481 }
3482 
3483 void TemplateTable::fast_accessfield(TosState state) {
3484   transition(atos, state);
3485 
3486   // Do the JVMTI work here to avoid disturbing the register state below
3487   if (JvmtiExport::can_post_field_access()) {
3488     // Check to see if a field access watch has been set before we
3489     // take the time to call into the VM.
3490     Label L1;
3491     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3492     __ testl(rcx, rcx);
3493     __ jcc(Assembler::zero, L1);
3494     // access constant pool cache entry
3495     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3496     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3497     __ verify_oop(rax);
3498     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3499     LP64_ONLY(__ mov(c_rarg1, rax));
3500     // c_rarg1: object pointer copied above
3501     // c_rarg2: cache entry pointer
3502     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3503     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3504     __ pop_ptr(rax); // restore object pointer
3505     __ bind(L1);
3506   }
3507 
3508   // access constant pool cache
3509   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3510   // replace index with field offset from cache entry
3511   // [jk] not needed currently
3512   // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3513   //                      in_bytes(ConstantPoolCache::base_offset() +
3514   //                               ConstantPoolCacheEntry::flags_offset())));
3515   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3516   // __ andl(rdx, 0x1);
3517   //
3518   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3519                          in_bytes(ConstantPoolCache::base_offset() +
3520                                   ConstantPoolCacheEntry::f2_offset())));
3521 
3522   // rax: object
3523   __ verify_oop(rax);
3524   __ null_check(rax);
3525   Address field(rax, rbx, Address::times_1);
3526 
3527   // access field
3528   switch (bytecode()) {
3529   case Bytecodes::_fast_agetfield:
3530     do_oop_load(_masm, field, rax);
3531     __ verify_oop(rax);
3532     break;
3533   case Bytecodes::_fast_lgetfield:
3534 #ifdef _LP64
3535     __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3536 #else
3537   __ stop("should not be rewritten");
3538 #endif
3539     break;
3540   case Bytecodes::_fast_igetfield:
3541     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3542     break;
3543   case Bytecodes::_fast_bgetfield:
3544     __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3545     break;
3546   case Bytecodes::_fast_sgetfield:
3547     __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3548     break;
3549   case Bytecodes::_fast_cgetfield:
3550     __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3551     break;
3552   case Bytecodes::_fast_fgetfield:
3553     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3554     break;
3555   case Bytecodes::_fast_dgetfield:
3556     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3557     break;
3558   default:
3559     ShouldNotReachHere();
3560   }
3561   // [jk] not needed currently
3562   //   Label notVolatile;
3563   //   __ testl(rdx, rdx);
3564   //   __ jcc(Assembler::zero, notVolatile);
3565   //   __ membar(Assembler::LoadLoad);
3566   //   __ bind(notVolatile);
3567 }
3568 
3569 void TemplateTable::fast_xaccess(TosState state) {
3570   transition(vtos, state);
3571 
3572   // get receiver
3573   __ movptr(rax, aaddress(0));
3574   // access constant pool cache
3575   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3576   __ movptr(rbx,
3577             Address(rcx, rdx, Address::times_ptr,
3578                     in_bytes(ConstantPoolCache::base_offset() +
3579                              ConstantPoolCacheEntry::f2_offset())));
3580   // make sure exception is reported in correct bcp range (getfield is
3581   // next instruction)
3582   __ increment(rbcp);
3583   __ null_check(rax);
3584   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3585   switch (state) {
3586   case itos:
3587     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3588     break;
3589   case atos:
3590     do_oop_load(_masm, field, rax);
3591     __ verify_oop(rax);
3592     break;
3593   case ftos:
3594     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3595     break;
3596   default:
3597     ShouldNotReachHere();
3598   }
3599 
3600   // [jk] not needed currently
3601   // Label notVolatile;
3602   // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3603   //                      in_bytes(ConstantPoolCache::base_offset() +
3604   //                               ConstantPoolCacheEntry::flags_offset())));
3605   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3606   // __ testl(rdx, 0x1);
3607   // __ jcc(Assembler::zero, notVolatile);
3608   // __ membar(Assembler::LoadLoad);
3609   // __ bind(notVolatile);
3610 
3611   __ decrement(rbcp);
3612 }
3613 
3614 //-----------------------------------------------------------------------------
3615 // Calls
3616 
3617 void TemplateTable::count_calls(Register method, Register temp) {
3618   // implemented elsewhere
3619   ShouldNotReachHere();
3620 }
3621 
3622 void TemplateTable::prepare_invoke(int byte_no,
3623                                    Register method,  // linked method (or i-klass)
3624                                    Register index,   // itable index, MethodType, etc.
3625                                    Register recv,    // if caller wants to see it
3626                                    Register flags    // if caller wants to test it
3627                                    ) {
3628   // determine flags
3629   const Bytecodes::Code code = bytecode();
3630   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3631   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3632   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3633   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3634   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3635   const bool load_receiver       = (recv  != noreg);
3636   const bool save_flags          = (flags != noreg);
3637   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3638   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3639   assert(flags == noreg || flags == rdx, "");
3640   assert(recv  == noreg || recv  == rcx, "");
3641 
3642   // setup registers & access constant pool cache
3643   if (recv  == noreg)  recv  = rcx;
3644   if (flags == noreg)  flags = rdx;
3645   assert_different_registers(method, index, recv, flags);
3646 
3647   // save 'interpreter return address'
3648   __ save_bcp();
3649 
3650   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3651 
3652   // maybe push appendix to arguments (just before return address)
3653   if (is_invokedynamic || is_invokehandle) {
3654     Label L_no_push;
3655     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3656     __ jcc(Assembler::zero, L_no_push);
3657     // Push the appendix as a trailing parameter.
3658     // This must be done before we get the receiver,
3659     // since the parameter_size includes it.
3660     __ push(rbx);
3661     __ mov(rbx, index);
3662     __ load_resolved_reference_at_index(index, rbx);
3663     __ pop(rbx);
3664     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3665     __ bind(L_no_push);
3666   }
3667 
3668   // load receiver if needed (after appendix is pushed so parameter size is correct)
3669   // Note: no return address pushed yet
3670   if (load_receiver) {
3671     __ movl(recv, flags);
3672     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3673     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3674     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3675     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3676     __ movptr(recv, recv_addr);
3677     __ verify_oop(recv);
3678   }
3679 
3680   if (save_flags) {
3681     __ movl(rbcp, flags);
3682   }
3683 
3684   // compute return type
3685   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3686   // Make sure we don't need to mask flags after the above shift
3687   ConstantPoolCacheEntry::verify_tos_state_shift();
3688   // load return address
3689   {
3690     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3691     ExternalAddress table(table_addr);
3692     LP64_ONLY(__ lea(rscratch1, table));
3693     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3694     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3695   }
3696 
3697   // push return address
3698   __ push(flags);
3699 
3700   // Restore flags value from the constant pool cache, and restore rsi
3701   // for later null checks.  r13 is the bytecode pointer
3702   if (save_flags) {
3703     __ movl(flags, rbcp);
3704     __ restore_bcp();
3705   }
3706 }
3707 
3708 void TemplateTable::invokevirtual_helper(Register index,
3709                                          Register recv,
3710                                          Register flags) {
3711   // Uses temporary registers rax, rdx
3712   assert_different_registers(index, recv, rax, rdx);
3713   assert(index == rbx, "");
3714   assert(recv  == rcx, "");
3715 
3716   // Test for an invoke of a final method
3717   Label notFinal;
3718   __ movl(rax, flags);
3719   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3720   __ jcc(Assembler::zero, notFinal);
3721 
3722   const Register method = index;  // method must be rbx
3723   assert(method == rbx,
3724          "Method* must be rbx for interpreter calling convention");
3725 
3726   // do the call - the index is actually the method to call
3727   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3728 
3729   // It's final, need a null check here!
3730   __ null_check(recv);
3731 
3732   // profile this call
3733   __ profile_final_call(rax);
3734   __ profile_arguments_type(rax, method, rbcp, true);
3735 
3736   __ jump_from_interpreted(method, rax);
3737 
3738   __ bind(notFinal);
3739 
3740   // get receiver klass
3741   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3742   __ load_klass(rax, recv);
3743 
3744   // profile this call
3745   __ profile_virtual_call(rax, rlocals, rdx);
3746   // get target Method* & entry point
3747   __ lookup_virtual_method(rax, index, method);
3748   __ profile_called_method(method, rdx, rbcp);
3749 
3750   __ profile_arguments_type(rdx, method, rbcp, true);
3751   __ jump_from_interpreted(method, rdx);
3752 }
3753 
3754 void TemplateTable::invokevirtual(int byte_no) {
3755   transition(vtos, vtos);
3756   assert(byte_no == f2_byte, "use this argument");
3757   prepare_invoke(byte_no,
3758                  rbx,    // method or vtable index
3759                  noreg,  // unused itable index
3760                  rcx, rdx); // recv, flags
3761 
3762   // rbx: index
3763   // rcx: receiver
3764   // rdx: flags
3765 
3766   invokevirtual_helper(rbx, rcx, rdx);
3767 }
3768 
3769 void TemplateTable::invokespecial(int byte_no) {
3770   transition(vtos, vtos);
3771   assert(byte_no == f1_byte, "use this argument");
3772   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3773                  rcx);  // get receiver also for null check
3774   __ verify_oop(rcx);
3775   __ null_check(rcx);
3776   // do the call
3777   __ profile_call(rax);
3778   __ profile_arguments_type(rax, rbx, rbcp, false);
3779   __ jump_from_interpreted(rbx, rax);
3780 }
3781 
3782 void TemplateTable::invokestatic(int byte_no) {
3783   transition(vtos, vtos);
3784   assert(byte_no == f1_byte, "use this argument");
3785   prepare_invoke(byte_no, rbx);  // get f1 Method*
3786   // do the call
3787   __ profile_call(rax);
3788   __ profile_arguments_type(rax, rbx, rbcp, false);
3789   __ jump_from_interpreted(rbx, rax);
3790 }
3791 
3792 
3793 void TemplateTable::fast_invokevfinal(int byte_no) {
3794   transition(vtos, vtos);
3795   assert(byte_no == f2_byte, "use this argument");
3796   __ stop("fast_invokevfinal not used on x86");
3797 }
3798 
3799 
3800 void TemplateTable::invokeinterface(int byte_no) {
3801   transition(vtos, vtos);
3802   assert(byte_no == f1_byte, "use this argument");
3803   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 Method*
3804                  rcx, rdx); // recv, flags
3805 
3806   // rax: reference klass (from f1) if interface method
3807   // rbx: method (from f2)
3808   // rcx: receiver
3809   // rdx: flags
3810 
3811   // First check for Object case, then private interface method,
3812   // then regular interface method.
3813 
3814   // Special case of invokeinterface called for virtual method of
3815   // java.lang.Object.  See cpCache.cpp for details.
3816   Label notObjectMethod;
3817   __ movl(rlocals, rdx);
3818   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3819   __ jcc(Assembler::zero, notObjectMethod);
3820   invokevirtual_helper(rbx, rcx, rdx);
3821   // no return from above
3822   __ bind(notObjectMethod);
3823 
3824   Label no_such_interface; // for receiver subtype check
3825   Register recvKlass; // used for exception processing
3826 
3827   // Check for private method invocation - indicated by vfinal
3828   Label notVFinal;
3829   __ movl(rlocals, rdx);
3830   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3831   __ jcc(Assembler::zero, notVFinal);
3832 
3833   // Get receiver klass into rlocals - also a null check
3834   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3835   __ load_klass(rlocals, rcx);
3836 
3837   Label subtype;
3838   __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3839   // If we get here the typecheck failed
3840   recvKlass = rdx;
3841   __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3842   __ jmp(no_such_interface);
3843 
3844   __ bind(subtype);
3845 
3846   // do the call - rbx is actually the method to call
3847 
3848   __ profile_final_call(rdx);
3849   __ profile_arguments_type(rdx, rbx, rbcp, true);
3850 
3851   __ jump_from_interpreted(rbx, rdx);
3852   // no return from above
3853   __ bind(notVFinal);
3854 
3855   // Get receiver klass into rdx - also a null check
3856   __ restore_locals();  // restore r14
3857   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3858   __ load_klass(rdx, rcx);
3859 
3860   Label no_such_method;
3861 
3862   // Preserve method for throw_AbstractMethodErrorVerbose.
3863   __ mov(rcx, rbx);
3864   // Receiver subtype check against REFC.
3865   // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3866   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3867                              rdx, rax, noreg,
3868                              // outputs: scan temp. reg, scan temp. reg
3869                              rbcp, rlocals,
3870                              no_such_interface,
3871                              /*return_method=*/false);
3872 
3873   // profile this call
3874   __ restore_bcp(); // rbcp was destroyed by receiver type check
3875   __ profile_virtual_call(rdx, rbcp, rlocals);
3876 
3877   // Get declaring interface class from method, and itable index
3878   __ movptr(rax, Address(rbx, Method::const_offset()));
3879   __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3880   __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3881   __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3882   __ subl(rbx, Method::itable_index_max);
3883   __ negl(rbx);
3884 
3885   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3886   __ mov(rlocals, rdx);
3887   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3888                              rlocals, rax, rbx,
3889                              // outputs: method, scan temp. reg
3890                              rbx, rbcp,
3891                              no_such_interface);
3892 
3893   // rbx: Method* to call
3894   // rcx: receiver
3895   // Check for abstract method error
3896   // Note: This should be done more efficiently via a throw_abstract_method_error
3897   //       interpreter entry point and a conditional jump to it in case of a null
3898   //       method.
3899   __ testptr(rbx, rbx);
3900   __ jcc(Assembler::zero, no_such_method);
3901 
3902   __ profile_called_method(rbx, rbcp, rdx);
3903   __ profile_arguments_type(rdx, rbx, rbcp, true);
3904 
3905   // do the call
3906   // rcx: receiver
3907   // rbx,: Method*
3908   __ jump_from_interpreted(rbx, rdx);
3909   __ should_not_reach_here();
3910 
3911   // exception handling code follows...
3912   // note: must restore interpreter registers to canonical
3913   //       state for exception handling to work correctly!
3914 
3915   __ bind(no_such_method);
3916   // throw exception
3917   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3918   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3919   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3920   // Pass arguments for generating a verbose error message.
3921 #ifdef _LP64
3922   recvKlass = c_rarg1;
3923   Register method    = c_rarg2;
3924   if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3925   if (method != rcx)    { __ movq(method, rcx);    }
3926 #else
3927   recvKlass = rdx;
3928   Register method    = rcx;
3929 #endif
3930   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3931              recvKlass, method);
3932   // The call_VM checks for exception, so we should never return here.
3933   __ should_not_reach_here();
3934 
3935   __ bind(no_such_interface);
3936   // throw exception
3937   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3938   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3939   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3940   // Pass arguments for generating a verbose error message.
3941   LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3942   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3943              recvKlass, rax);
3944   // the call_VM checks for exception, so we should never return here.
3945   __ should_not_reach_here();
3946 }
3947 
3948 void TemplateTable::invokehandle(int byte_no) {
3949   transition(vtos, vtos);
3950   assert(byte_no == f1_byte, "use this argument");
3951   const Register rbx_method = rbx;
3952   const Register rax_mtype  = rax;
3953   const Register rcx_recv   = rcx;
3954   const Register rdx_flags  = rdx;
3955 
3956   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3957   __ verify_method_ptr(rbx_method);
3958   __ verify_oop(rcx_recv);
3959   __ null_check(rcx_recv);
3960 
3961   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3962   // rbx: MH.invokeExact_MT method (from f2)
3963 
3964   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3965 
3966   // FIXME: profile the LambdaForm also
3967   __ profile_final_call(rax);
3968   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3969 
3970   __ jump_from_interpreted(rbx_method, rdx);
3971 }
3972 
3973 void TemplateTable::invokedynamic(int byte_no) {
3974   transition(vtos, vtos);
3975   assert(byte_no == f1_byte, "use this argument");
3976 
3977   const Register rbx_method   = rbx;
3978   const Register rax_callsite = rax;
3979 
3980   prepare_invoke(byte_no, rbx_method, rax_callsite);
3981 
3982   // rax: CallSite object (from cpool->resolved_references[f1])
3983   // rbx: MH.linkToCallSite method (from f2)
3984 
3985   // Note:  rax_callsite is already pushed by prepare_invoke
3986 
3987   // %%% should make a type profile for any invokedynamic that takes a ref argument
3988   // profile this call
3989   __ profile_call(rbcp);
3990   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3991 
3992   __ verify_oop(rax_callsite);
3993 
3994   __ jump_from_interpreted(rbx_method, rdx);
3995 }
3996 
3997 //-----------------------------------------------------------------------------
3998 // Allocation
3999 
4000 void TemplateTable::_new() {
4001   transition(vtos, atos);
4002   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4003   Label slow_case;
4004   Label slow_case_no_pop;
4005   Label done;
4006   Label initialize_header;
4007   Label initialize_object;  // including clearing the fields
4008 
4009   __ get_cpool_and_tags(rcx, rax);
4010 
4011   // Make sure the class we're about to instantiate has been resolved.
4012   // This is done before loading InstanceKlass to be consistent with the order
4013   // how Constant Pool is updated (see ConstantPool::klass_at_put)
4014   const int tags_offset = Array<u1>::base_offset_in_bytes();
4015   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4016   __ jcc(Assembler::notEqual, slow_case_no_pop);
4017 
4018   // get InstanceKlass
4019   __ load_resolved_klass_at_index(rcx, rdx, rcx);
4020   __ push(rcx);  // save the contexts of klass for initializing the header
4021 
4022   // make sure klass is initialized & doesn't have finalizer
4023   // make sure klass is fully initialized
4024   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4025   __ jcc(Assembler::notEqual, slow_case);
4026 
4027   // get instance_size in InstanceKlass (scaled to a count of bytes)
4028   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4029   // test to see if it has a finalizer or is malformed in some way
4030   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4031   __ jcc(Assembler::notZero, slow_case);
4032 
4033   // Allocate the instance:
4034   //  If TLAB is enabled:
4035   //    Try to allocate in the TLAB.
4036   //    If fails, go to the slow path.
4037   //  Else If inline contiguous allocations are enabled:
4038   //    Try to allocate in eden.
4039   //    If fails due to heap end, go to slow path.
4040   //
4041   //  If TLAB is enabled OR inline contiguous is enabled:
4042   //    Initialize the allocation.
4043   //    Exit.
4044   //
4045   //  Go to slow path.
4046 
4047   const bool allow_shared_alloc =
4048     Universe::heap()->supports_inline_contig_alloc();
4049 
4050   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4051 #ifndef _LP64
4052   if (UseTLAB || allow_shared_alloc) {
4053     __ get_thread(thread);
4054   }
4055 #endif // _LP64
4056 
4057   if (UseTLAB) {
4058     __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
4059     if (ZeroTLAB) {
4060       // the fields have been already cleared
4061       __ jmp(initialize_header);
4062     } else {
4063       // initialize both the header and fields
4064       __ jmp(initialize_object);
4065     }
4066   } else {
4067     // Allocation in the shared Eden, if allowed.
4068     //
4069     // rdx: instance size in bytes
4070     __ eden_allocate(thread, rax, rdx, 0, rbx, slow_case);
4071   }
4072 
4073   // If UseTLAB or allow_shared_alloc are true, the object is created above and
4074   // there is an initialize need. Otherwise, skip and go to the slow path.
4075   if (UseTLAB || allow_shared_alloc) {
4076     // The object is initialized before the header.  If the object size is
4077     // zero, go directly to the header initialization.
4078     __ bind(initialize_object);
4079     __ decrement(rdx, sizeof(oopDesc));
4080     __ jcc(Assembler::zero, initialize_header);
4081 
4082     // Initialize topmost object field, divide rdx by 8, check if odd and
4083     // test if zero.
4084     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
4085     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4086 
4087     // rdx must have been multiple of 8
4088 #ifdef ASSERT
4089     // make sure rdx was multiple of 8
4090     Label L;
4091     // Ignore partial flag stall after shrl() since it is debug VM
4092     __ jcc(Assembler::carryClear, L);
4093     __ stop("object size is not multiple of 2 - adjust this code");
4094     __ bind(L);
4095     // rdx must be > 0, no extra check needed here
4096 #endif
4097 
4098     // initialize remaining object fields: rdx was a multiple of 8
4099     { Label loop;
4100     __ bind(loop);
4101     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4102     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4103     __ decrement(rdx);
4104     __ jcc(Assembler::notZero, loop);
4105     }
4106 
4107     // initialize object header only.
4108     __ bind(initialize_header);
4109     if (UseBiasedLocking) {
4110       __ pop(rcx);   // get saved klass back in the register.
4111       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4112       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4113     } else {
4114       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4115                 (intptr_t)markOopDesc::prototype()); // header
4116       __ pop(rcx);   // get saved klass back in the register.
4117     }
4118 #ifdef _LP64
4119     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4120     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4121 #endif
4122     __ store_klass(rax, rcx);  // klass
4123 
4124     {
4125       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4126       // Trigger dtrace event for fastpath
4127       __ push(atos);
4128       __ call_VM_leaf(
4129            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4130       __ pop(atos);
4131     }
4132 
4133     __ jmp(done);
4134   }
4135 
4136   // slow case
4137   __ bind(slow_case);
4138   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4139   __ bind(slow_case_no_pop);
4140 
4141   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4142   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4143 
4144   __ get_constant_pool(rarg1);
4145   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4146   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4147    __ verify_oop(rax);
4148 
4149   // continue
4150   __ bind(done);
4151 }
4152 
4153 void TemplateTable::newarray() {
4154   transition(itos, atos);
4155   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4156   __ load_unsigned_byte(rarg1, at_bcp(1));
4157   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4158           rarg1, rax);
4159 }
4160 
4161 void TemplateTable::anewarray() {
4162   transition(itos, atos);
4163 
4164   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4165   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4166 
4167   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4168   __ get_constant_pool(rarg1);
4169   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4170           rarg1, rarg2, rax);
4171 }
4172 
4173 void TemplateTable::arraylength() {
4174   transition(atos, itos);
4175   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4176   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4177 }
4178 
4179 void TemplateTable::checkcast() {
4180   transition(atos, atos);
4181   Label done, is_null, ok_is_subtype, quicked, resolved;
4182   __ testptr(rax, rax); // object is in rax
4183   __ jcc(Assembler::zero, is_null);
4184 
4185   // Get cpool & tags index
4186   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4187   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4188   // See if bytecode has already been quicked
4189   __ cmpb(Address(rdx, rbx,
4190                   Address::times_1,
4191                   Array<u1>::base_offset_in_bytes()),
4192           JVM_CONSTANT_Class);
4193   __ jcc(Assembler::equal, quicked);
4194   __ push(atos); // save receiver for result, and for GC
4195   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4196 
4197   // vm_result_2 has metadata result
4198 #ifndef _LP64
4199   // borrow rdi from locals
4200   __ get_thread(rdi);
4201   __ get_vm_result_2(rax, rdi);
4202   __ restore_locals();
4203 #else
4204   __ get_vm_result_2(rax, r15_thread);
4205 #endif
4206 
4207   __ pop_ptr(rdx); // restore receiver
4208   __ jmpb(resolved);
4209 
4210   // Get superklass in rax and subklass in rbx
4211   __ bind(quicked);
4212   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4213   __ load_resolved_klass_at_index(rcx, rbx, rax);
4214 
4215   __ bind(resolved);
4216   __ load_klass(rbx, rdx);
4217 
4218   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4219   // Superklass in rax.  Subklass in rbx.
4220   __ gen_subtype_check(rbx, ok_is_subtype);
4221 
4222   // Come here on failure
4223   __ push_ptr(rdx);
4224   // object is at TOS
4225   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4226 
4227   // Come here on success
4228   __ bind(ok_is_subtype);
4229   __ mov(rax, rdx); // Restore object in rdx
4230 
4231   // Collect counts on whether this check-cast sees NULLs a lot or not.
4232   if (ProfileInterpreter) {
4233     __ jmp(done);
4234     __ bind(is_null);
4235     __ profile_null_seen(rcx);
4236   } else {
4237     __ bind(is_null);   // same as 'done'
4238   }
4239   __ bind(done);
4240 }
4241 
4242 void TemplateTable::instanceof() {
4243   transition(atos, itos);
4244   Label done, is_null, ok_is_subtype, quicked, resolved;
4245   __ testptr(rax, rax);
4246   __ jcc(Assembler::zero, is_null);
4247 
4248   // Get cpool & tags index
4249   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4250   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4251   // See if bytecode has already been quicked
4252   __ cmpb(Address(rdx, rbx,
4253                   Address::times_1,
4254                   Array<u1>::base_offset_in_bytes()),
4255           JVM_CONSTANT_Class);
4256   __ jcc(Assembler::equal, quicked);
4257 
4258   __ push(atos); // save receiver for result, and for GC
4259   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4260   // vm_result_2 has metadata result
4261 
4262 #ifndef _LP64
4263   // borrow rdi from locals
4264   __ get_thread(rdi);
4265   __ get_vm_result_2(rax, rdi);
4266   __ restore_locals();
4267 #else
4268   __ get_vm_result_2(rax, r15_thread);
4269 #endif
4270 
4271   __ pop_ptr(rdx); // restore receiver
4272   __ verify_oop(rdx);
4273   __ load_klass(rdx, rdx);
4274   __ jmpb(resolved);
4275 
4276   // Get superklass in rax and subklass in rdx
4277   __ bind(quicked);
4278   __ load_klass(rdx, rax);
4279   __ load_resolved_klass_at_index(rcx, rbx, rax);
4280 
4281   __ bind(resolved);
4282 
4283   // Generate subtype check.  Blows rcx, rdi
4284   // Superklass in rax.  Subklass in rdx.
4285   __ gen_subtype_check(rdx, ok_is_subtype);
4286 
4287   // Come here on failure
4288   __ xorl(rax, rax);
4289   __ jmpb(done);
4290   // Come here on success
4291   __ bind(ok_is_subtype);
4292   __ movl(rax, 1);
4293 
4294   // Collect counts on whether this test sees NULLs a lot or not.
4295   if (ProfileInterpreter) {
4296     __ jmp(done);
4297     __ bind(is_null);
4298     __ profile_null_seen(rcx);
4299   } else {
4300     __ bind(is_null);   // same as 'done'
4301   }
4302   __ bind(done);
4303   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4304   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4305 }
4306 
4307 
4308 //----------------------------------------------------------------------------------------------------
4309 // Breakpoints
4310 void TemplateTable::_breakpoint() {
4311   // Note: We get here even if we are single stepping..
4312   // jbug insists on setting breakpoints at every bytecode
4313   // even if we are in single step mode.
4314 
4315   transition(vtos, vtos);
4316 
4317   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4318 
4319   // get the unpatched byte code
4320   __ get_method(rarg);
4321   __ call_VM(noreg,
4322              CAST_FROM_FN_PTR(address,
4323                               InterpreterRuntime::get_original_bytecode_at),
4324              rarg, rbcp);
4325   __ mov(rbx, rax);  // why?
4326 
4327   // post the breakpoint event
4328   __ get_method(rarg);
4329   __ call_VM(noreg,
4330              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4331              rarg, rbcp);
4332 
4333   // complete the execution of original bytecode
4334   __ dispatch_only_normal(vtos);
4335 }
4336 
4337 //-----------------------------------------------------------------------------
4338 // Exceptions
4339 
4340 void TemplateTable::athrow() {
4341   transition(atos, vtos);
4342   __ null_check(rax);
4343   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4344 }
4345 
4346 //-----------------------------------------------------------------------------
4347 // Synchronization
4348 //
4349 // Note: monitorenter & exit are symmetric routines; which is reflected
4350 //       in the assembly code structure as well
4351 //
4352 // Stack layout:
4353 //
4354 // [expressions  ] <--- rsp               = expression stack top
4355 // ..
4356 // [expressions  ]
4357 // [monitor entry] <--- monitor block top = expression stack bot
4358 // ..
4359 // [monitor entry]
4360 // [frame data   ] <--- monitor block bot
4361 // ...
4362 // [saved rbp    ] <--- rbp
4363 void TemplateTable::monitorenter() {
4364   transition(atos, vtos);
4365 
4366   // check for NULL object
4367   __ null_check(rax);
4368 
4369   __ resolve(IS_NOT_NULL, rax);
4370 
4371   const Address monitor_block_top(
4372         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4373   const Address monitor_block_bot(
4374         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4375   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4376 
4377   Label allocated;
4378 
4379   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4380   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4381   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4382 
4383   // initialize entry pointer
4384   __ xorl(rmon, rmon); // points to free slot or NULL
4385 
4386   // find a free slot in the monitor block (result in rmon)
4387   {
4388     Label entry, loop, exit;
4389     __ movptr(rtop, monitor_block_top); // points to current entry,
4390                                         // starting with top-most entry
4391     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4392                                         // of monitor block
4393     __ jmpb(entry);
4394 
4395     __ bind(loop);
4396     // check if current entry is used
4397     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4398     // if not used then remember entry in rmon
4399     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4400     // check if current entry is for same object
4401     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4402     // if same object then stop searching
4403     __ jccb(Assembler::equal, exit);
4404     // otherwise advance to next entry
4405     __ addptr(rtop, entry_size);
4406     __ bind(entry);
4407     // check if bottom reached
4408     __ cmpptr(rtop, rbot);
4409     // if not at bottom then check this entry
4410     __ jcc(Assembler::notEqual, loop);
4411     __ bind(exit);
4412   }
4413 
4414   __ testptr(rmon, rmon); // check if a slot has been found
4415   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4416 
4417   // allocate one if there's no free slot
4418   {
4419     Label entry, loop;
4420     // 1. compute new pointers          // rsp: old expression stack top
4421     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4422     __ subptr(rsp, entry_size);         // move expression stack top
4423     __ subptr(rmon, entry_size);        // move expression stack bottom
4424     __ mov(rtop, rsp);                  // set start value for copy loop
4425     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4426     __ jmp(entry);
4427     // 2. move expression stack contents
4428     __ bind(loop);
4429     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4430                                                 // word from old location
4431     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4432     __ addptr(rtop, wordSize);                  // advance to next word
4433     __ bind(entry);
4434     __ cmpptr(rtop, rmon);                      // check if bottom reached
4435     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4436                                                 // copy next word
4437   }
4438 
4439   // call run-time routine
4440   // rmon: points to monitor entry
4441   __ bind(allocated);
4442 
4443   // Increment bcp to point to the next bytecode, so exception
4444   // handling for async. exceptions work correctly.
4445   // The object has already been poped from the stack, so the
4446   // expression stack looks correct.
4447   __ increment(rbcp);
4448 
4449   // store object
4450   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4451   __ lock_object(rmon);
4452 
4453   // check to make sure this monitor doesn't cause stack overflow after locking
4454   __ save_bcp();  // in case of exception
4455   __ generate_stack_overflow_check(0);
4456 
4457   // The bcp has already been incremented. Just need to dispatch to
4458   // next instruction.
4459   __ dispatch_next(vtos);
4460 }
4461 
4462 void TemplateTable::monitorexit() {
4463   transition(atos, vtos);
4464 
4465   // check for NULL object
4466   __ null_check(rax);
4467 
4468   __ resolve(IS_NOT_NULL, rax);
4469 
4470   const Address monitor_block_top(
4471         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4472   const Address monitor_block_bot(
4473         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4474   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4475 
4476   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4477   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4478 
4479   Label found;
4480 
4481   // find matching slot
4482   {
4483     Label entry, loop;
4484     __ movptr(rtop, monitor_block_top); // points to current entry,
4485                                         // starting with top-most entry
4486     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4487                                         // of monitor block
4488     __ jmpb(entry);
4489 
4490     __ bind(loop);
4491     // check if current entry is for same object
4492     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4493     // if same object then stop searching
4494     __ jcc(Assembler::equal, found);
4495     // otherwise advance to next entry
4496     __ addptr(rtop, entry_size);
4497     __ bind(entry);
4498     // check if bottom reached
4499     __ cmpptr(rtop, rbot);
4500     // if not at bottom then check this entry
4501     __ jcc(Assembler::notEqual, loop);
4502   }
4503 
4504   // error handling. Unlocking was not block-structured
4505   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4506                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4507   __ should_not_reach_here();
4508 
4509   // call run-time routine
4510   __ bind(found);
4511   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4512   __ unlock_object(rtop);
4513   __ pop_ptr(rax); // discard object
4514 }
4515 
4516 // Wide instructions
4517 void TemplateTable::wide() {
4518   transition(vtos, vtos);
4519   __ load_unsigned_byte(rbx, at_bcp(1));
4520   ExternalAddress wtable((address)Interpreter::_wentry_point);
4521   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4522   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4523 }
4524 
4525 // Multi arrays
4526 void TemplateTable::multianewarray() {
4527   transition(vtos, atos);
4528 
4529   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4530   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4531   // last dim is on top of stack; we want address of first one:
4532   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4533   // the latter wordSize to point to the beginning of the array.
4534   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4535   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4536   __ load_unsigned_byte(rbx, at_bcp(3));
4537   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4538 }