hotspot/src/cpu/x86/vm/templateTable_x86_32.cpp

Print this page
rev 611 : Merge
   1 #ifdef USE_PRAGMA_IDENT_SRC
   2 #pragma ident "@(#)templateTable_x86_32.cpp     1.323 07/09/17 09:26:00 JVM"
   3 #endif
   4 /*
   5  * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This code is free software; you can redistribute it and/or modify it
   9  * under the terms of the GNU General Public License version 2 only, as
  10  * published by the Free Software Foundation.
  11  *
  12  * This code is distributed in the hope that it will be useful, but WITHOUT
  13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15  * version 2 for more details (a copy is included in the LICENSE file that
  16  * accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License version
  19  * 2 along with this work; if not, write to the Free Software Foundation,
  20  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *  


  93 }
  94 
  95 // Condition conversion
  96 static Assembler::Condition j_not(TemplateTable::Condition cc) {
  97   switch (cc) {
  98     case TemplateTable::equal        : return Assembler::notEqual;
  99     case TemplateTable::not_equal    : return Assembler::equal;
 100     case TemplateTable::less         : return Assembler::greaterEqual;
 101     case TemplateTable::less_equal   : return Assembler::greater;
 102     case TemplateTable::greater      : return Assembler::lessEqual;
 103     case TemplateTable::greater_equal: return Assembler::less;
 104   }
 105   ShouldNotReachHere();
 106   return Assembler::zero;
 107 }
 108 
 109 
 110 //----------------------------------------------------------------------------------------------------
 111 // Miscelaneous helper routines
 112 








































































 113 Address TemplateTable::at_bcp(int offset) {
 114   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 115   return Address(rsi, offset);
 116 }
 117 
 118 
 119 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
 120                                    Register scratch,
 121                                    bool load_bc_into_scratch/*=true*/) {
 122                                    
 123   if (!RewriteBytecodes) return;
 124   // the pair bytecodes have already done the load.
 125   if (load_bc_into_scratch) __ movl(bc, bytecode);


 126   Label patch_done;
 127   if (JvmtiExport::can_post_breakpoint()) {
 128     Label fast_patch;
 129     // if a breakpoint is present we can't rewrite the stream directly
 130     __ movzxb(scratch, at_bcp(0));
 131     __ cmpl(scratch, Bytecodes::_breakpoint);
 132     __ jcc(Assembler::notEqual, fast_patch);
 133     __ get_method(scratch);
 134     // Let breakpoint table handling rewrite to quicker bytecode 
 135     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
 136 #ifndef ASSERT
 137     __ jmpb(patch_done);
 138     __ bind(fast_patch);
 139   }
 140 #else
 141     __ jmp(patch_done);
 142     __ bind(fast_patch);
 143   }
 144   Label okay;
 145   __ load_unsigned_byte(scratch, at_bcp(0));
 146   __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
 147   __ jccb(Assembler::equal, okay);
 148   __ cmpl(scratch, bc);
 149   __ jcc(Assembler::equal, okay);
 150   __ stop("patching the wrong bytecode");


 155   __ bind(patch_done);
 156 }
 157 
 158 //----------------------------------------------------------------------------------------------------
 159 // Individual instructions
 160 
 161 void TemplateTable::nop() {
 162   transition(vtos, vtos);
 163   // nothing to do
 164 }
 165 
 166 void TemplateTable::shouldnotreachhere() {
 167   transition(vtos, vtos);
 168   __ stop("shouldnotreachhere bytecode");
 169 }
 170 
 171 
 172 
 173 void TemplateTable::aconst_null() {
 174   transition(vtos, atos);
 175   __ xorl(rax, rax);
 176 }
 177 
 178 
 179 void TemplateTable::iconst(int value) {
 180   transition(vtos, itos);
 181   if (value == 0) {
 182     __ xorl(rax, rax);
 183   } else {
 184     __ movl(rax, value);
 185   }
 186 }
 187 
 188 
 189 void TemplateTable::lconst(int value) {
 190   transition(vtos, ltos);
 191   if (value == 0) {
 192     __ xorl(rax, rax);
 193   } else {
 194     __ movl(rax, value);
 195   }
 196   assert(value >= 0, "check this code");
 197   __ xorl(rdx, rdx);
 198 }
 199 
 200 
 201 void TemplateTable::fconst(int value) {
 202   transition(vtos, ftos);
 203          if (value == 0) { __ fldz();
 204   } else if (value == 1) { __ fld1();
 205   } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 206   } else                 { ShouldNotReachHere();
 207   }
 208 }
 209 
 210 
 211 void TemplateTable::dconst(int value) {
 212   transition(vtos, dtos);
 213          if (value == 0) { __ fldz();
 214   } else if (value == 1) { __ fld1();
 215   } else                 { ShouldNotReachHere();
 216   }
 217 }
 218 
 219 
 220 void TemplateTable::bipush() {
 221   transition(vtos, itos);
 222   __ load_signed_byte(rax, at_bcp(1));
 223 }
 224 
 225 
 226 void TemplateTable::sipush() {
 227   transition(vtos, itos);
 228   __ load_unsigned_word(rax, at_bcp(1));
 229   __ bswap(rax);
 230   __ sarl(rax, 16);
 231 }
 232 
 233 void TemplateTable::ldc(bool wide) {
 234   transition(vtos, vtos);
 235   Label call_ldc, notFloat, notClass, Done;
 236 
 237   if (wide) {
 238     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 239   } else {
 240     __ load_unsigned_byte(rbx, at_bcp(1));
 241   }
 242   __ get_cpool_and_tags(rcx, rax);
 243   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 244   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 245 
 246   // get type
 247   __ xorl(rdx, rdx);
 248   __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 249 
 250   // unresolved string - get the resolved string
 251   __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
 252   __ jccb(Assembler::equal, call_ldc);
 253 
 254   // unresolved class - get the resolved class
 255   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 256   __ jccb(Assembler::equal, call_ldc);
 257 
 258   // unresolved class in error (resolution failed) - call into runtime
 259   // so that the same error from first resolution attempt is thrown.
 260   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 261   __ jccb(Assembler::equal, call_ldc);
 262 
 263   // resolved class - need to call vm to get java mirror of the class
 264   __ cmpl(rdx, JVM_CONSTANT_Class);
 265   __ jcc(Assembler::notEqual, notClass);
 266 
 267   __ bind(call_ldc);
 268   __ movl(rcx, wide);
 269   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
 270   __ push(atos);
 271   __ jmp(Done);
 272 
 273   __ bind(notClass);
 274   __ cmpl(rdx, JVM_CONSTANT_Float);
 275   __ jccb(Assembler::notEqual, notFloat);
 276   // ftos
 277   __ fld_s(    Address(rcx, rbx, Address::times_4, base_offset));
 278   __ push(ftos);
 279   __ jmp(Done);
 280 
 281   __ bind(notFloat);
 282 #ifdef ASSERT
 283   { Label L;
 284     __ cmpl(rdx, JVM_CONSTANT_Integer);
 285     __ jcc(Assembler::equal, L);
 286     __ cmpl(rdx, JVM_CONSTANT_String);
 287     __ jcc(Assembler::equal, L);
 288     __ stop("unexpected tag type in ldc");
 289     __ bind(L);
 290   }
 291 #endif
 292   Label isOop;
 293   // atos and itos
 294   __ movl(rax, Address(rcx, rbx, Address::times_4, base_offset));
 295   // String is only oop type we will see here
 296   __ cmpl(rdx, JVM_CONSTANT_String);
 297   __ jccb(Assembler::equal, isOop);

 298   __ push(itos);
 299   __ jmp(Done);
 300   __ bind(isOop);

 301   __ push(atos);
 302 
 303   if (VerifyOops) {
 304     __ verify_oop(rax);
 305   }
 306   __ bind(Done);
 307 }
 308 
 309 void TemplateTable::ldc2_w() {
 310   transition(vtos, vtos);
 311   Label Long, Done;
 312   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 313 
 314   __ get_cpool_and_tags(rcx, rax);
 315   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 316   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 317 
 318   // get type
 319   __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
 320   __ jccb(Assembler::notEqual, Long);
 321   // dtos
 322   __ fld_d(    Address(rcx, rbx, Address::times_4, base_offset));
 323   __ push(dtos);
 324   __ jmpb(Done);
 325 
 326   __ bind(Long);
 327   // ltos
 328   __ movl(rax, Address(rcx, rbx, Address::times_4, base_offset + 0 * wordSize));
 329   __ movl(rdx, Address(rcx, rbx, Address::times_4, base_offset + 1 * wordSize));
 330 
 331   __ push(ltos);
 332 
 333   __ bind(Done);
 334 }
 335 
 336 
 337 void TemplateTable::locals_index(Register reg, int offset) {
 338   __ load_unsigned_byte(reg, at_bcp(offset));
 339   __ negl(reg); 
 340 }
 341 
 342 
 343 void TemplateTable::iload() {
 344   transition(vtos, itos);
 345   if (RewriteFrequentPairs) { 
 346     Label rewrite, done;
 347 
 348     // get next byte
 349     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 350     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 351     // last two iloads in a pair.  Comparing against fast_iload means that
 352     // the next bytecode is neither an iload or a caload, and therefore
 353     // an iload pair.
 354     __ cmpl(rbx, Bytecodes::_iload);
 355     __ jcc(Assembler::equal, done);
 356 
 357     __ cmpl(rbx, Bytecodes::_fast_iload);
 358     __ movl(rcx, Bytecodes::_fast_iload2);
 359     __ jccb(Assembler::equal, rewrite);


 385   locals_index(rbx);
 386   __ movl(rax, iaddress(rbx));
 387   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 388   __ push(itos);
 389   locals_index(rbx, 3);
 390   __ movl(rax, iaddress(rbx));
 391   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 392 }
 393   
 394 void TemplateTable::fast_iload() {
 395   transition(vtos, itos);
 396   locals_index(rbx);
 397   __ movl(rax, iaddress(rbx));
 398   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 399 }
 400 
 401 
 402 void TemplateTable::lload() {
 403   transition(vtos, ltos);
 404   locals_index(rbx);
 405   __ movl(rax, laddress(rbx));
 406   __ movl(rdx, haddress(rbx));
 407   debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 408 }
 409 
 410 
 411 void TemplateTable::fload() {
 412   transition(vtos, ftos);
 413   locals_index(rbx);
 414   __ fld_s(faddress(rbx));
 415   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 416 }
 417 
 418 
 419 void TemplateTable::dload() {
 420   transition(vtos, dtos);
 421   locals_index(rbx);
 422   if (TaggedStackInterpreter) {
 423     // Get double out of locals array, onto temp stack and load with
 424     // float instruction into ST0
 425     __ movl(rax, laddress(rbx));
 426     __ movl(rdx, haddress(rbx));
 427     __ pushl(rdx);  // push hi first
 428     __ pushl(rax);
 429     __ fld_d(Address(rsp, 0));
 430     __ addl(rsp, 2*wordSize);
 431     debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 432   } else {
 433     __ fld_d(daddress(rbx));
 434   }
 435 }
 436 
 437 
 438 void TemplateTable::aload() {
 439   transition(vtos, atos);
 440   locals_index(rbx);
 441   __ movl(rax, iaddress(rbx));
 442   debug_only(__ verify_local_tag(frame::TagReference, rbx));
 443 }
 444 
 445 
 446 void TemplateTable::locals_index_wide(Register reg) {
 447   __ movl(reg, at_bcp(2));
 448   __ bswap(reg);
 449   __ shrl(reg, 16);
 450   __ negl(reg); 
 451 }
 452 
 453 
 454 void TemplateTable::wide_iload() {
 455   transition(vtos, itos);
 456   locals_index_wide(rbx);
 457   __ movl(rax, iaddress(rbx));
 458   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 459 }
 460 
 461 
 462 void TemplateTable::wide_lload() {
 463   transition(vtos, ltos);
 464   locals_index_wide(rbx);
 465   __ movl(rax, laddress(rbx));
 466   __ movl(rdx, haddress(rbx));
 467   debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 468 }
 469 
 470 
 471 void TemplateTable::wide_fload() {
 472   transition(vtos, ftos);
 473   locals_index_wide(rbx);
 474   __ fld_s(faddress(rbx));
 475   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 476 }
 477 
 478 
 479 void TemplateTable::wide_dload() {
 480   transition(vtos, dtos);
 481   locals_index_wide(rbx);
 482   if (TaggedStackInterpreter) {
 483     // Get double out of locals array, onto temp stack and load with
 484     // float instruction into ST0
 485     __ movl(rax, laddress(rbx));
 486     __ movl(rdx, haddress(rbx));
 487     __ pushl(rdx);  // push hi first
 488     __ pushl(rax);
 489     __ fld_d(Address(rsp, 0));
 490     __ addl(rsp, 2*wordSize);
 491     debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 492   } else {
 493     __ fld_d(daddress(rbx));
 494   }
 495 }
 496 
 497 
 498 void TemplateTable::wide_aload() {
 499   transition(vtos, atos);
 500   locals_index_wide(rbx);
 501   __ movl(rax, iaddress(rbx));
 502   debug_only(__ verify_local_tag(frame::TagReference, rbx));
 503 }
 504 
 505 void TemplateTable::index_check(Register array, Register index) {
 506   // Pop ptr into array
 507   __ pop_ptr(array);
 508   index_check_without_pop(array, index);
 509 }
 510 
 511 void TemplateTable::index_check_without_pop(Register array, Register index) {
 512   // destroys rbx,
 513   // check array
 514   __ null_check(array, arrayOopDesc::length_offset_in_bytes());

 515   // check index
 516   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 517   if (index != rbx) {
 518     // ??? convention: move aberrant index into rbx, for exception message
 519     assert(rbx != array, "different registers");
 520     __ movl(rbx, index);
 521   }
 522   __ jump_cc(Assembler::aboveEqual,
 523              ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 524 }
 525 
 526 
 527 void TemplateTable::iaload() {
 528   transition(itos, itos);
 529   // rdx: array
 530   index_check(rdx, rax);  // kills rbx,
 531   // rax,: index
 532   __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
 533 }
 534 
 535 
 536 void TemplateTable::laload() {
 537   transition(itos, ltos);
 538   // rax,: index
 539   // rdx: array
 540   index_check(rdx, rax);
 541   __ movl(rbx, rax);
 542   // rbx,: index
 543   __ movl(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
 544   __ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize));
 545 }
 546 
 547 
 548 void TemplateTable::faload() {
 549   transition(itos, ftos);
 550   // rdx: array
 551   index_check(rdx, rax);  // kills rbx,
 552   // rax,: index
 553   __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 554 }
 555 
 556 
 557 void TemplateTable::daload() {
 558   transition(itos, dtos);
 559   // rdx: array
 560   index_check(rdx, rax);  // kills rbx,
 561   // rax,: index
 562   __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 563 }
 564 
 565 
 566 void TemplateTable::aaload() {
 567   transition(itos, atos);
 568   // rdx: array
 569   index_check(rdx, rax);  // kills rbx,
 570   // rax,: index
 571   __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 572 }
 573 
 574 
 575 void TemplateTable::baload() {
 576   transition(itos, itos);
 577   // rdx: array
 578   index_check(rdx, rax);  // kills rbx,
 579   // rax,: index
 580   // can do better code for P5 - fix this at some point
 581   __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 582   __ movl(rax, rbx);
 583 }
 584 
 585 
 586 void TemplateTable::caload() {
 587   transition(itos, itos);
 588   // rdx: array
 589   index_check(rdx, rax);  // kills rbx,
 590   // rax,: index
 591   // can do better code for P5 - may want to improve this at some point
 592   __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 593   __ movl(rax, rbx);
 594 }
 595 
 596 // iload followed by caload frequent pair
 597 void TemplateTable::fast_icaload() {
 598   transition(vtos, itos);
 599   // load index out of locals
 600   locals_index(rbx);
 601   __ movl(rax, iaddress(rbx));
 602   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 603 
 604   // rdx: array
 605   index_check(rdx, rax);
 606   // rax,: index
 607   __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 608   __ movl(rax, rbx);
 609 }
 610 
 611 void TemplateTable::saload() {
 612   transition(itos, itos);
 613   // rdx: array
 614   index_check(rdx, rax);  // kills rbx,
 615   // rax,: index
 616   // can do better code for P5 - may want to improve this at some point
 617   __ load_signed_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 618   __ movl(rax, rbx);
 619 }
 620 
 621 
 622 void TemplateTable::iload(int n) {
 623   transition(vtos, itos);
 624   __ movl(rax, iaddress(n));
 625   debug_only(__ verify_local_tag(frame::TagValue, n));
 626 }
 627 
 628 
 629 void TemplateTable::lload(int n) {
 630   transition(vtos, ltos);
 631   __ movl(rax, laddress(n));
 632   __ movl(rdx, haddress(n));
 633   debug_only(__ verify_local_tag(frame::TagCategory2, n));
 634 }
 635 
 636 
 637 void TemplateTable::fload(int n) {
 638   transition(vtos, ftos);
 639   __ fld_s(faddress(n));
 640   debug_only(__ verify_local_tag(frame::TagValue, n));
 641 }
 642 
 643 
 644 void TemplateTable::dload(int n) {
 645   transition(vtos, dtos);
 646   if (TaggedStackInterpreter) {
 647     // Get double out of locals array, onto temp stack and load with
 648     // float instruction into ST0
 649     __ movl(rax, laddress(n));
 650     __ movl(rdx, haddress(n));
 651     __ pushl(rdx);  // push hi first
 652     __ pushl(rax);
 653     __ fld_d(Address(rsp, 0));
 654     __ addl(rsp, 2*wordSize);  // reset rsp
 655     debug_only(__ verify_local_tag(frame::TagCategory2, n));
 656   } else {
 657     __ fld_d(daddress(n));
 658   }
 659 }
 660 
 661 
 662 void TemplateTable::aload(int n) {
 663   transition(vtos, atos);
 664   __ movl(rax, aaddress(n));
 665   debug_only(__ verify_local_tag(frame::TagReference, n));
 666 }
 667 
 668 
 669 void TemplateTable::aload_0() {
 670   transition(vtos, atos);
 671   // According to bytecode histograms, the pairs:
 672   //
 673   // _aload_0, _fast_igetfield
 674   // _aload_0, _fast_agetfield
 675   // _aload_0, _fast_fgetfield
 676   //
 677   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 678   // bytecode checks if the next bytecode is either _fast_igetfield, 
 679   // _fast_agetfield or _fast_fgetfield and then rewrites the
 680   // current bytecode into a pair bytecode; otherwise it rewrites the current
 681   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 682   //
 683   // Note: If the next bytecode is _getfield, the rewrite must be delayed,
 684   //       otherwise we may miss an opportunity for a pair.


 726     __ bind(rewrite);
 727     patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
 728 
 729     __ bind(done);
 730   } else {
 731     aload(0);
 732   }
 733 }
 734 
 735 void TemplateTable::istore() {
 736   transition(itos, vtos);
 737   locals_index(rbx);
 738   __ movl(iaddress(rbx), rax);
 739   __ tag_local(frame::TagValue, rbx);
 740 }
 741 
 742 
 743 void TemplateTable::lstore() {
 744   transition(ltos, vtos);
 745   locals_index(rbx);
 746   __ movl(laddress(rbx), rax);
 747   __ movl(haddress(rbx), rdx);
 748   __ tag_local(frame::TagCategory2, rbx);
 749 }
 750 
 751 
 752 void TemplateTable::fstore() {
 753   transition(ftos, vtos);
 754   locals_index(rbx);
 755   __ fstp_s(faddress(rbx));
 756   __ tag_local(frame::TagValue, rbx);
 757 }
 758 
 759 
 760 void TemplateTable::dstore() {
 761   transition(dtos, vtos);
 762   locals_index(rbx);
 763   if (TaggedStackInterpreter) {
 764     // Store double on stack and reload into locals nonadjacently
 765     __ subl(rsp, 2 * wordSize);
 766     __ fstp_d(Address(rsp, 0));
 767     __ popl(rax);
 768     __ popl(rdx);
 769     __ movl(laddress(rbx), rax);
 770     __ movl(haddress(rbx), rdx);
 771     __ tag_local(frame::TagCategory2, rbx);
 772   } else {
 773     __ fstp_d(daddress(rbx));
 774   }
 775 }
 776 
 777 
 778 void TemplateTable::astore() {
 779   transition(vtos, vtos);
 780   __ pop_ptr(rax, rdx);   // will need to pop tag too
 781   locals_index(rbx);
 782   __ movl(aaddress(rbx), rax);
 783   __ tag_local(rdx, rbx);    // need to store same tag in local may be returnAddr
 784 }
 785 
 786 
 787 void TemplateTable::wide_istore() {
 788   transition(vtos, vtos);
 789   __ pop_i(rax);
 790   locals_index_wide(rbx);
 791   __ movl(iaddress(rbx), rax);
 792   __ tag_local(frame::TagValue, rbx);
 793 }
 794 
 795 
 796 void TemplateTable::wide_lstore() {
 797   transition(vtos, vtos);
 798   __ pop_l(rax, rdx);
 799   locals_index_wide(rbx);
 800   __ movl(laddress(rbx), rax);
 801   __ movl(haddress(rbx), rdx);
 802   __ tag_local(frame::TagCategory2, rbx);
 803 }
 804 
 805 
 806 void TemplateTable::wide_fstore() {
 807   wide_istore();
 808 }
 809 
 810 
 811 void TemplateTable::wide_dstore() {
 812   wide_lstore();
 813 }
 814 
 815 
 816 void TemplateTable::wide_astore() {
 817   transition(vtos, vtos);
 818   __ pop_ptr(rax, rdx);
 819   locals_index_wide(rbx);
 820   __ movl(aaddress(rbx), rax);
 821   __ tag_local(rdx, rbx);
 822 }
 823 
 824 
 825 void TemplateTable::iastore() {
 826   transition(itos, vtos);
 827   __ pop_i(rbx);
 828   // rax,: value
 829   // rdx: array
 830   index_check(rdx, rbx);  // prefer index in rbx,
 831   // rbx,: index
 832   __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
 833 }
 834 
 835 
 836 void TemplateTable::lastore() {
 837   transition(ltos, vtos);
 838   __ pop_i(rbx);
 839   // rax,: low(value)
 840   // rcx: array
 841   // rdx: high(value)
 842   index_check(rcx, rbx);  // prefer index in rbx,
 843   // rbx,: index
 844   __ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
 845   __ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx);
 846 }
 847 
 848 
 849 void TemplateTable::fastore() {
 850   transition(ftos, vtos);
 851   __ pop_i(rbx);
 852   // rdx: array
 853   // st0: value
 854   index_check(rdx, rbx);  // prefer index in rbx,
 855   // rbx,: index
 856   __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 857 }
 858 
 859 
 860 void TemplateTable::dastore() {
 861   transition(dtos, vtos);
 862   __ pop_i(rbx);
 863   // rdx: array
 864   // st0: value
 865   index_check(rdx, rbx);  // prefer index in rbx,
 866   // rbx,: index
 867   __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 868 }
 869 
 870 
 871 void TemplateTable::aastore() {
 872   Label is_null, ok_is_subtype, done;
 873   transition(vtos, vtos);
 874   // stack: ..., array, index, value
 875   __ movl(rax, at_tos());     // Value
 876   __ movl(rcx, at_tos_p1());  // Index
 877   __ movl(rdx, at_tos_p2());  // Array


 878   index_check_without_pop(rdx, rcx);      // kills rbx,
 879   // do array store check - check for NULL value first
 880   __ testl(rax, rax);
 881   __ jcc(Assembler::zero, is_null);
 882 
 883   // Move subklass into EBX
 884   __ movl(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
 885   // Move superklass into EAX
 886   __ movl(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
 887   __ movl(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
 888   // Compress array+index*4+12 into a single register.  Frees ECX.
 889   __ leal(rdx, Address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 890 
 891   // Generate subtype check.  Blows ECX.  Resets EDI to locals.
 892   // Superklass in EAX.  Subklass in EBX.
 893   __ gen_subtype_check( rbx, ok_is_subtype );
 894 
 895   // Come here on failure
 896   // object is at TOS
 897   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
 898 
 899   // Come here on success
 900   __ bind(ok_is_subtype);
 901   __ movl(rax, at_rsp());     // Value
 902   __ movl(Address(rdx, 0), rax);
 903   __ store_check(rdx);
 904   __ jmpb(done);



 905 
 906   // Have a NULL in EAX, EDX=array, ECX=index.  Store NULL at ary[idx]
 907   __ bind(is_null);
 908   __ profile_null_seen(rbx);
 909   __ movl(Address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), rax);


 910 
 911   // Pop stack arguments
 912   __ bind(done);
 913   __ addl(rsp, 3 * Interpreter::stackElementSize());
 914 }
 915 
 916 
 917 void TemplateTable::bastore() {
 918   transition(itos, vtos);
 919   __ pop_i(rbx);
 920   // rax,: value
 921   // rdx: array
 922   index_check(rdx, rbx);  // prefer index in rbx,
 923   // rbx,: index
 924   __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
 925 }
 926 
 927 
 928 void TemplateTable::castore() {
 929   transition(itos, vtos);
 930   __ pop_i(rbx);
 931   // rax,: value
 932   // rdx: array
 933   index_check(rdx, rbx);  // prefer index in rbx,
 934   // rbx,: index
 935   __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
 936 }
 937 
 938 
 939 void TemplateTable::sastore() {
 940   castore();
 941 }
 942 
 943 
 944 void TemplateTable::istore(int n) {
 945   transition(itos, vtos);
 946   __ movl(iaddress(n), rax);
 947   __ tag_local(frame::TagValue, n);
 948 }
 949 
 950 
 951 void TemplateTable::lstore(int n) {
 952   transition(ltos, vtos);
 953   __ movl(laddress(n), rax);
 954   __ movl(haddress(n), rdx);
 955   __ tag_local(frame::TagCategory2, n);
 956 }
 957 
 958 
 959 void TemplateTable::fstore(int n) {
 960   transition(ftos, vtos);
 961   __ fstp_s(faddress(n));
 962   __ tag_local(frame::TagValue, n);
 963 }
 964 
 965 
 966 void TemplateTable::dstore(int n) {
 967   transition(dtos, vtos);
 968   if (TaggedStackInterpreter) {
 969     __ subl(rsp, 2 * wordSize);
 970     __ fstp_d(Address(rsp, 0));
 971     __ popl(rax);
 972     __ popl(rdx);
 973     __ movl(laddress(n), rax);
 974     __ movl(haddress(n), rdx);
 975     __ tag_local(frame::TagCategory2, n);
 976   } else {
 977     __ fstp_d(daddress(n));
 978   }
 979 }
 980 
 981 
 982 void TemplateTable::astore(int n) {
 983   transition(vtos, vtos);
 984   __ pop_ptr(rax, rdx);
 985   __ movl(aaddress(n), rax);
 986   __ tag_local(rdx, n);
 987 }
 988 
 989 
 990 void TemplateTable::pop() {
 991   transition(vtos, vtos);
 992   __ addl(rsp, Interpreter::stackElementSize());
 993 }
 994 
 995 
 996 void TemplateTable::pop2() {
 997   transition(vtos, vtos);
 998   __ addl(rsp, 2*Interpreter::stackElementSize());
 999 }
1000 
1001 
1002 void TemplateTable::dup() {
1003   transition(vtos, vtos);
1004   // stack: ..., a
1005   __ load_ptr_and_tag(0, rax, rdx);
1006   __ push_ptr(rax, rdx);
1007   // stack: ..., a, a
1008 }
1009 
1010 
1011 void TemplateTable::dup_x1() {
1012   transition(vtos, vtos);
1013   // stack: ..., a, b
1014   __ load_ptr_and_tag(0, rax, rdx);  // load b
1015   __ load_ptr_and_tag(1, rcx, rbx);  // load a
1016   __ store_ptr_and_tag(1, rax, rdx); // store b
1017   __ store_ptr_and_tag(0, rcx, rbx); // store a
1018   __ push_ptr(rax, rdx);             // push b


1085   // stack: ..., c, d, a, b, c, d
1086   // stack: ..., c, d, a, b, c, d
1087 }
1088 
1089 
1090 void TemplateTable::swap() {
1091   transition(vtos, vtos);
1092   // stack: ..., a, b
1093   __ load_ptr_and_tag(1, rcx, rbx);  // load a
1094   __ load_ptr_and_tag(0, rax, rdx);  // load b
1095   __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1096   __ store_ptr_and_tag(1, rax, rdx); // store b in a
1097   // stack: ..., b, a
1098 }
1099 
1100 
1101 void TemplateTable::iop2(Operation op) {
1102   transition(itos, itos);
1103   switch (op) {
1104     case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1105     case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1106     case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1107     case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1108     case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1109     case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1110     case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1111     case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1112     case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1113     default   : ShouldNotReachHere();
1114   }
1115 }
1116 
1117 
1118 void TemplateTable::lop2(Operation op) {
1119   transition(ltos, ltos);
1120   __ pop_l(rbx, rcx);
1121   switch (op) {
1122     case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1123     case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1124                __ movl(rax, rbx); __ movl(rdx, rcx); break;
1125     case _and: __ andl(rax, rbx); __ andl(rdx, rcx); break;
1126     case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1127     case _xor: __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1128     default : ShouldNotReachHere();
1129   }
1130 }
1131 
1132 
1133 void TemplateTable::idiv() {
1134   transition(itos, itos);
1135   __ movl(rcx, rax);
1136   __ pop_i(rax);
1137   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1138   //       they are not equal, one could do a normal division (no correction
1139   //       needed), which may speed up this implementation for the common case.
1140   //       (see also JVM spec., p.243 & p.271)
1141   __ corrected_idivl(rcx);
1142 }
1143 
1144 
1145 void TemplateTable::irem() {
1146   transition(itos, itos);
1147   __ movl(rcx, rax);
1148   __ pop_i(rax);
1149   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1150   //       they are not equal, one could do a normal division (no correction
1151   //       needed), which may speed up this implementation for the common case.
1152   //       (see also JVM spec., p.243 & p.271)
1153   __ corrected_idivl(rcx);
1154   __ movl(rax, rdx);
1155 }
1156 
1157 
1158 void TemplateTable::lmul() {
1159   transition(ltos, ltos);
1160   __ pop_l(rbx, rcx);
1161   __ pushl(rcx); __ pushl(rbx);
1162   __ pushl(rdx); __ pushl(rax);
1163   __ lmul(2 * wordSize, 0);
1164   __ addl(rsp, 4 * wordSize);  // take off temporaries
1165 }
1166 
1167 
1168 void TemplateTable::ldiv() {
1169   transition(ltos, ltos);
1170   __ pop_l(rbx, rcx);
1171   __ pushl(rcx); __ pushl(rbx);
1172   __ pushl(rdx); __ pushl(rax);
1173   // check if y = 0
1174   __ orl(rax, rdx);
1175   __ jump_cc(Assembler::zero,
1176              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1177   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1178   __ addl(rsp, 4 * wordSize);  // take off temporaries
1179 }
1180 
1181 
1182 void TemplateTable::lrem() {
1183   transition(ltos, ltos);
1184   __ pop_l(rbx, rcx);
1185   __ pushl(rcx); __ pushl(rbx);
1186   __ pushl(rdx); __ pushl(rax);
1187   // check if y = 0
1188   __ orl(rax, rdx);
1189   __ jump_cc(Assembler::zero,
1190              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1191   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1192   __ addl(rsp, 4 * wordSize);
1193 }
1194 
1195 
1196 void TemplateTable::lshl() {
1197   transition(itos, ltos);
1198   __ movl(rcx, rax);                             // get shift count
1199   __ pop_l(rax, rdx);                            // get shift value
1200   __ lshl(rdx, rax);
1201 }
1202 
1203 
1204 void TemplateTable::lshr() {
1205   transition(itos, ltos);
1206   __ movl(rcx, rax);                             // get shift count
1207   __ pop_l(rax, rdx);                            // get shift value
1208   __ lshr(rdx, rax, true);
1209 }
1210 
1211 
1212 void TemplateTable::lushr() {
1213   transition(itos, ltos);
1214   __ movl(rcx, rax);                             // get shift count
1215   __ pop_l(rax, rdx);                            // get shift value
1216   __ lshr(rdx, rax);
1217 }
1218 
1219 
1220 void TemplateTable::fop2(Operation op) {
1221   transition(ftos, ftos);
1222   __ pop_ftos_to_rsp();  // pop ftos into rsp
1223   switch (op) {
1224     case add: __ fadd_s (at_rsp());                break;
1225     case sub: __ fsubr_s(at_rsp());                break;
1226     case mul: __ fmul_s (at_rsp());                break;
1227     case div: __ fdivr_s(at_rsp());                break;
1228     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1229     default : ShouldNotReachHere();
1230   }
1231   __ f2ieee();
1232   __ popl(rax);  // pop float thing off
1233 }
1234 
1235 
1236 void TemplateTable::dop2(Operation op) {
1237   transition(dtos, dtos);
1238   __ pop_dtos_to_rsp();  // pop dtos into rsp
1239   
1240   switch (op) {
1241     case add: __ fadd_d (at_rsp());                break;
1242     case sub: __ fsubr_d(at_rsp());                break;
1243     case mul: {
1244       Label L_strict;
1245       Label L_join;
1246       const Address access_flags      (rcx, methodOopDesc::access_flags_offset());
1247       __ get_method(rcx);
1248       __ movl(rcx, access_flags);
1249       __ testl(rcx, JVM_ACC_STRICT);
1250       __ jccb(Assembler::notZero, L_strict);
1251       __ fmul_d (at_rsp());
1252       __ jmpb(L_join);


1266       __ get_method(rcx);
1267       __ movl(rcx, access_flags);
1268       __ testl(rcx, JVM_ACC_STRICT);
1269       __ jccb(Assembler::notZero, L_strict);
1270       __ fdivr_d(at_rsp());
1271       __ jmp(L_join);
1272       __ bind(L_strict);
1273       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1274       __ fmul_d (at_rsp());
1275       __ fdivrp();
1276       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1277       __ fmulp();
1278       __ bind(L_join);
1279       break;
1280     }
1281     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1282     default : ShouldNotReachHere();
1283   }
1284   __ d2ieee();
1285   // Pop double precision number from rsp.
1286   __ popl(rax);
1287   __ popl(rdx);
1288 }
1289 
1290 
1291 void TemplateTable::ineg() {
1292   transition(itos, itos);
1293   __ negl(rax);
1294 }
1295 
1296 
1297 void TemplateTable::lneg() {
1298   transition(ltos, ltos);
1299   __ lneg(rdx, rax);
1300 }
1301 
1302 
1303 void TemplateTable::fneg() {
1304   transition(ftos, ftos);
1305   __ fchs();
1306 }
1307 
1308 
1309 void TemplateTable::dneg() {
1310   transition(dtos, dtos);
1311   __ fchs();
1312 }
1313 
1314 
1315 void TemplateTable::iinc() {
1316   transition(vtos, vtos);
1317   __ load_signed_byte(rdx, at_bcp(2));           // get constant
1318   locals_index(rbx);
1319   __ addl(iaddress(rbx), rdx);
1320 }
1321 
1322 
1323 void TemplateTable::wide_iinc() {
1324   transition(vtos, vtos);
1325   __ movl(rdx, at_bcp(4));                       // get constant
1326   locals_index_wide(rbx);
1327   __ bswap(rdx);                                 // swap bytes & sign-extend constant
1328   __ sarl(rdx, 16);
1329   __ addl(iaddress(rbx), rdx);
1330   // Note: should probably use only one movl to get both
1331   //       the index and the constant -> fix this
1332 }
1333 
1334 
1335 void TemplateTable::convert() {
1336   // Checking
1337 #ifdef ASSERT
1338   { TosState tos_in  = ilgl;
1339     TosState tos_out = ilgl;
1340     switch (bytecode()) {
1341       case Bytecodes::_i2l: // fall through
1342       case Bytecodes::_i2f: // fall through
1343       case Bytecodes::_i2d: // fall through
1344       case Bytecodes::_i2b: // fall through
1345       case Bytecodes::_i2c: // fall through
1346       case Bytecodes::_i2s: tos_in = itos; break;
1347       case Bytecodes::_l2i: // fall through


1361       case Bytecodes::_d2i: // fall through
1362       case Bytecodes::_i2b: // fall through
1363       case Bytecodes::_i2c: // fall through
1364       case Bytecodes::_i2s: tos_out = itos; break;
1365       case Bytecodes::_i2l: // fall through
1366       case Bytecodes::_f2l: // fall through
1367       case Bytecodes::_d2l: tos_out = ltos; break;
1368       case Bytecodes::_i2f: // fall through
1369       case Bytecodes::_l2f: // fall through
1370       case Bytecodes::_d2f: tos_out = ftos; break;
1371       case Bytecodes::_i2d: // fall through
1372       case Bytecodes::_l2d: // fall through
1373       case Bytecodes::_f2d: tos_out = dtos; break;
1374       default             : ShouldNotReachHere();
1375     }
1376     transition(tos_in, tos_out);
1377   }
1378 #endif // ASSERT
1379 
1380   // Conversion
1381   // (Note: use pushl(rcx)/popl(rcx) for 1/2-word stack-ptr manipulation)
1382   switch (bytecode()) {
1383     case Bytecodes::_i2l:
1384       __ extend_sign(rdx, rax);
1385       break;
1386     case Bytecodes::_i2f:
1387       __ pushl(rax);         // store int on tos
1388       __ fild_s(at_rsp());   // load int to ST0
1389       __ f2ieee();           // truncate to float size
1390       __ popl(rcx);          // adjust rsp
1391       break;
1392     case Bytecodes::_i2d:
1393       __ pushl(rax);         // add one slot for d2ieee()
1394       __ pushl(rax);         // store int on tos
1395       __ fild_s(at_rsp());   // load int to ST0
1396       __ d2ieee();           // truncate to double size
1397       __ popl(rcx);          // adjust rsp
1398       __ popl(rcx);
1399       break;
1400     case Bytecodes::_i2b:
1401       __ shll(rax, 24);      // truncate upper 24 bits
1402       __ sarl(rax, 24);      // and sign-extend byte

1403       break;
1404     case Bytecodes::_i2c:
1405       __ andl(rax, 0xFFFF);  // truncate upper 16 bits

1406       break;
1407     case Bytecodes::_i2s:
1408       __ shll(rax, 16);      // truncate upper 16 bits
1409       __ sarl(rax, 16);      // and sign-extend short

1410       break;
1411     case Bytecodes::_l2i:
1412       /* nothing to do */
1413       break;
1414     case Bytecodes::_l2f:
1415       __ pushl(rdx);         // store long on tos
1416       __ pushl(rax);
1417       __ fild_d(at_rsp());   // load long to ST0
1418       __ f2ieee();           // truncate to float size
1419       __ popl(rcx);          // adjust rsp
1420       __ popl(rcx);
1421       break;
1422     case Bytecodes::_l2d:
1423       __ pushl(rdx);         // store long on tos
1424       __ pushl(rax);
1425       __ fild_d(at_rsp());   // load long to ST0
1426       __ d2ieee();           // truncate to double size
1427       __ popl(rcx);          // adjust rsp
1428       __ popl(rcx);
1429       break;
1430     case Bytecodes::_f2i:
1431       __ pushl(rcx);         // reserve space for argument
1432       __ fstp_s(at_rsp());   // pass float argument on stack
1433       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1434       break;
1435     case Bytecodes::_f2l:
1436       __ pushl(rcx);         // reserve space for argument
1437       __ fstp_s(at_rsp());   // pass float argument on stack
1438       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1439       break;
1440     case Bytecodes::_f2d:
1441       /* nothing to do */
1442       break;
1443     case Bytecodes::_d2i:
1444       __ pushl(rcx);         // reserve space for argument
1445       __ pushl(rcx);
1446       __ fstp_d(at_rsp());   // pass double argument on stack
1447       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1448       break;
1449     case Bytecodes::_d2l:
1450       __ pushl(rcx);         // reserve space for argument
1451       __ pushl(rcx);
1452       __ fstp_d(at_rsp());   // pass double argument on stack
1453       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1454       break;
1455     case Bytecodes::_d2f:
1456       __ pushl(rcx);         // reserve space for f2ieee()
1457       __ f2ieee();           // truncate to float size
1458       __ popl(rcx);          // adjust rsp
1459       break;
1460     default             :
1461       ShouldNotReachHere();
1462   }
1463 }
1464 
1465 
1466 void TemplateTable::lcmp() {
1467   transition(ltos, itos);
1468   // y = rdx:rax
1469   __ pop_l(rbx, rcx);             // get x = rcx:rbx
1470   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1471   __ movl(rax, rcx);
1472 }
1473 
1474 
1475 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1476   if (is_float) {
1477     __ pop_ftos_to_rsp();
1478     __ fld_s(at_rsp());
1479   } else {
1480     __ pop_dtos_to_rsp();
1481     __ fld_d(at_rsp());
1482     __ popl(rdx);
1483   }
1484   __ popl(rcx);
1485   __ fcmp2int(rax, unordered_result < 0);
1486 }
1487 
1488 
1489 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1490   __ get_method(rcx);           // ECX holds method
1491   __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1492 
1493   const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1494   const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1495   const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1496 
1497   // Load up EDX with the branch displacement
1498   __ movl(rdx, at_bcp(1));
1499   __ bswap(rdx);
1500   if (!is_wide) __ sarl(rdx, 16);


1501 
1502   // Handle all the JSR stuff here, then exit.
1503   // It's much shorter and cleaner than intermingling with the
1504   // non-JSR normal-branch stuff occuring below.
1505   if (is_jsr) {
1506     // Pre-load the next target bytecode into EBX
1507     __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1508 
1509     // compute return address as bci in rax,
1510     __ leal(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1511     __ subl(rax, Address(rcx, methodOopDesc::const_offset()));
1512     // Adjust the bcp in ESI by the displacement in EDX
1513     __ addl(rsi, rdx);
1514     // Push return address
1515     __ push_i(rax);
1516     // jsr returns vtos
1517     __ dispatch_only_noverify(vtos);
1518     return;
1519   }
1520 
1521   // Normal (non-jsr) branch handling
1522 
1523   // Adjust the bcp in ESI by the displacement in EDX
1524   __ addl(rsi, rdx);
1525 
1526   assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1527   Label backedge_counter_overflow;
1528   Label profile_method;
1529   Label dispatch;
1530   if (UseLoopCounter) {
1531     // increment backedge counter for backward branches
1532     // rax,: MDO
1533     // rbx,: MDO bumped taken-count
1534     // rcx: method
1535     // rdx: target offset
1536     // rsi: target bcp
1537     // rdi: locals pointer
1538     __ testl(rdx, rdx);             // check if forward or backward branch
1539     __ jcc(Assembler::positive, dispatch); // count only if backward branch
1540 
1541     // increment counter 
1542     __ movl(rax, Address(rcx, be_offset));        // load backedge counter
1543     __ increment(rax, InvocationCounter::count_increment); // increment counter
1544     __ movl(Address(rcx, be_offset), rax);        // store counter
1545 
1546     __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
1547     __ andl(rax, InvocationCounter::count_mask_value);     // and the status bits
1548     __ addl(rax, Address(rcx, be_offset));        // add both counters
1549 
1550     if (ProfileInterpreter) {
1551       // Test to see if we should create a method data oop
1552       __ cmp32(rax,
1553                ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1554       __ jcc(Assembler::less, dispatch);
1555 
1556       // if no method data exists, go to profile method
1557       __ test_method_data_pointer(rax, profile_method);
1558 
1559       if (UseOnStackReplacement) {
1560         // check for overflow against rbx, which is the MDO taken count
1561         __ cmp32(rbx,
1562                  ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1563         __ jcc(Assembler::below, dispatch);
1564 
1565         // When ProfileInterpreter is on, the backedge_count comes from the 
1566         // methodDataOop, which value does not get reset on the call to 
1567         // frequency_counter_overflow().  To avoid excessive calls to the overflow
1568         // routine while the method is being compiled, add a second test to make 
1569         // sure the overflow function is called only once every overflow_frequency.
1570         const int overflow_frequency = 1024;
1571         __ andl(rbx, overflow_frequency-1);
1572         __ jcc(Assembler::zero, backedge_counter_overflow);
1573 
1574       }
1575     } else {
1576       if (UseOnStackReplacement) {
1577         // check for overflow against rax, which is the sum of the counters
1578         __ cmp32(rax,
1579                  ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1580         __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1581 
1582       }
1583     }
1584     __ bind(dispatch);
1585   }
1586 
1587   // Pre-load the next target bytecode into EBX
1588   __ load_unsigned_byte(rbx, Address(rsi, 0));
1589 
1590   // continue with the bytecode @ target
1591   // rax,: return bci for jsr's, unused otherwise
1592   // rbx,: target bytecode
1593   // rsi: target bcp
1594   __ dispatch_only(vtos);
1595 
1596   if (UseLoopCounter) {
1597     if (ProfileInterpreter) {
1598       // Out-of-line code to allocate method data oop.
1599       __ bind(profile_method);
1600       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1601       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1602       __ movl(rcx, Address(rbp, method_offset));
1603       __ movl(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1604       __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1605       __ test_method_data_pointer(rcx, dispatch);
1606       // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1607       __ addl(rcx, in_bytes(methodDataOopDesc::data_offset()));
1608       __ addl(rcx, rax);
1609       __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1610       __ jmp(dispatch);
1611     }
1612 
1613     if (UseOnStackReplacement) {
1614 
1615       // invocation counter overflow
1616       __ bind(backedge_counter_overflow);
1617       __ negl(rdx);
1618       __ addl(rdx, rsi);        // branch bcp
1619       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1620       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1621 
1622       // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1623       // rbx,: target bytecode
1624       // rdx: scratch
1625       // rdi: locals pointer
1626       // rsi: bcp
1627       __ testl(rax, rax);                        // test result
1628       __ jcc(Assembler::zero, dispatch);         // no osr if null
1629       // nmethod may have been invalidated (VM may block upon call_VM return)
1630       __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1631       __ cmpl(rcx, InvalidOSREntryBci);
1632       __ jcc(Assembler::equal, dispatch);
1633       
1634       // We have the address of an on stack replacement routine in rax,        
1635       // We need to prepare to execute the OSR method. First we must
1636       // migrate the locals and monitors off of the stack.
1637 
1638       __ movl(rsi, rax);                             // save the nmethod
1639 
1640       const Register thread = rcx;
1641       __ get_thread(thread);
1642       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1643       // rax, is OSR buffer, move it to expected parameter location
1644       __ movl(rcx, rax);
1645 
1646       // pop the interpreter frame
1647       __ movl(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1648       __ leave();                                // remove frame anchor
1649       __ popl(rdi);                              // get return address
1650       __ movl(rsp, rdx);                         // set sp to sender sp
1651 
1652 
1653       Label skip;
1654       Label chkint;
1655 
1656       // The interpreter frame we have removed may be returning to
1657       // either the callstub or the interpreter. Since we will
1658       // now be returning from a compiled (OSR) nmethod we must
1659       // adjust the return to the return were it can handler compiled
1660       // results and clean the fpu stack. This is very similar to
1661       // what a i2c adapter must do.
1662 
1663       // Are we returning to the call stub?
1664 
1665       __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1666       __ jcc(Assembler::notEqual, chkint);
1667 
1668       // yes adjust to the specialized call stub  return.
1669       assert(StubRoutines::i486::get_call_stub_compiled_return() != NULL, "must be set");
1670       __ lea(rdi, ExternalAddress(StubRoutines::i486::get_call_stub_compiled_return()));
1671       __ jmp(skip);
1672 
1673       __ bind(chkint);
1674 
1675       // Are we returning to the interpreter? Look for sentinel
1676 
1677       __ cmpl(Address(rdi, -8), Interpreter::return_sentinel);
1678       __ jcc(Assembler::notEqual, skip);
1679 
1680       // Adjust to compiled return back to interpreter
1681 
1682       __ movl(rdi, Address(rdi, -4));
1683       __ bind(skip);
1684 
1685       // Align stack pointer for compiled code (note that caller is
1686       // responsible for undoing this fixup by remembering the old SP
1687       // in an rbp,-relative location)
1688       __ andl(rsp, -(StackAlignmentInBytes));
1689 
1690       // push the (possibly adjusted) return address
1691       __ pushl(rdi);
1692 
1693       // and begin the OSR nmethod
1694       __ jmp(Address(rsi, nmethod::osr_entry_point_offset()));
1695     }
1696   }
1697 }
1698 
1699 
1700 void TemplateTable::if_0cmp(Condition cc) {
1701   transition(itos, vtos);
1702   // assume branch is more often taken than not (loops use backward branches)
1703   Label not_taken;
1704   __ testl(rax, rax);
1705   __ jcc(j_not(cc), not_taken);
1706   branch(false, false);
1707   __ bind(not_taken);
1708   __ profile_not_taken_branch(rax);
1709 }
1710 
1711 
1712 void TemplateTable::if_icmp(Condition cc) {
1713   transition(itos, vtos);
1714   // assume branch is more often taken than not (loops use backward branches)
1715   Label not_taken;
1716   __ pop_i(rdx);
1717   __ cmpl(rdx, rax);
1718   __ jcc(j_not(cc), not_taken);
1719   branch(false, false);
1720   __ bind(not_taken);
1721   __ profile_not_taken_branch(rax);
1722 }
1723 
1724 
1725 void TemplateTable::if_nullcmp(Condition cc) {
1726   transition(atos, vtos);
1727   // assume branch is more often taken than not (loops use backward branches)
1728   Label not_taken;
1729   __ testl(rax, rax);
1730   __ jcc(j_not(cc), not_taken);
1731   branch(false, false);
1732   __ bind(not_taken);
1733   __ profile_not_taken_branch(rax);
1734 }
1735 
1736 
1737 void TemplateTable::if_acmp(Condition cc) {
1738   transition(atos, vtos);
1739   // assume branch is more often taken than not (loops use backward branches)
1740   Label not_taken;
1741   __ pop_ptr(rdx);
1742   __ cmpl(rdx, rax);
1743   __ jcc(j_not(cc), not_taken);
1744   branch(false, false);
1745   __ bind(not_taken);
1746   __ profile_not_taken_branch(rax);
1747 }
1748 
1749 
1750 void TemplateTable::ret() {
1751   transition(vtos, vtos);
1752   locals_index(rbx);
1753   __ movl(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1754   __ profile_ret(rbx, rcx);
1755   __ get_method(rax);
1756   __ movl(rsi, Address(rax, methodOopDesc::const_offset()));
1757   __ leal(rsi, Address(rsi, rbx, Address::times_1,
1758                        constMethodOopDesc::codes_offset()));
1759   __ dispatch_next(vtos);
1760 }
1761 
1762 
1763 void TemplateTable::wide_ret() {
1764   transition(vtos, vtos);
1765   locals_index_wide(rbx);
1766   __ movl(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1767   __ profile_ret(rbx, rcx);
1768   __ get_method(rax);
1769   __ movl(rsi, Address(rax, methodOopDesc::const_offset()));
1770   __ leal(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1771   __ dispatch_next(vtos);
1772 }
1773 
1774 
1775 void TemplateTable::tableswitch() {
1776   Label default_case, continue_execution;
1777   transition(itos, vtos);
1778   // align rsi
1779   __ leal(rbx, at_bcp(wordSize));
1780   __ andl(rbx, -wordSize);
1781   // load lo & hi
1782   __ movl(rcx, Address(rbx, 1 * wordSize));
1783   __ movl(rdx, Address(rbx, 2 * wordSize));
1784   __ bswap(rcx);
1785   __ bswap(rdx);
1786   // check against lo & hi
1787   __ cmpl(rax, rcx);
1788   __ jccb(Assembler::less, default_case);
1789   __ cmpl(rax, rdx);
1790   __ jccb(Assembler::greater, default_case);
1791   // lookup dispatch offset
1792   __ subl(rax, rcx);
1793   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * wordSize));
1794   __ profile_switch_case(rax, rbx, rcx);
1795   // continue execution
1796   __ bind(continue_execution);
1797   __ bswap(rdx);
1798   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1799   __ addl(rsi, rdx);
1800   __ dispatch_only(vtos);
1801   // handle default
1802   __ bind(default_case);
1803   __ profile_switch_default(rax);
1804   __ movl(rdx, Address(rbx, 0));
1805   __ jmp(continue_execution);
1806 }
1807 
1808 
1809 void TemplateTable::lookupswitch() {
1810   transition(itos, itos);
1811   __ stop("lookupswitch bytecode should have been rewritten");
1812 }
1813 
1814 
1815 void TemplateTable::fast_linearswitch() {
1816   transition(itos, vtos);
1817   Label loop_entry, loop, found, continue_execution;  
1818   // bswap rax, so we can avoid bswapping the table entries
1819   __ bswap(rax);
1820   // align rsi
1821   __ leal(rbx, at_bcp(wordSize));                // btw: should be able to get rid of this instruction (change offsets below)
1822   __ andl(rbx, -wordSize);
1823   // set counter
1824   __ movl(rcx, Address(rbx, wordSize));  
1825   __ bswap(rcx);
1826   __ jmpb(loop_entry);
1827   // table search
1828   __ bind(loop);
1829   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1830   __ jccb(Assembler::equal, found);
1831   __ bind(loop_entry);
1832   __ decrement(rcx);
1833   __ jcc(Assembler::greaterEqual, loop);
1834   // default case
1835   __ profile_switch_default(rax);
1836   __ movl(rdx, Address(rbx, 0));
1837   __ jmpb(continue_execution);
1838   // entry found -> get offset
1839   __ bind(found);
1840   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1841   __ profile_switch_case(rcx, rax, rbx);
1842   // continue execution
1843   __ bind(continue_execution);  
1844   __ bswap(rdx);
1845   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1846   __ addl(rsi, rdx);
1847   __ dispatch_only(vtos);
1848 }
1849 
1850 
1851 void TemplateTable::fast_binaryswitch() {
1852   transition(itos, vtos);
1853   // Implementation using the following core algorithm:
1854   //
1855   // int binary_search(int key, LookupswitchPair* array, int n) {
1856   //   // Binary search according to "Methodik des Programmierens" by
1857   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1858   //   int i = 0;
1859   //   int j = n;
1860   //   while (i+1 < j) {
1861   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1862   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1863   //     // where a stands for the array and assuming that the (inexisting)
1864   //     // element a[n] is infinitely big.
1865   //     int h = (i + j) >> 1;
1866   //     // i < h < j


1868   //       j = h;
1869   //     } else {
1870   //       i = h;
1871   //     }
1872   //   }
1873   //   // R: a[i] <= key < a[i+1] or Q
1874   //   // (i.e., if key is within array, i is the correct index)
1875   //   return i;
1876   // }
1877 
1878   // register allocation
1879   const Register key   = rax;                    // already set (tosca)
1880   const Register array = rbx;
1881   const Register i     = rcx;
1882   const Register j     = rdx;
1883   const Register h     = rdi;                    // needs to be restored
1884   const Register temp  = rsi;
1885   // setup array
1886   __ save_bcp();
1887 
1888   __ leal(array, at_bcp(3*wordSize));            // btw: should be able to get rid of this instruction (change offsets below)
1889   __ andl(array, -wordSize);
1890   // initialize i & j
1891   __ xorl(i, i);                                 // i = 0;
1892   __ movl(j, Address(array, -wordSize));         // j = length(array);    
1893   // Convert j into native byteordering  
1894   __ bswap(j);
1895   // and start
1896   Label entry;
1897   __ jmp(entry);
1898 
1899   // binary search loop
1900   { Label loop;
1901     __ bind(loop);
1902     // int h = (i + j) >> 1;
1903     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1904     __ sarl(h, 1);                               // h = (i + j) >> 1;
1905     // if (key < array[h].fast_match()) {
1906     //   j = h;
1907     // } else {
1908     //   i = h;
1909     // }
1910     // Convert array[h].match to native byte-ordering before compare
1911     __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1912     __ bswap(temp);
1913     __ cmpl(key, temp);
1914     if (VM_Version::supports_cmov()) {
1915       __ cmovl(Assembler::less        , j, h);   // j = h if (key <  array[h].fast_match())
1916       __ cmovl(Assembler::greaterEqual, i, h);   // i = h if (key >= array[h].fast_match())
1917     } else {
1918       Label set_i, end_of_if;
1919       __ jccb(Assembler::greaterEqual, set_i);    // {
1920       __ movl(j, h);                             //   j = h;
1921       __ jmp(end_of_if);                         // }
1922       __ bind(set_i);                            // else {
1923       __ movl(i, h);                             //   i = h;
1924       __ bind(end_of_if);                        // }
1925     }
1926     // while (i+1 < j)
1927     __ bind(entry);
1928     __ leal(h, Address(i, 1));                   // i+1
1929     __ cmpl(h, j);                               // i+1 < j
1930     __ jcc(Assembler::less, loop);
1931   }
1932 
1933   // end of binary search, result index is i (must check again!)
1934   Label default_case;
1935   // Convert array[i].match to native byte-ordering before compare
1936   __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1937   __ bswap(temp);
1938   __ cmpl(key, temp);
1939   __ jcc(Assembler::notEqual, default_case);
1940 
1941   // entry found -> j = offset
1942   __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1943   __ profile_switch_case(i, key, array);
1944   __ bswap(j);

1945   __ restore_bcp();
1946   __ restore_locals();                           // restore rdi
1947   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1948   
1949   __ addl(rsi, j);
1950   __ dispatch_only(vtos);
1951 
1952   // default case -> j = default offset
1953   __ bind(default_case);
1954   __ profile_switch_default(i);
1955   __ movl(j, Address(array, -2*wordSize));
1956   __ bswap(j);

1957   __ restore_bcp();
1958   __ restore_locals();                           // restore rdi
1959   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1960   __ addl(rsi, j);
1961   __ dispatch_only(vtos);
1962 }
1963 
1964 
1965 void TemplateTable::_return(TosState state) {
1966   transition(state, state);
1967   assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
1968 
1969   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1970     assert(state == vtos, "only valid state");
1971     __ movl(rax, aaddress(0));
1972     __ movl(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
1973     __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1974     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
1975     Label skip_register_finalizer;
1976     __ jcc(Assembler::zero, skip_register_finalizer);
1977 
1978     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
1979 
1980     __ bind(skip_register_finalizer);
1981   }
1982 
1983   __ remove_activation(state, rsi);
1984   __ jmp(rsi);
1985 }
1986 
1987 
1988 // ----------------------------------------------------------------------------
1989 // Volatile variables demand their effects be made known to all CPU's in
1990 // order.  Store buffers on most chips allow reads & writes to reorder; the
1991 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
1992 // memory barrier (i.e., it's not sufficient that the interpreter does not
1993 // reorder volatile references, the hardware also must not reorder them).
1994 // 
1995 // According to the new Java Memory Model (JMM):
1996 // (1) All volatiles are serialized wrt to each other.  
1997 // ALSO reads & writes act as aquire & release, so:
1998 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
1999 // the read float up to before the read.  It's OK for non-volatile memory refs
2000 // that happen before the volatile read to float down below it.
2001 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2002 // that happen BEFORE the write float down to after the write.  It's OK for
2003 // non-volatile memory refs that happen after the volatile write to float up
2004 // before it.
2005 //
2006 // We only put in barriers around volatile refs (they are expensive), not
2007 // _between_ memory refs (that would require us to track the flavor of the
2008 // previous memory refs).  Requirements (2) and (3) require some barriers
2009 // before volatile stores and after volatile loads.  These nearly cover
2010 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2011 // case is placed after volatile-stores although it could just as well go
2012 // before volatile-loads.
2013 void TemplateTable::volatile_barrier( ) {
2014   // Helper function to insert a is-volatile test and memory barrier
2015   if( !os::is_MP() ) return;    // Not needed on single CPU
2016   __ membar();
2017 }
2018 
2019 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
2020   assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
2021 
2022   Register temp = rbx;
2023 
2024   assert_different_registers(Rcache, index, temp);
2025 
2026   const int shift_count = (1 + byte_no)*BitsPerByte;
2027   Label resolved;
2028   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2029   __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));



2030   __ shrl(temp, shift_count);
2031   // have we resolved this bytecode? 
2032   __ andl(temp, 0xFF);
2033   __ cmpl(temp, (int)bytecode());
2034   __ jcc(Assembler::equal, resolved);
2035 
2036   // resolve first time through
2037   address entry;
2038   switch (bytecode()) {
2039     case Bytecodes::_getstatic      : // fall through
2040     case Bytecodes::_putstatic      : // fall through
2041     case Bytecodes::_getfield       : // fall through
2042     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2043     case Bytecodes::_invokevirtual  : // fall through
2044     case Bytecodes::_invokespecial  : // fall through
2045     case Bytecodes::_invokestatic   : // fall through
2046     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
2047     default                         : ShouldNotReachHere();                                 break;
2048   }
2049   __ movl(temp, (int)bytecode()); 
2050   __ call_VM(noreg, entry, temp);
2051   // Update registers with resolved info
2052   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2053   __ bind(resolved);
2054 }
2055 
2056 
2057 // The cache and index registers must be set before call
2058 void TemplateTable::load_field_cp_cache_entry(Register obj,
2059                                               Register cache,
2060                                               Register index,
2061                                               Register off,
2062                                               Register flags,
2063                                               bool is_static = false) {
2064   assert_different_registers(cache, index, flags, off);
2065 
2066   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2067   // Field offset
2068   __ movl(off, Address(cache, index, Address::times_4, 
2069            in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2070   // Flags    
2071   __ movl(flags, Address(cache, index, Address::times_4,
2072            in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2073 
2074   // klass     overwrite register
2075   if (is_static) {
2076     __ movl(obj, Address(cache, index, Address::times_4,
2077              in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2078   }
2079 }
2080 
2081 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2082                                                Register method,
2083                                                Register itable_index,
2084                                                Register flags,
2085                                                bool is_invokevirtual,
2086                                                bool is_invokevfinal /*unused*/) {
2087   // setup registers
2088   const Register cache = rcx;
2089   const Register index = rdx;
2090   assert_different_registers(method, flags);
2091   assert_different_registers(method, cache, index);
2092   assert_different_registers(itable_index, flags);
2093   assert_different_registers(itable_index, cache, index);
2094   // determine constant pool cache field offsets
2095   const int method_offset = in_bytes(
2096     constantPoolCacheOopDesc::base_offset() +
2097       (is_invokevirtual
2098        ? ConstantPoolCacheEntry::f2_offset()
2099        : ConstantPoolCacheEntry::f1_offset()
2100       )
2101     );
2102   const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2103                                     ConstantPoolCacheEntry::flags_offset());
2104   // access constant pool cache fields
2105   const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2106                                     ConstantPoolCacheEntry::f2_offset());
2107 
2108   resolve_cache_and_index(byte_no, cache, index);
2109 
2110   assert(wordSize == 4, "adjust code below");
2111   __ movl(method, Address(cache, index, Address::times_4, method_offset));
2112   if (itable_index != noreg) {
2113     __ movl(itable_index, Address(cache, index, Address::times_4, index_offset));
2114   }
2115   __ movl(flags , Address(cache, index, Address::times_4, flags_offset ));
2116 }
2117 
2118 
2119 // The registers cache and index expected to be set before call.
2120 // Correct values of the cache and index registers are preserved.
2121 void TemplateTable::jvmti_post_field_access(Register cache,
2122                                             Register index,
2123                                             bool is_static,
2124                                             bool has_tos) {
2125   if (JvmtiExport::can_post_field_access()) {
2126     // Check to see if a field access watch has been set before we take
2127     // the time to call into the VM.
2128     Label L1;
2129     assert_different_registers(cache, index, rax);
2130     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2131     __ testl(rax,rax);
2132     __ jcc(Assembler::zero, L1);
2133 
2134     // cache entry pointer
2135     __ addl(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2136     __ shll(index, LogBytesPerWord);
2137     __ addl(cache, index);
2138     if (is_static) {
2139       __ movl(rax, 0);      // NULL object reference
2140     } else {
2141       __ pop(atos);         // Get the object
2142       __ verify_oop(rax);
2143       __ push(atos);        // Restore stack state
2144     }
2145     // rax,:   object pointer or NULL
2146     // cache: cache entry pointer
2147     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2148                rax, cache);
2149     __ get_cache_and_index_at_bcp(cache, index, 1);
2150     __ bind(L1);
2151   } 
2152 }
2153 
2154 void TemplateTable::pop_and_check_object(Register r) {
2155   __ pop_ptr(r);
2156   __ null_check(r);  // for field access must check obj.
2157   __ verify_oop(r);
2158 }
2159 


2163   const Register cache = rcx;
2164   const Register index = rdx;
2165   const Register obj   = rcx;
2166   const Register off   = rbx;
2167   const Register flags = rax;
2168 
2169   resolve_cache_and_index(byte_no, cache, index);
2170   jvmti_post_field_access(cache, index, is_static, false);
2171   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2172 
2173   if (!is_static) pop_and_check_object(obj);
2174 
2175   const Address lo(obj, off, Address::times_1, 0*wordSize);
2176   const Address hi(obj, off, Address::times_1, 1*wordSize);
2177 
2178   Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2179 
2180   __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2181   assert(btos == 0, "change code, btos != 0");
2182   // btos
2183   __ andl(flags, 0x0f);
2184   __ jcc(Assembler::notZero, notByte);
2185 
2186   __ load_signed_byte(rax, lo ); 
2187   __ push(btos);
2188   // Rewrite bytecode to be faster
2189   if (!is_static) {
2190     patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2191   }
2192   __ jmp(Done);
2193 
2194   __ bind(notByte);
2195   // itos
2196   __ cmpl(flags, itos );
2197   __ jcc(Assembler::notEqual, notInt);
2198 
2199   __ movl(rax, lo );
2200   __ push(itos);
2201   // Rewrite bytecode to be faster
2202   if (!is_static) {
2203     patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);


2231   __ bind(notChar);
2232   // stos
2233   __ cmpl(flags, stos );
2234   __ jcc(Assembler::notEqual, notShort);
2235 
2236   __ load_signed_word(rax, lo );
2237   __ push(stos);
2238   if (!is_static) {
2239     patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2240   }
2241   __ jmp(Done);
2242 
2243   __ bind(notShort);
2244   // ltos
2245   __ cmpl(flags, ltos );
2246   __ jcc(Assembler::notEqual, notLong);
2247 
2248   // Generate code as if volatile.  There just aren't enough registers to
2249   // save that information and this code is faster than the test.
2250   __ fild_d(lo);                // Must load atomically
2251   __ subl(rsp,2*wordSize);      // Make space for store
2252   __ fistp_d(Address(rsp,0));
2253   __ popl(rax);
2254   __ popl(rdx);
2255 
2256   __ push(ltos);
2257   // Don't rewrite to _fast_lgetfield for potential volatile case.
2258   __ jmp(Done);
2259 
2260   __ bind(notLong);
2261   // ftos
2262   __ cmpl(flags, ftos );
2263   __ jcc(Assembler::notEqual, notFloat);
2264 
2265   __ fld_s(lo);
2266   __ push(ftos);
2267   if (!is_static) {
2268     patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2269   }
2270   __ jmp(Done);
2271 
2272   __ bind(notFloat);
2273   // dtos
2274   __ cmpl(flags, dtos );


2305 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2306 
2307   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2308 
2309   if (JvmtiExport::can_post_field_modification()) {
2310     // Check to see if a field modification watch has been set before we take
2311     // the time to call into the VM.
2312     Label L1;
2313     assert_different_registers(cache, index, rax);
2314     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2315     __ testl(rax, rax);
2316     __ jcc(Assembler::zero, L1);
2317 
2318     // The cache and index registers have been already set.
2319     // This allows to eliminate this call but the cache and index
2320     // registers have to be correspondingly used after this line.
2321     __ get_cache_and_index_at_bcp(rax, rdx, 1);
2322 
2323     if (is_static) {
2324       // Life is simple.  Null out the object pointer.
2325       __ xorl(rbx, rbx);
2326     } else {
2327       // Life is harder. The stack holds the value on top, followed by the object.
2328       // We don't know the size of the value, though; it could be one or two words
2329       // depending on its type. As a result, we must find the type to determine where
2330       // the object is.
2331       Label two_word, valsize_known;
2332       __ movl(rcx, Address(rax, rdx, Address::times_4, in_bytes(cp_base_offset +
2333                                    ConstantPoolCacheEntry::flags_offset())));
2334       __ movl(rbx, rsp);
2335       __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2336       // Make sure we don't need to mask rcx for tosBits after the above shift
2337       ConstantPoolCacheEntry::verify_tosBits();
2338       __ cmpl(rcx, ltos);
2339       __ jccb(Assembler::equal, two_word);
2340       __ cmpl(rcx, dtos);
2341       __ jccb(Assembler::equal, two_word);
2342       __ addl(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2343       __ jmpb(valsize_known);
2344 
2345       __ bind(two_word);
2346       __ addl(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2347     
2348       __ bind(valsize_known);
2349       // setup object pointer
2350       __ movl(rbx, Address(rbx, 0));
2351     }
2352     // cache entry pointer
2353     __ addl(rax, in_bytes(cp_base_offset));
2354     __ shll(rdx, LogBytesPerWord);
2355     __ addl(rax, rdx);
2356     // object (tos)
2357     __ movl(rcx, rsp);
2358     // rbx,: object pointer set up above (NULL if static)
2359     // rax,: cache entry pointer
2360     // rcx: jvalue object on the stack
2361     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2362                rbx, rax, rcx);
2363     __ get_cache_and_index_at_bcp(cache, index, 1);
2364     __ bind(L1);
2365   }
2366 }
2367 
2368 
2369 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2370   transition(vtos, vtos);
2371 
2372   const Register cache = rcx;
2373   const Register index = rdx;
2374   const Register obj   = rcx;
2375   const Register off   = rbx;
2376   const Register flags = rax;
2377 


2412   __ cmpl(flags, itos );
2413   __ jcc(Assembler::notEqual, notInt);
2414 
2415   __ pop(itos);
2416   if (!is_static) pop_and_check_object(obj);
2417 
2418   __ movl(lo, rax );
2419   if (!is_static) {
2420     patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2421   }
2422   __ jmp(Done);
2423   
2424   __ bind(notInt);
2425   // atos
2426   __ cmpl(flags, atos );
2427   __ jcc(Assembler::notEqual, notObj);
2428 
2429   __ pop(atos);
2430   if (!is_static) pop_and_check_object(obj);
2431 
2432   __ movl(lo, rax );
2433   __ store_check(obj, lo);  // Need to mark card
2434   if (!is_static) {
2435     patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2436   }

2437   __ jmp(Done);
2438 
2439   __ bind(notObj);
2440   // ctos
2441   __ cmpl(flags, ctos );
2442   __ jcc(Assembler::notEqual, notChar);
2443 
2444   __ pop(ctos);
2445   if (!is_static) pop_and_check_object(obj);
2446   __ movw(lo, rax );
2447   if (!is_static) {
2448     patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2449   }
2450   __ jmp(Done);
2451 
2452   __ bind(notChar);
2453   // stos
2454   __ cmpl(flags, stos );
2455   __ jcc(Assembler::notEqual, notShort);
2456 


2458   if (!is_static) pop_and_check_object(obj);
2459   __ movw(lo, rax );
2460   if (!is_static) {
2461     patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2462   }
2463   __ jmp(Done);
2464 
2465   __ bind(notShort);
2466   // ltos
2467   __ cmpl(flags, ltos );
2468   __ jcc(Assembler::notEqual, notLong);
2469 
2470   Label notVolatileLong;
2471   __ testl(rdx, rdx);
2472   __ jcc(Assembler::zero, notVolatileLong);
2473 
2474   __ pop(ltos);  // overwrites rdx, do this after testing volatile.
2475   if (!is_static) pop_and_check_object(obj);
2476   
2477   // Replace with real volatile test
2478   __ pushl(rdx);
2479   __ pushl(rax);                // Must update atomically with FIST
2480   __ fild_d(Address(rsp,0));    // So load into FPU register
2481   __ fistp_d(lo);               // and put into memory atomically
2482   __ addl(rsp,2*wordSize);
2483   volatile_barrier();


2484   // Don't rewrite volatile version
2485   __ jmp(notVolatile);
2486 
2487   __ bind(notVolatileLong);
2488 
2489   __ pop(ltos);  // overwrites rdx
2490   if (!is_static) pop_and_check_object(obj);
2491   __ movl(hi, rdx);
2492   __ movl(lo, rax);
2493   if (!is_static) {
2494     patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2495   }
2496   __ jmp(notVolatile);
2497 
2498   __ bind(notLong);
2499   // ftos
2500   __ cmpl(flags, ftos );
2501   __ jcc(Assembler::notEqual, notFloat);
2502 
2503   __ pop(ftos);
2504   if (!is_static) pop_and_check_object(obj);
2505   __ fstp_s(lo);
2506   if (!is_static) {
2507     patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2508   }
2509   __ jmp(Done);
2510 
2511   __ bind(notFloat);
2512   // dtos
2513   __ cmpl(flags, dtos );
2514   __ jcc(Assembler::notEqual, notDouble);
2515 
2516   __ pop(dtos);
2517   if (!is_static) pop_and_check_object(obj);
2518   __ fstp_d(lo);
2519   if (!is_static) {
2520     patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2521   }
2522   __ jmp(Done);
2523 
2524   __ bind(notDouble);
2525  
2526   __ stop("Bad state");
2527 
2528   __ bind(Done);
2529 
2530   // Check for volatile store
2531   __ testl(rdx, rdx);
2532   __ jcc(Assembler::zero, notVolatile);
2533   volatile_barrier( );

2534   __ bind(notVolatile);
2535 }
2536 
2537 
2538 void TemplateTable::putfield(int byte_no) {
2539   putfield_or_static(byte_no, false);
2540 }
2541 
2542 
2543 void TemplateTable::putstatic(int byte_no) {
2544   putfield_or_static(byte_no, true);
2545 }
2546 
2547 void TemplateTable::jvmti_post_fast_field_mod() {
2548   if (JvmtiExport::can_post_field_modification()) {
2549     // Check to see if a field modification watch has been set before we take
2550     // the time to call into the VM.
2551     Label L2;
2552     __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2553     __ testl(rcx,rcx);
2554     __ jcc(Assembler::zero, L2);
2555     __ pop_ptr(rbx);               // copy the object pointer from tos
2556     __ verify_oop(rbx);
2557     __ push_ptr(rbx);              // put the object pointer back on tos
2558     __ subl(rsp, sizeof(jvalue));  // add space for a jvalue object      
2559     __ movl(rcx, rsp);
2560     __ push_ptr(rbx);                 // save object pointer so we can steal rbx,
2561     __ movl(rbx, 0);
2562     const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2563     const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2564     switch (bytecode()) {          // load values into the jvalue object
2565     case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2566     case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2567     case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2568     case Bytecodes::_fast_iputfield: __ movl(lo_value, rax);                         break;
2569     case Bytecodes::_fast_lputfield: __ movl(hi_value, rdx); __ movl(lo_value, rax); break;




2570     // need to call fld_s() after fstp_s() to restore the value for below
2571     case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value);        break;

2572     // need to call fld_d() after fstp_d() to restore the value for below
2573     case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value);        break;

2574     // since rcx is not an object we don't call store_check() here
2575     case Bytecodes::_fast_aputfield: __ movl(lo_value, rax);                         break;

2576     default:  ShouldNotReachHere();
2577     }
2578     __ pop_ptr(rbx);  // restore copy of object pointer
2579 
2580     // Save rax, and sometimes rdx because call_VM() will clobber them,
2581     // then use them for JVM/DI purposes
2582     __ pushl(rax);
2583     if (bytecode() == Bytecodes::_fast_lputfield) __ pushl(rdx);
2584     // access constant pool cache entry
2585     __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2586     __ verify_oop(rbx);
2587     // rbx,: object pointer copied above
2588     // rax,: cache entry pointer
2589     // rcx: jvalue object on the stack
2590     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2591     if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx);  // restore high value
2592     __ popl(rax);     // restore lower value   
2593     __ addl(rsp, sizeof(jvalue));  // release jvalue object space
2594     __ bind(L2);
2595   }
2596 }
2597 
2598 void TemplateTable::fast_storefield(TosState state) {
2599   transition(state, vtos);
2600 
2601   ByteSize base = constantPoolCacheOopDesc::base_offset();
2602 
2603   jvmti_post_fast_field_mod();
2604 
2605   // access constant pool cache
2606   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2607 
2608   // test for volatile with rdx but rdx is tos register for lputfield.
2609   if (bytecode() == Bytecodes::_fast_lputfield) __ pushl(rdx);
2610   __ movl(rdx, Address(rcx, rbx, Address::times_4, in_bytes(base +
2611                        ConstantPoolCacheEntry::flags_offset())));
2612 
2613   // replace index with field offset from cache entry
2614   __ movl(rbx, Address(rcx, rbx, Address::times_4, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2615 
2616   // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2617   // volatile_barrier( );
2618 
2619   Label notVolatile, Done;
2620   __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2621   __ andl(rdx, 0x1);
2622   // Check for volatile store
2623   __ testl(rdx, rdx);
2624   __ jcc(Assembler::zero, notVolatile);
2625 
2626   if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx);
2627 
2628   // Get object from stack
2629   pop_and_check_object(rcx);
2630 
2631   // field addresses
2632   const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2633   const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2634 
2635   // access field
2636   switch (bytecode()) {
2637     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2638     case Bytecodes::_fast_sputfield: // fall through
2639     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2640     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2641     case Bytecodes::_fast_lputfield: __ movl(hi, rdx); __ movl(lo, rax);        break;



2642     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2643     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2644     case Bytecodes::_fast_aputfield: __ movl(lo, rax); __ store_check(rcx, lo); break;



2645     default:
2646       ShouldNotReachHere();
2647   }
2648 
2649   Label done;
2650   volatile_barrier( );
2651   __ jmpb(done);


2652 
2653   // Same code as above, but don't need rdx to test for volatile.
2654   __ bind(notVolatile);
2655 
2656   if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx);
2657 
2658   // Get object from stack
2659   pop_and_check_object(rcx);
2660 
2661   // access field
2662   switch (bytecode()) {
2663     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2664     case Bytecodes::_fast_sputfield: // fall through
2665     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2666     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2667     case Bytecodes::_fast_lputfield: __ movl(hi, rdx); __ movl(lo, rax);        break;



2668     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2669     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2670     case Bytecodes::_fast_aputfield: __ movl(lo, rax); __ store_check(rcx, lo); break;



2671     default:
2672       ShouldNotReachHere();
2673   }
2674   __ bind(done);
2675 }
2676 
2677 
2678 void TemplateTable::fast_accessfield(TosState state) {
2679   transition(atos, state);
2680 
2681   // do the JVMTI work here to avoid disturbing the register state below
2682   if (JvmtiExport::can_post_field_access()) {
2683     // Check to see if a field access watch has been set before we take
2684     // the time to call into the VM.
2685     Label L1;
2686     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2687     __ testl(rcx,rcx);
2688     __ jcc(Assembler::zero, L1);
2689     // access constant pool cache entry
2690     __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2691     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
2692     __ verify_oop(rax);
2693     // rax,: object pointer copied above
2694     // rcx: cache entry pointer
2695     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2696     __ pop_ptr(rax);   // restore object pointer
2697     __ bind(L1);
2698   }
2699 
2700   // access constant pool cache
2701   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2702   // replace index with field offset from cache entry
2703   __ movl(rbx, Address(rcx, rbx, Address::times_4, in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));



2704 
2705 
2706   // rax,: object
2707   __ verify_oop(rax);
2708   __ null_check(rax);
2709   // field addresses
2710   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2711   const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2712 
2713   // access field
2714   switch (bytecode()) {
2715     case Bytecodes::_fast_bgetfield: __ movsxb(rax, lo );                 break;
2716     case Bytecodes::_fast_sgetfield: __ load_signed_word(rax, lo );       break;
2717     case Bytecodes::_fast_cgetfield: __ load_unsigned_word(rax, lo );     break;
2718     case Bytecodes::_fast_igetfield: __ movl(rax, lo);                    break;
2719     case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten");  break;
2720     case Bytecodes::_fast_fgetfield: __ fld_s(lo);                        break;
2721     case Bytecodes::_fast_dgetfield: __ fld_d(lo);                        break;
2722     case Bytecodes::_fast_agetfield: __ movl(rax, lo); __ verify_oop(rax); break;
2723     default:
2724       ShouldNotReachHere();
2725   }
2726 
2727   // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2728   // volatile_barrier( );
2729 }
2730 
2731 void TemplateTable::fast_xaccess(TosState state) {
2732   transition(vtos, state);
2733   // get receiver
2734   __ movl(rax, aaddress(0));
2735   debug_only(__ verify_local_tag(frame::TagReference, 0));
2736   // access constant pool cache
2737   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2738   __ movl(rbx, Address(rcx, rdx, Address::times_4, in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));



2739   // make sure exception is reported in correct bcp range (getfield is next instruction)
2740   __ increment(rsi);
2741   __ null_check(rax);
2742   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2743   if (state == itos) {
2744     __ movl(rax, lo);
2745   } else if (state == atos) {
2746     __ movl(rax, lo);
2747     __ verify_oop(rax);
2748   } else if (state == ftos) {
2749     __ fld_s(lo);
2750   } else {
2751     ShouldNotReachHere();
2752   }
2753   __ decrement(rsi);
2754 }
2755 
2756 
2757 
2758 //----------------------------------------------------------------------------------------------------
2759 // Calls
2760 
2761 void TemplateTable::count_calls(Register method, Register temp) {  
2762   // implemented elsewhere
2763   ShouldNotReachHere();
2764 }
2765 
2766 


2770   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2771   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2772   const bool load_receiver       = code != Bytecodes::_invokestatic;
2773   const bool receiver_null_check = is_invokespecial;
2774   const bool save_flags = is_invokeinterface || is_invokevirtual;
2775   // setup registers & access constant pool cache
2776   const Register recv   = rcx;
2777   const Register flags  = rdx;  
2778   assert_different_registers(method, index, recv, flags);
2779 
2780   // save 'interpreter return address'
2781   __ save_bcp();
2782 
2783   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
2784 
2785   // load receiver if needed (note: no return address pushed yet)
2786   if (load_receiver) {
2787     __ movl(recv, flags);
2788     __ andl(recv, 0xFF);
2789     // recv count is 0 based?
2790     __ movl(recv, Address(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1)));
2791     __ verify_oop(recv);
2792   }
2793 
2794   // do null check if needed
2795   if (receiver_null_check) {
2796     __ null_check(recv);
2797   }
2798 
2799   if (save_flags) {
2800     __ movl(rsi, flags);
2801   }
2802 
2803   // compute return type
2804   __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2805   // Make sure we don't need to mask flags for tosBits after the above shift
2806   ConstantPoolCacheEntry::verify_tosBits();
2807   // load return address
2808   { const int table =
2809       is_invokeinterface
2810       ? (int)Interpreter::return_5_addrs_by_index_table()
2811       : (int)Interpreter::return_3_addrs_by_index_table();
2812     __ movl(flags, Address(noreg, flags, Address::times_4, table));
2813   }
2814 
2815   // push return address
2816   __ pushl(flags);
2817 
2818   // Restore flag value from the constant pool cache, and restore rsi
2819   // for later null checks.  rsi is the bytecode pointer
2820   if (save_flags) {
2821     __ movl(flags, rsi);
2822     __ restore_bcp();
2823   }
2824 }
2825 
2826 
2827 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2828                         Register flags) {
2829 
2830   // Uses temporary registers rax, rdx
2831   assert_different_registers(index, recv, rax, rdx);
2832 
2833   // Test for an invoke of a final method
2834   Label notFinal;
2835   __ movl(rax, flags);
2836   __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2837   __ jcc(Assembler::zero, notFinal);
2838 
2839   Register method = index;  // method must be rbx,
2840   assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2841 
2842   // do the call - the index is actually the method to call
2843   __ verify_oop(method);
2844 
2845   // It's final, need a null check here!
2846   __ null_check(recv);
2847 
2848   // profile this call
2849   __ profile_final_call(rax);
2850 
2851   __ jump_from_interpreted(method, rax);
2852 
2853   __ bind(notFinal);
2854 
2855   // get receiver klass
2856   __ null_check(recv, oopDesc::klass_offset_in_bytes());
2857   // Keep recv in rcx for callee expects it there
2858   __ movl(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2859   __ verify_oop(rax);
2860 
2861   // profile this call
2862   __ profile_virtual_call(rax, rdi, rdx);
2863 
2864   // get target methodOop & entry point
2865   const int base = instanceKlass::vtable_start_offset() * wordSize;    
2866   assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2867   __ movl(method, Address(rax, index, Address::times_4, base + vtableEntry::method_offset_in_bytes()));  
2868   __ jump_from_interpreted(method, rdx);
2869 }
2870 
2871 
2872 void TemplateTable::invokevirtual(int byte_no) {
2873   transition(vtos, vtos);
2874   prepare_invoke(rbx, noreg, byte_no, bytecode());
2875 
2876   // rbx,: index
2877   // rcx: receiver    
2878   // rdx: flags    
2879 
2880   invokevirtual_helper(rbx, rcx, rdx);
2881 }
2882 
2883 
2884 void TemplateTable::invokespecial(int byte_no) {
2885   transition(vtos, vtos);
2886   prepare_invoke(rbx, noreg, byte_no, bytecode());
2887   // do the call


2913   
2914   // rax,: Interface
2915   // rbx,: index
2916   // rcx: receiver    
2917   // rdx: flags
2918 
2919   // Special case of invokeinterface called for virtual method of
2920   // java.lang.Object.  See cpCacheOop.cpp for details.
2921   // This code isn't produced by javac, but could be produced by
2922   // another compliant java compiler.
2923   Label notMethod;
2924   __ movl(rdi, rdx);
2925   __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
2926   __ jcc(Assembler::zero, notMethod);
2927 
2928   invokevirtual_helper(rbx, rcx, rdx);
2929   __ bind(notMethod);
2930 
2931   // Get receiver klass into rdx - also a null check
2932   __ restore_locals();  // restore rdi
2933   __ movl(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
2934   __ verify_oop(rdx);
2935 
2936   // profile this call
2937   __ profile_virtual_call(rdx, rsi, rdi);
2938 
2939   __ movl(rdi, rdx); // Save klassOop in rdi
2940 
2941   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
2942   const int base = instanceKlass::vtable_start_offset() * wordSize;    
2943   assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2944   __ movl(rsi, Address(rdx, instanceKlass::vtable_length_offset() * wordSize)); // Get length of vtable
2945   __ leal(rdx, Address(rdx, rsi, Address::times_4, base));
2946   if (HeapWordsPerLong > 1) {
2947     // Round up to align_object_offset boundary
2948     __ round_to(rdx, BytesPerLong);
2949   }
2950 
2951   Label entry, search, interface_ok;
2952   
2953   __ jmpb(entry);   
2954   __ bind(search);
2955   __ addl(rdx, itableOffsetEntry::size() * wordSize);
2956   
2957   __ bind(entry);
2958 
2959   // Check that the entry is non-null.  A null entry means that the receiver
2960   // class doesn't implement the interface, and wasn't the same as the
2961   // receiver class checked when the interface was resolved.
2962   __ pushl(rdx);
2963   __ movl(rdx, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
2964   __ testl(rdx, rdx);
2965   __ jcc(Assembler::notZero, interface_ok);
2966   // throw exception
2967   __ popl(rdx);          // pop saved register first.
2968   __ popl(rbx);          // pop return address (pushed by prepare_invoke)
2969   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
2970   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
2971   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2972                    InterpreterRuntime::throw_IncompatibleClassChangeError));
2973   // the call_VM checks for exception, so we should never return here.
2974   __ should_not_reach_here();
2975   __ bind(interface_ok);
2976 
2977     __ popl(rdx);
2978 
2979     __ cmpl(rax, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
2980     __ jcc(Assembler::notEqual, search);
2981         
2982     __ movl(rdx, Address(rdx, itableOffsetEntry::offset_offset_in_bytes()));      
2983     __ addl(rdx, rdi); // Add offset to klassOop
2984     assert(itableMethodEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2985     __ movl(rbx, Address(rdx, rbx, Address::times_4));
2986     // rbx,: methodOop to call
2987     // rcx: receiver
2988     // Check for abstract method error
2989     // Note: This should be done more efficiently via a throw_abstract_method_error
2990     //       interpreter entry point and a conditional jump to it in case of a null
2991     //       method.
2992     { Label L;
2993       __ testl(rbx, rbx);
2994       __ jcc(Assembler::notZero, L);
2995       // throw exception
2996           // note: must restore interpreter registers to canonical
2997           //       state for exception handling to work correctly!
2998           __ popl(rbx);          // pop return address (pushed by prepare_invoke)
2999           __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
3000           __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3001       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3002       // the call_VM checks for exception, so we should never return here.
3003       __ should_not_reach_here();
3004       __ bind(L);
3005     }
3006 
3007   // do the call
3008   // rcx: receiver
3009   // rbx,: methodOop
3010   __ jump_from_interpreted(rbx, rdx);
3011 }
3012 
3013 //----------------------------------------------------------------------------------------------------
3014 // Allocation
3015 
3016 void TemplateTable::_new() {
3017   transition(vtos, atos);
3018   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3019   Label slow_case;
3020   Label done;
3021   Label initialize_header;
3022   Label initialize_object;  // including clearing the fields
3023   Label allocate_shared;
3024 
3025   ExternalAddress heap_top((address)Universe::heap()->top_addr());
3026 
3027   __ get_cpool_and_tags(rcx, rax);
3028   // get instanceKlass
3029   __ movl(rcx, Address(rcx, rdx, Address::times_4, sizeof(constantPoolOopDesc)));
3030   __ pushl(rcx);  // save the contexts of klass for initializing the header
3031 
3032   // make sure the class we're about to instantiate has been resolved. 
3033   // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
3034   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3035   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3036   __ jcc(Assembler::notEqual, slow_case);
3037 
3038   // make sure klass is initialized & doesn't have finalizer
3039   // make sure klass is fully initialized
3040   __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3041   __ jcc(Assembler::notEqual, slow_case);
3042 
3043   // get instance_size in instanceKlass (scaled to a count of bytes)
3044   __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3045   // test to see if it has a finalizer or is malformed in some way
3046   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3047   __ jcc(Assembler::notZero, slow_case);
3048 
3049   // 
3050   // Allocate the instance
3051   // 1) Try to allocate in the TLAB
3052   // 2) if fail and the object is large allocate in the shared Eden
3053   // 3) if the above fails (or is not applicable), go to a slow case
3054   // (creates a new TLAB, etc.)
3055 
3056   const bool allow_shared_alloc =
3057     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3058 
3059   if (UseTLAB) {
3060     const Register thread = rcx;
3061 
3062     __ get_thread(thread);
3063     __ movl(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3064     __ leal(rbx, Address(rax, rdx, Address::times_1));
3065     __ cmpl(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3066     __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3067     __ movl(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3068     if (ZeroTLAB) {
3069       // the fields have been already cleared
3070       __ jmp(initialize_header);
3071     } else {
3072       // initialize both the header and fields
3073       __ jmp(initialize_object);
3074     }
3075   }
3076 
3077   // Allocation in the shared Eden, if allowed.
3078   //
3079   // rdx: instance size in bytes
3080   if (allow_shared_alloc) {
3081     __ bind(allocate_shared);
3082 


3083     Label retry;
3084     __ bind(retry);
3085     __ mov32(rax, heap_top);
3086     __ leal(rbx, Address(rax, rdx, Address::times_1));
3087     __ cmp32(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3088     __ jcc(Assembler::above, slow_case);
3089 
3090     // Compare rax, with the top addr, and if still equal, store the new
3091     // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3092     // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3093     //
3094     // rax,: object begin
3095     // rbx,: object end
3096     // rdx: instance size in bytes
3097     if (os::is_MP()) __ lock();
3098     __ cmpxchgptr(rbx, heap_top);
3099 
3100     // if someone beat us on the allocation, try again, otherwise continue 
3101     __ jcc(Assembler::notEqual, retry);
3102   }
3103 
3104   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3105     // The object is initialized before the header.  If the object size is
3106     // zero, go directly to the header initialization.
3107     __ bind(initialize_object);
3108     __ decrement(rdx, sizeof(oopDesc));
3109     __ jcc(Assembler::zero, initialize_header);
3110 
3111   // Initialize topmost object field, divide rdx by 8, check if odd and
3112   // test if zero.
3113     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
3114     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3115 
3116   // rdx must have been multiple of 8
3117 #ifdef ASSERT
3118     // make sure rdx was multiple of 8
3119     Label L;
3120     // Ignore partial flag stall after shrl() since it is debug VM
3121     __ jccb(Assembler::carryClear, L);
3122     __ stop("object size is not multiple of 2 - adjust this code");
3123     __ bind(L);
3124     // rdx must be > 0, no extra check needed here
3125 #endif
3126 
3127     // initialize remaining object fields: rdx was a multiple of 8
3128     { Label loop;
3129     __ bind(loop);
3130     __ movl(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3131     __ movl(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx);
3132     __ decrement(rdx);
3133     __ jcc(Assembler::notZero, loop);
3134     }
3135 
3136     // initialize object header only.
3137     __ bind(initialize_header);
3138     if (UseBiasedLocking) {
3139       __ popl(rcx);   // get saved klass back in the register.
3140       __ movl(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3141       __ movl(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3142     } else {
3143       __ movl(Address(rax, oopDesc::mark_offset_in_bytes ()),
3144               (int)markOopDesc::prototype()); // header
3145       __ popl(rcx);   // get saved klass back in the register.
3146     }
3147     __ movl(Address(rax, oopDesc::klass_offset_in_bytes()), rcx);  // klass
3148 
3149     {
3150       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3151       // Trigger dtrace event for fastpath
3152       __ push(atos);
3153       __ call_VM_leaf(
3154            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3155       __ pop(atos);
3156     }
3157 
3158     __ jmp(done);
3159   }
3160 
3161   // slow case
3162   __ bind(slow_case);
3163   __ popl(rcx);   // restore stack pointer to what it was when we came in.
3164   __ get_constant_pool(rax);
3165   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3166   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3167 
3168   // continue
3169   __ bind(done);
3170 }
3171 
3172 
3173 void TemplateTable::newarray() {
3174   transition(itos, atos);
3175   __ push_i(rax);                                 // make sure everything is on the stack
3176   __ load_unsigned_byte(rdx, at_bcp(1));
3177   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3178   __ pop_i(rdx);                                  // discard size
3179 }
3180 
3181 
3182 void TemplateTable::anewarray() {
3183   transition(itos, atos);
3184   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3185   __ get_constant_pool(rcx);
3186   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3187 }
3188 
3189 
3190 void TemplateTable::arraylength() {
3191   transition(atos, itos);
3192   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3193   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3194 }
3195 
3196 
3197 void TemplateTable::checkcast() {
3198   transition(atos, atos);
3199   Label done, is_null, ok_is_subtype, quicked, resolved;
3200   __ testl(rax, rax);   // Object is in EAX
3201   __ jcc(Assembler::zero, is_null);
3202 
3203   // Get cpool & tags index
3204   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3205   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3206   // See if bytecode has already been quicked
3207   __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3208   __ jcc(Assembler::equal, quicked);
3209 
3210   __ push(atos);
3211   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3212   __ pop_ptr(rdx);
3213   __ jmpb(resolved);
3214 
3215   // Get superklass in EAX and subklass in EBX
3216   __ bind(quicked);
3217   __ movl(rdx, rax);          // Save object in EDX; EAX needed for subtype check
3218   __ movl(rax, Address(rcx, rbx, Address::times_4, sizeof(constantPoolOopDesc)));
3219 
3220   __ bind(resolved);
3221   __ movl(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3222 
3223   // Generate subtype check.  Blows ECX.  Resets EDI.  Object in EDX.
3224   // Superklass in EAX.  Subklass in EBX.
3225   __ gen_subtype_check( rbx, ok_is_subtype );
3226 
3227   // Come here on failure
3228   __ pushl(rdx);
3229   // object is at TOS
3230   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3231 
3232   // Come here on success
3233   __ bind(ok_is_subtype);
3234   __ movl(rax,rdx);           // Restore object in EDX
3235 
3236   // Collect counts on whether this check-cast sees NULLs a lot or not.
3237   if (ProfileInterpreter) {
3238     __ jmp(done);
3239     __ bind(is_null);
3240     __ profile_null_seen(rcx);
3241   } else {
3242     __ bind(is_null);   // same as 'done'
3243   }
3244   __ bind(done);
3245 }
3246 
3247 
3248 void TemplateTable::instanceof() {
3249   transition(atos, itos);
3250   Label done, is_null, ok_is_subtype, quicked, resolved;
3251   __ testl(rax, rax);
3252   __ jcc(Assembler::zero, is_null);
3253 
3254   // Get cpool & tags index
3255   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3256   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3257   // See if bytecode has already been quicked
3258   __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3259   __ jcc(Assembler::equal, quicked);
3260 
3261   __ push(atos);
3262   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3263   __ pop_ptr(rdx);
3264   __ movl(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3265   __ jmp(resolved);
3266 
3267   // Get superklass in EAX and subklass in EDX
3268   __ bind(quicked);
3269   __ movl(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3270   __ movl(rax, Address(rcx, rbx, Address::times_4, sizeof(constantPoolOopDesc)));
3271 
3272   __ bind(resolved);
3273 
3274   // Generate subtype check.  Blows ECX.  Resets EDI.
3275   // Superklass in EAX.  Subklass in EDX.
3276   __ gen_subtype_check( rdx, ok_is_subtype );
3277 
3278   // Come here on failure
3279   __ xorl(rax,rax);
3280   __ jmpb(done);
3281   // Come here on success
3282   __ bind(ok_is_subtype);
3283   __ movl(rax, 1);
3284 
3285   // Collect counts on whether this test sees NULLs a lot or not.
3286   if (ProfileInterpreter) {
3287     __ jmp(done);
3288     __ bind(is_null);
3289     __ profile_null_seen(rcx);
3290   } else {


3292   }
3293   __ bind(done);
3294   // rax, = 0: obj == NULL or  obj is not an instanceof the specified klass
3295   // rax, = 1: obj != NULL and obj is     an instanceof the specified klass
3296 }
3297 
3298 
3299 //----------------------------------------------------------------------------------------------------
3300 // Breakpoints
3301 void TemplateTable::_breakpoint() {
3302   
3303   // Note: We get here even if we are single stepping..
3304   // jbug inists on setting breakpoints at every bytecode 
3305   // even if we are in single step mode.  
3306  
3307   transition(vtos, vtos);
3308 
3309   // get the unpatched byte code
3310   __ get_method(rcx);
3311   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3312   __ movl(rbx, rax);
3313 
3314   // post the breakpoint event
3315   __ get_method(rcx);
3316   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3317 
3318   // complete the execution of original bytecode
3319   __ dispatch_only_normal(vtos);
3320 } 
3321 
3322 
3323 //----------------------------------------------------------------------------------------------------
3324 // Exceptions
3325 
3326 void TemplateTable::athrow() {
3327   transition(atos, vtos);
3328   __ null_check(rax);
3329   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3330 }
3331 
3332 


3348 // ...
3349 // [saved rbp,    ] <--- rbp,
3350 
3351 
3352 void TemplateTable::monitorenter() {
3353   transition(atos, vtos);
3354 
3355   // check for NULL object
3356   __ null_check(rax);
3357 
3358   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3359   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3360   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3361   Label allocated;
3362 
3363   // initialize entry pointer
3364   __ xorl(rdx, rdx);                             // points to free slot or NULL
3365 
3366   // find a free slot in the monitor block (result in rdx)
3367   { Label entry, loop, exit;
3368     __ movl(rcx, monitor_block_top);             // points to current entry, starting with top-most entry
3369     __ leal(rbx, monitor_block_bot);             // points to word before bottom of monitor block
3370     __ jmpb(entry);
3371 
3372     __ bind(loop);
3373     __ cmpl(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);  // check if current entry is used
3374 
3375 // TODO - need new func here - kbt
3376     if (VM_Version::supports_cmov()) {
3377       __ cmovl(Assembler::equal, rdx, rcx);      // if not used then remember entry in rdx
3378     } else {
3379       Label L;
3380       __ jccb(Assembler::notEqual, L);
3381       __ movl(rdx, rcx);                         // if not used then remember entry in rdx
3382       __ bind(L);
3383     }
3384     __ cmpl(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3385     __ jccb(Assembler::equal, exit);              // if same object then stop searching
3386     __ addl(rcx, entry_size);                    // otherwise advance to next entry
3387     __ bind(entry);
3388     __ cmpl(rcx, rbx);                           // check if bottom reached
3389     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3390     __ bind(exit);
3391   }
3392 
3393   __ testl(rdx, rdx);                            // check if a slot has been found
3394   __ jccb(Assembler::notZero, allocated);         // if found, continue with that one
3395 
3396   // allocate one if there's no free slot
3397   { Label entry, loop;
3398     // 1. compute new pointers                   // rsp: old expression stack top
3399     __ movl(rdx, monitor_block_bot);             // rdx: old expression stack bottom
3400     __ subl(rsp, entry_size);                    // move expression stack top
3401     __ subl(rdx, entry_size);                    // move expression stack bottom
3402     __ movl(rcx, rsp);                           // set start value for copy loop
3403     __ movl(monitor_block_bot, rdx);             // set new monitor block top
3404     __ jmp(entry);
3405     // 2. move expression stack contents
3406     __ bind(loop);
3407     __ movl(rbx, Address(rcx, entry_size));      // load expression stack word from old location
3408     __ movl(Address(rcx, 0), rbx);               // and store it at new location
3409     __ addl(rcx, wordSize);                      // advance to next word
3410     __ bind(entry);
3411     __ cmpl(rcx, rdx);                           // check if bottom reached
3412     __ jcc(Assembler::notEqual, loop);           // if not at bottom then copy next word
3413   }
3414   
3415   // call run-time routine
3416   // rdx: points to monitor entry
3417   __ bind(allocated);
3418 
3419   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly. 
3420   // The object has already been poped from the stack, so the expression stack looks correct.
3421   __ increment(rsi);
3422 
3423   __ movl(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax);     // store object  
3424   __ lock_object(rdx);  
3425 
3426   // check to make sure this monitor doesn't cause stack overflow after locking
3427   __ save_bcp();  // in case of exception
3428   __ generate_stack_overflow_check(0);
3429 
3430   // The bcp has already been incremented. Just need to dispatch to next instruction.
3431   __ dispatch_next(vtos);
3432 }
3433 
3434 
3435 void TemplateTable::monitorexit() {
3436   transition(atos, vtos);
3437 
3438   // check for NULL object
3439   __ null_check(rax);
3440 
3441   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3442   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3443   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3444   Label found;
3445 
3446   // find matching slot
3447   { Label entry, loop;
3448     __ movl(rdx, monitor_block_top);             // points to current entry, starting with top-most entry
3449     __ leal(rbx, monitor_block_bot);             // points to word before bottom of monitor block
3450     __ jmpb(entry);
3451 
3452     __ bind(loop);
3453     __ cmpl(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3454     __ jcc(Assembler::equal, found);             // if same object then stop searching
3455     __ addl(rdx, entry_size);                    // otherwise advance to next entry
3456     __ bind(entry);
3457     __ cmpl(rdx, rbx);                           // check if bottom reached
3458     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3459   }
3460 
3461   // error handling. Unlocking was not block-structured
3462   Label end;
3463   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3464   __ should_not_reach_here();
3465 
3466   // call run-time routine
3467   // rcx: points to monitor entry
3468   __ bind(found);
3469   __ push_ptr(rax);                                 // make sure object is on stack (contract with oopMaps)  
3470   __ unlock_object(rdx);    
3471   __ pop_ptr(rax);                                  // discard object  
3472   __ bind(end);
3473 }
3474 
3475 
3476 //----------------------------------------------------------------------------------------------------
3477 // Wide instructions
3478 
3479 void TemplateTable::wide() {
3480   transition(vtos, vtos);
3481   __ load_unsigned_byte(rbx, at_bcp(1));
3482   __ jmp(Address(noreg, rbx, Address::times_4, int(Interpreter::_wentry_point)));

3483   // Note: the rsi increment step is part of the individual wide bytecode implementations
3484 }
3485 
3486 
3487 //----------------------------------------------------------------------------------------------------
3488 // Multi arrays
3489 
3490 void TemplateTable::multianewarray() {
3491   transition(vtos, atos);
3492   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3493   // last dim is on top of stack; we want address of first one:
3494   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3495   // the latter wordSize to point to the beginning of the array.
3496   __ leal(  rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3497   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax);     // pass in rax,
3498   __ load_unsigned_byte(rbx, at_bcp(3));
3499   __ leal(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
3500 }
3501 
3502 #endif /* !CC_INTERP */



   1 /*
   2  * Copyright 1997-2008 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *


  90 }
  91 
  92 // Condition conversion
  93 static Assembler::Condition j_not(TemplateTable::Condition cc) {
  94   switch (cc) {
  95     case TemplateTable::equal        : return Assembler::notEqual;
  96     case TemplateTable::not_equal    : return Assembler::equal;
  97     case TemplateTable::less         : return Assembler::greaterEqual;
  98     case TemplateTable::less_equal   : return Assembler::greater;
  99     case TemplateTable::greater      : return Assembler::lessEqual;
 100     case TemplateTable::greater_equal: return Assembler::less;
 101   }
 102   ShouldNotReachHere();
 103   return Assembler::zero;
 104 }
 105 
 106 
 107 //----------------------------------------------------------------------------------------------------
 108 // Miscelaneous helper routines
 109 
 110 // Store an oop (or NULL) at the address described by obj.
 111 // If val == noreg this means store a NULL
 112 
 113 static void do_oop_store(InterpreterMacroAssembler* _masm,
 114                          Address obj,
 115                          Register val,
 116                          BarrierSet::Name barrier,
 117                          bool precise) {
 118   assert(val == noreg || val == rax, "parameter is just for looks");
 119   switch (barrier) {
 120 #ifndef SERIALGC
 121     case BarrierSet::G1SATBCT:
 122     case BarrierSet::G1SATBCTLogging:
 123       {
 124         // flatten object address if needed
 125         // We do it regardless of precise because we need the registers
 126         if (obj.index() == noreg && obj.disp() == 0) {
 127           if (obj.base() != rdx) {
 128             __ movl(rdx, obj.base());
 129           }
 130         } else {
 131           __ leal(rdx, obj);
 132         }
 133         __ get_thread(rcx);
 134         __ save_bcp();
 135         __ g1_write_barrier_pre(rdx, rcx, rsi, rbx, val != noreg);
 136 
 137         // Do the actual store
 138         // noreg means NULL
 139         if (val == noreg) {
 140           __ movl(Address(rdx, 0), NULL_WORD);
 141           // No post barrier for NULL
 142         } else {
 143           __ movl(Address(rdx, 0), val);
 144           __ g1_write_barrier_post(rdx, rax, rcx, rbx, rsi);
 145         }
 146         __ restore_bcp();
 147 
 148       }
 149       break;
 150 #endif // SERIALGC
 151     case BarrierSet::CardTableModRef:
 152     case BarrierSet::CardTableExtension:
 153       {
 154         if (val == noreg) {
 155           __ movl(obj, NULL_WORD);
 156         } else {
 157           __ movl(obj, val);
 158           // flatten object address if needed
 159           if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
 160             __ store_check(obj.base());
 161           } else {
 162             __ leal(rdx, obj);
 163             __ store_check(rdx);
 164           }
 165         }
 166       }
 167       break;
 168     case BarrierSet::ModRef:
 169     case BarrierSet::Other:
 170       if (val == noreg) {
 171         __ movl(obj, NULL_WORD);
 172       } else {
 173         __ movl(obj, val);
 174       }
 175       break;
 176     default      :
 177       ShouldNotReachHere();
 178 
 179   }
 180 }
 181 
 182 Address TemplateTable::at_bcp(int offset) {
 183   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 184   return Address(rsi, offset);
 185 }
 186 
 187 
 188 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
 189                                    Register scratch,
 190                                    bool load_bc_into_scratch/*=true*/) {
 191 
 192   if (!RewriteBytecodes) return;
 193   // the pair bytecodes have already done the load.
 194   if (load_bc_into_scratch) {
 195     __ movl(bc, bytecode);
 196   }
 197   Label patch_done;
 198   if (JvmtiExport::can_post_breakpoint()) {
 199     Label fast_patch;
 200     // if a breakpoint is present we can't rewrite the stream directly
 201     __ movzbl(scratch, at_bcp(0));
 202     __ cmpl(scratch, Bytecodes::_breakpoint);
 203     __ jcc(Assembler::notEqual, fast_patch);
 204     __ get_method(scratch);
 205     // Let breakpoint table handling rewrite to quicker bytecode
 206     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
 207 #ifndef ASSERT
 208     __ jmpb(patch_done);
 209     __ bind(fast_patch);
 210   }
 211 #else
 212     __ jmp(patch_done);
 213     __ bind(fast_patch);
 214   }
 215   Label okay;
 216   __ load_unsigned_byte(scratch, at_bcp(0));
 217   __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
 218   __ jccb(Assembler::equal, okay);
 219   __ cmpl(scratch, bc);
 220   __ jcc(Assembler::equal, okay);
 221   __ stop("patching the wrong bytecode");


 226   __ bind(patch_done);
 227 }
 228 
 229 //----------------------------------------------------------------------------------------------------
 230 // Individual instructions
 231 
 232 void TemplateTable::nop() {
 233   transition(vtos, vtos);
 234   // nothing to do
 235 }
 236 
 237 void TemplateTable::shouldnotreachhere() {
 238   transition(vtos, vtos);
 239   __ stop("shouldnotreachhere bytecode");
 240 }
 241 
 242 
 243 
 244 void TemplateTable::aconst_null() {
 245   transition(vtos, atos);
 246   __ xorptr(rax, rax);
 247 }
 248 
 249 
 250 void TemplateTable::iconst(int value) {
 251   transition(vtos, itos);
 252   if (value == 0) {
 253     __ xorptr(rax, rax);
 254   } else {
 255     __ movptr(rax, value);
 256   }
 257 }
 258 
 259 
 260 void TemplateTable::lconst(int value) {
 261   transition(vtos, ltos);
 262   if (value == 0) {
 263     __ xorptr(rax, rax);
 264   } else {
 265     __ movptr(rax, value);
 266   }
 267   assert(value >= 0, "check this code");
 268   __ xorptr(rdx, rdx);
 269 }
 270 
 271 
 272 void TemplateTable::fconst(int value) {
 273   transition(vtos, ftos);
 274          if (value == 0) { __ fldz();
 275   } else if (value == 1) { __ fld1();
 276   } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 277   } else                 { ShouldNotReachHere();
 278   }
 279 }
 280 
 281 
 282 void TemplateTable::dconst(int value) {
 283   transition(vtos, dtos);
 284          if (value == 0) { __ fldz();
 285   } else if (value == 1) { __ fld1();
 286   } else                 { ShouldNotReachHere();
 287   }
 288 }
 289 
 290 
 291 void TemplateTable::bipush() {
 292   transition(vtos, itos);
 293   __ load_signed_byte(rax, at_bcp(1));
 294 }
 295 
 296 
 297 void TemplateTable::sipush() {
 298   transition(vtos, itos);
 299   __ load_unsigned_word(rax, at_bcp(1));
 300   __ bswapl(rax);
 301   __ sarl(rax, 16);
 302 }
 303 
 304 void TemplateTable::ldc(bool wide) {
 305   transition(vtos, vtos);
 306   Label call_ldc, notFloat, notClass, Done;
 307 
 308   if (wide) {
 309     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 310   } else {
 311     __ load_unsigned_byte(rbx, at_bcp(1));
 312   }
 313   __ get_cpool_and_tags(rcx, rax);
 314   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 315   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 316 
 317   // get type
 318   __ xorptr(rdx, rdx);
 319   __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 320 
 321   // unresolved string - get the resolved string
 322   __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
 323   __ jccb(Assembler::equal, call_ldc);
 324 
 325   // unresolved class - get the resolved class
 326   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 327   __ jccb(Assembler::equal, call_ldc);
 328 
 329   // unresolved class in error (resolution failed) - call into runtime
 330   // so that the same error from first resolution attempt is thrown.
 331   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 332   __ jccb(Assembler::equal, call_ldc);
 333 
 334   // resolved class - need to call vm to get java mirror of the class
 335   __ cmpl(rdx, JVM_CONSTANT_Class);
 336   __ jcc(Assembler::notEqual, notClass);
 337 
 338   __ bind(call_ldc);
 339   __ movl(rcx, wide);
 340   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
 341   __ push(atos);
 342   __ jmp(Done);
 343 
 344   __ bind(notClass);
 345   __ cmpl(rdx, JVM_CONSTANT_Float);
 346   __ jccb(Assembler::notEqual, notFloat);
 347   // ftos
 348   __ fld_s(    Address(rcx, rbx, Address::times_ptr, base_offset));
 349   __ push(ftos);
 350   __ jmp(Done);
 351 
 352   __ bind(notFloat);
 353 #ifdef ASSERT
 354   { Label L;
 355     __ cmpl(rdx, JVM_CONSTANT_Integer);
 356     __ jcc(Assembler::equal, L);
 357     __ cmpl(rdx, JVM_CONSTANT_String);
 358     __ jcc(Assembler::equal, L);
 359     __ stop("unexpected tag type in ldc");
 360     __ bind(L);
 361   }
 362 #endif
 363   Label isOop;
 364   // atos and itos

 365   // String is only oop type we will see here
 366   __ cmpl(rdx, JVM_CONSTANT_String);
 367   __ jccb(Assembler::equal, isOop);
 368   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 369   __ push(itos);
 370   __ jmp(Done);
 371   __ bind(isOop);
 372   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 373   __ push(atos);
 374 
 375   if (VerifyOops) {
 376     __ verify_oop(rax);
 377   }
 378   __ bind(Done);
 379 }
 380 
 381 void TemplateTable::ldc2_w() {
 382   transition(vtos, vtos);
 383   Label Long, Done;
 384   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 385 
 386   __ get_cpool_and_tags(rcx, rax);
 387   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 388   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 389 
 390   // get type
 391   __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
 392   __ jccb(Assembler::notEqual, Long);
 393   // dtos
 394   __ fld_d(    Address(rcx, rbx, Address::times_ptr, base_offset));
 395   __ push(dtos);
 396   __ jmpb(Done);
 397 
 398   __ bind(Long);
 399   // ltos
 400   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 401   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 402 
 403   __ push(ltos);
 404 
 405   __ bind(Done);
 406 }
 407 
 408 
 409 void TemplateTable::locals_index(Register reg, int offset) {
 410   __ load_unsigned_byte(reg, at_bcp(offset));
 411   __ negptr(reg);
 412 }
 413 
 414 
 415 void TemplateTable::iload() {
 416   transition(vtos, itos);
 417   if (RewriteFrequentPairs) {
 418     Label rewrite, done;
 419 
 420     // get next byte
 421     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 422     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 423     // last two iloads in a pair.  Comparing against fast_iload means that
 424     // the next bytecode is neither an iload or a caload, and therefore
 425     // an iload pair.
 426     __ cmpl(rbx, Bytecodes::_iload);
 427     __ jcc(Assembler::equal, done);
 428 
 429     __ cmpl(rbx, Bytecodes::_fast_iload);
 430     __ movl(rcx, Bytecodes::_fast_iload2);
 431     __ jccb(Assembler::equal, rewrite);


 457   locals_index(rbx);
 458   __ movl(rax, iaddress(rbx));
 459   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 460   __ push(itos);
 461   locals_index(rbx, 3);
 462   __ movl(rax, iaddress(rbx));
 463   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 464 }
 465 
 466 void TemplateTable::fast_iload() {
 467   transition(vtos, itos);
 468   locals_index(rbx);
 469   __ movl(rax, iaddress(rbx));
 470   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 471 }
 472 
 473 
 474 void TemplateTable::lload() {
 475   transition(vtos, ltos);
 476   locals_index(rbx);
 477   __ movptr(rax, laddress(rbx));
 478   NOT_LP64(__ movl(rdx, haddress(rbx)));
 479   debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 480 }
 481 
 482 
 483 void TemplateTable::fload() {
 484   transition(vtos, ftos);
 485   locals_index(rbx);
 486   __ fld_s(faddress(rbx));
 487   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 488 }
 489 
 490 
 491 void TemplateTable::dload() {
 492   transition(vtos, dtos);
 493   locals_index(rbx);
 494   if (TaggedStackInterpreter) {
 495     // Get double out of locals array, onto temp stack and load with
 496     // float instruction into ST0
 497     __ movl(rax, laddress(rbx));
 498     __ movl(rdx, haddress(rbx));
 499     __ push(rdx);  // push hi first
 500     __ push(rax);
 501     __ fld_d(Address(rsp, 0));
 502     __ addptr(rsp, 2*wordSize);
 503     debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 504   } else {
 505     __ fld_d(daddress(rbx));
 506   }
 507 }
 508 
 509 
 510 void TemplateTable::aload() {
 511   transition(vtos, atos);
 512   locals_index(rbx);
 513   __ movptr(rax, aaddress(rbx));
 514   debug_only(__ verify_local_tag(frame::TagReference, rbx));
 515 }
 516 
 517 
 518 void TemplateTable::locals_index_wide(Register reg) {
 519   __ movl(reg, at_bcp(2));
 520   __ bswapl(reg);
 521   __ shrl(reg, 16);
 522   __ negptr(reg);
 523 }
 524 
 525 
 526 void TemplateTable::wide_iload() {
 527   transition(vtos, itos);
 528   locals_index_wide(rbx);
 529   __ movl(rax, iaddress(rbx));
 530   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 531 }
 532 
 533 
 534 void TemplateTable::wide_lload() {
 535   transition(vtos, ltos);
 536   locals_index_wide(rbx);
 537   __ movptr(rax, laddress(rbx));
 538   NOT_LP64(__ movl(rdx, haddress(rbx)));
 539   debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 540 }
 541 
 542 
 543 void TemplateTable::wide_fload() {
 544   transition(vtos, ftos);
 545   locals_index_wide(rbx);
 546   __ fld_s(faddress(rbx));
 547   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 548 }
 549 
 550 
 551 void TemplateTable::wide_dload() {
 552   transition(vtos, dtos);
 553   locals_index_wide(rbx);
 554   if (TaggedStackInterpreter) {
 555     // Get double out of locals array, onto temp stack and load with
 556     // float instruction into ST0
 557     __ movl(rax, laddress(rbx));
 558     __ movl(rdx, haddress(rbx));
 559     __ push(rdx);  // push hi first
 560     __ push(rax);
 561     __ fld_d(Address(rsp, 0));
 562     __ addl(rsp, 2*wordSize);
 563     debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
 564   } else {
 565     __ fld_d(daddress(rbx));
 566   }
 567 }
 568 
 569 
 570 void TemplateTable::wide_aload() {
 571   transition(vtos, atos);
 572   locals_index_wide(rbx);
 573   __ movptr(rax, aaddress(rbx));
 574   debug_only(__ verify_local_tag(frame::TagReference, rbx));
 575 }
 576 
 577 void TemplateTable::index_check(Register array, Register index) {
 578   // Pop ptr into array
 579   __ pop_ptr(array);
 580   index_check_without_pop(array, index);
 581 }
 582 
 583 void TemplateTable::index_check_without_pop(Register array, Register index) {
 584   // destroys rbx,
 585   // check array
 586   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 587   LP64_ONLY(__ movslq(index, index));
 588   // check index
 589   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 590   if (index != rbx) {
 591     // ??? convention: move aberrant index into rbx, for exception message
 592     assert(rbx != array, "different registers");
 593     __ mov(rbx, index);
 594   }
 595   __ jump_cc(Assembler::aboveEqual,
 596              ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 597 }
 598 
 599 
 600 void TemplateTable::iaload() {
 601   transition(itos, itos);
 602   // rdx: array
 603   index_check(rdx, rax);  // kills rbx,
 604   // rax,: index
 605   __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
 606 }
 607 
 608 
 609 void TemplateTable::laload() {
 610   transition(itos, ltos);
 611   // rax,: index
 612   // rdx: array
 613   index_check(rdx, rax);
 614   __ mov(rbx, rax);
 615   // rbx,: index
 616   __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
 617   NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
 618 }
 619 
 620 
 621 void TemplateTable::faload() {
 622   transition(itos, ftos);
 623   // rdx: array
 624   index_check(rdx, rax);  // kills rbx,
 625   // rax,: index
 626   __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 627 }
 628 
 629 
 630 void TemplateTable::daload() {
 631   transition(itos, dtos);
 632   // rdx: array
 633   index_check(rdx, rax);  // kills rbx,
 634   // rax,: index
 635   __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 636 }
 637 
 638 
 639 void TemplateTable::aaload() {
 640   transition(itos, atos);
 641   // rdx: array
 642   index_check(rdx, rax);  // kills rbx,
 643   // rax,: index
 644   __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 645 }
 646 
 647 
 648 void TemplateTable::baload() {
 649   transition(itos, itos);
 650   // rdx: array
 651   index_check(rdx, rax);  // kills rbx,
 652   // rax,: index
 653   // can do better code for P5 - fix this at some point
 654   __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 655   __ mov(rax, rbx);
 656 }
 657 
 658 
 659 void TemplateTable::caload() {
 660   transition(itos, itos);
 661   // rdx: array
 662   index_check(rdx, rax);  // kills rbx,
 663   // rax,: index
 664   // can do better code for P5 - may want to improve this at some point
 665   __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 666   __ mov(rax, rbx);
 667 }
 668 
 669 // iload followed by caload frequent pair
 670 void TemplateTable::fast_icaload() {
 671   transition(vtos, itos);
 672   // load index out of locals
 673   locals_index(rbx);
 674   __ movl(rax, iaddress(rbx));
 675   debug_only(__ verify_local_tag(frame::TagValue, rbx));
 676 
 677   // rdx: array
 678   index_check(rdx, rax);
 679   // rax,: index
 680   __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 681   __ mov(rax, rbx);
 682 }
 683 
 684 void TemplateTable::saload() {
 685   transition(itos, itos);
 686   // rdx: array
 687   index_check(rdx, rax);  // kills rbx,
 688   // rax,: index
 689   // can do better code for P5 - may want to improve this at some point
 690   __ load_signed_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 691   __ mov(rax, rbx);
 692 }
 693 
 694 
 695 void TemplateTable::iload(int n) {
 696   transition(vtos, itos);
 697   __ movl(rax, iaddress(n));
 698   debug_only(__ verify_local_tag(frame::TagValue, n));
 699 }
 700 
 701 
 702 void TemplateTable::lload(int n) {
 703   transition(vtos, ltos);
 704   __ movptr(rax, laddress(n));
 705   NOT_LP64(__ movptr(rdx, haddress(n)));
 706   debug_only(__ verify_local_tag(frame::TagCategory2, n));
 707 }
 708 
 709 
 710 void TemplateTable::fload(int n) {
 711   transition(vtos, ftos);
 712   __ fld_s(faddress(n));
 713   debug_only(__ verify_local_tag(frame::TagValue, n));
 714 }
 715 
 716 
 717 void TemplateTable::dload(int n) {
 718   transition(vtos, dtos);
 719   if (TaggedStackInterpreter) {
 720     // Get double out of locals array, onto temp stack and load with
 721     // float instruction into ST0
 722     __ movl(rax, laddress(n));
 723     __ movl(rdx, haddress(n));
 724     __ push(rdx);  // push hi first
 725     __ push(rax);
 726     __ fld_d(Address(rsp, 0));
 727     __ addptr(rsp, 2*wordSize);  // reset rsp
 728     debug_only(__ verify_local_tag(frame::TagCategory2, n));
 729   } else {
 730     __ fld_d(daddress(n));
 731   }
 732 }
 733 
 734 
 735 void TemplateTable::aload(int n) {
 736   transition(vtos, atos);
 737   __ movptr(rax, aaddress(n));
 738   debug_only(__ verify_local_tag(frame::TagReference, n));
 739 }
 740 
 741 
 742 void TemplateTable::aload_0() {
 743   transition(vtos, atos);
 744   // According to bytecode histograms, the pairs:
 745   //
 746   // _aload_0, _fast_igetfield
 747   // _aload_0, _fast_agetfield
 748   // _aload_0, _fast_fgetfield
 749   //
 750   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 751   // bytecode checks if the next bytecode is either _fast_igetfield,
 752   // _fast_agetfield or _fast_fgetfield and then rewrites the
 753   // current bytecode into a pair bytecode; otherwise it rewrites the current
 754   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 755   //
 756   // Note: If the next bytecode is _getfield, the rewrite must be delayed,
 757   //       otherwise we may miss an opportunity for a pair.


 799     __ bind(rewrite);
 800     patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
 801 
 802     __ bind(done);
 803   } else {
 804     aload(0);
 805   }
 806 }
 807 
 808 void TemplateTable::istore() {
 809   transition(itos, vtos);
 810   locals_index(rbx);
 811   __ movl(iaddress(rbx), rax);
 812   __ tag_local(frame::TagValue, rbx);
 813 }
 814 
 815 
 816 void TemplateTable::lstore() {
 817   transition(ltos, vtos);
 818   locals_index(rbx);
 819   __ movptr(laddress(rbx), rax);
 820   NOT_LP64(__ movptr(haddress(rbx), rdx));
 821   __ tag_local(frame::TagCategory2, rbx);
 822 }
 823 
 824 
 825 void TemplateTable::fstore() {
 826   transition(ftos, vtos);
 827   locals_index(rbx);
 828   __ fstp_s(faddress(rbx));
 829   __ tag_local(frame::TagValue, rbx);
 830 }
 831 
 832 
 833 void TemplateTable::dstore() {
 834   transition(dtos, vtos);
 835   locals_index(rbx);
 836   if (TaggedStackInterpreter) {
 837     // Store double on stack and reload into locals nonadjacently
 838     __ subptr(rsp, 2 * wordSize);
 839     __ fstp_d(Address(rsp, 0));
 840     __ pop(rax);
 841     __ pop(rdx);
 842     __ movptr(laddress(rbx), rax);
 843     __ movptr(haddress(rbx), rdx);
 844     __ tag_local(frame::TagCategory2, rbx);
 845   } else {
 846     __ fstp_d(daddress(rbx));
 847   }
 848 }
 849 
 850 
 851 void TemplateTable::astore() {
 852   transition(vtos, vtos);
 853   __ pop_ptr(rax, rdx);   // will need to pop tag too
 854   locals_index(rbx);
 855   __ movptr(aaddress(rbx), rax);
 856   __ tag_local(rdx, rbx);    // need to store same tag in local may be returnAddr
 857 }
 858 
 859 
 860 void TemplateTable::wide_istore() {
 861   transition(vtos, vtos);
 862   __ pop_i(rax);
 863   locals_index_wide(rbx);
 864   __ movl(iaddress(rbx), rax);
 865   __ tag_local(frame::TagValue, rbx);
 866 }
 867 
 868 
 869 void TemplateTable::wide_lstore() {
 870   transition(vtos, vtos);
 871   __ pop_l(rax, rdx);
 872   locals_index_wide(rbx);
 873   __ movptr(laddress(rbx), rax);
 874   NOT_LP64(__ movl(haddress(rbx), rdx));
 875   __ tag_local(frame::TagCategory2, rbx);
 876 }
 877 
 878 
 879 void TemplateTable::wide_fstore() {
 880   wide_istore();
 881 }
 882 
 883 
 884 void TemplateTable::wide_dstore() {
 885   wide_lstore();
 886 }
 887 
 888 
 889 void TemplateTable::wide_astore() {
 890   transition(vtos, vtos);
 891   __ pop_ptr(rax, rdx);
 892   locals_index_wide(rbx);
 893   __ movptr(aaddress(rbx), rax);
 894   __ tag_local(rdx, rbx);
 895 }
 896 
 897 
 898 void TemplateTable::iastore() {
 899   transition(itos, vtos);
 900   __ pop_i(rbx);
 901   // rax,: value
 902   // rdx: array
 903   index_check(rdx, rbx);  // prefer index in rbx,
 904   // rbx,: index
 905   __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
 906 }
 907 
 908 
 909 void TemplateTable::lastore() {
 910   transition(ltos, vtos);
 911   __ pop_i(rbx);
 912   // rax,: low(value)
 913   // rcx: array
 914   // rdx: high(value)
 915   index_check(rcx, rbx);  // prefer index in rbx,
 916   // rbx,: index
 917   __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
 918   NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
 919 }
 920 
 921 
 922 void TemplateTable::fastore() {
 923   transition(ftos, vtos);
 924   __ pop_i(rbx);
 925   // rdx: array
 926   // st0: value
 927   index_check(rdx, rbx);  // prefer index in rbx,
 928   // rbx,: index
 929   __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 930 }
 931 
 932 
 933 void TemplateTable::dastore() {
 934   transition(dtos, vtos);
 935   __ pop_i(rbx);
 936   // rdx: array
 937   // st0: value
 938   index_check(rdx, rbx);  // prefer index in rbx,
 939   // rbx,: index
 940   __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 941 }
 942 
 943 
 944 void TemplateTable::aastore() {
 945   Label is_null, ok_is_subtype, done;
 946   transition(vtos, vtos);
 947   // stack: ..., array, index, value
 948   __ movptr(rax, at_tos());     // Value
 949   __ movl(rcx, at_tos_p1());  // Index
 950   __ movptr(rdx, at_tos_p2());  // Array
 951 
 952   Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 953   index_check_without_pop(rdx, rcx);      // kills rbx,
 954   // do array store check - check for NULL value first
 955   __ testptr(rax, rax);
 956   __ jcc(Assembler::zero, is_null);
 957 
 958   // Move subklass into EBX
 959   __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
 960   // Move superklass into EAX
 961   __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
 962   __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
 963   // Compress array+index*wordSize+12 into a single register.  Frees ECX.
 964   __ lea(rdx, element_address);
 965 
 966   // Generate subtype check.  Blows ECX.  Resets EDI to locals.
 967   // Superklass in EAX.  Subklass in EBX.
 968   __ gen_subtype_check( rbx, ok_is_subtype );
 969 
 970   // Come here on failure
 971   // object is at TOS
 972   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
 973 
 974   // Come here on success
 975   __ bind(ok_is_subtype);
 976 
 977   // Get the value to store
 978   __ movptr(rax, at_rsp());
 979   // and store it with appropriate barrier
 980   do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
 981 
 982   __ jmp(done);
 983 
 984   // Have a NULL in EAX, EDX=array, ECX=index.  Store NULL at ary[idx]
 985   __ bind(is_null);
 986   __ profile_null_seen(rbx);
 987 
 988   // Store NULL, (noreg means NULL to do_oop_store)
 989   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
 990 
 991   // Pop stack arguments
 992   __ bind(done);
 993   __ addptr(rsp, 3 * Interpreter::stackElementSize());
 994 }
 995 
 996 
 997 void TemplateTable::bastore() {
 998   transition(itos, vtos);
 999   __ pop_i(rbx);
1000   // rax,: value
1001   // rdx: array
1002   index_check(rdx, rbx);  // prefer index in rbx,
1003   // rbx,: index
1004   __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
1005 }
1006 
1007 
1008 void TemplateTable::castore() {
1009   transition(itos, vtos);
1010   __ pop_i(rbx);
1011   // rax,: value
1012   // rdx: array
1013   index_check(rdx, rbx);  // prefer index in rbx,
1014   // rbx,: index
1015   __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
1016 }
1017 
1018 
1019 void TemplateTable::sastore() {
1020   castore();
1021 }
1022 
1023 
1024 void TemplateTable::istore(int n) {
1025   transition(itos, vtos);
1026   __ movl(iaddress(n), rax);
1027   __ tag_local(frame::TagValue, n);
1028 }
1029 
1030 
1031 void TemplateTable::lstore(int n) {
1032   transition(ltos, vtos);
1033   __ movptr(laddress(n), rax);
1034   NOT_LP64(__ movptr(haddress(n), rdx));
1035   __ tag_local(frame::TagCategory2, n);
1036 }
1037 
1038 
1039 void TemplateTable::fstore(int n) {
1040   transition(ftos, vtos);
1041   __ fstp_s(faddress(n));
1042   __ tag_local(frame::TagValue, n);
1043 }
1044 
1045 
1046 void TemplateTable::dstore(int n) {
1047   transition(dtos, vtos);
1048   if (TaggedStackInterpreter) {
1049     __ subptr(rsp, 2 * wordSize);
1050     __ fstp_d(Address(rsp, 0));
1051     __ pop(rax);
1052     __ pop(rdx);
1053     __ movl(laddress(n), rax);
1054     __ movl(haddress(n), rdx);
1055     __ tag_local(frame::TagCategory2, n);
1056   } else {
1057     __ fstp_d(daddress(n));
1058   }
1059 }
1060 
1061 
1062 void TemplateTable::astore(int n) {
1063   transition(vtos, vtos);
1064   __ pop_ptr(rax, rdx);
1065   __ movptr(aaddress(n), rax);
1066   __ tag_local(rdx, n);
1067 }
1068 
1069 
1070 void TemplateTable::pop() {
1071   transition(vtos, vtos);
1072   __ addptr(rsp, Interpreter::stackElementSize());
1073 }
1074 
1075 
1076 void TemplateTable::pop2() {
1077   transition(vtos, vtos);
1078   __ addptr(rsp, 2*Interpreter::stackElementSize());
1079 }
1080 
1081 
1082 void TemplateTable::dup() {
1083   transition(vtos, vtos);
1084   // stack: ..., a
1085   __ load_ptr_and_tag(0, rax, rdx);
1086   __ push_ptr(rax, rdx);
1087   // stack: ..., a, a
1088 }
1089 
1090 
1091 void TemplateTable::dup_x1() {
1092   transition(vtos, vtos);
1093   // stack: ..., a, b
1094   __ load_ptr_and_tag(0, rax, rdx);  // load b
1095   __ load_ptr_and_tag(1, rcx, rbx);  // load a
1096   __ store_ptr_and_tag(1, rax, rdx); // store b
1097   __ store_ptr_and_tag(0, rcx, rbx); // store a
1098   __ push_ptr(rax, rdx);             // push b


1165   // stack: ..., c, d, a, b, c, d
1166   // stack: ..., c, d, a, b, c, d
1167 }
1168 
1169 
1170 void TemplateTable::swap() {
1171   transition(vtos, vtos);
1172   // stack: ..., a, b
1173   __ load_ptr_and_tag(1, rcx, rbx);  // load a
1174   __ load_ptr_and_tag(0, rax, rdx);  // load b
1175   __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1176   __ store_ptr_and_tag(1, rax, rdx); // store b in a
1177   // stack: ..., b, a
1178 }
1179 
1180 
1181 void TemplateTable::iop2(Operation op) {
1182   transition(itos, itos);
1183   switch (op) {
1184     case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1185     case sub  : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1186     case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1187     case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1188     case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1189     case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1190     case shl  : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1191     case shr  : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1192     case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax);      break; // implicit masking of lower 5 bits by Intel shift instr.
1193     default   : ShouldNotReachHere();
1194   }
1195 }
1196 
1197 
1198 void TemplateTable::lop2(Operation op) {
1199   transition(ltos, ltos);
1200   __ pop_l(rbx, rcx);
1201   switch (op) {
1202     case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1203     case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1204                __ mov(rax, rbx); __ mov(rdx, rcx); break;
1205     case _and: __ andl(rax, rbx); __ andl(rdx, rcx); break;
1206     case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1207     case _xor: __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1208     default : ShouldNotReachHere();
1209   }
1210 }
1211 
1212 
1213 void TemplateTable::idiv() {
1214   transition(itos, itos);
1215   __ mov(rcx, rax);
1216   __ pop_i(rax);
1217   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1218   //       they are not equal, one could do a normal division (no correction
1219   //       needed), which may speed up this implementation for the common case.
1220   //       (see also JVM spec., p.243 & p.271)
1221   __ corrected_idivl(rcx);
1222 }
1223 
1224 
1225 void TemplateTable::irem() {
1226   transition(itos, itos);
1227   __ mov(rcx, rax);
1228   __ pop_i(rax);
1229   // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1230   //       they are not equal, one could do a normal division (no correction
1231   //       needed), which may speed up this implementation for the common case.
1232   //       (see also JVM spec., p.243 & p.271)
1233   __ corrected_idivl(rcx);
1234   __ mov(rax, rdx);
1235 }
1236 
1237 
1238 void TemplateTable::lmul() {
1239   transition(ltos, ltos);
1240   __ pop_l(rbx, rcx);
1241   __ push(rcx); __ push(rbx);
1242   __ push(rdx); __ push(rax);
1243   __ lmul(2 * wordSize, 0);
1244   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1245 }
1246 
1247 
1248 void TemplateTable::ldiv() {
1249   transition(ltos, ltos);
1250   __ pop_l(rbx, rcx);
1251   __ push(rcx); __ push(rbx);
1252   __ push(rdx); __ push(rax);
1253   // check if y = 0
1254   __ orl(rax, rdx);
1255   __ jump_cc(Assembler::zero,
1256              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1257   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1258   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1259 }
1260 
1261 
1262 void TemplateTable::lrem() {
1263   transition(ltos, ltos);
1264   __ pop_l(rbx, rcx);
1265   __ push(rcx); __ push(rbx);
1266   __ push(rdx); __ push(rax);
1267   // check if y = 0
1268   __ orl(rax, rdx);
1269   __ jump_cc(Assembler::zero,
1270              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1271   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1272   __ addptr(rsp, 4 * wordSize);
1273 }
1274 
1275 
1276 void TemplateTable::lshl() {
1277   transition(itos, ltos);
1278   __ movl(rcx, rax);                             // get shift count
1279   __ pop_l(rax, rdx);                            // get shift value
1280   __ lshl(rdx, rax);
1281 }
1282 
1283 
1284 void TemplateTable::lshr() {
1285   transition(itos, ltos);
1286   __ mov(rcx, rax);                              // get shift count
1287   __ pop_l(rax, rdx);                            // get shift value
1288   __ lshr(rdx, rax, true);
1289 }
1290 
1291 
1292 void TemplateTable::lushr() {
1293   transition(itos, ltos);
1294   __ mov(rcx, rax);                              // get shift count
1295   __ pop_l(rax, rdx);                            // get shift value
1296   __ lshr(rdx, rax);
1297 }
1298 
1299 
1300 void TemplateTable::fop2(Operation op) {
1301   transition(ftos, ftos);
1302   __ pop_ftos_to_rsp();  // pop ftos into rsp
1303   switch (op) {
1304     case add: __ fadd_s (at_rsp());                break;
1305     case sub: __ fsubr_s(at_rsp());                break;
1306     case mul: __ fmul_s (at_rsp());                break;
1307     case div: __ fdivr_s(at_rsp());                break;
1308     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1309     default : ShouldNotReachHere();
1310   }
1311   __ f2ieee();
1312   __ pop(rax);  // pop float thing off
1313 }
1314 
1315 
1316 void TemplateTable::dop2(Operation op) {
1317   transition(dtos, dtos);
1318   __ pop_dtos_to_rsp();  // pop dtos into rsp
1319 
1320   switch (op) {
1321     case add: __ fadd_d (at_rsp());                break;
1322     case sub: __ fsubr_d(at_rsp());                break;
1323     case mul: {
1324       Label L_strict;
1325       Label L_join;
1326       const Address access_flags      (rcx, methodOopDesc::access_flags_offset());
1327       __ get_method(rcx);
1328       __ movl(rcx, access_flags);
1329       __ testl(rcx, JVM_ACC_STRICT);
1330       __ jccb(Assembler::notZero, L_strict);
1331       __ fmul_d (at_rsp());
1332       __ jmpb(L_join);


1346       __ get_method(rcx);
1347       __ movl(rcx, access_flags);
1348       __ testl(rcx, JVM_ACC_STRICT);
1349       __ jccb(Assembler::notZero, L_strict);
1350       __ fdivr_d(at_rsp());
1351       __ jmp(L_join);
1352       __ bind(L_strict);
1353       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1354       __ fmul_d (at_rsp());
1355       __ fdivrp();
1356       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1357       __ fmulp();
1358       __ bind(L_join);
1359       break;
1360     }
1361     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1362     default : ShouldNotReachHere();
1363   }
1364   __ d2ieee();
1365   // Pop double precision number from rsp.
1366   __ pop(rax);
1367   __ pop(rdx);
1368 }
1369 
1370 
1371 void TemplateTable::ineg() {
1372   transition(itos, itos);
1373   __ negl(rax);
1374 }
1375 
1376 
1377 void TemplateTable::lneg() {
1378   transition(ltos, ltos);
1379   __ lneg(rdx, rax);
1380 }
1381 
1382 
1383 void TemplateTable::fneg() {
1384   transition(ftos, ftos);
1385   __ fchs();
1386 }
1387 
1388 
1389 void TemplateTable::dneg() {
1390   transition(dtos, dtos);
1391   __ fchs();
1392 }
1393 
1394 
1395 void TemplateTable::iinc() {
1396   transition(vtos, vtos);
1397   __ load_signed_byte(rdx, at_bcp(2));           // get constant
1398   locals_index(rbx);
1399   __ addl(iaddress(rbx), rdx);
1400 }
1401 
1402 
1403 void TemplateTable::wide_iinc() {
1404   transition(vtos, vtos);
1405   __ movl(rdx, at_bcp(4));                       // get constant
1406   locals_index_wide(rbx);
1407   __ bswapl(rdx);                                 // swap bytes & sign-extend constant
1408   __ sarl(rdx, 16);
1409   __ addl(iaddress(rbx), rdx);
1410   // Note: should probably use only one movl to get both
1411   //       the index and the constant -> fix this
1412 }
1413 
1414 
1415 void TemplateTable::convert() {
1416   // Checking
1417 #ifdef ASSERT
1418   { TosState tos_in  = ilgl;
1419     TosState tos_out = ilgl;
1420     switch (bytecode()) {
1421       case Bytecodes::_i2l: // fall through
1422       case Bytecodes::_i2f: // fall through
1423       case Bytecodes::_i2d: // fall through
1424       case Bytecodes::_i2b: // fall through
1425       case Bytecodes::_i2c: // fall through
1426       case Bytecodes::_i2s: tos_in = itos; break;
1427       case Bytecodes::_l2i: // fall through


1441       case Bytecodes::_d2i: // fall through
1442       case Bytecodes::_i2b: // fall through
1443       case Bytecodes::_i2c: // fall through
1444       case Bytecodes::_i2s: tos_out = itos; break;
1445       case Bytecodes::_i2l: // fall through
1446       case Bytecodes::_f2l: // fall through
1447       case Bytecodes::_d2l: tos_out = ltos; break;
1448       case Bytecodes::_i2f: // fall through
1449       case Bytecodes::_l2f: // fall through
1450       case Bytecodes::_d2f: tos_out = ftos; break;
1451       case Bytecodes::_i2d: // fall through
1452       case Bytecodes::_l2d: // fall through
1453       case Bytecodes::_f2d: tos_out = dtos; break;
1454       default             : ShouldNotReachHere();
1455     }
1456     transition(tos_in, tos_out);
1457   }
1458 #endif // ASSERT
1459 
1460   // Conversion
1461   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1462   switch (bytecode()) {
1463     case Bytecodes::_i2l:
1464       __ extend_sign(rdx, rax);
1465       break;
1466     case Bytecodes::_i2f:
1467       __ push(rax);          // store int on tos
1468       __ fild_s(at_rsp());   // load int to ST0
1469       __ f2ieee();           // truncate to float size
1470       __ pop(rcx);           // adjust rsp
1471       break;
1472     case Bytecodes::_i2d:
1473       __ push(rax);          // add one slot for d2ieee()
1474       __ push(rax);          // store int on tos
1475       __ fild_s(at_rsp());   // load int to ST0
1476       __ d2ieee();           // truncate to double size
1477       __ pop(rcx);           // adjust rsp
1478       __ pop(rcx);
1479       break;
1480     case Bytecodes::_i2b:
1481       __ shll(rax, 24);      // truncate upper 24 bits
1482       __ sarl(rax, 24);      // and sign-extend byte
1483       LP64_ONLY(__ movsbl(rax, rax));
1484       break;
1485     case Bytecodes::_i2c:
1486       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1487       LP64_ONLY(__ movzwl(rax, rax));
1488       break;
1489     case Bytecodes::_i2s:
1490       __ shll(rax, 16);      // truncate upper 16 bits
1491       __ sarl(rax, 16);      // and sign-extend short
1492       LP64_ONLY(__ movswl(rax, rax));
1493       break;
1494     case Bytecodes::_l2i:
1495       /* nothing to do */
1496       break;
1497     case Bytecodes::_l2f:
1498       __ push(rdx);          // store long on tos
1499       __ push(rax);
1500       __ fild_d(at_rsp());   // load long to ST0
1501       __ f2ieee();           // truncate to float size
1502       __ pop(rcx);           // adjust rsp
1503       __ pop(rcx);
1504       break;
1505     case Bytecodes::_l2d:
1506       __ push(rdx);          // store long on tos
1507       __ push(rax);
1508       __ fild_d(at_rsp());   // load long to ST0
1509       __ d2ieee();           // truncate to double size
1510       __ pop(rcx);           // adjust rsp
1511       __ pop(rcx);
1512       break;
1513     case Bytecodes::_f2i:
1514       __ push(rcx);          // reserve space for argument
1515       __ fstp_s(at_rsp());   // pass float argument on stack
1516       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1517       break;
1518     case Bytecodes::_f2l:
1519       __ push(rcx);          // reserve space for argument
1520       __ fstp_s(at_rsp());   // pass float argument on stack
1521       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1522       break;
1523     case Bytecodes::_f2d:
1524       /* nothing to do */
1525       break;
1526     case Bytecodes::_d2i:
1527       __ push(rcx);          // reserve space for argument
1528       __ push(rcx);
1529       __ fstp_d(at_rsp());   // pass double argument on stack
1530       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1531       break;
1532     case Bytecodes::_d2l:
1533       __ push(rcx);          // reserve space for argument
1534       __ push(rcx);
1535       __ fstp_d(at_rsp());   // pass double argument on stack
1536       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1537       break;
1538     case Bytecodes::_d2f:
1539       __ push(rcx);          // reserve space for f2ieee()
1540       __ f2ieee();           // truncate to float size
1541       __ pop(rcx);           // adjust rsp
1542       break;
1543     default             :
1544       ShouldNotReachHere();
1545   }
1546 }
1547 
1548 
1549 void TemplateTable::lcmp() {
1550   transition(ltos, itos);
1551   // y = rdx:rax
1552   __ pop_l(rbx, rcx);             // get x = rcx:rbx
1553   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1554   __ mov(rax, rcx);
1555 }
1556 
1557 
1558 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1559   if (is_float) {
1560     __ pop_ftos_to_rsp();
1561     __ fld_s(at_rsp());
1562   } else {
1563     __ pop_dtos_to_rsp();
1564     __ fld_d(at_rsp());
1565     __ pop(rdx);
1566   }
1567   __ pop(rcx);
1568   __ fcmp2int(rax, unordered_result < 0);
1569 }
1570 
1571 
1572 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1573   __ get_method(rcx);           // ECX holds method
1574   __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1575 
1576   const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1577   const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1578   const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1579 
1580   // Load up EDX with the branch displacement
1581   __ movl(rdx, at_bcp(1));
1582   __ bswapl(rdx);
1583   if (!is_wide) __ sarl(rdx, 16);
1584   LP64_ONLY(__ movslq(rdx, rdx));
1585 
1586 
1587   // Handle all the JSR stuff here, then exit.
1588   // It's much shorter and cleaner than intermingling with the
1589   // non-JSR normal-branch stuff occuring below.
1590   if (is_jsr) {
1591     // Pre-load the next target bytecode into EBX
1592     __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1593 
1594     // compute return address as bci in rax,
1595     __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1596     __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1597     // Adjust the bcp in RSI by the displacement in EDX
1598     __ addptr(rsi, rdx);
1599     // Push return address
1600     __ push_i(rax);
1601     // jsr returns vtos
1602     __ dispatch_only_noverify(vtos);
1603     return;
1604   }
1605 
1606   // Normal (non-jsr) branch handling
1607 
1608   // Adjust the bcp in RSI by the displacement in EDX
1609   __ addptr(rsi, rdx);
1610 
1611   assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1612   Label backedge_counter_overflow;
1613   Label profile_method;
1614   Label dispatch;
1615   if (UseLoopCounter) {
1616     // increment backedge counter for backward branches
1617     // rax,: MDO
1618     // rbx,: MDO bumped taken-count
1619     // rcx: method
1620     // rdx: target offset
1621     // rsi: target bcp
1622     // rdi: locals pointer
1623     __ testl(rdx, rdx);             // check if forward or backward branch
1624     __ jcc(Assembler::positive, dispatch); // count only if backward branch
1625 
1626     // increment counter
1627     __ movl(rax, Address(rcx, be_offset));        // load backedge counter
1628     __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1629     __ movl(Address(rcx, be_offset), rax);        // store counter
1630 
1631     __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
1632     __ andl(rax, InvocationCounter::count_mask_value);     // and the status bits
1633     __ addl(rax, Address(rcx, be_offset));        // add both counters
1634 
1635     if (ProfileInterpreter) {
1636       // Test to see if we should create a method data oop
1637       __ cmp32(rax,
1638                ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1639       __ jcc(Assembler::less, dispatch);
1640 
1641       // if no method data exists, go to profile method
1642       __ test_method_data_pointer(rax, profile_method);
1643 
1644       if (UseOnStackReplacement) {
1645         // check for overflow against rbx, which is the MDO taken count
1646         __ cmp32(rbx,
1647                  ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1648         __ jcc(Assembler::below, dispatch);
1649 
1650         // When ProfileInterpreter is on, the backedge_count comes from the
1651         // methodDataOop, which value does not get reset on the call to
1652         // frequency_counter_overflow().  To avoid excessive calls to the overflow
1653         // routine while the method is being compiled, add a second test to make
1654         // sure the overflow function is called only once every overflow_frequency.
1655         const int overflow_frequency = 1024;
1656         __ andptr(rbx, overflow_frequency-1);
1657         __ jcc(Assembler::zero, backedge_counter_overflow);
1658 
1659       }
1660     } else {
1661       if (UseOnStackReplacement) {
1662         // check for overflow against rax, which is the sum of the counters
1663         __ cmp32(rax,
1664                  ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1665         __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1666 
1667       }
1668     }
1669     __ bind(dispatch);
1670   }
1671 
1672   // Pre-load the next target bytecode into EBX
1673   __ load_unsigned_byte(rbx, Address(rsi, 0));
1674 
1675   // continue with the bytecode @ target
1676   // rax,: return bci for jsr's, unused otherwise
1677   // rbx,: target bytecode
1678   // rsi: target bcp
1679   __ dispatch_only(vtos);
1680 
1681   if (UseLoopCounter) {
1682     if (ProfileInterpreter) {
1683       // Out-of-line code to allocate method data oop.
1684       __ bind(profile_method);
1685       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1686       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1687       __ movptr(rcx, Address(rbp, method_offset));
1688       __ movptr(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1689       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1690       __ test_method_data_pointer(rcx, dispatch);
1691       // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1692       __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1693       __ addptr(rcx, rax);
1694       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1695       __ jmp(dispatch);
1696     }
1697 
1698     if (UseOnStackReplacement) {
1699 
1700       // invocation counter overflow
1701       __ bind(backedge_counter_overflow);
1702       __ negptr(rdx);
1703       __ addptr(rdx, rsi);        // branch bcp
1704       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1705       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
1706 
1707       // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1708       // rbx,: target bytecode
1709       // rdx: scratch
1710       // rdi: locals pointer
1711       // rsi: bcp
1712       __ testptr(rax, rax);                      // test result
1713       __ jcc(Assembler::zero, dispatch);         // no osr if null
1714       // nmethod may have been invalidated (VM may block upon call_VM return)
1715       __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1716       __ cmpl(rcx, InvalidOSREntryBci);
1717       __ jcc(Assembler::equal, dispatch);
1718 
1719       // We have the address of an on stack replacement routine in rax,
1720       // We need to prepare to execute the OSR method. First we must
1721       // migrate the locals and monitors off of the stack.
1722 
1723       __ mov(rbx, rax);                             // save the nmethod
1724 
1725       const Register thread = rcx;
1726       __ get_thread(thread);
1727       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1728       // rax, is OSR buffer, move it to expected parameter location
1729       __ mov(rcx, rax);
1730 
1731       // pop the interpreter frame
1732       __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1733       __ leave();                                // remove frame anchor
1734       __ pop(rdi);                               // get return address
1735       __ mov(rsp, rdx);                          // set sp to sender sp
1736 
1737 
1738       Label skip;
1739       Label chkint;
1740 
1741       // The interpreter frame we have removed may be returning to
1742       // either the callstub or the interpreter. Since we will
1743       // now be returning from a compiled (OSR) nmethod we must
1744       // adjust the return to the return were it can handler compiled
1745       // results and clean the fpu stack. This is very similar to
1746       // what a i2c adapter must do.
1747 
1748       // Are we returning to the call stub?
1749 
1750       __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1751       __ jcc(Assembler::notEqual, chkint);
1752 
1753       // yes adjust to the specialized call stub  return.
1754       assert(StubRoutines::x86::get_call_stub_compiled_return() != NULL, "must be set");
1755       __ lea(rdi, ExternalAddress(StubRoutines::x86::get_call_stub_compiled_return()));
1756       __ jmp(skip);
1757 
1758       __ bind(chkint);
1759 
1760       // Are we returning to the interpreter? Look for sentinel
1761 
1762       __ cmpl(Address(rdi, -2*wordSize), Interpreter::return_sentinel);
1763       __ jcc(Assembler::notEqual, skip);
1764 
1765       // Adjust to compiled return back to interpreter
1766 
1767       __ movptr(rdi, Address(rdi, -wordSize));
1768       __ bind(skip);
1769 
1770       // Align stack pointer for compiled code (note that caller is
1771       // responsible for undoing this fixup by remembering the old SP
1772       // in an rbp,-relative location)
1773       __ andptr(rsp, -(StackAlignmentInBytes));
1774 
1775       // push the (possibly adjusted) return address
1776       __ push(rdi);
1777 
1778       // and begin the OSR nmethod
1779       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1780     }
1781   }
1782 }
1783 
1784 
1785 void TemplateTable::if_0cmp(Condition cc) {
1786   transition(itos, vtos);
1787   // assume branch is more often taken than not (loops use backward branches)
1788   Label not_taken;
1789   __ testl(rax, rax);
1790   __ jcc(j_not(cc), not_taken);
1791   branch(false, false);
1792   __ bind(not_taken);
1793   __ profile_not_taken_branch(rax);
1794 }
1795 
1796 
1797 void TemplateTable::if_icmp(Condition cc) {
1798   transition(itos, vtos);
1799   // assume branch is more often taken than not (loops use backward branches)
1800   Label not_taken;
1801   __ pop_i(rdx);
1802   __ cmpl(rdx, rax);
1803   __ jcc(j_not(cc), not_taken);
1804   branch(false, false);
1805   __ bind(not_taken);
1806   __ profile_not_taken_branch(rax);
1807 }
1808 
1809 
1810 void TemplateTable::if_nullcmp(Condition cc) {
1811   transition(atos, vtos);
1812   // assume branch is more often taken than not (loops use backward branches)
1813   Label not_taken;
1814   __ testptr(rax, rax);
1815   __ jcc(j_not(cc), not_taken);
1816   branch(false, false);
1817   __ bind(not_taken);
1818   __ profile_not_taken_branch(rax);
1819 }
1820 
1821 
1822 void TemplateTable::if_acmp(Condition cc) {
1823   transition(atos, vtos);
1824   // assume branch is more often taken than not (loops use backward branches)
1825   Label not_taken;
1826   __ pop_ptr(rdx);
1827   __ cmpptr(rdx, rax);
1828   __ jcc(j_not(cc), not_taken);
1829   branch(false, false);
1830   __ bind(not_taken);
1831   __ profile_not_taken_branch(rax);
1832 }
1833 
1834 
1835 void TemplateTable::ret() {
1836   transition(vtos, vtos);
1837   locals_index(rbx);
1838   __ movptr(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1839   __ profile_ret(rbx, rcx);
1840   __ get_method(rax);
1841   __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1842   __ lea(rsi, Address(rsi, rbx, Address::times_1,
1843                       constMethodOopDesc::codes_offset()));
1844   __ dispatch_next(vtos);
1845 }
1846 
1847 
1848 void TemplateTable::wide_ret() {
1849   transition(vtos, vtos);
1850   locals_index_wide(rbx);
1851   __ movptr(rbx, iaddress(rbx));                   // get return bci, compute return bcp
1852   __ profile_ret(rbx, rcx);
1853   __ get_method(rax);
1854   __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1855   __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1856   __ dispatch_next(vtos);
1857 }
1858 
1859 
1860 void TemplateTable::tableswitch() {
1861   Label default_case, continue_execution;
1862   transition(itos, vtos);
1863   // align rsi
1864   __ lea(rbx, at_bcp(wordSize));
1865   __ andptr(rbx, -wordSize);
1866   // load lo & hi
1867   __ movl(rcx, Address(rbx, 1 * wordSize));
1868   __ movl(rdx, Address(rbx, 2 * wordSize));
1869   __ bswapl(rcx);
1870   __ bswapl(rdx);
1871   // check against lo & hi
1872   __ cmpl(rax, rcx);
1873   __ jccb(Assembler::less, default_case);
1874   __ cmpl(rax, rdx);
1875   __ jccb(Assembler::greater, default_case);
1876   // lookup dispatch offset
1877   __ subl(rax, rcx);
1878   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1879   __ profile_switch_case(rax, rbx, rcx);
1880   // continue execution
1881   __ bind(continue_execution);
1882   __ bswapl(rdx);
1883   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1884   __ addptr(rsi, rdx);
1885   __ dispatch_only(vtos);
1886   // handle default
1887   __ bind(default_case);
1888   __ profile_switch_default(rax);
1889   __ movl(rdx, Address(rbx, 0));
1890   __ jmp(continue_execution);
1891 }
1892 
1893 
1894 void TemplateTable::lookupswitch() {
1895   transition(itos, itos);
1896   __ stop("lookupswitch bytecode should have been rewritten");
1897 }
1898 
1899 
1900 void TemplateTable::fast_linearswitch() {
1901   transition(itos, vtos);
1902   Label loop_entry, loop, found, continue_execution;
1903   // bswapl rax, so we can avoid bswapping the table entries
1904   __ bswapl(rax);
1905   // align rsi
1906   __ lea(rbx, at_bcp(wordSize));                // btw: should be able to get rid of this instruction (change offsets below)
1907   __ andptr(rbx, -wordSize);
1908   // set counter
1909   __ movl(rcx, Address(rbx, wordSize));
1910   __ bswapl(rcx);
1911   __ jmpb(loop_entry);
1912   // table search
1913   __ bind(loop);
1914   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1915   __ jccb(Assembler::equal, found);
1916   __ bind(loop_entry);
1917   __ decrementl(rcx);
1918   __ jcc(Assembler::greaterEqual, loop);
1919   // default case
1920   __ profile_switch_default(rax);
1921   __ movl(rdx, Address(rbx, 0));
1922   __ jmpb(continue_execution);
1923   // entry found -> get offset
1924   __ bind(found);
1925   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1926   __ profile_switch_case(rcx, rax, rbx);
1927   // continue execution
1928   __ bind(continue_execution);
1929   __ bswapl(rdx);
1930   __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1931   __ addptr(rsi, rdx);
1932   __ dispatch_only(vtos);
1933 }
1934 
1935 
1936 void TemplateTable::fast_binaryswitch() {
1937   transition(itos, vtos);
1938   // Implementation using the following core algorithm:
1939   //
1940   // int binary_search(int key, LookupswitchPair* array, int n) {
1941   //   // Binary search according to "Methodik des Programmierens" by
1942   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1943   //   int i = 0;
1944   //   int j = n;
1945   //   while (i+1 < j) {
1946   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1947   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1948   //     // where a stands for the array and assuming that the (inexisting)
1949   //     // element a[n] is infinitely big.
1950   //     int h = (i + j) >> 1;
1951   //     // i < h < j


1953   //       j = h;
1954   //     } else {
1955   //       i = h;
1956   //     }
1957   //   }
1958   //   // R: a[i] <= key < a[i+1] or Q
1959   //   // (i.e., if key is within array, i is the correct index)
1960   //   return i;
1961   // }
1962 
1963   // register allocation
1964   const Register key   = rax;                    // already set (tosca)
1965   const Register array = rbx;
1966   const Register i     = rcx;
1967   const Register j     = rdx;
1968   const Register h     = rdi;                    // needs to be restored
1969   const Register temp  = rsi;
1970   // setup array
1971   __ save_bcp();
1972 
1973   __ lea(array, at_bcp(3*wordSize));             // btw: should be able to get rid of this instruction (change offsets below)
1974   __ andptr(array, -wordSize);
1975   // initialize i & j
1976   __ xorl(i, i);                                 // i = 0;
1977   __ movl(j, Address(array, -wordSize));         // j = length(array);
1978   // Convert j into native byteordering
1979   __ bswapl(j);
1980   // and start
1981   Label entry;
1982   __ jmp(entry);
1983 
1984   // binary search loop
1985   { Label loop;
1986     __ bind(loop);
1987     // int h = (i + j) >> 1;
1988     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1989     __ sarl(h, 1);                               // h = (i + j) >> 1;
1990     // if (key < array[h].fast_match()) {
1991     //   j = h;
1992     // } else {
1993     //   i = h;
1994     // }
1995     // Convert array[h].match to native byte-ordering before compare
1996     __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1997     __ bswapl(temp);
1998     __ cmpl(key, temp);
1999     if (VM_Version::supports_cmov()) {
2000       __ cmovl(Assembler::less        , j, h);   // j = h if (key <  array[h].fast_match())
2001       __ cmovl(Assembler::greaterEqual, i, h);   // i = h if (key >= array[h].fast_match())
2002     } else {
2003       Label set_i, end_of_if;
2004       __ jccb(Assembler::greaterEqual, set_i);     // {
2005       __ mov(j, h);                                //   j = h;
2006       __ jmp(end_of_if);                           // }
2007       __ bind(set_i);                              // else {
2008       __ mov(i, h);                                //   i = h;
2009       __ bind(end_of_if);                          // }
2010     }
2011     // while (i+1 < j)
2012     __ bind(entry);
2013     __ leal(h, Address(i, 1));                   // i+1
2014     __ cmpl(h, j);                               // i+1 < j
2015     __ jcc(Assembler::less, loop);
2016   }
2017 
2018   // end of binary search, result index is i (must check again!)
2019   Label default_case;
2020   // Convert array[i].match to native byte-ordering before compare
2021   __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
2022   __ bswapl(temp);
2023   __ cmpl(key, temp);
2024   __ jcc(Assembler::notEqual, default_case);
2025 
2026   // entry found -> j = offset
2027   __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
2028   __ profile_switch_case(i, key, array);
2029   __ bswapl(j);
2030   LP64_ONLY(__ movslq(j, j));
2031   __ restore_bcp();
2032   __ restore_locals();                           // restore rdi
2033   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2034 
2035   __ addptr(rsi, j);
2036   __ dispatch_only(vtos);
2037 
2038   // default case -> j = default offset
2039   __ bind(default_case);
2040   __ profile_switch_default(i);
2041   __ movl(j, Address(array, -2*wordSize));
2042   __ bswapl(j);
2043   LP64_ONLY(__ movslq(j, j));
2044   __ restore_bcp();
2045   __ restore_locals();                           // restore rdi
2046   __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2047   __ addptr(rsi, j);
2048   __ dispatch_only(vtos);
2049 }
2050 
2051 
2052 void TemplateTable::_return(TosState state) {
2053   transition(state, state);
2054   assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2055 
2056   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2057     assert(state == vtos, "only valid state");
2058     __ movptr(rax, aaddress(0));
2059     __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
2060     __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2061     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2062     Label skip_register_finalizer;
2063     __ jcc(Assembler::zero, skip_register_finalizer);
2064 
2065     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2066 
2067     __ bind(skip_register_finalizer);
2068   }
2069 
2070   __ remove_activation(state, rsi);
2071   __ jmp(rsi);
2072 }
2073 
2074 
2075 // ----------------------------------------------------------------------------
2076 // Volatile variables demand their effects be made known to all CPU's in
2077 // order.  Store buffers on most chips allow reads & writes to reorder; the
2078 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2079 // memory barrier (i.e., it's not sufficient that the interpreter does not
2080 // reorder volatile references, the hardware also must not reorder them).
2081 //
2082 // According to the new Java Memory Model (JMM):
2083 // (1) All volatiles are serialized wrt to each other.
2084 // ALSO reads & writes act as aquire & release, so:
2085 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2086 // the read float up to before the read.  It's OK for non-volatile memory refs
2087 // that happen before the volatile read to float down below it.
2088 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2089 // that happen BEFORE the write float down to after the write.  It's OK for
2090 // non-volatile memory refs that happen after the volatile write to float up
2091 // before it.
2092 //
2093 // We only put in barriers around volatile refs (they are expensive), not
2094 // _between_ memory refs (that would require us to track the flavor of the
2095 // previous memory refs).  Requirements (2) and (3) require some barriers
2096 // before volatile stores and after volatile loads.  These nearly cover
2097 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2098 // case is placed after volatile-stores although it could just as well go
2099 // before volatile-loads.
2100 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2101   // Helper function to insert a is-volatile test and memory barrier
2102   if( !os::is_MP() ) return;    // Not needed on single CPU
2103   __ membar(order_constraint);
2104 }
2105 
2106 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
2107   assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
2108 
2109   Register temp = rbx;
2110 
2111   assert_different_registers(Rcache, index, temp);
2112 
2113   const int shift_count = (1 + byte_no)*BitsPerByte;
2114   Label resolved;
2115   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2116   __ movl(temp, Address(Rcache,
2117                           index,
2118                           Address::times_ptr,
2119                           constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2120   __ shrl(temp, shift_count);
2121   // have we resolved this bytecode?
2122   __ andptr(temp, 0xFF);
2123   __ cmpl(temp, (int)bytecode());
2124   __ jcc(Assembler::equal, resolved);
2125 
2126   // resolve first time through
2127   address entry;
2128   switch (bytecode()) {
2129     case Bytecodes::_getstatic      : // fall through
2130     case Bytecodes::_putstatic      : // fall through
2131     case Bytecodes::_getfield       : // fall through
2132     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2133     case Bytecodes::_invokevirtual  : // fall through
2134     case Bytecodes::_invokespecial  : // fall through
2135     case Bytecodes::_invokestatic   : // fall through
2136     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
2137     default                         : ShouldNotReachHere();                                 break;
2138   }
2139   __ movl(temp, (int)bytecode());
2140   __ call_VM(noreg, entry, temp);
2141   // Update registers with resolved info
2142   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2143   __ bind(resolved);
2144 }
2145 
2146 
2147 // The cache and index registers must be set before call
2148 void TemplateTable::load_field_cp_cache_entry(Register obj,
2149                                               Register cache,
2150                                               Register index,
2151                                               Register off,
2152                                               Register flags,
2153                                               bool is_static = false) {
2154   assert_different_registers(cache, index, flags, off);
2155 
2156   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2157   // Field offset
2158   __ movptr(off, Address(cache, index, Address::times_ptr,
2159                          in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2160   // Flags
2161   __ movl(flags, Address(cache, index, Address::times_ptr,
2162            in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2163 
2164   // klass     overwrite register
2165   if (is_static) {
2166     __ movptr(obj, Address(cache, index, Address::times_ptr,
2167                            in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2168   }
2169 }
2170 
2171 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2172                                                Register method,
2173                                                Register itable_index,
2174                                                Register flags,
2175                                                bool is_invokevirtual,
2176                                                bool is_invokevfinal /*unused*/) {
2177   // setup registers
2178   const Register cache = rcx;
2179   const Register index = rdx;
2180   assert_different_registers(method, flags);
2181   assert_different_registers(method, cache, index);
2182   assert_different_registers(itable_index, flags);
2183   assert_different_registers(itable_index, cache, index);
2184   // determine constant pool cache field offsets
2185   const int method_offset = in_bytes(
2186     constantPoolCacheOopDesc::base_offset() +
2187       (is_invokevirtual
2188        ? ConstantPoolCacheEntry::f2_offset()
2189        : ConstantPoolCacheEntry::f1_offset()
2190       )
2191     );
2192   const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2193                                     ConstantPoolCacheEntry::flags_offset());
2194   // access constant pool cache fields
2195   const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2196                                     ConstantPoolCacheEntry::f2_offset());
2197 
2198   resolve_cache_and_index(byte_no, cache, index);
2199 
2200   __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));

2201   if (itable_index != noreg) {
2202     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2203   }
2204   __ movl(flags , Address(cache, index, Address::times_ptr, flags_offset ));
2205 }
2206 
2207 
2208 // The registers cache and index expected to be set before call.
2209 // Correct values of the cache and index registers are preserved.
2210 void TemplateTable::jvmti_post_field_access(Register cache,
2211                                             Register index,
2212                                             bool is_static,
2213                                             bool has_tos) {
2214   if (JvmtiExport::can_post_field_access()) {
2215     // Check to see if a field access watch has been set before we take
2216     // the time to call into the VM.
2217     Label L1;
2218     assert_different_registers(cache, index, rax);
2219     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2220     __ testl(rax,rax);
2221     __ jcc(Assembler::zero, L1);
2222 
2223     // cache entry pointer
2224     __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2225     __ shll(index, LogBytesPerWord);
2226     __ addptr(cache, index);
2227     if (is_static) {
2228       __ xorptr(rax, rax);      // NULL object reference
2229     } else {
2230       __ pop(atos);         // Get the object
2231       __ verify_oop(rax);
2232       __ push(atos);        // Restore stack state
2233     }
2234     // rax,:   object pointer or NULL
2235     // cache: cache entry pointer
2236     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2237                rax, cache);
2238     __ get_cache_and_index_at_bcp(cache, index, 1);
2239     __ bind(L1);
2240   }
2241 }
2242 
2243 void TemplateTable::pop_and_check_object(Register r) {
2244   __ pop_ptr(r);
2245   __ null_check(r);  // for field access must check obj.
2246   __ verify_oop(r);
2247 }
2248 


2252   const Register cache = rcx;
2253   const Register index = rdx;
2254   const Register obj   = rcx;
2255   const Register off   = rbx;
2256   const Register flags = rax;
2257 
2258   resolve_cache_and_index(byte_no, cache, index);
2259   jvmti_post_field_access(cache, index, is_static, false);
2260   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2261 
2262   if (!is_static) pop_and_check_object(obj);
2263 
2264   const Address lo(obj, off, Address::times_1, 0*wordSize);
2265   const Address hi(obj, off, Address::times_1, 1*wordSize);
2266 
2267   Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2268 
2269   __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2270   assert(btos == 0, "change code, btos != 0");
2271   // btos
2272   __ andptr(flags, 0x0f);
2273   __ jcc(Assembler::notZero, notByte);
2274 
2275   __ load_signed_byte(rax, lo );
2276   __ push(btos);
2277   // Rewrite bytecode to be faster
2278   if (!is_static) {
2279     patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2280   }
2281   __ jmp(Done);
2282 
2283   __ bind(notByte);
2284   // itos
2285   __ cmpl(flags, itos );
2286   __ jcc(Assembler::notEqual, notInt);
2287 
2288   __ movl(rax, lo );
2289   __ push(itos);
2290   // Rewrite bytecode to be faster
2291   if (!is_static) {
2292     patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);


2320   __ bind(notChar);
2321   // stos
2322   __ cmpl(flags, stos );
2323   __ jcc(Assembler::notEqual, notShort);
2324 
2325   __ load_signed_word(rax, lo );
2326   __ push(stos);
2327   if (!is_static) {
2328     patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2329   }
2330   __ jmp(Done);
2331 
2332   __ bind(notShort);
2333   // ltos
2334   __ cmpl(flags, ltos );
2335   __ jcc(Assembler::notEqual, notLong);
2336 
2337   // Generate code as if volatile.  There just aren't enough registers to
2338   // save that information and this code is faster than the test.
2339   __ fild_d(lo);                // Must load atomically
2340   __ subptr(rsp,2*wordSize);    // Make space for store
2341   __ fistp_d(Address(rsp,0));
2342   __ pop(rax);
2343   __ pop(rdx);
2344 
2345   __ push(ltos);
2346   // Don't rewrite to _fast_lgetfield for potential volatile case.
2347   __ jmp(Done);
2348 
2349   __ bind(notLong);
2350   // ftos
2351   __ cmpl(flags, ftos );
2352   __ jcc(Assembler::notEqual, notFloat);
2353 
2354   __ fld_s(lo);
2355   __ push(ftos);
2356   if (!is_static) {
2357     patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2358   }
2359   __ jmp(Done);
2360 
2361   __ bind(notFloat);
2362   // dtos
2363   __ cmpl(flags, dtos );


2394 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2395 
2396   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2397 
2398   if (JvmtiExport::can_post_field_modification()) {
2399     // Check to see if a field modification watch has been set before we take
2400     // the time to call into the VM.
2401     Label L1;
2402     assert_different_registers(cache, index, rax);
2403     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2404     __ testl(rax, rax);
2405     __ jcc(Assembler::zero, L1);
2406 
2407     // The cache and index registers have been already set.
2408     // This allows to eliminate this call but the cache and index
2409     // registers have to be correspondingly used after this line.
2410     __ get_cache_and_index_at_bcp(rax, rdx, 1);
2411 
2412     if (is_static) {
2413       // Life is simple.  Null out the object pointer.
2414       __ xorptr(rbx, rbx);
2415     } else {
2416       // Life is harder. The stack holds the value on top, followed by the object.
2417       // We don't know the size of the value, though; it could be one or two words
2418       // depending on its type. As a result, we must find the type to determine where
2419       // the object is.
2420       Label two_word, valsize_known;
2421       __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2422                                    ConstantPoolCacheEntry::flags_offset())));
2423       __ mov(rbx, rsp);
2424       __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2425       // Make sure we don't need to mask rcx for tosBits after the above shift
2426       ConstantPoolCacheEntry::verify_tosBits();
2427       __ cmpl(rcx, ltos);
2428       __ jccb(Assembler::equal, two_word);
2429       __ cmpl(rcx, dtos);
2430       __ jccb(Assembler::equal, two_word);
2431       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2432       __ jmpb(valsize_known);
2433 
2434       __ bind(two_word);
2435       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2436 
2437       __ bind(valsize_known);
2438       // setup object pointer
2439       __ movptr(rbx, Address(rbx, 0));
2440     }
2441     // cache entry pointer
2442     __ addptr(rax, in_bytes(cp_base_offset));
2443     __ shll(rdx, LogBytesPerWord);
2444     __ addptr(rax, rdx);
2445     // object (tos)
2446     __ mov(rcx, rsp);
2447     // rbx,: object pointer set up above (NULL if static)
2448     // rax,: cache entry pointer
2449     // rcx: jvalue object on the stack
2450     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2451                rbx, rax, rcx);
2452     __ get_cache_and_index_at_bcp(cache, index, 1);
2453     __ bind(L1);
2454   }
2455 }
2456 
2457 
2458 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2459   transition(vtos, vtos);
2460 
2461   const Register cache = rcx;
2462   const Register index = rdx;
2463   const Register obj   = rcx;
2464   const Register off   = rbx;
2465   const Register flags = rax;
2466 


2501   __ cmpl(flags, itos );
2502   __ jcc(Assembler::notEqual, notInt);
2503 
2504   __ pop(itos);
2505   if (!is_static) pop_and_check_object(obj);
2506 
2507   __ movl(lo, rax );
2508   if (!is_static) {
2509     patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2510   }
2511   __ jmp(Done);
2512 
2513   __ bind(notInt);
2514   // atos
2515   __ cmpl(flags, atos );
2516   __ jcc(Assembler::notEqual, notObj);
2517 
2518   __ pop(atos);
2519   if (!is_static) pop_and_check_object(obj);
2520 
2521   do_oop_store(_masm, lo, rax, _bs->kind(), false);
2522 
2523   if (!is_static) {
2524     patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2525   }
2526 
2527   __ jmp(Done);
2528 
2529   __ bind(notObj);
2530   // ctos
2531   __ cmpl(flags, ctos );
2532   __ jcc(Assembler::notEqual, notChar);
2533 
2534   __ pop(ctos);
2535   if (!is_static) pop_and_check_object(obj);
2536   __ movw(lo, rax );
2537   if (!is_static) {
2538     patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2539   }
2540   __ jmp(Done);
2541 
2542   __ bind(notChar);
2543   // stos
2544   __ cmpl(flags, stos );
2545   __ jcc(Assembler::notEqual, notShort);
2546 


2548   if (!is_static) pop_and_check_object(obj);
2549   __ movw(lo, rax );
2550   if (!is_static) {
2551     patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2552   }
2553   __ jmp(Done);
2554 
2555   __ bind(notShort);
2556   // ltos
2557   __ cmpl(flags, ltos );
2558   __ jcc(Assembler::notEqual, notLong);
2559 
2560   Label notVolatileLong;
2561   __ testl(rdx, rdx);
2562   __ jcc(Assembler::zero, notVolatileLong);
2563 
2564   __ pop(ltos);  // overwrites rdx, do this after testing volatile.
2565   if (!is_static) pop_and_check_object(obj);
2566 
2567   // Replace with real volatile test
2568   __ push(rdx);
2569   __ push(rax);                 // Must update atomically with FIST
2570   __ fild_d(Address(rsp,0));    // So load into FPU register
2571   __ fistp_d(lo);               // and put into memory atomically
2572   __ addptr(rsp, 2*wordSize);
2573   // volatile_barrier();
2574   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2575                                                Assembler::StoreStore));
2576   // Don't rewrite volatile version
2577   __ jmp(notVolatile);
2578 
2579   __ bind(notVolatileLong);
2580 
2581   __ pop(ltos);  // overwrites rdx
2582   if (!is_static) pop_and_check_object(obj);
2583   NOT_LP64(__ movptr(hi, rdx));
2584   __ movptr(lo, rax);
2585   if (!is_static) {
2586     patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2587   }
2588   __ jmp(notVolatile);
2589 
2590   __ bind(notLong);
2591   // ftos
2592   __ cmpl(flags, ftos );
2593   __ jcc(Assembler::notEqual, notFloat);
2594 
2595   __ pop(ftos);
2596   if (!is_static) pop_and_check_object(obj);
2597   __ fstp_s(lo);
2598   if (!is_static) {
2599     patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2600   }
2601   __ jmp(Done);
2602 
2603   __ bind(notFloat);
2604   // dtos
2605   __ cmpl(flags, dtos );
2606   __ jcc(Assembler::notEqual, notDouble);
2607 
2608   __ pop(dtos);
2609   if (!is_static) pop_and_check_object(obj);
2610   __ fstp_d(lo);
2611   if (!is_static) {
2612     patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2613   }
2614   __ jmp(Done);
2615 
2616   __ bind(notDouble);
2617 
2618   __ stop("Bad state");
2619 
2620   __ bind(Done);
2621 
2622   // Check for volatile store
2623   __ testl(rdx, rdx);
2624   __ jcc(Assembler::zero, notVolatile);
2625   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2626                                                Assembler::StoreStore));
2627   __ bind(notVolatile);
2628 }
2629 
2630 
2631 void TemplateTable::putfield(int byte_no) {
2632   putfield_or_static(byte_no, false);
2633 }
2634 
2635 
2636 void TemplateTable::putstatic(int byte_no) {
2637   putfield_or_static(byte_no, true);
2638 }
2639 
2640 void TemplateTable::jvmti_post_fast_field_mod() {
2641   if (JvmtiExport::can_post_field_modification()) {
2642     // Check to see if a field modification watch has been set before we take
2643     // the time to call into the VM.
2644     Label L2;
2645     __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2646     __ testl(rcx,rcx);
2647     __ jcc(Assembler::zero, L2);
2648     __ pop_ptr(rbx);               // copy the object pointer from tos
2649     __ verify_oop(rbx);
2650     __ push_ptr(rbx);              // put the object pointer back on tos
2651     __ subptr(rsp, sizeof(jvalue));  // add space for a jvalue object
2652     __ mov(rcx, rsp);
2653     __ push_ptr(rbx);                 // save object pointer so we can steal rbx,
2654     __ xorptr(rbx, rbx);
2655     const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2656     const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2657     switch (bytecode()) {          // load values into the jvalue object
2658     case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2659     case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2660     case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2661     case Bytecodes::_fast_iputfield: __ movl(lo_value, rax);                         break;
2662     case Bytecodes::_fast_lputfield:
2663       NOT_LP64(__ movptr(hi_value, rdx));
2664       __ movptr(lo_value, rax);
2665       break;
2666 
2667     // need to call fld_s() after fstp_s() to restore the value for below
2668     case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value);        break;
2669 
2670     // need to call fld_d() after fstp_d() to restore the value for below
2671     case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value);        break;
2672 
2673     // since rcx is not an object we don't call store_check() here
2674     case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax);                       break;
2675 
2676     default:  ShouldNotReachHere();
2677     }
2678     __ pop_ptr(rbx);  // restore copy of object pointer
2679 
2680     // Save rax, and sometimes rdx because call_VM() will clobber them,
2681     // then use them for JVM/DI purposes
2682     __ push(rax);
2683     if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2684     // access constant pool cache entry
2685     __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2686     __ verify_oop(rbx);
2687     // rbx,: object pointer copied above
2688     // rax,: cache entry pointer
2689     // rcx: jvalue object on the stack
2690     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2691     if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);  // restore high value
2692     __ pop(rax);     // restore lower value
2693     __ addptr(rsp, sizeof(jvalue));  // release jvalue object space
2694     __ bind(L2);
2695   }
2696 }
2697 
2698 void TemplateTable::fast_storefield(TosState state) {
2699   transition(state, vtos);
2700 
2701   ByteSize base = constantPoolCacheOopDesc::base_offset();
2702 
2703   jvmti_post_fast_field_mod();
2704 
2705   // access constant pool cache
2706   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2707 
2708   // test for volatile with rdx but rdx is tos register for lputfield.
2709   if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2710   __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2711                        ConstantPoolCacheEntry::flags_offset())));
2712 
2713   // replace index with field offset from cache entry
2714   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2715 
2716   // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2717   // volatile_barrier( );
2718 
2719   Label notVolatile, Done;
2720   __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2721   __ andl(rdx, 0x1);
2722   // Check for volatile store
2723   __ testl(rdx, rdx);
2724   __ jcc(Assembler::zero, notVolatile);
2725 
2726   if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2727 
2728   // Get object from stack
2729   pop_and_check_object(rcx);
2730 
2731   // field addresses
2732   const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2733   const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2734 
2735   // access field
2736   switch (bytecode()) {
2737     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2738     case Bytecodes::_fast_sputfield: // fall through
2739     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2740     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2741     case Bytecodes::_fast_lputfield:
2742       NOT_LP64(__ movptr(hi, rdx));
2743       __ movptr(lo, rax);
2744       break;
2745     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2746     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2747     case Bytecodes::_fast_aputfield: {
2748       do_oop_store(_masm, lo, rax, _bs->kind(), false);
2749       break;
2750     }
2751     default:
2752       ShouldNotReachHere();
2753   }
2754 
2755   Label done;
2756   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2757                                                Assembler::StoreStore));
2758   // Barriers are so large that short branch doesn't reach!
2759   __ jmp(done);
2760 
2761   // Same code as above, but don't need rdx to test for volatile.
2762   __ bind(notVolatile);
2763 
2764   if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2765 
2766   // Get object from stack
2767   pop_and_check_object(rcx);
2768 
2769   // access field
2770   switch (bytecode()) {
2771     case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2772     case Bytecodes::_fast_sputfield: // fall through
2773     case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2774     case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2775     case Bytecodes::_fast_lputfield:
2776       NOT_LP64(__ movptr(hi, rdx));
2777       __ movptr(lo, rax);
2778       break;
2779     case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2780     case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2781     case Bytecodes::_fast_aputfield: {
2782       do_oop_store(_masm, lo, rax, _bs->kind(), false);
2783       break;
2784     }
2785     default:
2786       ShouldNotReachHere();
2787   }
2788   __ bind(done);
2789 }
2790 
2791 
2792 void TemplateTable::fast_accessfield(TosState state) {
2793   transition(atos, state);
2794 
2795   // do the JVMTI work here to avoid disturbing the register state below
2796   if (JvmtiExport::can_post_field_access()) {
2797     // Check to see if a field access watch has been set before we take
2798     // the time to call into the VM.
2799     Label L1;
2800     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2801     __ testl(rcx,rcx);
2802     __ jcc(Assembler::zero, L1);
2803     // access constant pool cache entry
2804     __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2805     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
2806     __ verify_oop(rax);
2807     // rax,: object pointer copied above
2808     // rcx: cache entry pointer
2809     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2810     __ pop_ptr(rax);   // restore object pointer
2811     __ bind(L1);
2812   }
2813 
2814   // access constant pool cache
2815   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2816   // replace index with field offset from cache entry
2817   __ movptr(rbx, Address(rcx,
2818                          rbx,
2819                          Address::times_ptr,
2820                          in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2821 
2822 
2823   // rax,: object
2824   __ verify_oop(rax);
2825   __ null_check(rax);
2826   // field addresses
2827   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2828   const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2829 
2830   // access field
2831   switch (bytecode()) {
2832     case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo );                 break;
2833     case Bytecodes::_fast_sgetfield: __ load_signed_word(rax, lo );       break;
2834     case Bytecodes::_fast_cgetfield: __ load_unsigned_word(rax, lo );     break;
2835     case Bytecodes::_fast_igetfield: __ movl(rax, lo);                    break;
2836     case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten");  break;
2837     case Bytecodes::_fast_fgetfield: __ fld_s(lo);                        break;
2838     case Bytecodes::_fast_dgetfield: __ fld_d(lo);                        break;
2839     case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2840     default:
2841       ShouldNotReachHere();
2842   }
2843 
2844   // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2845   // volatile_barrier( );
2846 }
2847 
2848 void TemplateTable::fast_xaccess(TosState state) {
2849   transition(vtos, state);
2850   // get receiver
2851   __ movptr(rax, aaddress(0));
2852   debug_only(__ verify_local_tag(frame::TagReference, 0));
2853   // access constant pool cache
2854   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2855   __ movptr(rbx, Address(rcx,
2856                          rdx,
2857                          Address::times_ptr,
2858                          in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2859   // make sure exception is reported in correct bcp range (getfield is next instruction)
2860   __ increment(rsi);
2861   __ null_check(rax);
2862   const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2863   if (state == itos) {
2864     __ movl(rax, lo);
2865   } else if (state == atos) {
2866     __ movptr(rax, lo);
2867     __ verify_oop(rax);
2868   } else if (state == ftos) {
2869     __ fld_s(lo);
2870   } else {
2871     ShouldNotReachHere();
2872   }
2873   __ decrement(rsi);
2874 }
2875 
2876 
2877 
2878 //----------------------------------------------------------------------------------------------------
2879 // Calls
2880 
2881 void TemplateTable::count_calls(Register method, Register temp) {
2882   // implemented elsewhere
2883   ShouldNotReachHere();
2884 }
2885 
2886 


2890   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2891   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2892   const bool load_receiver       = code != Bytecodes::_invokestatic;
2893   const bool receiver_null_check = is_invokespecial;
2894   const bool save_flags = is_invokeinterface || is_invokevirtual;
2895   // setup registers & access constant pool cache
2896   const Register recv   = rcx;
2897   const Register flags  = rdx;
2898   assert_different_registers(method, index, recv, flags);
2899 
2900   // save 'interpreter return address'
2901   __ save_bcp();
2902 
2903   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
2904 
2905   // load receiver if needed (note: no return address pushed yet)
2906   if (load_receiver) {
2907     __ movl(recv, flags);
2908     __ andl(recv, 0xFF);
2909     // recv count is 0 based?
2910     __ movptr(recv, Address(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1)));
2911     __ verify_oop(recv);
2912   }
2913 
2914   // do null check if needed
2915   if (receiver_null_check) {
2916     __ null_check(recv);
2917   }
2918 
2919   if (save_flags) {
2920     __ mov(rsi, flags);
2921   }
2922 
2923   // compute return type
2924   __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2925   // Make sure we don't need to mask flags for tosBits after the above shift
2926   ConstantPoolCacheEntry::verify_tosBits();
2927   // load return address
2928   {
2929     ExternalAddress table(is_invokeinterface ? (address)Interpreter::return_5_addrs_by_index_table() :
2930                                                (address)Interpreter::return_3_addrs_by_index_table());
2931     __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));

2932   }
2933 
2934   // push return address
2935   __ push(flags);
2936 
2937   // Restore flag value from the constant pool cache, and restore rsi
2938   // for later null checks.  rsi is the bytecode pointer
2939   if (save_flags) {
2940     __ mov(flags, rsi);
2941     __ restore_bcp();
2942   }
2943 }
2944 
2945 
2946 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2947                         Register flags) {
2948 
2949   // Uses temporary registers rax, rdx
2950   assert_different_registers(index, recv, rax, rdx);
2951 
2952   // Test for an invoke of a final method
2953   Label notFinal;
2954   __ movl(rax, flags);
2955   __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2956   __ jcc(Assembler::zero, notFinal);
2957 
2958   Register method = index;  // method must be rbx,
2959   assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2960 
2961   // do the call - the index is actually the method to call
2962   __ verify_oop(method);
2963 
2964   // It's final, need a null check here!
2965   __ null_check(recv);
2966 
2967   // profile this call
2968   __ profile_final_call(rax);
2969 
2970   __ jump_from_interpreted(method, rax);
2971 
2972   __ bind(notFinal);
2973 
2974   // get receiver klass
2975   __ null_check(recv, oopDesc::klass_offset_in_bytes());
2976   // Keep recv in rcx for callee expects it there
2977   __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2978   __ verify_oop(rax);
2979 
2980   // profile this call
2981   __ profile_virtual_call(rax, rdi, rdx);
2982 
2983   // get target methodOop & entry point
2984   const int base = instanceKlass::vtable_start_offset() * wordSize;
2985   assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2986   __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2987   __ jump_from_interpreted(method, rdx);
2988 }
2989 
2990 
2991 void TemplateTable::invokevirtual(int byte_no) {
2992   transition(vtos, vtos);
2993   prepare_invoke(rbx, noreg, byte_no, bytecode());
2994 
2995   // rbx,: index
2996   // rcx: receiver
2997   // rdx: flags
2998 
2999   invokevirtual_helper(rbx, rcx, rdx);
3000 }
3001 
3002 
3003 void TemplateTable::invokespecial(int byte_no) {
3004   transition(vtos, vtos);
3005   prepare_invoke(rbx, noreg, byte_no, bytecode());
3006   // do the call


3032 
3033   // rax,: Interface
3034   // rbx,: index
3035   // rcx: receiver
3036   // rdx: flags
3037 
3038   // Special case of invokeinterface called for virtual method of
3039   // java.lang.Object.  See cpCacheOop.cpp for details.
3040   // This code isn't produced by javac, but could be produced by
3041   // another compliant java compiler.
3042   Label notMethod;
3043   __ movl(rdi, rdx);
3044   __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
3045   __ jcc(Assembler::zero, notMethod);
3046 
3047   invokevirtual_helper(rbx, rcx, rdx);
3048   __ bind(notMethod);
3049 
3050   // Get receiver klass into rdx - also a null check
3051   __ restore_locals();  // restore rdi
3052   __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
3053   __ verify_oop(rdx);
3054 
3055   // profile this call
3056   __ profile_virtual_call(rdx, rsi, rdi);
3057 
3058   __ mov(rdi, rdx); // Save klassOop in rdi
3059 
3060   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
3061   const int base = instanceKlass::vtable_start_offset() * wordSize;
3062   assert(vtableEntry::size() * wordSize == (1 << (int)Address::times_ptr), "adjust the scaling in the code below");
3063   __ movl(rsi, Address(rdx, instanceKlass::vtable_length_offset() * wordSize)); // Get length of vtable
3064   __ lea(rdx, Address(rdx, rsi, Address::times_4, base));
3065   if (HeapWordsPerLong > 1) {
3066     // Round up to align_object_offset boundary
3067     __ round_to(rdx, BytesPerLong);
3068   }
3069 
3070   Label entry, search, interface_ok;
3071 
3072   __ jmpb(entry);
3073   __ bind(search);
3074   __ addptr(rdx, itableOffsetEntry::size() * wordSize);
3075 
3076   __ bind(entry);
3077 
3078   // Check that the entry is non-null.  A null entry means that the receiver
3079   // class doesn't implement the interface, and wasn't the same as the
3080   // receiver class checked when the interface was resolved.
3081   __ push(rdx);
3082   __ movptr(rdx, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
3083   __ testptr(rdx, rdx);
3084   __ jcc(Assembler::notZero, interface_ok);
3085   // throw exception
3086   __ pop(rdx);           // pop saved register first.
3087   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3088   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
3089   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3090   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3091                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3092   // the call_VM checks for exception, so we should never return here.
3093   __ should_not_reach_here();
3094   __ bind(interface_ok);
3095 
3096     __ pop(rdx);
3097 
3098     __ cmpptr(rax, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
3099     __ jcc(Assembler::notEqual, search);
3100 
3101     __ movl(rdx, Address(rdx, itableOffsetEntry::offset_offset_in_bytes()));
3102     __ addptr(rdx, rdi); // Add offset to klassOop
3103     assert(itableMethodEntry::size() * wordSize == (1 << (int)Address::times_ptr), "adjust the scaling in the code below");
3104     __ movptr(rbx, Address(rdx, rbx, Address::times_ptr));
3105     // rbx,: methodOop to call
3106     // rcx: receiver
3107     // Check for abstract method error
3108     // Note: This should be done more efficiently via a throw_abstract_method_error
3109     //       interpreter entry point and a conditional jump to it in case of a null
3110     //       method.
3111     { Label L;
3112       __ testptr(rbx, rbx);
3113       __ jcc(Assembler::notZero, L);
3114       // throw exception
3115           // note: must restore interpreter registers to canonical
3116           //       state for exception handling to work correctly!
3117           __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3118           __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
3119           __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3120       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3121       // the call_VM checks for exception, so we should never return here.
3122       __ should_not_reach_here();
3123       __ bind(L);
3124     }
3125 
3126   // do the call
3127   // rcx: receiver
3128   // rbx,: methodOop
3129   __ jump_from_interpreted(rbx, rdx);
3130 }
3131 
3132 //----------------------------------------------------------------------------------------------------
3133 // Allocation
3134 
3135 void TemplateTable::_new() {
3136   transition(vtos, atos);
3137   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3138   Label slow_case;
3139   Label done;
3140   Label initialize_header;
3141   Label initialize_object;  // including clearing the fields
3142   Label allocate_shared;
3143 


3144   __ get_cpool_and_tags(rcx, rax);
3145   // get instanceKlass
3146   __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3147   __ push(rcx);  // save the contexts of klass for initializing the header
3148 
3149   // make sure the class we're about to instantiate has been resolved.
3150   // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
3151   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3152   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3153   __ jcc(Assembler::notEqual, slow_case);
3154 
3155   // make sure klass is initialized & doesn't have finalizer
3156   // make sure klass is fully initialized
3157   __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3158   __ jcc(Assembler::notEqual, slow_case);
3159 
3160   // get instance_size in instanceKlass (scaled to a count of bytes)
3161   __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3162   // test to see if it has a finalizer or is malformed in some way
3163   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3164   __ jcc(Assembler::notZero, slow_case);
3165 
3166   //
3167   // Allocate the instance
3168   // 1) Try to allocate in the TLAB
3169   // 2) if fail and the object is large allocate in the shared Eden
3170   // 3) if the above fails (or is not applicable), go to a slow case
3171   // (creates a new TLAB, etc.)
3172 
3173   const bool allow_shared_alloc =
3174     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3175 
3176   if (UseTLAB) {
3177     const Register thread = rcx;
3178 
3179     __ get_thread(thread);
3180     __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3181     __ lea(rbx, Address(rax, rdx, Address::times_1));
3182     __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3183     __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3184     __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3185     if (ZeroTLAB) {
3186       // the fields have been already cleared
3187       __ jmp(initialize_header);
3188     } else {
3189       // initialize both the header and fields
3190       __ jmp(initialize_object);
3191     }
3192   }
3193 
3194   // Allocation in the shared Eden, if allowed.
3195   //
3196   // rdx: instance size in bytes
3197   if (allow_shared_alloc) {
3198     __ bind(allocate_shared);
3199 
3200     ExternalAddress heap_top((address)Universe::heap()->top_addr());
3201 
3202     Label retry;
3203     __ bind(retry);
3204     __ movptr(rax, heap_top);
3205     __ lea(rbx, Address(rax, rdx, Address::times_1));
3206     __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3207     __ jcc(Assembler::above, slow_case);
3208 
3209     // Compare rax, with the top addr, and if still equal, store the new
3210     // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3211     // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3212     //
3213     // rax,: object begin
3214     // rbx,: object end
3215     // rdx: instance size in bytes
3216     __ locked_cmpxchgptr(rbx, heap_top);

3217 
3218     // if someone beat us on the allocation, try again, otherwise continue
3219     __ jcc(Assembler::notEqual, retry);
3220   }
3221 
3222   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3223     // The object is initialized before the header.  If the object size is
3224     // zero, go directly to the header initialization.
3225     __ bind(initialize_object);
3226     __ decrement(rdx, sizeof(oopDesc));
3227     __ jcc(Assembler::zero, initialize_header);
3228 
3229   // Initialize topmost object field, divide rdx by 8, check if odd and
3230   // test if zero.
3231     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
3232     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3233 
3234   // rdx must have been multiple of 8
3235 #ifdef ASSERT
3236     // make sure rdx was multiple of 8
3237     Label L;
3238     // Ignore partial flag stall after shrl() since it is debug VM
3239     __ jccb(Assembler::carryClear, L);
3240     __ stop("object size is not multiple of 2 - adjust this code");
3241     __ bind(L);
3242     // rdx must be > 0, no extra check needed here
3243 #endif
3244 
3245     // initialize remaining object fields: rdx was a multiple of 8
3246     { Label loop;
3247     __ bind(loop);
3248     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3249     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3250     __ decrement(rdx);
3251     __ jcc(Assembler::notZero, loop);
3252     }
3253 
3254     // initialize object header only.
3255     __ bind(initialize_header);
3256     if (UseBiasedLocking) {
3257       __ pop(rcx);   // get saved klass back in the register.
3258       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3259       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3260     } else {
3261       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3262                 (int32_t)markOopDesc::prototype()); // header
3263       __ pop(rcx);   // get saved klass back in the register.
3264     }
3265     __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx);  // klass
3266 
3267     {
3268       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3269       // Trigger dtrace event for fastpath
3270       __ push(atos);
3271       __ call_VM_leaf(
3272            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3273       __ pop(atos);
3274     }
3275 
3276     __ jmp(done);
3277   }
3278 
3279   // slow case
3280   __ bind(slow_case);
3281   __ pop(rcx);   // restore stack pointer to what it was when we came in.
3282   __ get_constant_pool(rax);
3283   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3284   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3285 
3286   // continue
3287   __ bind(done);
3288 }
3289 
3290 
3291 void TemplateTable::newarray() {
3292   transition(itos, atos);
3293   __ push_i(rax);                                 // make sure everything is on the stack
3294   __ load_unsigned_byte(rdx, at_bcp(1));
3295   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3296   __ pop_i(rdx);                                  // discard size
3297 }
3298 
3299 
3300 void TemplateTable::anewarray() {
3301   transition(itos, atos);
3302   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3303   __ get_constant_pool(rcx);
3304   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3305 }
3306 
3307 
3308 void TemplateTable::arraylength() {
3309   transition(atos, itos);
3310   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3311   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3312 }
3313 
3314 
3315 void TemplateTable::checkcast() {
3316   transition(atos, atos);
3317   Label done, is_null, ok_is_subtype, quicked, resolved;
3318   __ testptr(rax, rax);   // Object is in EAX
3319   __ jcc(Assembler::zero, is_null);
3320 
3321   // Get cpool & tags index
3322   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3323   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3324   // See if bytecode has already been quicked
3325   __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3326   __ jcc(Assembler::equal, quicked);
3327 
3328   __ push(atos);
3329   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3330   __ pop_ptr(rdx);
3331   __ jmpb(resolved);
3332 
3333   // Get superklass in EAX and subklass in EBX
3334   __ bind(quicked);
3335   __ mov(rdx, rax);          // Save object in EDX; EAX needed for subtype check
3336   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3337 
3338   __ bind(resolved);
3339   __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3340 
3341   // Generate subtype check.  Blows ECX.  Resets EDI.  Object in EDX.
3342   // Superklass in EAX.  Subklass in EBX.
3343   __ gen_subtype_check( rbx, ok_is_subtype );
3344 
3345   // Come here on failure
3346   __ push(rdx);
3347   // object is at TOS
3348   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3349 
3350   // Come here on success
3351   __ bind(ok_is_subtype);
3352   __ mov(rax,rdx);           // Restore object in EDX
3353 
3354   // Collect counts on whether this check-cast sees NULLs a lot or not.
3355   if (ProfileInterpreter) {
3356     __ jmp(done);
3357     __ bind(is_null);
3358     __ profile_null_seen(rcx);
3359   } else {
3360     __ bind(is_null);   // same as 'done'
3361   }
3362   __ bind(done);
3363 }
3364 
3365 
3366 void TemplateTable::instanceof() {
3367   transition(atos, itos);
3368   Label done, is_null, ok_is_subtype, quicked, resolved;
3369   __ testptr(rax, rax);
3370   __ jcc(Assembler::zero, is_null);
3371 
3372   // Get cpool & tags index
3373   __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3374   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3375   // See if bytecode has already been quicked
3376   __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3377   __ jcc(Assembler::equal, quicked);
3378 
3379   __ push(atos);
3380   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3381   __ pop_ptr(rdx);
3382   __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3383   __ jmp(resolved);
3384 
3385   // Get superklass in EAX and subklass in EDX
3386   __ bind(quicked);
3387   __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3388   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3389 
3390   __ bind(resolved);
3391 
3392   // Generate subtype check.  Blows ECX.  Resets EDI.
3393   // Superklass in EAX.  Subklass in EDX.
3394   __ gen_subtype_check( rdx, ok_is_subtype );
3395 
3396   // Come here on failure
3397   __ xorl(rax,rax);
3398   __ jmpb(done);
3399   // Come here on success
3400   __ bind(ok_is_subtype);
3401   __ movl(rax, 1);
3402 
3403   // Collect counts on whether this test sees NULLs a lot or not.
3404   if (ProfileInterpreter) {
3405     __ jmp(done);
3406     __ bind(is_null);
3407     __ profile_null_seen(rcx);
3408   } else {


3410   }
3411   __ bind(done);
3412   // rax, = 0: obj == NULL or  obj is not an instanceof the specified klass
3413   // rax, = 1: obj != NULL and obj is     an instanceof the specified klass
3414 }
3415 
3416 
3417 //----------------------------------------------------------------------------------------------------
3418 // Breakpoints
3419 void TemplateTable::_breakpoint() {
3420 
3421   // Note: We get here even if we are single stepping..
3422   // jbug inists on setting breakpoints at every bytecode
3423   // even if we are in single step mode.
3424 
3425   transition(vtos, vtos);
3426 
3427   // get the unpatched byte code
3428   __ get_method(rcx);
3429   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3430   __ mov(rbx, rax);
3431 
3432   // post the breakpoint event
3433   __ get_method(rcx);
3434   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3435 
3436   // complete the execution of original bytecode
3437   __ dispatch_only_normal(vtos);
3438 }
3439 
3440 
3441 //----------------------------------------------------------------------------------------------------
3442 // Exceptions
3443 
3444 void TemplateTable::athrow() {
3445   transition(atos, vtos);
3446   __ null_check(rax);
3447   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3448 }
3449 
3450 


3466 // ...
3467 // [saved rbp,    ] <--- rbp,
3468 
3469 
3470 void TemplateTable::monitorenter() {
3471   transition(atos, vtos);
3472 
3473   // check for NULL object
3474   __ null_check(rax);
3475 
3476   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3477   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3478   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3479   Label allocated;
3480 
3481   // initialize entry pointer
3482   __ xorl(rdx, rdx);                             // points to free slot or NULL
3483 
3484   // find a free slot in the monitor block (result in rdx)
3485   { Label entry, loop, exit;
3486     __ movptr(rcx, monitor_block_top);            // points to current entry, starting with top-most entry
3487     __ lea(rbx, monitor_block_bot);               // points to word before bottom of monitor block
3488     __ jmpb(entry);
3489 
3490     __ bind(loop);
3491     __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);  // check if current entry is used
3492 
3493 // TODO - need new func here - kbt
3494     if (VM_Version::supports_cmov()) {
3495       __ cmov(Assembler::equal, rdx, rcx);       // if not used then remember entry in rdx
3496     } else {
3497       Label L;
3498       __ jccb(Assembler::notEqual, L);
3499       __ mov(rdx, rcx);                          // if not used then remember entry in rdx
3500       __ bind(L);
3501     }
3502     __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3503     __ jccb(Assembler::equal, exit);             // if same object then stop searching
3504     __ addptr(rcx, entry_size);                  // otherwise advance to next entry
3505     __ bind(entry);
3506     __ cmpptr(rcx, rbx);                         // check if bottom reached
3507     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3508     __ bind(exit);
3509   }
3510 
3511   __ testptr(rdx, rdx);                          // check if a slot has been found
3512   __ jccb(Assembler::notZero, allocated);        // if found, continue with that one
3513 
3514   // allocate one if there's no free slot
3515   { Label entry, loop;
3516     // 1. compute new pointers                   // rsp: old expression stack top
3517     __ movptr(rdx, monitor_block_bot);           // rdx: old expression stack bottom
3518     __ subptr(rsp, entry_size);                  // move expression stack top
3519     __ subptr(rdx, entry_size);                  // move expression stack bottom
3520     __ mov(rcx, rsp);                            // set start value for copy loop
3521     __ movptr(monitor_block_bot, rdx);           // set new monitor block top
3522     __ jmp(entry);
3523     // 2. move expression stack contents
3524     __ bind(loop);
3525     __ movptr(rbx, Address(rcx, entry_size));    // load expression stack word from old location
3526     __ movptr(Address(rcx, 0), rbx);             // and store it at new location
3527     __ addptr(rcx, wordSize);                    // advance to next word
3528     __ bind(entry);
3529     __ cmpptr(rcx, rdx);                         // check if bottom reached
3530     __ jcc(Assembler::notEqual, loop);           // if not at bottom then copy next word
3531   }
3532 
3533   // call run-time routine
3534   // rdx: points to monitor entry
3535   __ bind(allocated);
3536 
3537   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3538   // The object has already been poped from the stack, so the expression stack looks correct.
3539   __ increment(rsi);
3540 
3541   __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax);     // store object
3542   __ lock_object(rdx);
3543 
3544   // check to make sure this monitor doesn't cause stack overflow after locking
3545   __ save_bcp();  // in case of exception
3546   __ generate_stack_overflow_check(0);
3547 
3548   // The bcp has already been incremented. Just need to dispatch to next instruction.
3549   __ dispatch_next(vtos);
3550 }
3551 
3552 
3553 void TemplateTable::monitorexit() {
3554   transition(atos, vtos);
3555 
3556   // check for NULL object
3557   __ null_check(rax);
3558 
3559   const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3560   const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
3561   const int entry_size =         (     frame::interpreter_frame_monitor_size()           * wordSize);
3562   Label found;
3563 
3564   // find matching slot
3565   { Label entry, loop;
3566     __ movptr(rdx, monitor_block_top);           // points to current entry, starting with top-most entry
3567     __ lea(rbx, monitor_block_bot);             // points to word before bottom of monitor block
3568     __ jmpb(entry);
3569 
3570     __ bind(loop);
3571     __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));   // check if current entry is for same object
3572     __ jcc(Assembler::equal, found);             // if same object then stop searching
3573     __ addptr(rdx, entry_size);                  // otherwise advance to next entry
3574     __ bind(entry);
3575     __ cmpptr(rdx, rbx);                         // check if bottom reached
3576     __ jcc(Assembler::notEqual, loop);           // if not at bottom then check this entry
3577   }
3578 
3579   // error handling. Unlocking was not block-structured
3580   Label end;
3581   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3582   __ should_not_reach_here();
3583 
3584   // call run-time routine
3585   // rcx: points to monitor entry
3586   __ bind(found);
3587   __ push_ptr(rax);                                 // make sure object is on stack (contract with oopMaps)
3588   __ unlock_object(rdx);
3589   __ pop_ptr(rax);                                  // discard object
3590   __ bind(end);
3591 }
3592 
3593 
3594 //----------------------------------------------------------------------------------------------------
3595 // Wide instructions
3596 
3597 void TemplateTable::wide() {
3598   transition(vtos, vtos);
3599   __ load_unsigned_byte(rbx, at_bcp(1));
3600   ExternalAddress wtable((address)Interpreter::_wentry_point);
3601   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3602   // Note: the rsi increment step is part of the individual wide bytecode implementations
3603 }
3604 
3605 
3606 //----------------------------------------------------------------------------------------------------
3607 // Multi arrays
3608 
3609 void TemplateTable::multianewarray() {
3610   transition(vtos, atos);
3611   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3612   // last dim is on top of stack; we want address of first one:
3613   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3614   // the latter wordSize to point to the beginning of the array.
3615   __ lea(  rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3616   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax);     // pass in rax,
3617   __ load_unsigned_byte(rbx, at_bcp(3));
3618   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
3619 }
3620 
3621 #endif /* !CC_INTERP */