1 /*
   2  * Copyright (c) 1997, 2022, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "compiler/disassembler.hpp"
  28 #include "gc/shared/collectedHeap.hpp"
  29 #include "gc/shared/tlab_globals.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateTable.hpp"
  34 #include "memory/universe.hpp"
  35 #include "oops/methodData.hpp"
  36 #include "oops/objArrayKlass.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "prims/jvmtiExport.hpp"
  39 #include "prims/methodHandles.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/safepointMechanism.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "utilities/macros.hpp"
  46 
  47 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  48 
  49 // Global Register Names
  50 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  51 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  52 
  53 // Address Computation: local variables
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 #ifndef _LP64
  63 static inline Address haddress(int n) {
  64   return iaddress(n + 0);
  65 }
  66 #endif
  67 
  68 static inline Address faddress(int n) {
  69   return iaddress(n);
  70 }
  71 
  72 static inline Address daddress(int n) {
  73   return laddress(n);
  74 }
  75 
  76 static inline Address aaddress(int n) {
  77   return iaddress(n);
  78 }
  79 
  80 static inline Address iaddress(Register r) {
  81   return Address(rlocals, r, Address::times_ptr);
  82 }
  83 
  84 static inline Address laddress(Register r) {
  85   return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
  86 }
  87 
  88 #ifndef _LP64
  89 static inline Address haddress(Register r)       {
  90   return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  91 }
  92 #endif
  93 
  94 static inline Address faddress(Register r) {
  95   return iaddress(r);
  96 }
  97 
  98 static inline Address daddress(Register r) {
  99   return laddress(r);
 100 }
 101 
 102 static inline Address aaddress(Register r) {
 103   return iaddress(r);
 104 }
 105 
 106 
 107 // expression stack
 108 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
 109 // data beyond the rsp which is potentially unsafe in an MT environment;
 110 // an interrupt may overwrite that data.)
 111 static inline Address at_rsp   () {
 112   return Address(rsp, 0);
 113 }
 114 
 115 // At top of Java expression stack which may be different than esp().  It
 116 // isn't for category 1 objects.
 117 static inline Address at_tos   () {
 118   return Address(rsp,  Interpreter::expr_offset_in_bytes(0));
 119 }
 120 
 121 static inline Address at_tos_p1() {
 122   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
 123 }
 124 
 125 static inline Address at_tos_p2() {
 126   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
 127 }
 128 
 129 // Condition conversion
 130 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 131   switch (cc) {
 132   case TemplateTable::equal        : return Assembler::notEqual;
 133   case TemplateTable::not_equal    : return Assembler::equal;
 134   case TemplateTable::less         : return Assembler::greaterEqual;
 135   case TemplateTable::less_equal   : return Assembler::greater;
 136   case TemplateTable::greater      : return Assembler::lessEqual;
 137   case TemplateTable::greater_equal: return Assembler::less;
 138   }
 139   ShouldNotReachHere();
 140   return Assembler::zero;
 141 }
 142 
 143 
 144 
 145 // Miscellaneous helper routines
 146 // Store an oop (or NULL) at the address described by obj.
 147 // If val == noreg this means store a NULL
 148 
 149 
 150 static void do_oop_store(InterpreterMacroAssembler* _masm,
 151                          Address dst,
 152                          Register val,
 153                          DecoratorSet decorators = 0) {
 154   assert(val == noreg || val == rax, "parameter is just for looks");
 155   __ store_heap_oop(dst, val, rdx, rbx, LP64_ONLY(r8) NOT_LP64(rsi), decorators);
 156 }
 157 
 158 static void do_oop_load(InterpreterMacroAssembler* _masm,
 159                         Address src,
 160                         Register dst,
 161                         DecoratorSet decorators = 0) {
 162   __ load_heap_oop(dst, src, rdx, rbx, decorators);
 163 }
 164 
 165 Address TemplateTable::at_bcp(int offset) {
 166   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 167   return Address(rbcp, offset);
 168 }
 169 
 170 
 171 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 172                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 173                                    int byte_no) {
 174   if (!RewriteBytecodes)  return;
 175   Label L_patch_done;
 176 
 177   switch (bc) {
 178   case Bytecodes::_fast_aputfield:
 179   case Bytecodes::_fast_bputfield:
 180   case Bytecodes::_fast_zputfield:
 181   case Bytecodes::_fast_cputfield:
 182   case Bytecodes::_fast_dputfield:
 183   case Bytecodes::_fast_fputfield:
 184   case Bytecodes::_fast_iputfield:
 185   case Bytecodes::_fast_lputfield:
 186   case Bytecodes::_fast_sputfield:
 187     {
 188       // We skip bytecode quickening for putfield instructions when
 189       // the put_code written to the constant pool cache is zero.
 190       // This is required so that every execution of this instruction
 191       // calls out to InterpreterRuntime::resolve_get_put to do
 192       // additional, required work.
 193       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 194       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 195       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 196       __ movl(bc_reg, bc);
 197       __ cmpl(temp_reg, (int) 0);
 198       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 199     }
 200     break;
 201   default:
 202     assert(byte_no == -1, "sanity");
 203     // the pair bytecodes have already done the load.
 204     if (load_bc_into_bc_reg) {
 205       __ movl(bc_reg, bc);
 206     }
 207   }
 208 
 209   if (JvmtiExport::can_post_breakpoint()) {
 210     Label L_fast_patch;
 211     // if a breakpoint is present we can't rewrite the stream directly
 212     __ movzbl(temp_reg, at_bcp(0));
 213     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 214     __ jcc(Assembler::notEqual, L_fast_patch);
 215     __ get_method(temp_reg);
 216     // Let breakpoint table handling rewrite to quicker bytecode
 217     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 218 #ifndef ASSERT
 219     __ jmpb(L_patch_done);
 220 #else
 221     __ jmp(L_patch_done);
 222 #endif
 223     __ bind(L_fast_patch);
 224   }
 225 
 226 #ifdef ASSERT
 227   Label L_okay;
 228   __ load_unsigned_byte(temp_reg, at_bcp(0));
 229   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 230   __ jcc(Assembler::equal, L_okay);
 231   __ cmpl(temp_reg, bc_reg);
 232   __ jcc(Assembler::equal, L_okay);
 233   __ stop("patching the wrong bytecode");
 234   __ bind(L_okay);
 235 #endif
 236 
 237   // patch bytecode
 238   __ movb(at_bcp(0), bc_reg);
 239   __ bind(L_patch_done);
 240 }
 241 // Individual instructions
 242 
 243 
 244 void TemplateTable::nop() {
 245   transition(vtos, vtos);
 246   // nothing to do
 247 }
 248 
 249 void TemplateTable::shouldnotreachhere() {
 250   transition(vtos, vtos);
 251   __ stop("shouldnotreachhere bytecode");
 252 }
 253 
 254 void TemplateTable::aconst_null() {
 255   transition(vtos, atos);
 256   __ xorl(rax, rax);
 257 }
 258 
 259 void TemplateTable::iconst(int value) {
 260   transition(vtos, itos);
 261   if (value == 0) {
 262     __ xorl(rax, rax);
 263   } else {
 264     __ movl(rax, value);
 265   }
 266 }
 267 
 268 void TemplateTable::lconst(int value) {
 269   transition(vtos, ltos);
 270   if (value == 0) {
 271     __ xorl(rax, rax);
 272   } else {
 273     __ movl(rax, value);
 274   }
 275 #ifndef _LP64
 276   assert(value >= 0, "check this code");
 277   __ xorptr(rdx, rdx);
 278 #endif
 279 }
 280 
 281 
 282 
 283 void TemplateTable::fconst(int value) {
 284   transition(vtos, ftos);
 285   if (UseSSE >= 1) {
 286     static float one = 1.0f, two = 2.0f;
 287     switch (value) {
 288     case 0:
 289       __ xorps(xmm0, xmm0);
 290       break;
 291     case 1:
 292       __ movflt(xmm0, ExternalAddress((address) &one), rscratch1);
 293       break;
 294     case 2:
 295       __ movflt(xmm0, ExternalAddress((address) &two), rscratch1);
 296       break;
 297     default:
 298       ShouldNotReachHere();
 299       break;
 300     }
 301   } else {
 302 #ifdef _LP64
 303     ShouldNotReachHere();
 304 #else
 305            if (value == 0) { __ fldz();
 306     } else if (value == 1) { __ fld1();
 307     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 308     } else                 { ShouldNotReachHere();
 309     }
 310 #endif // _LP64
 311   }
 312 }
 313 
 314 void TemplateTable::dconst(int value) {
 315   transition(vtos, dtos);
 316   if (UseSSE >= 2) {
 317     static double one = 1.0;
 318     switch (value) {
 319     case 0:
 320       __ xorpd(xmm0, xmm0);
 321       break;
 322     case 1:
 323       __ movdbl(xmm0, ExternalAddress((address) &one), rscratch1);
 324       break;
 325     default:
 326       ShouldNotReachHere();
 327       break;
 328     }
 329   } else {
 330 #ifdef _LP64
 331     ShouldNotReachHere();
 332 #else
 333            if (value == 0) { __ fldz();
 334     } else if (value == 1) { __ fld1();
 335     } else                 { ShouldNotReachHere();
 336     }
 337 #endif
 338   }
 339 }
 340 
 341 void TemplateTable::bipush() {
 342   transition(vtos, itos);
 343   __ load_signed_byte(rax, at_bcp(1));
 344 }
 345 
 346 void TemplateTable::sipush() {
 347   transition(vtos, itos);
 348   __ load_unsigned_short(rax, at_bcp(1));
 349   __ bswapl(rax);
 350   __ sarl(rax, 16);
 351 }
 352 
 353 void TemplateTable::ldc(bool wide) {
 354   transition(vtos, vtos);
 355   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 356   Label call_ldc, notFloat, notClass, notInt, Done;
 357 
 358   if (wide) {
 359     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 360   } else {
 361     __ load_unsigned_byte(rbx, at_bcp(1));
 362   }
 363 
 364   __ get_cpool_and_tags(rcx, rax);
 365   const int base_offset = ConstantPool::header_size() * wordSize;
 366   const int tags_offset = Array<u1>::base_offset_in_bytes();
 367 
 368   // get type
 369   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 370 
 371   // unresolved class - get the resolved class
 372   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 373   __ jccb(Assembler::equal, call_ldc);
 374 
 375   // unresolved class in error state - call into runtime to throw the error
 376   // from the first resolution attempt
 377   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 378   __ jccb(Assembler::equal, call_ldc);
 379 
 380   // resolved class - need to call vm to get java mirror of the class
 381   __ cmpl(rdx, JVM_CONSTANT_Class);
 382   __ jcc(Assembler::notEqual, notClass);
 383 
 384   __ bind(call_ldc);
 385 
 386   __ movl(rarg, wide);
 387   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 388 
 389   __ push(atos);
 390   __ jmp(Done);
 391 
 392   __ bind(notClass);
 393   __ cmpl(rdx, JVM_CONSTANT_Float);
 394   __ jccb(Assembler::notEqual, notFloat);
 395 
 396   // ftos
 397   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 398   __ push(ftos);
 399   __ jmp(Done);
 400 
 401   __ bind(notFloat);
 402   __ cmpl(rdx, JVM_CONSTANT_Integer);
 403   __ jccb(Assembler::notEqual, notInt);
 404 
 405   // itos
 406   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 407   __ push(itos);
 408   __ jmp(Done);
 409 
 410   // assume the tag is for condy; if not, the VM runtime will tell us
 411   __ bind(notInt);
 412   condy_helper(Done);
 413 
 414   __ bind(Done);
 415 }
 416 
 417 // Fast path for caching oop constants.
 418 void TemplateTable::fast_aldc(bool wide) {
 419   transition(vtos, atos);
 420 
 421   Register result = rax;
 422   Register tmp = rdx;
 423   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 424   int index_size = wide ? sizeof(u2) : sizeof(u1);
 425 
 426   Label resolved;
 427 
 428   // We are resolved if the resolved reference cache entry contains a
 429   // non-null object (String, MethodType, etc.)
 430   assert_different_registers(result, tmp);
 431   __ get_cache_index_at_bcp(tmp, 1, index_size);
 432   __ load_resolved_reference_at_index(result, tmp);
 433   __ testptr(result, result);
 434   __ jcc(Assembler::notZero, resolved);
 435 
 436   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 437 
 438   // first time invocation - must resolve first
 439   __ movl(rarg, (int)bytecode());
 440   __ call_VM(result, entry, rarg);
 441   __ bind(resolved);
 442 
 443   { // Check for the null sentinel.
 444     // If we just called the VM, it already did the mapping for us,
 445     // but it's harmless to retry.
 446     Label notNull;
 447     ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
 448     __ movptr(tmp, null_sentinel);
 449     __ resolve_oop_handle(tmp, rscratch2);
 450     __ cmpoop(tmp, result);
 451     __ jccb(Assembler::notEqual, notNull);
 452     __ xorptr(result, result);  // NULL object reference
 453     __ bind(notNull);
 454   }
 455 
 456   if (VerifyOops) {
 457     __ verify_oop(result);
 458   }
 459 }
 460 
 461 void TemplateTable::ldc2_w() {
 462   transition(vtos, vtos);
 463   Label notDouble, notLong, Done;
 464   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 465 
 466   __ get_cpool_and_tags(rcx, rax);
 467   const int base_offset = ConstantPool::header_size() * wordSize;
 468   const int tags_offset = Array<u1>::base_offset_in_bytes();
 469 
 470   // get type
 471   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 472   __ cmpl(rdx, JVM_CONSTANT_Double);
 473   __ jccb(Assembler::notEqual, notDouble);
 474 
 475   // dtos
 476   __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
 477   __ push(dtos);
 478 
 479   __ jmp(Done);
 480   __ bind(notDouble);
 481   __ cmpl(rdx, JVM_CONSTANT_Long);
 482   __ jccb(Assembler::notEqual, notLong);
 483 
 484   // ltos
 485   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 486   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 487   __ push(ltos);
 488   __ jmp(Done);
 489 
 490   __ bind(notLong);
 491   condy_helper(Done);
 492 
 493   __ bind(Done);
 494 }
 495 
 496 void TemplateTable::condy_helper(Label& Done) {
 497   const Register obj = rax;
 498   const Register off = rbx;
 499   const Register flags = rcx;
 500   const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 501   __ movl(rarg, (int)bytecode());
 502   call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
 503 #ifndef _LP64
 504   // borrow rdi from locals
 505   __ get_thread(rdi);
 506   __ get_vm_result_2(flags, rdi);
 507   __ restore_locals();
 508 #else
 509   __ get_vm_result_2(flags, r15_thread);
 510 #endif
 511   // VMr = obj = base address to find primitive value to push
 512   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 513   __ movl(off, flags);
 514   __ andl(off, ConstantPoolCacheEntry::field_index_mask);
 515   const Address field(obj, off, Address::times_1, 0*wordSize);
 516 
 517   // What sort of thing are we loading?
 518   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
 519   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
 520 
 521   switch (bytecode()) {
 522   case Bytecodes::_ldc:
 523   case Bytecodes::_ldc_w:
 524     {
 525       // tos in (itos, ftos, stos, btos, ctos, ztos)
 526       Label notInt, notFloat, notShort, notByte, notChar, notBool;
 527       __ cmpl(flags, itos);
 528       __ jccb(Assembler::notEqual, notInt);
 529       // itos
 530       __ movl(rax, field);
 531       __ push(itos);
 532       __ jmp(Done);
 533 
 534       __ bind(notInt);
 535       __ cmpl(flags, ftos);
 536       __ jccb(Assembler::notEqual, notFloat);
 537       // ftos
 538       __ load_float(field);
 539       __ push(ftos);
 540       __ jmp(Done);
 541 
 542       __ bind(notFloat);
 543       __ cmpl(flags, stos);
 544       __ jccb(Assembler::notEqual, notShort);
 545       // stos
 546       __ load_signed_short(rax, field);
 547       __ push(stos);
 548       __ jmp(Done);
 549 
 550       __ bind(notShort);
 551       __ cmpl(flags, btos);
 552       __ jccb(Assembler::notEqual, notByte);
 553       // btos
 554       __ load_signed_byte(rax, field);
 555       __ push(btos);
 556       __ jmp(Done);
 557 
 558       __ bind(notByte);
 559       __ cmpl(flags, ctos);
 560       __ jccb(Assembler::notEqual, notChar);
 561       // ctos
 562       __ load_unsigned_short(rax, field);
 563       __ push(ctos);
 564       __ jmp(Done);
 565 
 566       __ bind(notChar);
 567       __ cmpl(flags, ztos);
 568       __ jccb(Assembler::notEqual, notBool);
 569       // ztos
 570       __ load_signed_byte(rax, field);
 571       __ push(ztos);
 572       __ jmp(Done);
 573 
 574       __ bind(notBool);
 575       break;
 576     }
 577 
 578   case Bytecodes::_ldc2_w:
 579     {
 580       Label notLong, notDouble;
 581       __ cmpl(flags, ltos);
 582       __ jccb(Assembler::notEqual, notLong);
 583       // ltos
 584       // Loading high word first because movptr clobbers rax
 585       NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
 586       __ movptr(rax, field);
 587       __ push(ltos);
 588       __ jmp(Done);
 589 
 590       __ bind(notLong);
 591       __ cmpl(flags, dtos);
 592       __ jccb(Assembler::notEqual, notDouble);
 593       // dtos
 594       __ load_double(field);
 595       __ push(dtos);
 596       __ jmp(Done);
 597 
 598       __ bind(notDouble);
 599       break;
 600     }
 601 
 602   default:
 603     ShouldNotReachHere();
 604   }
 605 
 606   __ stop("bad ldc/condy");
 607 }
 608 
 609 void TemplateTable::locals_index(Register reg, int offset) {
 610   __ load_unsigned_byte(reg, at_bcp(offset));
 611   __ negptr(reg);
 612 }
 613 
 614 void TemplateTable::iload() {
 615   iload_internal();
 616 }
 617 
 618 void TemplateTable::nofast_iload() {
 619   iload_internal(may_not_rewrite);
 620 }
 621 
 622 void TemplateTable::iload_internal(RewriteControl rc) {
 623   transition(vtos, itos);
 624   if (RewriteFrequentPairs && rc == may_rewrite) {
 625     Label rewrite, done;
 626     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 627     LP64_ONLY(assert(rbx != bc, "register damaged"));
 628 
 629     // get next byte
 630     __ load_unsigned_byte(rbx,
 631                           at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 632     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 633     // last two iloads in a pair.  Comparing against fast_iload means that
 634     // the next bytecode is neither an iload or a caload, and therefore
 635     // an iload pair.
 636     __ cmpl(rbx, Bytecodes::_iload);
 637     __ jcc(Assembler::equal, done);
 638 
 639     __ cmpl(rbx, Bytecodes::_fast_iload);
 640     __ movl(bc, Bytecodes::_fast_iload2);
 641 
 642     __ jccb(Assembler::equal, rewrite);
 643 
 644     // if _caload, rewrite to fast_icaload
 645     __ cmpl(rbx, Bytecodes::_caload);
 646     __ movl(bc, Bytecodes::_fast_icaload);
 647     __ jccb(Assembler::equal, rewrite);
 648 
 649     // rewrite so iload doesn't check again.
 650     __ movl(bc, Bytecodes::_fast_iload);
 651 
 652     // rewrite
 653     // bc: fast bytecode
 654     __ bind(rewrite);
 655     patch_bytecode(Bytecodes::_iload, bc, rbx, false);
 656     __ bind(done);
 657   }
 658 
 659   // Get the local value into tos
 660   locals_index(rbx);
 661   __ movl(rax, iaddress(rbx));
 662 }
 663 
 664 void TemplateTable::fast_iload2() {
 665   transition(vtos, itos);
 666   locals_index(rbx);
 667   __ movl(rax, iaddress(rbx));
 668   __ push(itos);
 669   locals_index(rbx, 3);
 670   __ movl(rax, iaddress(rbx));
 671 }
 672 
 673 void TemplateTable::fast_iload() {
 674   transition(vtos, itos);
 675   locals_index(rbx);
 676   __ movl(rax, iaddress(rbx));
 677 }
 678 
 679 void TemplateTable::lload() {
 680   transition(vtos, ltos);
 681   locals_index(rbx);
 682   __ movptr(rax, laddress(rbx));
 683   NOT_LP64(__ movl(rdx, haddress(rbx)));
 684 }
 685 
 686 void TemplateTable::fload() {
 687   transition(vtos, ftos);
 688   locals_index(rbx);
 689   __ load_float(faddress(rbx));
 690 }
 691 
 692 void TemplateTable::dload() {
 693   transition(vtos, dtos);
 694   locals_index(rbx);
 695   __ load_double(daddress(rbx));
 696 }
 697 
 698 void TemplateTable::aload() {
 699   transition(vtos, atos);
 700   locals_index(rbx);
 701   __ movptr(rax, aaddress(rbx));
 702 }
 703 
 704 void TemplateTable::locals_index_wide(Register reg) {
 705   __ load_unsigned_short(reg, at_bcp(2));
 706   __ bswapl(reg);
 707   __ shrl(reg, 16);
 708   __ negptr(reg);
 709 }
 710 
 711 void TemplateTable::wide_iload() {
 712   transition(vtos, itos);
 713   locals_index_wide(rbx);
 714   __ movl(rax, iaddress(rbx));
 715 }
 716 
 717 void TemplateTable::wide_lload() {
 718   transition(vtos, ltos);
 719   locals_index_wide(rbx);
 720   __ movptr(rax, laddress(rbx));
 721   NOT_LP64(__ movl(rdx, haddress(rbx)));
 722 }
 723 
 724 void TemplateTable::wide_fload() {
 725   transition(vtos, ftos);
 726   locals_index_wide(rbx);
 727   __ load_float(faddress(rbx));
 728 }
 729 
 730 void TemplateTable::wide_dload() {
 731   transition(vtos, dtos);
 732   locals_index_wide(rbx);
 733   __ load_double(daddress(rbx));
 734 }
 735 
 736 void TemplateTable::wide_aload() {
 737   transition(vtos, atos);
 738   locals_index_wide(rbx);
 739   __ movptr(rax, aaddress(rbx));
 740 }
 741 
 742 void TemplateTable::index_check(Register array, Register index) {
 743   // Pop ptr into array
 744   __ pop_ptr(array);
 745   index_check_without_pop(array, index);
 746 }
 747 
 748 void TemplateTable::index_check_without_pop(Register array, Register index) {
 749   // destroys rbx
 750   // check array
 751   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 752   // sign extend index for use by indexed load
 753   __ movl2ptr(index, index);
 754   // check index
 755   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 756   if (index != rbx) {
 757     // ??? convention: move aberrant index into rbx for exception message
 758     assert(rbx != array, "different registers");
 759     __ movl(rbx, index);
 760   }
 761   Label skip;
 762   __ jccb(Assembler::below, skip);
 763   // Pass array to create more detailed exceptions.
 764   __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
 765   __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 766   __ bind(skip);
 767 }
 768 
 769 void TemplateTable::iaload() {
 770   transition(itos, itos);
 771   // rax: index
 772   // rdx: array
 773   index_check(rdx, rax); // kills rbx
 774   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
 775                     Address(rdx, rax, Address::times_4,
 776                             arrayOopDesc::base_offset_in_bytes(T_INT)),
 777                     noreg, noreg);
 778 }
 779 
 780 void TemplateTable::laload() {
 781   transition(itos, ltos);
 782   // rax: index
 783   // rdx: array
 784   index_check(rdx, rax); // kills rbx
 785   NOT_LP64(__ mov(rbx, rax));
 786   // rbx,: index
 787   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
 788                     Address(rdx, rbx, Address::times_8,
 789                             arrayOopDesc::base_offset_in_bytes(T_LONG)),
 790                     noreg, noreg);
 791 }
 792 
 793 
 794 
 795 void TemplateTable::faload() {
 796   transition(itos, ftos);
 797   // rax: index
 798   // rdx: array
 799   index_check(rdx, rax); // kills rbx
 800   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
 801                     Address(rdx, rax,
 802                             Address::times_4,
 803                             arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
 804                     noreg, noreg);
 805 }
 806 
 807 void TemplateTable::daload() {
 808   transition(itos, dtos);
 809   // rax: index
 810   // rdx: array
 811   index_check(rdx, rax); // kills rbx
 812   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
 813                     Address(rdx, rax,
 814                             Address::times_8,
 815                             arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
 816                     noreg, noreg);
 817 }
 818 
 819 void TemplateTable::aaload() {
 820   transition(itos, atos);
 821   // rax: index
 822   // rdx: array
 823   index_check(rdx, rax); // kills rbx
 824   do_oop_load(_masm,
 825               Address(rdx, rax,
 826                       UseCompressedOops ? Address::times_4 : Address::times_ptr,
 827                       arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
 828               rax,
 829               IS_ARRAY);
 830 }
 831 
 832 void TemplateTable::baload() {
 833   transition(itos, itos);
 834   // rax: index
 835   // rdx: array
 836   index_check(rdx, rax); // kills rbx
 837   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
 838                     Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
 839                     noreg, noreg);
 840 }
 841 
 842 void TemplateTable::caload() {
 843   transition(itos, itos);
 844   // rax: index
 845   // rdx: array
 846   index_check(rdx, rax); // kills rbx
 847   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 848                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 849                     noreg, noreg);
 850 }
 851 
 852 // iload followed by caload frequent pair
 853 void TemplateTable::fast_icaload() {
 854   transition(vtos, itos);
 855   // load index out of locals
 856   locals_index(rbx);
 857   __ movl(rax, iaddress(rbx));
 858 
 859   // rax: index
 860   // rdx: array
 861   index_check(rdx, rax); // kills rbx
 862   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 863                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 864                     noreg, noreg);
 865 }
 866 
 867 
 868 void TemplateTable::saload() {
 869   transition(itos, itos);
 870   // rax: index
 871   // rdx: array
 872   index_check(rdx, rax); // kills rbx
 873   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
 874                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
 875                     noreg, noreg);
 876 }
 877 
 878 void TemplateTable::iload(int n) {
 879   transition(vtos, itos);
 880   __ movl(rax, iaddress(n));
 881 }
 882 
 883 void TemplateTable::lload(int n) {
 884   transition(vtos, ltos);
 885   __ movptr(rax, laddress(n));
 886   NOT_LP64(__ movptr(rdx, haddress(n)));
 887 }
 888 
 889 void TemplateTable::fload(int n) {
 890   transition(vtos, ftos);
 891   __ load_float(faddress(n));
 892 }
 893 
 894 void TemplateTable::dload(int n) {
 895   transition(vtos, dtos);
 896   __ load_double(daddress(n));
 897 }
 898 
 899 void TemplateTable::aload(int n) {
 900   transition(vtos, atos);
 901   __ movptr(rax, aaddress(n));
 902 }
 903 
 904 void TemplateTable::aload_0() {
 905   aload_0_internal();
 906 }
 907 
 908 void TemplateTable::nofast_aload_0() {
 909   aload_0_internal(may_not_rewrite);
 910 }
 911 
 912 void TemplateTable::aload_0_internal(RewriteControl rc) {
 913   transition(vtos, atos);
 914   // According to bytecode histograms, the pairs:
 915   //
 916   // _aload_0, _fast_igetfield
 917   // _aload_0, _fast_agetfield
 918   // _aload_0, _fast_fgetfield
 919   //
 920   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 921   // _aload_0 bytecode checks if the next bytecode is either
 922   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 923   // rewrites the current bytecode into a pair bytecode; otherwise it
 924   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 925   // the pair check anymore.
 926   //
 927   // Note: If the next bytecode is _getfield, the rewrite must be
 928   //       delayed, otherwise we may miss an opportunity for a pair.
 929   //
 930   // Also rewrite frequent pairs
 931   //   aload_0, aload_1
 932   //   aload_0, iload_1
 933   // These bytecodes with a small amount of code are most profitable
 934   // to rewrite
 935   if (RewriteFrequentPairs && rc == may_rewrite) {
 936     Label rewrite, done;
 937 
 938     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 939     LP64_ONLY(assert(rbx != bc, "register damaged"));
 940 
 941     // get next byte
 942     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 943 
 944     // if _getfield then wait with rewrite
 945     __ cmpl(rbx, Bytecodes::_getfield);
 946     __ jcc(Assembler::equal, done);
 947 
 948     // if _igetfield then rewrite to _fast_iaccess_0
 949     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 950     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 951     __ movl(bc, Bytecodes::_fast_iaccess_0);
 952     __ jccb(Assembler::equal, rewrite);
 953 
 954     // if _agetfield then rewrite to _fast_aaccess_0
 955     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 956     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 957     __ movl(bc, Bytecodes::_fast_aaccess_0);
 958     __ jccb(Assembler::equal, rewrite);
 959 
 960     // if _fgetfield then rewrite to _fast_faccess_0
 961     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 962     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 963     __ movl(bc, Bytecodes::_fast_faccess_0);
 964     __ jccb(Assembler::equal, rewrite);
 965 
 966     // else rewrite to _fast_aload0
 967     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 968     __ movl(bc, Bytecodes::_fast_aload_0);
 969 
 970     // rewrite
 971     // bc: fast bytecode
 972     __ bind(rewrite);
 973     patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
 974 
 975     __ bind(done);
 976   }
 977 
 978   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 979   aload(0);
 980 }
 981 
 982 void TemplateTable::istore() {
 983   transition(itos, vtos);
 984   locals_index(rbx);
 985   __ movl(iaddress(rbx), rax);
 986 }
 987 
 988 
 989 void TemplateTable::lstore() {
 990   transition(ltos, vtos);
 991   locals_index(rbx);
 992   __ movptr(laddress(rbx), rax);
 993   NOT_LP64(__ movptr(haddress(rbx), rdx));
 994 }
 995 
 996 void TemplateTable::fstore() {
 997   transition(ftos, vtos);
 998   locals_index(rbx);
 999   __ store_float(faddress(rbx));
1000 }
1001 
1002 void TemplateTable::dstore() {
1003   transition(dtos, vtos);
1004   locals_index(rbx);
1005   __ store_double(daddress(rbx));
1006 }
1007 
1008 void TemplateTable::astore() {
1009   transition(vtos, vtos);
1010   __ pop_ptr(rax);
1011   locals_index(rbx);
1012   __ movptr(aaddress(rbx), rax);
1013 }
1014 
1015 void TemplateTable::wide_istore() {
1016   transition(vtos, vtos);
1017   __ pop_i();
1018   locals_index_wide(rbx);
1019   __ movl(iaddress(rbx), rax);
1020 }
1021 
1022 void TemplateTable::wide_lstore() {
1023   transition(vtos, vtos);
1024   NOT_LP64(__ pop_l(rax, rdx));
1025   LP64_ONLY(__ pop_l());
1026   locals_index_wide(rbx);
1027   __ movptr(laddress(rbx), rax);
1028   NOT_LP64(__ movl(haddress(rbx), rdx));
1029 }
1030 
1031 void TemplateTable::wide_fstore() {
1032 #ifdef _LP64
1033   transition(vtos, vtos);
1034   __ pop_f(xmm0);
1035   locals_index_wide(rbx);
1036   __ movflt(faddress(rbx), xmm0);
1037 #else
1038   wide_istore();
1039 #endif
1040 }
1041 
1042 void TemplateTable::wide_dstore() {
1043 #ifdef _LP64
1044   transition(vtos, vtos);
1045   __ pop_d(xmm0);
1046   locals_index_wide(rbx);
1047   __ movdbl(daddress(rbx), xmm0);
1048 #else
1049   wide_lstore();
1050 #endif
1051 }
1052 
1053 void TemplateTable::wide_astore() {
1054   transition(vtos, vtos);
1055   __ pop_ptr(rax);
1056   locals_index_wide(rbx);
1057   __ movptr(aaddress(rbx), rax);
1058 }
1059 
1060 void TemplateTable::iastore() {
1061   transition(itos, vtos);
1062   __ pop_i(rbx);
1063   // rax: value
1064   // rbx: index
1065   // rdx: array
1066   index_check(rdx, rbx); // prefer index in rbx
1067   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1068                      Address(rdx, rbx, Address::times_4,
1069                              arrayOopDesc::base_offset_in_bytes(T_INT)),
1070                      rax, noreg, noreg, noreg);
1071 }
1072 
1073 void TemplateTable::lastore() {
1074   transition(ltos, vtos);
1075   __ pop_i(rbx);
1076   // rax,: low(value)
1077   // rcx: array
1078   // rdx: high(value)
1079   index_check(rcx, rbx);  // prefer index in rbx,
1080   // rbx,: index
1081   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1082                      Address(rcx, rbx, Address::times_8,
1083                              arrayOopDesc::base_offset_in_bytes(T_LONG)),
1084                      noreg /* ltos */, noreg, noreg, noreg);
1085 }
1086 
1087 
1088 void TemplateTable::fastore() {
1089   transition(ftos, vtos);
1090   __ pop_i(rbx);
1091   // value is in UseSSE >= 1 ? xmm0 : ST(0)
1092   // rbx:  index
1093   // rdx:  array
1094   index_check(rdx, rbx); // prefer index in rbx
1095   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1096                      Address(rdx, rbx, Address::times_4,
1097                              arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1098                      noreg /* ftos */, noreg, noreg, noreg);
1099 }
1100 
1101 void TemplateTable::dastore() {
1102   transition(dtos, vtos);
1103   __ pop_i(rbx);
1104   // value is in UseSSE >= 2 ? xmm0 : ST(0)
1105   // rbx:  index
1106   // rdx:  array
1107   index_check(rdx, rbx); // prefer index in rbx
1108   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1109                      Address(rdx, rbx, Address::times_8,
1110                              arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1111                      noreg /* dtos */, noreg, noreg, noreg);
1112 }
1113 
1114 void TemplateTable::aastore() {
1115   Label is_null, ok_is_subtype, done;
1116   transition(vtos, vtos);
1117   // stack: ..., array, index, value
1118   __ movptr(rax, at_tos());    // value
1119   __ movl(rcx, at_tos_p1()); // index
1120   __ movptr(rdx, at_tos_p2()); // array
1121 
1122   Address element_address(rdx, rcx,
1123                           UseCompressedOops? Address::times_4 : Address::times_ptr,
1124                           arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1125 
1126   index_check_without_pop(rdx, rcx);     // kills rbx
1127   __ testptr(rax, rax);
1128   __ jcc(Assembler::zero, is_null);
1129 
1130   // Move subklass into rbx
1131   __ load_klass(rbx, rax, rscratch1);
1132   // Move superklass into rax
1133   __ load_klass(rax, rdx, rscratch1);
1134   __ movptr(rax, Address(rax,
1135                          ObjArrayKlass::element_klass_offset()));
1136 
1137   // Generate subtype check.  Blows rcx, rdi
1138   // Superklass in rax.  Subklass in rbx.
1139   __ gen_subtype_check(rbx, ok_is_subtype);
1140 
1141   // Come here on failure
1142   // object is at TOS
1143   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1144 
1145   // Come here on success
1146   __ bind(ok_is_subtype);
1147 
1148   // Get the value we will store
1149   __ movptr(rax, at_tos());
1150   __ movl(rcx, at_tos_p1()); // index
1151   // Now store using the appropriate barrier
1152   do_oop_store(_masm, element_address, rax, IS_ARRAY);
1153   __ jmp(done);
1154 
1155   // Have a NULL in rax, rdx=array, ecx=index.  Store NULL at ary[idx]
1156   __ bind(is_null);
1157   __ profile_null_seen(rbx);
1158 
1159   // Store a NULL
1160   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1161 
1162   // Pop stack arguments
1163   __ bind(done);
1164   __ addptr(rsp, 3 * Interpreter::stackElementSize);
1165 }
1166 
1167 void TemplateTable::bastore() {
1168   transition(itos, vtos);
1169   __ pop_i(rbx);
1170   // rax: value
1171   // rbx: index
1172   // rdx: array
1173   index_check(rdx, rbx); // prefer index in rbx
1174   // Need to check whether array is boolean or byte
1175   // since both types share the bastore bytecode.
1176   __ load_klass(rcx, rdx, rscratch1);
1177   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1178   int diffbit = Klass::layout_helper_boolean_diffbit();
1179   __ testl(rcx, diffbit);
1180   Label L_skip;
1181   __ jccb(Assembler::zero, L_skip);
1182   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1183   __ bind(L_skip);
1184   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1185                      Address(rdx, rbx,Address::times_1,
1186                              arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1187                      rax, noreg, noreg, noreg);
1188 }
1189 
1190 void TemplateTable::castore() {
1191   transition(itos, vtos);
1192   __ pop_i(rbx);
1193   // rax: value
1194   // rbx: index
1195   // rdx: array
1196   index_check(rdx, rbx);  // prefer index in rbx
1197   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1198                      Address(rdx, rbx, Address::times_2,
1199                              arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1200                      rax, noreg, noreg, noreg);
1201 }
1202 
1203 
1204 void TemplateTable::sastore() {
1205   castore();
1206 }
1207 
1208 void TemplateTable::istore(int n) {
1209   transition(itos, vtos);
1210   __ movl(iaddress(n), rax);
1211 }
1212 
1213 void TemplateTable::lstore(int n) {
1214   transition(ltos, vtos);
1215   __ movptr(laddress(n), rax);
1216   NOT_LP64(__ movptr(haddress(n), rdx));
1217 }
1218 
1219 void TemplateTable::fstore(int n) {
1220   transition(ftos, vtos);
1221   __ store_float(faddress(n));
1222 }
1223 
1224 void TemplateTable::dstore(int n) {
1225   transition(dtos, vtos);
1226   __ store_double(daddress(n));
1227 }
1228 
1229 
1230 void TemplateTable::astore(int n) {
1231   transition(vtos, vtos);
1232   __ pop_ptr(rax);
1233   __ movptr(aaddress(n), rax);
1234 }
1235 
1236 void TemplateTable::pop() {
1237   transition(vtos, vtos);
1238   __ addptr(rsp, Interpreter::stackElementSize);
1239 }
1240 
1241 void TemplateTable::pop2() {
1242   transition(vtos, vtos);
1243   __ addptr(rsp, 2 * Interpreter::stackElementSize);
1244 }
1245 
1246 
1247 void TemplateTable::dup() {
1248   transition(vtos, vtos);
1249   __ load_ptr(0, rax);
1250   __ push_ptr(rax);
1251   // stack: ..., a, a
1252 }
1253 
1254 void TemplateTable::dup_x1() {
1255   transition(vtos, vtos);
1256   // stack: ..., a, b
1257   __ load_ptr( 0, rax);  // load b
1258   __ load_ptr( 1, rcx);  // load a
1259   __ store_ptr(1, rax);  // store b
1260   __ store_ptr(0, rcx);  // store a
1261   __ push_ptr(rax);      // push b
1262   // stack: ..., b, a, b
1263 }
1264 
1265 void TemplateTable::dup_x2() {
1266   transition(vtos, vtos);
1267   // stack: ..., a, b, c
1268   __ load_ptr( 0, rax);  // load c
1269   __ load_ptr( 2, rcx);  // load a
1270   __ store_ptr(2, rax);  // store c in a
1271   __ push_ptr(rax);      // push c
1272   // stack: ..., c, b, c, c
1273   __ load_ptr( 2, rax);  // load b
1274   __ store_ptr(2, rcx);  // store a in b
1275   // stack: ..., c, a, c, c
1276   __ store_ptr(1, rax);  // store b in c
1277   // stack: ..., c, a, b, c
1278 }
1279 
1280 void TemplateTable::dup2() {
1281   transition(vtos, vtos);
1282   // stack: ..., a, b
1283   __ load_ptr(1, rax);  // load a
1284   __ push_ptr(rax);     // push a
1285   __ load_ptr(1, rax);  // load b
1286   __ push_ptr(rax);     // push b
1287   // stack: ..., a, b, a, b
1288 }
1289 
1290 
1291 void TemplateTable::dup2_x1() {
1292   transition(vtos, vtos);
1293   // stack: ..., a, b, c
1294   __ load_ptr( 0, rcx);  // load c
1295   __ load_ptr( 1, rax);  // load b
1296   __ push_ptr(rax);      // push b
1297   __ push_ptr(rcx);      // push c
1298   // stack: ..., a, b, c, b, c
1299   __ store_ptr(3, rcx);  // store c in b
1300   // stack: ..., a, c, c, b, c
1301   __ load_ptr( 4, rcx);  // load a
1302   __ store_ptr(2, rcx);  // store a in 2nd c
1303   // stack: ..., a, c, a, b, c
1304   __ store_ptr(4, rax);  // store b in a
1305   // stack: ..., b, c, a, b, c
1306 }
1307 
1308 void TemplateTable::dup2_x2() {
1309   transition(vtos, vtos);
1310   // stack: ..., a, b, c, d
1311   __ load_ptr( 0, rcx);  // load d
1312   __ load_ptr( 1, rax);  // load c
1313   __ push_ptr(rax);      // push c
1314   __ push_ptr(rcx);      // push d
1315   // stack: ..., a, b, c, d, c, d
1316   __ load_ptr( 4, rax);  // load b
1317   __ store_ptr(2, rax);  // store b in d
1318   __ store_ptr(4, rcx);  // store d in b
1319   // stack: ..., a, d, c, b, c, d
1320   __ load_ptr( 5, rcx);  // load a
1321   __ load_ptr( 3, rax);  // load c
1322   __ store_ptr(3, rcx);  // store a in c
1323   __ store_ptr(5, rax);  // store c in a
1324   // stack: ..., c, d, a, b, c, d
1325 }
1326 
1327 void TemplateTable::swap() {
1328   transition(vtos, vtos);
1329   // stack: ..., a, b
1330   __ load_ptr( 1, rcx);  // load a
1331   __ load_ptr( 0, rax);  // load b
1332   __ store_ptr(0, rcx);  // store a in b
1333   __ store_ptr(1, rax);  // store b in a
1334   // stack: ..., b, a
1335 }
1336 
1337 void TemplateTable::iop2(Operation op) {
1338   transition(itos, itos);
1339   switch (op) {
1340   case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1341   case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1342   case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1343   case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1344   case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1345   case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1346   case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break;
1347   case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break;
1348   case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break;
1349   default   : ShouldNotReachHere();
1350   }
1351 }
1352 
1353 void TemplateTable::lop2(Operation op) {
1354   transition(ltos, ltos);
1355 #ifdef _LP64
1356   switch (op) {
1357   case add  :                    __ pop_l(rdx); __ addptr(rax, rdx); break;
1358   case sub  : __ mov(rdx, rax);  __ pop_l(rax); __ subptr(rax, rdx); break;
1359   case _and :                    __ pop_l(rdx); __ andptr(rax, rdx); break;
1360   case _or  :                    __ pop_l(rdx); __ orptr (rax, rdx); break;
1361   case _xor :                    __ pop_l(rdx); __ xorptr(rax, rdx); break;
1362   default   : ShouldNotReachHere();
1363   }
1364 #else
1365   __ pop_l(rbx, rcx);
1366   switch (op) {
1367     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1368     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1369                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1370     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1371     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1372     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1373     default   : ShouldNotReachHere();
1374   }
1375 #endif
1376 }
1377 
1378 void TemplateTable::idiv() {
1379   transition(itos, itos);
1380   __ movl(rcx, rax);
1381   __ pop_i(rax);
1382   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1383   //       they are not equal, one could do a normal division (no correction
1384   //       needed), which may speed up this implementation for the common case.
1385   //       (see also JVM spec., p.243 & p.271)
1386   __ corrected_idivl(rcx);
1387 }
1388 
1389 void TemplateTable::irem() {
1390   transition(itos, itos);
1391   __ movl(rcx, rax);
1392   __ pop_i(rax);
1393   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1394   //       they are not equal, one could do a normal division (no correction
1395   //       needed), which may speed up this implementation for the common case.
1396   //       (see also JVM spec., p.243 & p.271)
1397   __ corrected_idivl(rcx);
1398   __ movl(rax, rdx);
1399 }
1400 
1401 void TemplateTable::lmul() {
1402   transition(ltos, ltos);
1403 #ifdef _LP64
1404   __ pop_l(rdx);
1405   __ imulq(rax, rdx);
1406 #else
1407   __ pop_l(rbx, rcx);
1408   __ push(rcx); __ push(rbx);
1409   __ push(rdx); __ push(rax);
1410   __ lmul(2 * wordSize, 0);
1411   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1412 #endif
1413 }
1414 
1415 void TemplateTable::ldiv() {
1416   transition(ltos, ltos);
1417 #ifdef _LP64
1418   __ mov(rcx, rax);
1419   __ pop_l(rax);
1420   // generate explicit div0 check
1421   __ testq(rcx, rcx);
1422   __ jump_cc(Assembler::zero,
1423              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1424   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1425   //       they are not equal, one could do a normal division (no correction
1426   //       needed), which may speed up this implementation for the common case.
1427   //       (see also JVM spec., p.243 & p.271)
1428   __ corrected_idivq(rcx); // kills rbx
1429 #else
1430   __ pop_l(rbx, rcx);
1431   __ push(rcx); __ push(rbx);
1432   __ push(rdx); __ push(rax);
1433   // check if y = 0
1434   __ orl(rax, rdx);
1435   __ jump_cc(Assembler::zero,
1436              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1437   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1438   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1439 #endif
1440 }
1441 
1442 void TemplateTable::lrem() {
1443   transition(ltos, ltos);
1444 #ifdef _LP64
1445   __ mov(rcx, rax);
1446   __ pop_l(rax);
1447   __ testq(rcx, rcx);
1448   __ jump_cc(Assembler::zero,
1449              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1450   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1451   //       they are not equal, one could do a normal division (no correction
1452   //       needed), which may speed up this implementation for the common case.
1453   //       (see also JVM spec., p.243 & p.271)
1454   __ corrected_idivq(rcx); // kills rbx
1455   __ mov(rax, rdx);
1456 #else
1457   __ pop_l(rbx, rcx);
1458   __ push(rcx); __ push(rbx);
1459   __ push(rdx); __ push(rax);
1460   // check if y = 0
1461   __ orl(rax, rdx);
1462   __ jump_cc(Assembler::zero,
1463              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1464   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1465   __ addptr(rsp, 4 * wordSize);
1466 #endif
1467 }
1468 
1469 void TemplateTable::lshl() {
1470   transition(itos, ltos);
1471   __ movl(rcx, rax);                             // get shift count
1472   #ifdef _LP64
1473   __ pop_l(rax);                                 // get shift value
1474   __ shlq(rax);
1475 #else
1476   __ pop_l(rax, rdx);                            // get shift value
1477   __ lshl(rdx, rax);
1478 #endif
1479 }
1480 
1481 void TemplateTable::lshr() {
1482 #ifdef _LP64
1483   transition(itos, ltos);
1484   __ movl(rcx, rax);                             // get shift count
1485   __ pop_l(rax);                                 // get shift value
1486   __ sarq(rax);
1487 #else
1488   transition(itos, ltos);
1489   __ mov(rcx, rax);                              // get shift count
1490   __ pop_l(rax, rdx);                            // get shift value
1491   __ lshr(rdx, rax, true);
1492 #endif
1493 }
1494 
1495 void TemplateTable::lushr() {
1496   transition(itos, ltos);
1497 #ifdef _LP64
1498   __ movl(rcx, rax);                             // get shift count
1499   __ pop_l(rax);                                 // get shift value
1500   __ shrq(rax);
1501 #else
1502   __ mov(rcx, rax);                              // get shift count
1503   __ pop_l(rax, rdx);                            // get shift value
1504   __ lshr(rdx, rax);
1505 #endif
1506 }
1507 
1508 void TemplateTable::fop2(Operation op) {
1509   transition(ftos, ftos);
1510 
1511   if (UseSSE >= 1) {
1512     switch (op) {
1513     case add:
1514       __ addss(xmm0, at_rsp());
1515       __ addptr(rsp, Interpreter::stackElementSize);
1516       break;
1517     case sub:
1518       __ movflt(xmm1, xmm0);
1519       __ pop_f(xmm0);
1520       __ subss(xmm0, xmm1);
1521       break;
1522     case mul:
1523       __ mulss(xmm0, at_rsp());
1524       __ addptr(rsp, Interpreter::stackElementSize);
1525       break;
1526     case div:
1527       __ movflt(xmm1, xmm0);
1528       __ pop_f(xmm0);
1529       __ divss(xmm0, xmm1);
1530       break;
1531     case rem:
1532       // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1533       // modulo operation. The frem method calls the function
1534       // double fmod(double x, double y) in math.h. The documentation of fmod states:
1535       // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1536       // (signalling or quiet) is returned.
1537       //
1538       // On x86_32 platforms the FPU is used to perform the modulo operation. The
1539       // reason is that on 32-bit Windows the sign of modulo operations diverges from
1540       // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1541       // The fprem instruction used on x86_32 is functionally equivalent to
1542       // SharedRuntime::frem in that it returns a NaN.
1543 #ifdef _LP64
1544       __ movflt(xmm1, xmm0);
1545       __ pop_f(xmm0);
1546       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1547 #else // !_LP64
1548       __ push_f(xmm0);
1549       __ pop_f();
1550       __ fld_s(at_rsp());
1551       __ fremr(rax);
1552       __ f2ieee();
1553       __ pop(rax);  // pop second operand off the stack
1554       __ push_f();
1555       __ pop_f(xmm0);
1556 #endif // _LP64
1557       break;
1558     default:
1559       ShouldNotReachHere();
1560       break;
1561     }
1562   } else {
1563 #ifdef _LP64
1564     ShouldNotReachHere();
1565 #else // !_LP64
1566     switch (op) {
1567     case add: __ fadd_s (at_rsp());                break;
1568     case sub: __ fsubr_s(at_rsp());                break;
1569     case mul: __ fmul_s (at_rsp());                break;
1570     case div: __ fdivr_s(at_rsp());                break;
1571     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1572     default : ShouldNotReachHere();
1573     }
1574     __ f2ieee();
1575     __ pop(rax);  // pop second operand off the stack
1576 #endif // _LP64
1577   }
1578 }
1579 
1580 void TemplateTable::dop2(Operation op) {
1581   transition(dtos, dtos);
1582   if (UseSSE >= 2) {
1583     switch (op) {
1584     case add:
1585       __ addsd(xmm0, at_rsp());
1586       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1587       break;
1588     case sub:
1589       __ movdbl(xmm1, xmm0);
1590       __ pop_d(xmm0);
1591       __ subsd(xmm0, xmm1);
1592       break;
1593     case mul:
1594       __ mulsd(xmm0, at_rsp());
1595       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1596       break;
1597     case div:
1598       __ movdbl(xmm1, xmm0);
1599       __ pop_d(xmm0);
1600       __ divsd(xmm0, xmm1);
1601       break;
1602     case rem:
1603       // Similar to fop2(), the modulo operation is performed using the
1604       // SharedRuntime::drem method (on x86_64 platforms) or using the
1605       // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1606 #ifdef _LP64
1607       __ movdbl(xmm1, xmm0);
1608       __ pop_d(xmm0);
1609       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1610 #else // !_LP64
1611       __ push_d(xmm0);
1612       __ pop_d();
1613       __ fld_d(at_rsp());
1614       __ fremr(rax);
1615       __ d2ieee();
1616       __ pop(rax);
1617       __ pop(rdx);
1618       __ push_d();
1619       __ pop_d(xmm0);
1620 #endif // _LP64
1621       break;
1622     default:
1623       ShouldNotReachHere();
1624       break;
1625     }
1626   } else {
1627 #ifdef _LP64
1628     ShouldNotReachHere();
1629 #else // !_LP64
1630     switch (op) {
1631     case add: __ fadd_d (at_rsp());                break;
1632     case sub: __ fsubr_d(at_rsp());                break;
1633     case mul: {
1634       // strict semantics
1635       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1636       __ fmulp();
1637       __ fmul_d (at_rsp());
1638       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1639       __ fmulp();
1640       break;
1641     }
1642     case div: {
1643       // strict semantics
1644       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1645       __ fmul_d (at_rsp());
1646       __ fdivrp();
1647       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1648       __ fmulp();
1649       break;
1650     }
1651     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1652     default : ShouldNotReachHere();
1653     }
1654     __ d2ieee();
1655     // Pop double precision number from rsp.
1656     __ pop(rax);
1657     __ pop(rdx);
1658 #endif // _LP64
1659   }
1660 }
1661 
1662 void TemplateTable::ineg() {
1663   transition(itos, itos);
1664   __ negl(rax);
1665 }
1666 
1667 void TemplateTable::lneg() {
1668   transition(ltos, ltos);
1669   LP64_ONLY(__ negq(rax));
1670   NOT_LP64(__ lneg(rdx, rax));
1671 }
1672 
1673 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1674 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1675   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1676   // of 128-bits operands for SSE instructions.
1677   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1678   // Store the value to a 128-bits operand.
1679   operand[0] = lo;
1680   operand[1] = hi;
1681   return operand;
1682 }
1683 
1684 // Buffer for 128-bits masks used by SSE instructions.
1685 static jlong float_signflip_pool[2*2];
1686 static jlong double_signflip_pool[2*2];
1687 
1688 void TemplateTable::fneg() {
1689   transition(ftos, ftos);
1690   if (UseSSE >= 1) {
1691     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1692     __ xorps(xmm0, ExternalAddress((address) float_signflip), rscratch1);
1693   } else {
1694     LP64_ONLY(ShouldNotReachHere());
1695     NOT_LP64(__ fchs());
1696   }
1697 }
1698 
1699 void TemplateTable::dneg() {
1700   transition(dtos, dtos);
1701   if (UseSSE >= 2) {
1702     static jlong *double_signflip =
1703       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1704     __ xorpd(xmm0, ExternalAddress((address) double_signflip), rscratch1);
1705   } else {
1706 #ifdef _LP64
1707     ShouldNotReachHere();
1708 #else
1709     __ fchs();
1710 #endif
1711   }
1712 }
1713 
1714 void TemplateTable::iinc() {
1715   transition(vtos, vtos);
1716   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1717   locals_index(rbx);
1718   __ addl(iaddress(rbx), rdx);
1719 }
1720 
1721 void TemplateTable::wide_iinc() {
1722   transition(vtos, vtos);
1723   __ movl(rdx, at_bcp(4)); // get constant
1724   locals_index_wide(rbx);
1725   __ bswapl(rdx); // swap bytes & sign-extend constant
1726   __ sarl(rdx, 16);
1727   __ addl(iaddress(rbx), rdx);
1728   // Note: should probably use only one movl to get both
1729   //       the index and the constant -> fix this
1730 }
1731 
1732 void TemplateTable::convert() {
1733 #ifdef _LP64
1734   // Checking
1735 #ifdef ASSERT
1736   {
1737     TosState tos_in  = ilgl;
1738     TosState tos_out = ilgl;
1739     switch (bytecode()) {
1740     case Bytecodes::_i2l: // fall through
1741     case Bytecodes::_i2f: // fall through
1742     case Bytecodes::_i2d: // fall through
1743     case Bytecodes::_i2b: // fall through
1744     case Bytecodes::_i2c: // fall through
1745     case Bytecodes::_i2s: tos_in = itos; break;
1746     case Bytecodes::_l2i: // fall through
1747     case Bytecodes::_l2f: // fall through
1748     case Bytecodes::_l2d: tos_in = ltos; break;
1749     case Bytecodes::_f2i: // fall through
1750     case Bytecodes::_f2l: // fall through
1751     case Bytecodes::_f2d: tos_in = ftos; break;
1752     case Bytecodes::_d2i: // fall through
1753     case Bytecodes::_d2l: // fall through
1754     case Bytecodes::_d2f: tos_in = dtos; break;
1755     default             : ShouldNotReachHere();
1756     }
1757     switch (bytecode()) {
1758     case Bytecodes::_l2i: // fall through
1759     case Bytecodes::_f2i: // fall through
1760     case Bytecodes::_d2i: // fall through
1761     case Bytecodes::_i2b: // fall through
1762     case Bytecodes::_i2c: // fall through
1763     case Bytecodes::_i2s: tos_out = itos; break;
1764     case Bytecodes::_i2l: // fall through
1765     case Bytecodes::_f2l: // fall through
1766     case Bytecodes::_d2l: tos_out = ltos; break;
1767     case Bytecodes::_i2f: // fall through
1768     case Bytecodes::_l2f: // fall through
1769     case Bytecodes::_d2f: tos_out = ftos; break;
1770     case Bytecodes::_i2d: // fall through
1771     case Bytecodes::_l2d: // fall through
1772     case Bytecodes::_f2d: tos_out = dtos; break;
1773     default             : ShouldNotReachHere();
1774     }
1775     transition(tos_in, tos_out);
1776   }
1777 #endif // ASSERT
1778 
1779   static const int64_t is_nan = 0x8000000000000000L;
1780 
1781   // Conversion
1782   switch (bytecode()) {
1783   case Bytecodes::_i2l:
1784     __ movslq(rax, rax);
1785     break;
1786   case Bytecodes::_i2f:
1787     __ cvtsi2ssl(xmm0, rax);
1788     break;
1789   case Bytecodes::_i2d:
1790     __ cvtsi2sdl(xmm0, rax);
1791     break;
1792   case Bytecodes::_i2b:
1793     __ movsbl(rax, rax);
1794     break;
1795   case Bytecodes::_i2c:
1796     __ movzwl(rax, rax);
1797     break;
1798   case Bytecodes::_i2s:
1799     __ movswl(rax, rax);
1800     break;
1801   case Bytecodes::_l2i:
1802     __ movl(rax, rax);
1803     break;
1804   case Bytecodes::_l2f:
1805     __ cvtsi2ssq(xmm0, rax);
1806     break;
1807   case Bytecodes::_l2d:
1808     __ cvtsi2sdq(xmm0, rax);
1809     break;
1810   case Bytecodes::_f2i:
1811   {
1812     Label L;
1813     __ cvttss2sil(rax, xmm0);
1814     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1815     __ jcc(Assembler::notEqual, L);
1816     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1817     __ bind(L);
1818   }
1819     break;
1820   case Bytecodes::_f2l:
1821   {
1822     Label L;
1823     __ cvttss2siq(rax, xmm0);
1824     // NaN or overflow/underflow?
1825     __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1826     __ jcc(Assembler::notEqual, L);
1827     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1828     __ bind(L);
1829   }
1830     break;
1831   case Bytecodes::_f2d:
1832     __ cvtss2sd(xmm0, xmm0);
1833     break;
1834   case Bytecodes::_d2i:
1835   {
1836     Label L;
1837     __ cvttsd2sil(rax, xmm0);
1838     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1839     __ jcc(Assembler::notEqual, L);
1840     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1841     __ bind(L);
1842   }
1843     break;
1844   case Bytecodes::_d2l:
1845   {
1846     Label L;
1847     __ cvttsd2siq(rax, xmm0);
1848     // NaN or overflow/underflow?
1849     __ cmp64(rax, ExternalAddress((address) &is_nan), rscratch1);
1850     __ jcc(Assembler::notEqual, L);
1851     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1852     __ bind(L);
1853   }
1854     break;
1855   case Bytecodes::_d2f:
1856     __ cvtsd2ss(xmm0, xmm0);
1857     break;
1858   default:
1859     ShouldNotReachHere();
1860   }
1861 #else // !_LP64
1862   // Checking
1863 #ifdef ASSERT
1864   { TosState tos_in  = ilgl;
1865     TosState tos_out = ilgl;
1866     switch (bytecode()) {
1867       case Bytecodes::_i2l: // fall through
1868       case Bytecodes::_i2f: // fall through
1869       case Bytecodes::_i2d: // fall through
1870       case Bytecodes::_i2b: // fall through
1871       case Bytecodes::_i2c: // fall through
1872       case Bytecodes::_i2s: tos_in = itos; break;
1873       case Bytecodes::_l2i: // fall through
1874       case Bytecodes::_l2f: // fall through
1875       case Bytecodes::_l2d: tos_in = ltos; break;
1876       case Bytecodes::_f2i: // fall through
1877       case Bytecodes::_f2l: // fall through
1878       case Bytecodes::_f2d: tos_in = ftos; break;
1879       case Bytecodes::_d2i: // fall through
1880       case Bytecodes::_d2l: // fall through
1881       case Bytecodes::_d2f: tos_in = dtos; break;
1882       default             : ShouldNotReachHere();
1883     }
1884     switch (bytecode()) {
1885       case Bytecodes::_l2i: // fall through
1886       case Bytecodes::_f2i: // fall through
1887       case Bytecodes::_d2i: // fall through
1888       case Bytecodes::_i2b: // fall through
1889       case Bytecodes::_i2c: // fall through
1890       case Bytecodes::_i2s: tos_out = itos; break;
1891       case Bytecodes::_i2l: // fall through
1892       case Bytecodes::_f2l: // fall through
1893       case Bytecodes::_d2l: tos_out = ltos; break;
1894       case Bytecodes::_i2f: // fall through
1895       case Bytecodes::_l2f: // fall through
1896       case Bytecodes::_d2f: tos_out = ftos; break;
1897       case Bytecodes::_i2d: // fall through
1898       case Bytecodes::_l2d: // fall through
1899       case Bytecodes::_f2d: tos_out = dtos; break;
1900       default             : ShouldNotReachHere();
1901     }
1902     transition(tos_in, tos_out);
1903   }
1904 #endif // ASSERT
1905 
1906   // Conversion
1907   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1908   switch (bytecode()) {
1909     case Bytecodes::_i2l:
1910       __ extend_sign(rdx, rax);
1911       break;
1912     case Bytecodes::_i2f:
1913       if (UseSSE >= 1) {
1914         __ cvtsi2ssl(xmm0, rax);
1915       } else {
1916         __ push(rax);          // store int on tos
1917         __ fild_s(at_rsp());   // load int to ST0
1918         __ f2ieee();           // truncate to float size
1919         __ pop(rcx);           // adjust rsp
1920       }
1921       break;
1922     case Bytecodes::_i2d:
1923       if (UseSSE >= 2) {
1924         __ cvtsi2sdl(xmm0, rax);
1925       } else {
1926       __ push(rax);          // add one slot for d2ieee()
1927       __ push(rax);          // store int on tos
1928       __ fild_s(at_rsp());   // load int to ST0
1929       __ d2ieee();           // truncate to double size
1930       __ pop(rcx);           // adjust rsp
1931       __ pop(rcx);
1932       }
1933       break;
1934     case Bytecodes::_i2b:
1935       __ shll(rax, 24);      // truncate upper 24 bits
1936       __ sarl(rax, 24);      // and sign-extend byte
1937       LP64_ONLY(__ movsbl(rax, rax));
1938       break;
1939     case Bytecodes::_i2c:
1940       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1941       LP64_ONLY(__ movzwl(rax, rax));
1942       break;
1943     case Bytecodes::_i2s:
1944       __ shll(rax, 16);      // truncate upper 16 bits
1945       __ sarl(rax, 16);      // and sign-extend short
1946       LP64_ONLY(__ movswl(rax, rax));
1947       break;
1948     case Bytecodes::_l2i:
1949       /* nothing to do */
1950       break;
1951     case Bytecodes::_l2f:
1952       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1953       // 64-bit long values to floats. On 32-bit platforms it is not possible
1954       // to use that instruction with 64-bit operands, therefore the FPU is
1955       // used to perform the conversion.
1956       __ push(rdx);          // store long on tos
1957       __ push(rax);
1958       __ fild_d(at_rsp());   // load long to ST0
1959       __ f2ieee();           // truncate to float size
1960       __ pop(rcx);           // adjust rsp
1961       __ pop(rcx);
1962       if (UseSSE >= 1) {
1963         __ push_f();
1964         __ pop_f(xmm0);
1965       }
1966       break;
1967     case Bytecodes::_l2d:
1968       // On 32-bit platforms the FPU is used for conversion because on
1969       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1970       // instruction with 64-bit operands.
1971       __ push(rdx);          // store long on tos
1972       __ push(rax);
1973       __ fild_d(at_rsp());   // load long to ST0
1974       __ d2ieee();           // truncate to double size
1975       __ pop(rcx);           // adjust rsp
1976       __ pop(rcx);
1977       if (UseSSE >= 2) {
1978         __ push_d();
1979         __ pop_d(xmm0);
1980       }
1981       break;
1982     case Bytecodes::_f2i:
1983       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1984       // as it returns 0 for any NaN.
1985       if (UseSSE >= 1) {
1986         __ push_f(xmm0);
1987       } else {
1988         __ push(rcx);          // reserve space for argument
1989         __ fstp_s(at_rsp());   // pass float argument on stack
1990       }
1991       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1992       break;
1993     case Bytecodes::_f2l:
1994       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1995       // as it returns 0 for any NaN.
1996       if (UseSSE >= 1) {
1997        __ push_f(xmm0);
1998       } else {
1999         __ push(rcx);          // reserve space for argument
2000         __ fstp_s(at_rsp());   // pass float argument on stack
2001       }
2002       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2003       break;
2004     case Bytecodes::_f2d:
2005       if (UseSSE < 1) {
2006         /* nothing to do */
2007       } else if (UseSSE == 1) {
2008         __ push_f(xmm0);
2009         __ pop_f();
2010       } else { // UseSSE >= 2
2011         __ cvtss2sd(xmm0, xmm0);
2012       }
2013       break;
2014     case Bytecodes::_d2i:
2015       if (UseSSE >= 2) {
2016         __ push_d(xmm0);
2017       } else {
2018         __ push(rcx);          // reserve space for argument
2019         __ push(rcx);
2020         __ fstp_d(at_rsp());   // pass double argument on stack
2021       }
2022       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2023       break;
2024     case Bytecodes::_d2l:
2025       if (UseSSE >= 2) {
2026         __ push_d(xmm0);
2027       } else {
2028         __ push(rcx);          // reserve space for argument
2029         __ push(rcx);
2030         __ fstp_d(at_rsp());   // pass double argument on stack
2031       }
2032       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2033       break;
2034     case Bytecodes::_d2f:
2035       if (UseSSE <= 1) {
2036         __ push(rcx);          // reserve space for f2ieee()
2037         __ f2ieee();           // truncate to float size
2038         __ pop(rcx);           // adjust rsp
2039         if (UseSSE == 1) {
2040           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2041           // the conversion is performed using the FPU in this case.
2042           __ push_f();
2043           __ pop_f(xmm0);
2044         }
2045       } else { // UseSSE >= 2
2046         __ cvtsd2ss(xmm0, xmm0);
2047       }
2048       break;
2049     default             :
2050       ShouldNotReachHere();
2051   }
2052 #endif // _LP64
2053 }
2054 
2055 void TemplateTable::lcmp() {
2056   transition(ltos, itos);
2057 #ifdef _LP64
2058   Label done;
2059   __ pop_l(rdx);
2060   __ cmpq(rdx, rax);
2061   __ movl(rax, -1);
2062   __ jccb(Assembler::less, done);
2063   __ setb(Assembler::notEqual, rax);
2064   __ movzbl(rax, rax);
2065   __ bind(done);
2066 #else
2067 
2068   // y = rdx:rax
2069   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2070   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2071   __ mov(rax, rcx);
2072 #endif
2073 }
2074 
2075 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2076   if ((is_float && UseSSE >= 1) ||
2077       (!is_float && UseSSE >= 2)) {
2078     Label done;
2079     if (is_float) {
2080       // XXX get rid of pop here, use ... reg, mem32
2081       __ pop_f(xmm1);
2082       __ ucomiss(xmm1, xmm0);
2083     } else {
2084       // XXX get rid of pop here, use ... reg, mem64
2085       __ pop_d(xmm1);
2086       __ ucomisd(xmm1, xmm0);
2087     }
2088     if (unordered_result < 0) {
2089       __ movl(rax, -1);
2090       __ jccb(Assembler::parity, done);
2091       __ jccb(Assembler::below, done);
2092       __ setb(Assembler::notEqual, rdx);
2093       __ movzbl(rax, rdx);
2094     } else {
2095       __ movl(rax, 1);
2096       __ jccb(Assembler::parity, done);
2097       __ jccb(Assembler::above, done);
2098       __ movl(rax, 0);
2099       __ jccb(Assembler::equal, done);
2100       __ decrementl(rax);
2101     }
2102     __ bind(done);
2103   } else {
2104 #ifdef _LP64
2105     ShouldNotReachHere();
2106 #else // !_LP64
2107     if (is_float) {
2108       __ fld_s(at_rsp());
2109     } else {
2110       __ fld_d(at_rsp());
2111       __ pop(rdx);
2112     }
2113     __ pop(rcx);
2114     __ fcmp2int(rax, unordered_result < 0);
2115 #endif // _LP64
2116   }
2117 }
2118 
2119 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2120   __ get_method(rcx); // rcx holds method
2121   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2122                                      // holds bumped taken count
2123 
2124   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2125                              InvocationCounter::counter_offset();
2126   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2127                               InvocationCounter::counter_offset();
2128 
2129   // Load up edx with the branch displacement
2130   if (is_wide) {
2131     __ movl(rdx, at_bcp(1));
2132   } else {
2133     __ load_signed_short(rdx, at_bcp(1));
2134   }
2135   __ bswapl(rdx);
2136 
2137   if (!is_wide) {
2138     __ sarl(rdx, 16);
2139   }
2140   LP64_ONLY(__ movl2ptr(rdx, rdx));
2141 
2142   // Handle all the JSR stuff here, then exit.
2143   // It's much shorter and cleaner than intermingling with the non-JSR
2144   // normal-branch stuff occurring below.
2145   if (is_jsr) {
2146     // Pre-load the next target bytecode into rbx
2147     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2148 
2149     // compute return address as bci in rax
2150     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2151                         in_bytes(ConstMethod::codes_offset())));
2152     __ subptr(rax, Address(rcx, Method::const_offset()));
2153     // Adjust the bcp in r13 by the displacement in rdx
2154     __ addptr(rbcp, rdx);
2155     // jsr returns atos that is not an oop
2156     __ push_i(rax);
2157     __ dispatch_only(vtos, true);
2158     return;
2159   }
2160 
2161   // Normal (non-jsr) branch handling
2162 
2163   // Adjust the bcp in r13 by the displacement in rdx
2164   __ addptr(rbcp, rdx);
2165 
2166   assert(UseLoopCounter || !UseOnStackReplacement,
2167          "on-stack-replacement requires loop counters");
2168   Label backedge_counter_overflow;
2169   Label dispatch;
2170   if (UseLoopCounter) {
2171     // increment backedge counter for backward branches
2172     // rax: MDO
2173     // rbx: MDO bumped taken-count
2174     // rcx: method
2175     // rdx: target offset
2176     // r13: target bcp
2177     // r14: locals pointer
2178     __ testl(rdx, rdx);             // check if forward or backward branch
2179     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2180 
2181     // check if MethodCounters exists
2182     Label has_counters;
2183     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2184     __ testptr(rax, rax);
2185     __ jcc(Assembler::notZero, has_counters);
2186     __ push(rdx);
2187     __ push(rcx);
2188     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2189                rcx);
2190     __ pop(rcx);
2191     __ pop(rdx);
2192     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2193     __ testptr(rax, rax);
2194     __ jcc(Assembler::zero, dispatch);
2195     __ bind(has_counters);
2196 
2197     Label no_mdo;
2198     if (ProfileInterpreter) {
2199       // Are we profiling?
2200       __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2201       __ testptr(rbx, rbx);
2202       __ jccb(Assembler::zero, no_mdo);
2203       // Increment the MDO backedge counter
2204       const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2205           in_bytes(InvocationCounter::counter_offset()));
2206       const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2207       __ increment_mask_and_jump(mdo_backedge_counter, mask, rax,
2208           UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2209       __ jmp(dispatch);
2210     }
2211     __ bind(no_mdo);
2212     // Increment backedge counter in MethodCounters*
2213     __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2214     const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2215     __ increment_mask_and_jump(Address(rcx, be_offset), mask, rax,
2216         UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2217     __ bind(dispatch);
2218   }
2219 
2220   // Pre-load the next target bytecode into rbx
2221   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2222 
2223   // continue with the bytecode @ target
2224   // rax: return bci for jsr's, unused otherwise
2225   // rbx: target bytecode
2226   // r13: target bcp
2227   __ dispatch_only(vtos, true);
2228 
2229   if (UseLoopCounter) {
2230     if (UseOnStackReplacement) {
2231       Label set_mdp;
2232       // invocation counter overflow
2233       __ bind(backedge_counter_overflow);
2234       __ negptr(rdx);
2235       __ addptr(rdx, rbcp); // branch bcp
2236       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2237       __ call_VM(noreg,
2238                  CAST_FROM_FN_PTR(address,
2239                                   InterpreterRuntime::frequency_counter_overflow),
2240                  rdx);
2241 
2242       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2243       // rdx: scratch
2244       // r14: locals pointer
2245       // r13: bcp
2246       __ testptr(rax, rax);                        // test result
2247       __ jcc(Assembler::zero, dispatch);         // no osr if null
2248       // nmethod may have been invalidated (VM may block upon call_VM return)
2249       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2250       __ jcc(Assembler::notEqual, dispatch);
2251 
2252       // We have the address of an on stack replacement routine in rax.
2253       // In preparation of invoking it, first we must migrate the locals
2254       // and monitors from off the interpreter frame on the stack.
2255       // Ensure to save the osr nmethod over the migration call,
2256       // it will be preserved in rbx.
2257       __ mov(rbx, rax);
2258 
2259       NOT_LP64(__ get_thread(rcx));
2260 
2261       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2262 
2263       // rax is OSR buffer, move it to expected parameter location
2264       LP64_ONLY(__ mov(j_rarg0, rax));
2265       NOT_LP64(__ mov(rcx, rax));
2266       // We use j_rarg definitions here so that registers don't conflict as parameter
2267       // registers change across platforms as we are in the midst of a calling
2268       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2269 
2270       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2271       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2272 
2273       // pop the interpreter frame
2274       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2275       __ leave();                                // remove frame anchor
2276       __ pop(retaddr);                           // get return address
2277       __ mov(rsp, sender_sp);                   // set sp to sender sp
2278       // Ensure compiled code always sees stack at proper alignment
2279       __ andptr(rsp, -(StackAlignmentInBytes));
2280 
2281       // unlike x86 we need no specialized return from compiled code
2282       // to the interpreter or the call stub.
2283 
2284       // push the return address
2285       __ push(retaddr);
2286 
2287       // and begin the OSR nmethod
2288       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2289     }
2290   }
2291 }
2292 
2293 void TemplateTable::if_0cmp(Condition cc) {
2294   transition(itos, vtos);
2295   // assume branch is more often taken than not (loops use backward branches)
2296   Label not_taken;
2297   __ testl(rax, rax);
2298   __ jcc(j_not(cc), not_taken);
2299   branch(false, false);
2300   __ bind(not_taken);
2301   __ profile_not_taken_branch(rax);
2302 }
2303 
2304 void TemplateTable::if_icmp(Condition cc) {
2305   transition(itos, vtos);
2306   // assume branch is more often taken than not (loops use backward branches)
2307   Label not_taken;
2308   __ pop_i(rdx);
2309   __ cmpl(rdx, rax);
2310   __ jcc(j_not(cc), not_taken);
2311   branch(false, false);
2312   __ bind(not_taken);
2313   __ profile_not_taken_branch(rax);
2314 }
2315 
2316 void TemplateTable::if_nullcmp(Condition cc) {
2317   transition(atos, vtos);
2318   // assume branch is more often taken than not (loops use backward branches)
2319   Label not_taken;
2320   __ testptr(rax, rax);
2321   __ jcc(j_not(cc), not_taken);
2322   branch(false, false);
2323   __ bind(not_taken);
2324   __ profile_not_taken_branch(rax);
2325 }
2326 
2327 void TemplateTable::if_acmp(Condition cc) {
2328   transition(atos, vtos);
2329   // assume branch is more often taken than not (loops use backward branches)
2330   Label not_taken;
2331   __ pop_ptr(rdx);
2332   __ cmpoop(rdx, rax);
2333   __ jcc(j_not(cc), not_taken);
2334   branch(false, false);
2335   __ bind(not_taken);
2336   __ profile_not_taken_branch(rax);
2337 }
2338 
2339 void TemplateTable::ret() {
2340   transition(vtos, vtos);
2341   locals_index(rbx);
2342   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2343   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2344   __ profile_ret(rbx, rcx);
2345   __ get_method(rax);
2346   __ movptr(rbcp, Address(rax, Method::const_offset()));
2347   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2348                       ConstMethod::codes_offset()));
2349   __ dispatch_next(vtos, 0, true);
2350 }
2351 
2352 void TemplateTable::wide_ret() {
2353   transition(vtos, vtos);
2354   locals_index_wide(rbx);
2355   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2356   __ profile_ret(rbx, rcx);
2357   __ get_method(rax);
2358   __ movptr(rbcp, Address(rax, Method::const_offset()));
2359   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2360   __ dispatch_next(vtos, 0, true);
2361 }
2362 
2363 void TemplateTable::tableswitch() {
2364   Label default_case, continue_execution;
2365   transition(itos, vtos);
2366 
2367   // align r13/rsi
2368   __ lea(rbx, at_bcp(BytesPerInt));
2369   __ andptr(rbx, -BytesPerInt);
2370   // load lo & hi
2371   __ movl(rcx, Address(rbx, BytesPerInt));
2372   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2373   __ bswapl(rcx);
2374   __ bswapl(rdx);
2375   // check against lo & hi
2376   __ cmpl(rax, rcx);
2377   __ jcc(Assembler::less, default_case);
2378   __ cmpl(rax, rdx);
2379   __ jcc(Assembler::greater, default_case);
2380   // lookup dispatch offset
2381   __ subl(rax, rcx);
2382   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2383   __ profile_switch_case(rax, rbx, rcx);
2384   // continue execution
2385   __ bind(continue_execution);
2386   __ bswapl(rdx);
2387   LP64_ONLY(__ movl2ptr(rdx, rdx));
2388   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2389   __ addptr(rbcp, rdx);
2390   __ dispatch_only(vtos, true);
2391   // handle default
2392   __ bind(default_case);
2393   __ profile_switch_default(rax);
2394   __ movl(rdx, Address(rbx, 0));
2395   __ jmp(continue_execution);
2396 }
2397 
2398 void TemplateTable::lookupswitch() {
2399   transition(itos, itos);
2400   __ stop("lookupswitch bytecode should have been rewritten");
2401 }
2402 
2403 void TemplateTable::fast_linearswitch() {
2404   transition(itos, vtos);
2405   Label loop_entry, loop, found, continue_execution;
2406   // bswap rax so we can avoid bswapping the table entries
2407   __ bswapl(rax);
2408   // align r13
2409   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2410                                     // this instruction (change offsets
2411                                     // below)
2412   __ andptr(rbx, -BytesPerInt);
2413   // set counter
2414   __ movl(rcx, Address(rbx, BytesPerInt));
2415   __ bswapl(rcx);
2416   __ jmpb(loop_entry);
2417   // table search
2418   __ bind(loop);
2419   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2420   __ jcc(Assembler::equal, found);
2421   __ bind(loop_entry);
2422   __ decrementl(rcx);
2423   __ jcc(Assembler::greaterEqual, loop);
2424   // default case
2425   __ profile_switch_default(rax);
2426   __ movl(rdx, Address(rbx, 0));
2427   __ jmp(continue_execution);
2428   // entry found -> get offset
2429   __ bind(found);
2430   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2431   __ profile_switch_case(rcx, rax, rbx);
2432   // continue execution
2433   __ bind(continue_execution);
2434   __ bswapl(rdx);
2435   __ movl2ptr(rdx, rdx);
2436   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2437   __ addptr(rbcp, rdx);
2438   __ dispatch_only(vtos, true);
2439 }
2440 
2441 void TemplateTable::fast_binaryswitch() {
2442   transition(itos, vtos);
2443   // Implementation using the following core algorithm:
2444   //
2445   // int binary_search(int key, LookupswitchPair* array, int n) {
2446   //   // Binary search according to "Methodik des Programmierens" by
2447   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2448   //   int i = 0;
2449   //   int j = n;
2450   //   while (i+1 < j) {
2451   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2452   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2453   //     // where a stands for the array and assuming that the (inexisting)
2454   //     // element a[n] is infinitely big.
2455   //     int h = (i + j) >> 1;
2456   //     // i < h < j
2457   //     if (key < array[h].fast_match()) {
2458   //       j = h;
2459   //     } else {
2460   //       i = h;
2461   //     }
2462   //   }
2463   //   // R: a[i] <= key < a[i+1] or Q
2464   //   // (i.e., if key is within array, i is the correct index)
2465   //   return i;
2466   // }
2467 
2468   // Register allocation
2469   const Register key   = rax; // already set (tosca)
2470   const Register array = rbx;
2471   const Register i     = rcx;
2472   const Register j     = rdx;
2473   const Register h     = rdi;
2474   const Register temp  = rsi;
2475 
2476   // Find array start
2477   NOT_LP64(__ save_bcp());
2478 
2479   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2480                                           // get rid of this
2481                                           // instruction (change
2482                                           // offsets below)
2483   __ andptr(array, -BytesPerInt);
2484 
2485   // Initialize i & j
2486   __ xorl(i, i);                            // i = 0;
2487   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2488 
2489   // Convert j into native byteordering
2490   __ bswapl(j);
2491 
2492   // And start
2493   Label entry;
2494   __ jmp(entry);
2495 
2496   // binary search loop
2497   {
2498     Label loop;
2499     __ bind(loop);
2500     // int h = (i + j) >> 1;
2501     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2502     __ sarl(h, 1);                               // h = (i + j) >> 1;
2503     // if (key < array[h].fast_match()) {
2504     //   j = h;
2505     // } else {
2506     //   i = h;
2507     // }
2508     // Convert array[h].match to native byte-ordering before compare
2509     __ movl(temp, Address(array, h, Address::times_8));
2510     __ bswapl(temp);
2511     __ cmpl(key, temp);
2512     // j = h if (key <  array[h].fast_match())
2513     __ cmov32(Assembler::less, j, h);
2514     // i = h if (key >= array[h].fast_match())
2515     __ cmov32(Assembler::greaterEqual, i, h);
2516     // while (i+1 < j)
2517     __ bind(entry);
2518     __ leal(h, Address(i, 1)); // i+1
2519     __ cmpl(h, j);             // i+1 < j
2520     __ jcc(Assembler::less, loop);
2521   }
2522 
2523   // end of binary search, result index is i (must check again!)
2524   Label default_case;
2525   // Convert array[i].match to native byte-ordering before compare
2526   __ movl(temp, Address(array, i, Address::times_8));
2527   __ bswapl(temp);
2528   __ cmpl(key, temp);
2529   __ jcc(Assembler::notEqual, default_case);
2530 
2531   // entry found -> j = offset
2532   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2533   __ profile_switch_case(i, key, array);
2534   __ bswapl(j);
2535   LP64_ONLY(__ movslq(j, j));
2536 
2537   NOT_LP64(__ restore_bcp());
2538   NOT_LP64(__ restore_locals());                           // restore rdi
2539 
2540   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2541   __ addptr(rbcp, j);
2542   __ dispatch_only(vtos, true);
2543 
2544   // default case -> j = default offset
2545   __ bind(default_case);
2546   __ profile_switch_default(i);
2547   __ movl(j, Address(array, -2 * BytesPerInt));
2548   __ bswapl(j);
2549   LP64_ONLY(__ movslq(j, j));
2550 
2551   NOT_LP64(__ restore_bcp());
2552   NOT_LP64(__ restore_locals());
2553 
2554   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2555   __ addptr(rbcp, j);
2556   __ dispatch_only(vtos, true);
2557 }
2558 
2559 void TemplateTable::_return(TosState state) {
2560   transition(state, state);
2561 
2562   assert(_desc->calls_vm(),
2563          "inconsistent calls_vm information"); // call in remove_activation
2564 
2565   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2566     assert(state == vtos, "only valid state");
2567     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2568     __ movptr(robj, aaddress(0));
2569     __ load_klass(rdi, robj, rscratch1);
2570     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2571     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2572     Label skip_register_finalizer;
2573     __ jcc(Assembler::zero, skip_register_finalizer);
2574 
2575     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2576 
2577     __ bind(skip_register_finalizer);
2578   }
2579 
2580   if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2581     Label no_safepoint;
2582     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2583 #ifdef _LP64
2584     __ testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2585 #else
2586     const Register thread = rdi;
2587     __ get_thread(thread);
2588     __ testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2589 #endif
2590     __ jcc(Assembler::zero, no_safepoint);
2591     __ push(state);
2592     __ push_cont_fastpath();
2593     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2594                                        InterpreterRuntime::at_safepoint));
2595     __ pop_cont_fastpath();
2596     __ pop(state);
2597     __ bind(no_safepoint);
2598   }
2599 
2600   // Narrow result if state is itos but result type is smaller.
2601   // Need to narrow in the return bytecode rather than in generate_return_entry
2602   // since compiled code callers expect the result to already be narrowed.
2603   if (state == itos) {
2604     __ narrow(rax);
2605   }
2606   __ remove_activation(state, rbcp);
2607 
2608   __ jmp(rbcp);
2609 }
2610 
2611 // ----------------------------------------------------------------------------
2612 // Volatile variables demand their effects be made known to all CPU's
2613 // in order.  Store buffers on most chips allow reads & writes to
2614 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2615 // without some kind of memory barrier (i.e., it's not sufficient that
2616 // the interpreter does not reorder volatile references, the hardware
2617 // also must not reorder them).
2618 //
2619 // According to the new Java Memory Model (JMM):
2620 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2621 //     writes act as acquire & release, so:
2622 // (2) A read cannot let unrelated NON-volatile memory refs that
2623 //     happen after the read float up to before the read.  It's OK for
2624 //     non-volatile memory refs that happen before the volatile read to
2625 //     float down below it.
2626 // (3) Similar a volatile write cannot let unrelated NON-volatile
2627 //     memory refs that happen BEFORE the write float down to after the
2628 //     write.  It's OK for non-volatile memory refs that happen after the
2629 //     volatile write to float up before it.
2630 //
2631 // We only put in barriers around volatile refs (they are expensive),
2632 // not _between_ memory refs (that would require us to track the
2633 // flavor of the previous memory refs).  Requirements (2) and (3)
2634 // require some barriers before volatile stores and after volatile
2635 // loads.  These nearly cover requirement (1) but miss the
2636 // volatile-store-volatile-load case.  This final case is placed after
2637 // volatile-stores although it could just as well go before
2638 // volatile-loads.
2639 
2640 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2641   // Helper function to insert a is-volatile test and memory barrier
2642   __ membar(order_constraint);
2643 }
2644 
2645 void TemplateTable::resolve_cache_and_index(int byte_no,
2646                                             Register cache,
2647                                             Register index,
2648                                             size_t index_size) {
2649   const Register temp = rbx;
2650   assert_different_registers(cache, index, temp);
2651 
2652   Label L_clinit_barrier_slow;
2653   Label resolved;
2654 
2655   Bytecodes::Code code = bytecode();
2656   switch (code) {
2657   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2658   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2659   default: break;
2660   }
2661 
2662   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2663   __ get_cache_and_index_and_bytecode_at_bcp(cache, index, temp, byte_no, 1, index_size);
2664   __ cmpl(temp, code);  // have we resolved this bytecode?
2665   __ jcc(Assembler::equal, resolved);
2666 
2667   // resolve first time through
2668   // Class initialization barrier slow path lands here as well.
2669   __ bind(L_clinit_barrier_slow);
2670   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2671   __ movl(temp, code);
2672   __ call_VM(noreg, entry, temp);
2673   // Update registers with resolved info
2674   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
2675 
2676   __ bind(resolved);
2677 
2678   // Class initialization barrier for static methods
2679   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2680     const Register method = temp;
2681     const Register klass  = temp;
2682     const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2683     assert(thread != noreg, "x86_32 not supported");
2684 
2685     __ load_resolved_method_at_index(byte_no, method, cache, index);
2686     __ load_method_holder(klass, method);
2687     __ clinit_barrier(klass, thread, NULL /*L_fast_path*/, &L_clinit_barrier_slow);
2688   }
2689 }
2690 
2691 // The cache and index registers must be set before call
2692 void TemplateTable::load_field_cp_cache_entry(Register obj,
2693                                               Register cache,
2694                                               Register index,
2695                                               Register off,
2696                                               Register flags,
2697                                               bool is_static = false) {
2698   assert_different_registers(cache, index, flags, off);
2699 
2700   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2701   // Field offset
2702   __ movptr(off, Address(cache, index, Address::times_ptr,
2703                          in_bytes(cp_base_offset +
2704                                   ConstantPoolCacheEntry::f2_offset())));
2705   // Flags
2706   __ movl(flags, Address(cache, index, Address::times_ptr,
2707                          in_bytes(cp_base_offset +
2708                                   ConstantPoolCacheEntry::flags_offset())));
2709 
2710   // klass overwrite register
2711   if (is_static) {
2712     __ movptr(obj, Address(cache, index, Address::times_ptr,
2713                            in_bytes(cp_base_offset +
2714                                     ConstantPoolCacheEntry::f1_offset())));
2715     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2716     __ movptr(obj, Address(obj, mirror_offset));
2717     __ resolve_oop_handle(obj, rscratch2);
2718   }
2719 }
2720 
2721 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2722                                                Register method,
2723                                                Register itable_index,
2724                                                Register flags,
2725                                                bool is_invokevirtual,
2726                                                bool is_invokevfinal, /*unused*/
2727                                                bool is_invokedynamic) {
2728   // setup registers
2729   const Register cache = rcx;
2730   const Register index = rdx;
2731   assert_different_registers(method, flags);
2732   assert_different_registers(method, cache, index);
2733   assert_different_registers(itable_index, flags);
2734   assert_different_registers(itable_index, cache, index);
2735   // determine constant pool cache field offsets
2736   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2737   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2738                                     ConstantPoolCacheEntry::flags_offset());
2739   // access constant pool cache fields
2740   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2741                                     ConstantPoolCacheEntry::f2_offset());
2742 
2743   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2744   resolve_cache_and_index(byte_no, cache, index, index_size);
2745   __ load_resolved_method_at_index(byte_no, method, cache, index);
2746 
2747   if (itable_index != noreg) {
2748     // pick up itable or appendix index from f2 also:
2749     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2750   }
2751   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2752 }
2753 
2754 // The registers cache and index expected to be set before call.
2755 // Correct values of the cache and index registers are preserved.
2756 void TemplateTable::jvmti_post_field_access(Register cache,
2757                                             Register index,
2758                                             bool is_static,
2759                                             bool has_tos) {
2760   if (JvmtiExport::can_post_field_access()) {
2761     // Check to see if a field access watch has been set before we take
2762     // the time to call into the VM.
2763     Label L1;
2764     assert_different_registers(cache, index, rax);
2765     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2766     __ testl(rax,rax);
2767     __ jcc(Assembler::zero, L1);
2768 
2769     // cache entry pointer
2770     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2771     __ shll(index, LogBytesPerWord);
2772     __ addptr(cache, index);
2773     if (is_static) {
2774       __ xorptr(rax, rax);      // NULL object reference
2775     } else {
2776       __ pop(atos);         // Get the object
2777       __ verify_oop(rax);
2778       __ push(atos);        // Restore stack state
2779     }
2780     // rax,:   object pointer or NULL
2781     // cache: cache entry pointer
2782     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2783                rax, cache);
2784     __ get_cache_and_index_at_bcp(cache, index, 1);
2785     __ bind(L1);
2786   }
2787 }
2788 
2789 void TemplateTable::pop_and_check_object(Register r) {
2790   __ pop_ptr(r);
2791   __ null_check(r);  // for field access must check obj.
2792   __ verify_oop(r);
2793 }
2794 
2795 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2796   transition(vtos, vtos);
2797 
2798   const Register cache = rcx;
2799   const Register index = rdx;
2800   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2801   const Register off   = rbx;
2802   const Register flags = rax;
2803   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2804 
2805   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2806   jvmti_post_field_access(cache, index, is_static, false);
2807   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2808 
2809   if (!is_static) pop_and_check_object(obj);
2810 
2811   const Address field(obj, off, Address::times_1, 0*wordSize);
2812 
2813   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2814 
2815   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2816   // Make sure we don't need to mask edx after the above shift
2817   assert(btos == 0, "change code, btos != 0");
2818 
2819   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2820 
2821   __ jcc(Assembler::notZero, notByte);
2822   // btos
2823   __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2824   __ push(btos);
2825   // Rewrite bytecode to be faster
2826   if (!is_static && rc == may_rewrite) {
2827     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2828   }
2829   __ jmp(Done);
2830 
2831   __ bind(notByte);
2832   __ cmpl(flags, ztos);
2833   __ jcc(Assembler::notEqual, notBool);
2834 
2835   // ztos (same code as btos)
2836   __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2837   __ push(ztos);
2838   // Rewrite bytecode to be faster
2839   if (!is_static && rc == may_rewrite) {
2840     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2841     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2842   }
2843   __ jmp(Done);
2844 
2845   __ bind(notBool);
2846   __ cmpl(flags, atos);
2847   __ jcc(Assembler::notEqual, notObj);
2848   // atos
2849   do_oop_load(_masm, field, rax);
2850   __ push(atos);
2851   if (!is_static && rc == may_rewrite) {
2852     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2853   }
2854   __ jmp(Done);
2855 
2856   __ bind(notObj);
2857   __ cmpl(flags, itos);
2858   __ jcc(Assembler::notEqual, notInt);
2859   // itos
2860   __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2861   __ push(itos);
2862   // Rewrite bytecode to be faster
2863   if (!is_static && rc == may_rewrite) {
2864     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2865   }
2866   __ jmp(Done);
2867 
2868   __ bind(notInt);
2869   __ cmpl(flags, ctos);
2870   __ jcc(Assembler::notEqual, notChar);
2871   // ctos
2872   __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2873   __ push(ctos);
2874   // Rewrite bytecode to be faster
2875   if (!is_static && rc == may_rewrite) {
2876     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2877   }
2878   __ jmp(Done);
2879 
2880   __ bind(notChar);
2881   __ cmpl(flags, stos);
2882   __ jcc(Assembler::notEqual, notShort);
2883   // stos
2884   __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2885   __ push(stos);
2886   // Rewrite bytecode to be faster
2887   if (!is_static && rc == may_rewrite) {
2888     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2889   }
2890   __ jmp(Done);
2891 
2892   __ bind(notShort);
2893   __ cmpl(flags, ltos);
2894   __ jcc(Assembler::notEqual, notLong);
2895   // ltos
2896     // Generate code as if volatile (x86_32).  There just aren't enough registers to
2897     // save that information and this code is faster than the test.
2898   __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2899   __ push(ltos);
2900   // Rewrite bytecode to be faster
2901   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2902   __ jmp(Done);
2903 
2904   __ bind(notLong);
2905   __ cmpl(flags, ftos);
2906   __ jcc(Assembler::notEqual, notFloat);
2907   // ftos
2908 
2909   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2910   __ push(ftos);
2911   // Rewrite bytecode to be faster
2912   if (!is_static && rc == may_rewrite) {
2913     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2914   }
2915   __ jmp(Done);
2916 
2917   __ bind(notFloat);
2918 #ifdef ASSERT
2919   Label notDouble;
2920   __ cmpl(flags, dtos);
2921   __ jcc(Assembler::notEqual, notDouble);
2922 #endif
2923   // dtos
2924   // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
2925   __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
2926   __ push(dtos);
2927   // Rewrite bytecode to be faster
2928   if (!is_static && rc == may_rewrite) {
2929     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2930   }
2931 #ifdef ASSERT
2932   __ jmp(Done);
2933 
2934   __ bind(notDouble);
2935   __ stop("Bad state");
2936 #endif
2937 
2938   __ bind(Done);
2939   // [jk] not needed currently
2940   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2941   //                                              Assembler::LoadStore));
2942 }
2943 
2944 void TemplateTable::getfield(int byte_no) {
2945   getfield_or_static(byte_no, false);
2946 }
2947 
2948 void TemplateTable::nofast_getfield(int byte_no) {
2949   getfield_or_static(byte_no, false, may_not_rewrite);
2950 }
2951 
2952 void TemplateTable::getstatic(int byte_no) {
2953   getfield_or_static(byte_no, true);
2954 }
2955 
2956 
2957 // The registers cache and index expected to be set before call.
2958 // The function may destroy various registers, just not the cache and index registers.
2959 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2960 
2961   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
2962   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
2963   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
2964   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2965 
2966   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2967 
2968   if (JvmtiExport::can_post_field_modification()) {
2969     // Check to see if a field modification watch has been set before
2970     // we take the time to call into the VM.
2971     Label L1;
2972     assert_different_registers(cache, index, rax);
2973     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2974     __ testl(rax, rax);
2975     __ jcc(Assembler::zero, L1);
2976 
2977     __ get_cache_and_index_at_bcp(robj, RDX, 1);
2978 
2979 
2980     if (is_static) {
2981       // Life is simple.  Null out the object pointer.
2982       __ xorl(RBX, RBX);
2983 
2984     } else {
2985       // Life is harder. The stack holds the value on top, followed by
2986       // the object.  We don't know the size of the value, though; it
2987       // could be one or two words depending on its type. As a result,
2988       // we must find the type to determine where the object is.
2989 #ifndef _LP64
2990       Label two_word, valsize_known;
2991 #endif
2992       __ movl(RCX, Address(robj, RDX,
2993                            Address::times_ptr,
2994                            in_bytes(cp_base_offset +
2995                                      ConstantPoolCacheEntry::flags_offset())));
2996       NOT_LP64(__ mov(rbx, rsp));
2997       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2998 
2999       // Make sure we don't need to mask rcx after the above shift
3000       ConstantPoolCacheEntry::verify_tos_state_shift();
3001 #ifdef _LP64
3002       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3003       __ cmpl(c_rarg3, ltos);
3004       __ cmovptr(Assembler::equal,
3005                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3006       __ cmpl(c_rarg3, dtos);
3007       __ cmovptr(Assembler::equal,
3008                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3009 #else
3010       __ cmpl(rcx, ltos);
3011       __ jccb(Assembler::equal, two_word);
3012       __ cmpl(rcx, dtos);
3013       __ jccb(Assembler::equal, two_word);
3014       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3015       __ jmpb(valsize_known);
3016 
3017       __ bind(two_word);
3018       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3019 
3020       __ bind(valsize_known);
3021       // setup object pointer
3022       __ movptr(rbx, Address(rbx, 0));
3023 #endif
3024     }
3025     // cache entry pointer
3026     __ addptr(robj, in_bytes(cp_base_offset));
3027     __ shll(RDX, LogBytesPerWord);
3028     __ addptr(robj, RDX);
3029     // object (tos)
3030     __ mov(RCX, rsp);
3031     // c_rarg1: object pointer set up above (NULL if static)
3032     // c_rarg2: cache entry pointer
3033     // c_rarg3: jvalue object on the stack
3034     __ call_VM(noreg,
3035                CAST_FROM_FN_PTR(address,
3036                                 InterpreterRuntime::post_field_modification),
3037                RBX, robj, RCX);
3038     __ get_cache_and_index_at_bcp(cache, index, 1);
3039     __ bind(L1);
3040   }
3041 }
3042 
3043 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3044   transition(vtos, vtos);
3045 
3046   const Register cache = rcx;
3047   const Register index = rdx;
3048   const Register obj   = rcx;
3049   const Register off   = rbx;
3050   const Register flags = rax;
3051 
3052   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3053   jvmti_post_field_mod(cache, index, is_static);
3054   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3055 
3056   // [jk] not needed currently
3057   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3058   //                                              Assembler::StoreStore));
3059 
3060   Label notVolatile, Done;
3061   __ movl(rdx, flags);
3062   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3063   __ andl(rdx, 0x1);
3064 
3065   // Check for volatile store
3066   __ testl(rdx, rdx);
3067   __ jcc(Assembler::zero, notVolatile);
3068 
3069   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3070   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3071                                                Assembler::StoreStore));
3072   __ jmp(Done);
3073   __ bind(notVolatile);
3074 
3075   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3076 
3077   __ bind(Done);
3078 }
3079 
3080 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3081                                               Register obj, Register off, Register flags) {
3082 
3083   // field addresses
3084   const Address field(obj, off, Address::times_1, 0*wordSize);
3085   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3086 
3087   Label notByte, notBool, notInt, notShort, notChar,
3088         notLong, notFloat, notObj;
3089   Label Done;
3090 
3091   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3092 
3093   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3094 
3095   assert(btos == 0, "change code, btos != 0");
3096   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3097   __ jcc(Assembler::notZero, notByte);
3098 
3099   // btos
3100   {
3101     __ pop(btos);
3102     if (!is_static) pop_and_check_object(obj);
3103     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3104     if (!is_static && rc == may_rewrite) {
3105       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3106     }
3107     __ jmp(Done);
3108   }
3109 
3110   __ bind(notByte);
3111   __ cmpl(flags, ztos);
3112   __ jcc(Assembler::notEqual, notBool);
3113 
3114   // ztos
3115   {
3116     __ pop(ztos);
3117     if (!is_static) pop_and_check_object(obj);
3118     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3119     if (!is_static && rc == may_rewrite) {
3120       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3121     }
3122     __ jmp(Done);
3123   }
3124 
3125   __ bind(notBool);
3126   __ cmpl(flags, atos);
3127   __ jcc(Assembler::notEqual, notObj);
3128 
3129   // atos
3130   {
3131     __ pop(atos);
3132     if (!is_static) pop_and_check_object(obj);
3133     // Store into the field
3134     do_oop_store(_masm, field, rax);
3135     if (!is_static && rc == may_rewrite) {
3136       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3137     }
3138     __ jmp(Done);
3139   }
3140 
3141   __ bind(notObj);
3142   __ cmpl(flags, itos);
3143   __ jcc(Assembler::notEqual, notInt);
3144 
3145   // itos
3146   {
3147     __ pop(itos);
3148     if (!is_static) pop_and_check_object(obj);
3149     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3150     if (!is_static && rc == may_rewrite) {
3151       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3152     }
3153     __ jmp(Done);
3154   }
3155 
3156   __ bind(notInt);
3157   __ cmpl(flags, ctos);
3158   __ jcc(Assembler::notEqual, notChar);
3159 
3160   // ctos
3161   {
3162     __ pop(ctos);
3163     if (!is_static) pop_and_check_object(obj);
3164     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3165     if (!is_static && rc == may_rewrite) {
3166       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3167     }
3168     __ jmp(Done);
3169   }
3170 
3171   __ bind(notChar);
3172   __ cmpl(flags, stos);
3173   __ jcc(Assembler::notEqual, notShort);
3174 
3175   // stos
3176   {
3177     __ pop(stos);
3178     if (!is_static) pop_and_check_object(obj);
3179     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3180     if (!is_static && rc == may_rewrite) {
3181       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3182     }
3183     __ jmp(Done);
3184   }
3185 
3186   __ bind(notShort);
3187   __ cmpl(flags, ltos);
3188   __ jcc(Assembler::notEqual, notLong);
3189 
3190   // ltos
3191   {
3192     __ pop(ltos);
3193     if (!is_static) pop_and_check_object(obj);
3194     // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3195     __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg, noreg);
3196 #ifdef _LP64
3197     if (!is_static && rc == may_rewrite) {
3198       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3199     }
3200 #endif // _LP64
3201     __ jmp(Done);
3202   }
3203 
3204   __ bind(notLong);
3205   __ cmpl(flags, ftos);
3206   __ jcc(Assembler::notEqual, notFloat);
3207 
3208   // ftos
3209   {
3210     __ pop(ftos);
3211     if (!is_static) pop_and_check_object(obj);
3212     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3213     if (!is_static && rc == may_rewrite) {
3214       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3215     }
3216     __ jmp(Done);
3217   }
3218 
3219   __ bind(notFloat);
3220 #ifdef ASSERT
3221   Label notDouble;
3222   __ cmpl(flags, dtos);
3223   __ jcc(Assembler::notEqual, notDouble);
3224 #endif
3225 
3226   // dtos
3227   {
3228     __ pop(dtos);
3229     if (!is_static) pop_and_check_object(obj);
3230     // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3231     __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg, noreg);
3232     if (!is_static && rc == may_rewrite) {
3233       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3234     }
3235   }
3236 
3237 #ifdef ASSERT
3238   __ jmp(Done);
3239 
3240   __ bind(notDouble);
3241   __ stop("Bad state");
3242 #endif
3243 
3244   __ bind(Done);
3245 }
3246 
3247 void TemplateTable::putfield(int byte_no) {
3248   putfield_or_static(byte_no, false);
3249 }
3250 
3251 void TemplateTable::nofast_putfield(int byte_no) {
3252   putfield_or_static(byte_no, false, may_not_rewrite);
3253 }
3254 
3255 void TemplateTable::putstatic(int byte_no) {
3256   putfield_or_static(byte_no, true);
3257 }
3258 
3259 void TemplateTable::jvmti_post_fast_field_mod() {
3260 
3261   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3262 
3263   if (JvmtiExport::can_post_field_modification()) {
3264     // Check to see if a field modification watch has been set before
3265     // we take the time to call into the VM.
3266     Label L2;
3267     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3268     __ testl(scratch, scratch);
3269     __ jcc(Assembler::zero, L2);
3270     __ pop_ptr(rbx);                  // copy the object pointer from tos
3271     __ verify_oop(rbx);
3272     __ push_ptr(rbx);                 // put the object pointer back on tos
3273     // Save tos values before call_VM() clobbers them. Since we have
3274     // to do it for every data type, we use the saved values as the
3275     // jvalue object.
3276     switch (bytecode()) {          // load values into the jvalue object
3277     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3278     case Bytecodes::_fast_bputfield: // fall through
3279     case Bytecodes::_fast_zputfield: // fall through
3280     case Bytecodes::_fast_sputfield: // fall through
3281     case Bytecodes::_fast_cputfield: // fall through
3282     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3283     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3284     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3285     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3286 
3287     default:
3288       ShouldNotReachHere();
3289     }
3290     __ mov(scratch, rsp);             // points to jvalue on the stack
3291     // access constant pool cache entry
3292     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3293     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3294     __ verify_oop(rbx);
3295     // rbx: object pointer copied above
3296     // c_rarg2: cache entry pointer
3297     // c_rarg3: jvalue object on the stack
3298     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3299     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3300 
3301     switch (bytecode()) {             // restore tos values
3302     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3303     case Bytecodes::_fast_bputfield: // fall through
3304     case Bytecodes::_fast_zputfield: // fall through
3305     case Bytecodes::_fast_sputfield: // fall through
3306     case Bytecodes::_fast_cputfield: // fall through
3307     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3308     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3309     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3310     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3311     default: break;
3312     }
3313     __ bind(L2);
3314   }
3315 }
3316 
3317 void TemplateTable::fast_storefield(TosState state) {
3318   transition(state, vtos);
3319 
3320   ByteSize base = ConstantPoolCache::base_offset();
3321 
3322   jvmti_post_fast_field_mod();
3323 
3324   // access constant pool cache
3325   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3326 
3327   // test for volatile with rdx but rdx is tos register for lputfield.
3328   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3329                        in_bytes(base +
3330                                 ConstantPoolCacheEntry::flags_offset())));
3331 
3332   // replace index with field offset from cache entry
3333   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3334                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3335 
3336   // [jk] not needed currently
3337   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3338   //                                              Assembler::StoreStore));
3339 
3340   Label notVolatile, Done;
3341   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3342   __ andl(rdx, 0x1);
3343 
3344   // Get object from stack
3345   pop_and_check_object(rcx);
3346 
3347   // field address
3348   const Address field(rcx, rbx, Address::times_1);
3349 
3350   // Check for volatile store
3351   __ testl(rdx, rdx);
3352   __ jcc(Assembler::zero, notVolatile);
3353 
3354   fast_storefield_helper(field, rax);
3355   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3356                                                Assembler::StoreStore));
3357   __ jmp(Done);
3358   __ bind(notVolatile);
3359 
3360   fast_storefield_helper(field, rax);
3361 
3362   __ bind(Done);
3363 }
3364 
3365 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3366 
3367   // access field
3368   switch (bytecode()) {
3369   case Bytecodes::_fast_aputfield:
3370     do_oop_store(_masm, field, rax);
3371     break;
3372   case Bytecodes::_fast_lputfield:
3373 #ifdef _LP64
3374     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg, noreg);
3375 #else
3376   __ stop("should not be rewritten");
3377 #endif
3378     break;
3379   case Bytecodes::_fast_iputfield:
3380     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3381     break;
3382   case Bytecodes::_fast_zputfield:
3383     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3384     break;
3385   case Bytecodes::_fast_bputfield:
3386     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3387     break;
3388   case Bytecodes::_fast_sputfield:
3389     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3390     break;
3391   case Bytecodes::_fast_cputfield:
3392     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3393     break;
3394   case Bytecodes::_fast_fputfield:
3395     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg, noreg);
3396     break;
3397   case Bytecodes::_fast_dputfield:
3398     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg, noreg);
3399     break;
3400   default:
3401     ShouldNotReachHere();
3402   }
3403 }
3404 
3405 void TemplateTable::fast_accessfield(TosState state) {
3406   transition(atos, state);
3407 
3408   // Do the JVMTI work here to avoid disturbing the register state below
3409   if (JvmtiExport::can_post_field_access()) {
3410     // Check to see if a field access watch has been set before we
3411     // take the time to call into the VM.
3412     Label L1;
3413     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3414     __ testl(rcx, rcx);
3415     __ jcc(Assembler::zero, L1);
3416     // access constant pool cache entry
3417     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3418     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3419     __ verify_oop(rax);
3420     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3421     LP64_ONLY(__ mov(c_rarg1, rax));
3422     // c_rarg1: object pointer copied above
3423     // c_rarg2: cache entry pointer
3424     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3425     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3426     __ pop_ptr(rax); // restore object pointer
3427     __ bind(L1);
3428   }
3429 
3430   // access constant pool cache
3431   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3432   // replace index with field offset from cache entry
3433   // [jk] not needed currently
3434   // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3435   //                      in_bytes(ConstantPoolCache::base_offset() +
3436   //                               ConstantPoolCacheEntry::flags_offset())));
3437   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3438   // __ andl(rdx, 0x1);
3439   //
3440   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3441                          in_bytes(ConstantPoolCache::base_offset() +
3442                                   ConstantPoolCacheEntry::f2_offset())));
3443 
3444   // rax: object
3445   __ verify_oop(rax);
3446   __ null_check(rax);
3447   Address field(rax, rbx, Address::times_1);
3448 
3449   // access field
3450   switch (bytecode()) {
3451   case Bytecodes::_fast_agetfield:
3452     do_oop_load(_masm, field, rax);
3453     __ verify_oop(rax);
3454     break;
3455   case Bytecodes::_fast_lgetfield:
3456 #ifdef _LP64
3457     __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3458 #else
3459   __ stop("should not be rewritten");
3460 #endif
3461     break;
3462   case Bytecodes::_fast_igetfield:
3463     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3464     break;
3465   case Bytecodes::_fast_bgetfield:
3466     __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3467     break;
3468   case Bytecodes::_fast_sgetfield:
3469     __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3470     break;
3471   case Bytecodes::_fast_cgetfield:
3472     __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3473     break;
3474   case Bytecodes::_fast_fgetfield:
3475     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3476     break;
3477   case Bytecodes::_fast_dgetfield:
3478     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3479     break;
3480   default:
3481     ShouldNotReachHere();
3482   }
3483   // [jk] not needed currently
3484   //   Label notVolatile;
3485   //   __ testl(rdx, rdx);
3486   //   __ jcc(Assembler::zero, notVolatile);
3487   //   __ membar(Assembler::LoadLoad);
3488   //   __ bind(notVolatile);
3489 }
3490 
3491 void TemplateTable::fast_xaccess(TosState state) {
3492   transition(vtos, state);
3493 
3494   // get receiver
3495   __ movptr(rax, aaddress(0));
3496   // access constant pool cache
3497   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3498   __ movptr(rbx,
3499             Address(rcx, rdx, Address::times_ptr,
3500                     in_bytes(ConstantPoolCache::base_offset() +
3501                              ConstantPoolCacheEntry::f2_offset())));
3502   // make sure exception is reported in correct bcp range (getfield is
3503   // next instruction)
3504   __ increment(rbcp);
3505   __ null_check(rax);
3506   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3507   switch (state) {
3508   case itos:
3509     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3510     break;
3511   case atos:
3512     do_oop_load(_masm, field, rax);
3513     __ verify_oop(rax);
3514     break;
3515   case ftos:
3516     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3517     break;
3518   default:
3519     ShouldNotReachHere();
3520   }
3521 
3522   // [jk] not needed currently
3523   // Label notVolatile;
3524   // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3525   //                      in_bytes(ConstantPoolCache::base_offset() +
3526   //                               ConstantPoolCacheEntry::flags_offset())));
3527   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3528   // __ testl(rdx, 0x1);
3529   // __ jcc(Assembler::zero, notVolatile);
3530   // __ membar(Assembler::LoadLoad);
3531   // __ bind(notVolatile);
3532 
3533   __ decrement(rbcp);
3534 }
3535 
3536 //-----------------------------------------------------------------------------
3537 // Calls
3538 
3539 void TemplateTable::prepare_invoke(int byte_no,
3540                                    Register method,  // linked method (or i-klass)
3541                                    Register index,   // itable index, MethodType, etc.
3542                                    Register recv,    // if caller wants to see it
3543                                    Register flags    // if caller wants to test it
3544                                    ) {
3545   // determine flags
3546   const Bytecodes::Code code = bytecode();
3547   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3548   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3549   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3550   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3551   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3552   const bool load_receiver       = (recv  != noreg);
3553   const bool save_flags          = (flags != noreg);
3554   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3555   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3556   assert(flags == noreg || flags == rdx, "");
3557   assert(recv  == noreg || recv  == rcx, "");
3558 
3559   // setup registers & access constant pool cache
3560   if (recv  == noreg)  recv  = rcx;
3561   if (flags == noreg)  flags = rdx;
3562   assert_different_registers(method, index, recv, flags);
3563 
3564   // save 'interpreter return address'
3565   __ save_bcp();
3566 
3567   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3568 
3569   // maybe push appendix to arguments (just before return address)
3570   if (is_invokedynamic || is_invokehandle) {
3571     Label L_no_push;
3572     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3573     __ jcc(Assembler::zero, L_no_push);
3574     // Push the appendix as a trailing parameter.
3575     // This must be done before we get the receiver,
3576     // since the parameter_size includes it.
3577     __ push(rbx);
3578     __ mov(rbx, index);
3579     __ load_resolved_reference_at_index(index, rbx);
3580     __ pop(rbx);
3581     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3582     __ bind(L_no_push);
3583   }
3584 
3585   // load receiver if needed (after appendix is pushed so parameter size is correct)
3586   // Note: no return address pushed yet
3587   if (load_receiver) {
3588     __ movl(recv, flags);
3589     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3590     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3591     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3592     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3593     __ movptr(recv, recv_addr);
3594     __ verify_oop(recv);
3595   }
3596 
3597   if (save_flags) {
3598     __ movl(rbcp, flags);
3599   }
3600 
3601   // compute return type
3602   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3603   // Make sure we don't need to mask flags after the above shift
3604   ConstantPoolCacheEntry::verify_tos_state_shift();
3605   // load return address
3606   {
3607     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3608     ExternalAddress table(table_addr);
3609 #ifdef _LP64
3610     __ lea(rscratch1, table);
3611     __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
3612 #else
3613     __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
3614 #endif // _LP64
3615   }
3616 
3617   // push return address
3618   __ push(flags);
3619 
3620   // Restore flags value from the constant pool cache, and restore rsi
3621   // for later null checks.  r13 is the bytecode pointer
3622   if (save_flags) {
3623     __ movl(flags, rbcp);
3624     __ restore_bcp();
3625   }
3626 }
3627 
3628 void TemplateTable::invokevirtual_helper(Register index,
3629                                          Register recv,
3630                                          Register flags) {
3631   // Uses temporary registers rax, rdx
3632   assert_different_registers(index, recv, rax, rdx);
3633   assert(index == rbx, "");
3634   assert(recv  == rcx, "");
3635 
3636   // Test for an invoke of a final method
3637   Label notFinal;
3638   __ movl(rax, flags);
3639   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3640   __ jcc(Assembler::zero, notFinal);
3641 
3642   const Register method = index;  // method must be rbx
3643   assert(method == rbx,
3644          "Method* must be rbx for interpreter calling convention");
3645 
3646   // do the call - the index is actually the method to call
3647   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3648 
3649   // It's final, need a null check here!
3650   __ null_check(recv);
3651 
3652   // profile this call
3653   __ profile_final_call(rax);
3654   __ profile_arguments_type(rax, method, rbcp, true);
3655 
3656   __ jump_from_interpreted(method, rax);
3657 
3658   __ bind(notFinal);
3659 
3660   // get receiver klass
3661   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3662   __ load_klass(rax, recv, rscratch1);
3663 
3664   // profile this call
3665   __ profile_virtual_call(rax, rlocals, rdx);
3666   // get target Method* & entry point
3667   __ lookup_virtual_method(rax, index, method);
3668 
3669   __ profile_arguments_type(rdx, method, rbcp, true);
3670   __ jump_from_interpreted(method, rdx);
3671 }
3672 
3673 void TemplateTable::invokevirtual(int byte_no) {
3674   transition(vtos, vtos);
3675   assert(byte_no == f2_byte, "use this argument");
3676   prepare_invoke(byte_no,
3677                  rbx,    // method or vtable index
3678                  noreg,  // unused itable index
3679                  rcx, rdx); // recv, flags
3680 
3681   // rbx: index
3682   // rcx: receiver
3683   // rdx: flags
3684 
3685   invokevirtual_helper(rbx, rcx, rdx);
3686 }
3687 
3688 void TemplateTable::invokespecial(int byte_no) {
3689   transition(vtos, vtos);
3690   assert(byte_no == f1_byte, "use this argument");
3691   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3692                  rcx);  // get receiver also for null check
3693   __ verify_oop(rcx);
3694   __ null_check(rcx);
3695   // do the call
3696   __ profile_call(rax);
3697   __ profile_arguments_type(rax, rbx, rbcp, false);
3698   __ jump_from_interpreted(rbx, rax);
3699 }
3700 
3701 void TemplateTable::invokestatic(int byte_no) {
3702   transition(vtos, vtos);
3703   assert(byte_no == f1_byte, "use this argument");
3704   prepare_invoke(byte_no, rbx);  // get f1 Method*
3705   // do the call
3706   __ profile_call(rax);
3707   __ profile_arguments_type(rax, rbx, rbcp, false);
3708   __ jump_from_interpreted(rbx, rax);
3709 }
3710 
3711 
3712 void TemplateTable::fast_invokevfinal(int byte_no) {
3713   transition(vtos, vtos);
3714   assert(byte_no == f2_byte, "use this argument");
3715   __ stop("fast_invokevfinal not used on x86");
3716 }
3717 
3718 
3719 void TemplateTable::invokeinterface(int byte_no) {
3720   transition(vtos, vtos);
3721   assert(byte_no == f1_byte, "use this argument");
3722   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 Method*
3723                  rcx, rdx); // recv, flags
3724 
3725   // rax: reference klass (from f1) if interface method
3726   // rbx: method (from f2)
3727   // rcx: receiver
3728   // rdx: flags
3729 
3730   // First check for Object case, then private interface method,
3731   // then regular interface method.
3732 
3733   // Special case of invokeinterface called for virtual method of
3734   // java.lang.Object.  See cpCache.cpp for details.
3735   Label notObjectMethod;
3736   __ movl(rlocals, rdx);
3737   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3738   __ jcc(Assembler::zero, notObjectMethod);
3739   invokevirtual_helper(rbx, rcx, rdx);
3740   // no return from above
3741   __ bind(notObjectMethod);
3742 
3743   Label no_such_interface; // for receiver subtype check
3744   Register recvKlass; // used for exception processing
3745 
3746   // Check for private method invocation - indicated by vfinal
3747   Label notVFinal;
3748   __ movl(rlocals, rdx);
3749   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3750   __ jcc(Assembler::zero, notVFinal);
3751 
3752   // Get receiver klass into rlocals - also a null check
3753   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3754   __ load_klass(rlocals, rcx, rscratch1);
3755 
3756   Label subtype;
3757   __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3758   // If we get here the typecheck failed
3759   recvKlass = rdx;
3760   __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3761   __ jmp(no_such_interface);
3762 
3763   __ bind(subtype);
3764 
3765   // do the call - rbx is actually the method to call
3766 
3767   __ profile_final_call(rdx);
3768   __ profile_arguments_type(rdx, rbx, rbcp, true);
3769 
3770   __ jump_from_interpreted(rbx, rdx);
3771   // no return from above
3772   __ bind(notVFinal);
3773 
3774   // Get receiver klass into rdx - also a null check
3775   __ restore_locals();  // restore r14
3776   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3777   __ load_klass(rdx, rcx, rscratch1);
3778 
3779   Label no_such_method;
3780 
3781   // Preserve method for throw_AbstractMethodErrorVerbose.
3782   __ mov(rcx, rbx);
3783   // Receiver subtype check against REFC.
3784   // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3785   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3786                              rdx, rax, noreg,
3787                              // outputs: scan temp. reg, scan temp. reg
3788                              rbcp, rlocals,
3789                              no_such_interface,
3790                              /*return_method=*/false);
3791 
3792   // profile this call
3793   __ restore_bcp(); // rbcp was destroyed by receiver type check
3794   __ profile_virtual_call(rdx, rbcp, rlocals);
3795 
3796   // Get declaring interface class from method, and itable index
3797   __ load_method_holder(rax, rbx);
3798   __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3799   __ subl(rbx, Method::itable_index_max);
3800   __ negl(rbx);
3801 
3802   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3803   __ mov(rlocals, rdx);
3804   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3805                              rlocals, rax, rbx,
3806                              // outputs: method, scan temp. reg
3807                              rbx, rbcp,
3808                              no_such_interface);
3809 
3810   // rbx: Method* to call
3811   // rcx: receiver
3812   // Check for abstract method error
3813   // Note: This should be done more efficiently via a throw_abstract_method_error
3814   //       interpreter entry point and a conditional jump to it in case of a null
3815   //       method.
3816   __ testptr(rbx, rbx);
3817   __ jcc(Assembler::zero, no_such_method);
3818 
3819   __ profile_arguments_type(rdx, rbx, rbcp, true);
3820 
3821   // do the call
3822   // rcx: receiver
3823   // rbx,: Method*
3824   __ jump_from_interpreted(rbx, rdx);
3825   __ should_not_reach_here();
3826 
3827   // exception handling code follows...
3828   // note: must restore interpreter registers to canonical
3829   //       state for exception handling to work correctly!
3830 
3831   __ bind(no_such_method);
3832   // throw exception
3833   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3834   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3835   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3836   // Pass arguments for generating a verbose error message.
3837 #ifdef _LP64
3838   recvKlass = c_rarg1;
3839   Register method    = c_rarg2;
3840   if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3841   if (method != rcx)    { __ movq(method, rcx);    }
3842 #else
3843   recvKlass = rdx;
3844   Register method    = rcx;
3845 #endif
3846   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3847              recvKlass, method);
3848   // The call_VM checks for exception, so we should never return here.
3849   __ should_not_reach_here();
3850 
3851   __ bind(no_such_interface);
3852   // throw exception
3853   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3854   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3855   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3856   // Pass arguments for generating a verbose error message.
3857   LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3858   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3859              recvKlass, rax);
3860   // the call_VM checks for exception, so we should never return here.
3861   __ should_not_reach_here();
3862 }
3863 
3864 void TemplateTable::invokehandle(int byte_no) {
3865   transition(vtos, vtos);
3866   assert(byte_no == f1_byte, "use this argument");
3867   const Register rbx_method = rbx;
3868   const Register rax_mtype  = rax;
3869   const Register rcx_recv   = rcx;
3870   const Register rdx_flags  = rdx;
3871 
3872   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3873   __ verify_method_ptr(rbx_method);
3874   __ verify_oop(rcx_recv);
3875   __ null_check(rcx_recv);
3876 
3877   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3878   // rbx: MH.invokeExact_MT method (from f2)
3879 
3880   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3881 
3882   // FIXME: profile the LambdaForm also
3883   __ profile_final_call(rax);
3884   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3885 
3886   __ jump_from_interpreted(rbx_method, rdx);
3887 }
3888 
3889 void TemplateTable::invokedynamic(int byte_no) {
3890   transition(vtos, vtos);
3891   assert(byte_no == f1_byte, "use this argument");
3892 
3893   const Register rbx_method   = rbx;
3894   const Register rax_callsite = rax;
3895 
3896   prepare_invoke(byte_no, rbx_method, rax_callsite);
3897 
3898   // rax: CallSite object (from cpool->resolved_references[f1])
3899   // rbx: MH.linkToCallSite method (from f2)
3900 
3901   // Note:  rax_callsite is already pushed by prepare_invoke
3902 
3903   // %%% should make a type profile for any invokedynamic that takes a ref argument
3904   // profile this call
3905   __ profile_call(rbcp);
3906   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3907 
3908   __ verify_oop(rax_callsite);
3909 
3910   __ jump_from_interpreted(rbx_method, rdx);
3911 }
3912 
3913 //-----------------------------------------------------------------------------
3914 // Allocation
3915 
3916 void TemplateTable::_new() {
3917   transition(vtos, atos);
3918   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3919   Label slow_case;
3920   Label slow_case_no_pop;
3921   Label done;
3922   Label initialize_header;
3923 
3924   __ get_cpool_and_tags(rcx, rax);
3925 
3926   // Make sure the class we're about to instantiate has been resolved.
3927   // This is done before loading InstanceKlass to be consistent with the order
3928   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3929   const int tags_offset = Array<u1>::base_offset_in_bytes();
3930   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3931   __ jcc(Assembler::notEqual, slow_case_no_pop);
3932 
3933   // get InstanceKlass
3934   __ load_resolved_klass_at_index(rcx, rcx, rdx);
3935   __ push(rcx);  // save the contexts of klass for initializing the header
3936 
3937   // make sure klass is initialized & doesn't have finalizer
3938   // make sure klass is fully initialized
3939   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3940   __ jcc(Assembler::notEqual, slow_case);
3941 
3942   // get instance_size in InstanceKlass (scaled to a count of bytes)
3943   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3944   // test to see if it has a finalizer or is malformed in some way
3945   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3946   __ jcc(Assembler::notZero, slow_case);
3947 
3948   // Allocate the instance:
3949   //  If TLAB is enabled:
3950   //    Try to allocate in the TLAB.
3951   //    If fails, go to the slow path.
3952   //    Initialize the allocation.
3953   //    Exit.
3954   //
3955   //  Go to slow path.
3956 
3957   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3958 
3959   if (UseTLAB) {
3960     NOT_LP64(__ get_thread(thread);)
3961     __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
3962     if (ZeroTLAB) {
3963       // the fields have been already cleared
3964       __ jmp(initialize_header);
3965     }
3966 
3967     // The object is initialized before the header.  If the object size is
3968     // zero, go directly to the header initialization.
3969     __ decrement(rdx, sizeof(oopDesc));
3970     __ jcc(Assembler::zero, initialize_header);
3971 
3972     // Initialize topmost object field, divide rdx by 8, check if odd and
3973     // test if zero.
3974     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
3975     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3976 
3977     // rdx must have been multiple of 8
3978 #ifdef ASSERT
3979     // make sure rdx was multiple of 8
3980     Label L;
3981     // Ignore partial flag stall after shrl() since it is debug VM
3982     __ jcc(Assembler::carryClear, L);
3983     __ stop("object size is not multiple of 2 - adjust this code");
3984     __ bind(L);
3985     // rdx must be > 0, no extra check needed here
3986 #endif
3987 
3988     // initialize remaining object fields: rdx was a multiple of 8
3989     { Label loop;
3990     __ bind(loop);
3991     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3992     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3993     __ decrement(rdx);
3994     __ jcc(Assembler::notZero, loop);
3995     }
3996 
3997     // initialize object header only.
3998     __ bind(initialize_header);
3999     __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
4000               (intptr_t)markWord::prototype().value()); // header
4001     __ pop(rcx);   // get saved klass back in the register.
4002 #ifdef _LP64
4003     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4004     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4005 #endif
4006     __ store_klass(rax, rcx, rscratch1);  // klass
4007 
4008     {
4009       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0, rscratch1);
4010       // Trigger dtrace event for fastpath
4011       __ push(atos);
4012       __ call_VM_leaf(
4013            CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), rax);
4014       __ pop(atos);
4015     }
4016 
4017     __ jmp(done);
4018   }
4019 
4020   // slow case
4021   __ bind(slow_case);
4022   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4023   __ bind(slow_case_no_pop);
4024 
4025   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4026   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4027 
4028   __ get_constant_pool(rarg1);
4029   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4030   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4031    __ verify_oop(rax);
4032 
4033   // continue
4034   __ bind(done);
4035 }
4036 
4037 void TemplateTable::newarray() {
4038   transition(itos, atos);
4039   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4040   __ load_unsigned_byte(rarg1, at_bcp(1));
4041   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4042           rarg1, rax);
4043 }
4044 
4045 void TemplateTable::anewarray() {
4046   transition(itos, atos);
4047 
4048   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4049   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4050 
4051   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4052   __ get_constant_pool(rarg1);
4053   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4054           rarg1, rarg2, rax);
4055 }
4056 
4057 void TemplateTable::arraylength() {
4058   transition(atos, itos);
4059   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4060   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4061 }
4062 
4063 void TemplateTable::checkcast() {
4064   transition(atos, atos);
4065   Label done, is_null, ok_is_subtype, quicked, resolved;
4066   __ testptr(rax, rax); // object is in rax
4067   __ jcc(Assembler::zero, is_null);
4068 
4069   // Get cpool & tags index
4070   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4071   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4072   // See if bytecode has already been quicked
4073   __ cmpb(Address(rdx, rbx,
4074                   Address::times_1,
4075                   Array<u1>::base_offset_in_bytes()),
4076           JVM_CONSTANT_Class);
4077   __ jcc(Assembler::equal, quicked);
4078   __ push(atos); // save receiver for result, and for GC
4079   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4080 
4081   // vm_result_2 has metadata result
4082 #ifndef _LP64
4083   // borrow rdi from locals
4084   __ get_thread(rdi);
4085   __ get_vm_result_2(rax, rdi);
4086   __ restore_locals();
4087 #else
4088   __ get_vm_result_2(rax, r15_thread);
4089 #endif
4090 
4091   __ pop_ptr(rdx); // restore receiver
4092   __ jmpb(resolved);
4093 
4094   // Get superklass in rax and subklass in rbx
4095   __ bind(quicked);
4096   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4097   __ load_resolved_klass_at_index(rax, rcx, rbx);
4098 
4099   __ bind(resolved);
4100   __ load_klass(rbx, rdx, rscratch1);
4101 
4102   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4103   // Superklass in rax.  Subklass in rbx.
4104   __ gen_subtype_check(rbx, ok_is_subtype);
4105 
4106   // Come here on failure
4107   __ push_ptr(rdx);
4108   // object is at TOS
4109   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4110 
4111   // Come here on success
4112   __ bind(ok_is_subtype);
4113   __ mov(rax, rdx); // Restore object in rdx
4114 
4115   // Collect counts on whether this check-cast sees NULLs a lot or not.
4116   if (ProfileInterpreter) {
4117     __ jmp(done);
4118     __ bind(is_null);
4119     __ profile_null_seen(rcx);
4120   } else {
4121     __ bind(is_null);   // same as 'done'
4122   }
4123   __ bind(done);
4124 }
4125 
4126 void TemplateTable::instanceof() {
4127   transition(atos, itos);
4128   Label done, is_null, ok_is_subtype, quicked, resolved;
4129   __ testptr(rax, rax);
4130   __ jcc(Assembler::zero, is_null);
4131 
4132   // Get cpool & tags index
4133   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4134   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4135   // See if bytecode has already been quicked
4136   __ cmpb(Address(rdx, rbx,
4137                   Address::times_1,
4138                   Array<u1>::base_offset_in_bytes()),
4139           JVM_CONSTANT_Class);
4140   __ jcc(Assembler::equal, quicked);
4141 
4142   __ push(atos); // save receiver for result, and for GC
4143   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4144   // vm_result_2 has metadata result
4145 
4146 #ifndef _LP64
4147   // borrow rdi from locals
4148   __ get_thread(rdi);
4149   __ get_vm_result_2(rax, rdi);
4150   __ restore_locals();
4151 #else
4152   __ get_vm_result_2(rax, r15_thread);
4153 #endif
4154 
4155   __ pop_ptr(rdx); // restore receiver
4156   __ verify_oop(rdx);
4157   __ load_klass(rdx, rdx, rscratch1);
4158   __ jmpb(resolved);
4159 
4160   // Get superklass in rax and subklass in rdx
4161   __ bind(quicked);
4162   __ load_klass(rdx, rax, rscratch1);
4163   __ load_resolved_klass_at_index(rax, rcx, rbx);
4164 
4165   __ bind(resolved);
4166 
4167   // Generate subtype check.  Blows rcx, rdi
4168   // Superklass in rax.  Subklass in rdx.
4169   __ gen_subtype_check(rdx, ok_is_subtype);
4170 
4171   // Come here on failure
4172   __ xorl(rax, rax);
4173   __ jmpb(done);
4174   // Come here on success
4175   __ bind(ok_is_subtype);
4176   __ movl(rax, 1);
4177 
4178   // Collect counts on whether this test sees NULLs a lot or not.
4179   if (ProfileInterpreter) {
4180     __ jmp(done);
4181     __ bind(is_null);
4182     __ profile_null_seen(rcx);
4183   } else {
4184     __ bind(is_null);   // same as 'done'
4185   }
4186   __ bind(done);
4187   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4188   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4189 }
4190 
4191 
4192 //----------------------------------------------------------------------------------------------------
4193 // Breakpoints
4194 void TemplateTable::_breakpoint() {
4195   // Note: We get here even if we are single stepping..
4196   // jbug insists on setting breakpoints at every bytecode
4197   // even if we are in single step mode.
4198 
4199   transition(vtos, vtos);
4200 
4201   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4202 
4203   // get the unpatched byte code
4204   __ get_method(rarg);
4205   __ call_VM(noreg,
4206              CAST_FROM_FN_PTR(address,
4207                               InterpreterRuntime::get_original_bytecode_at),
4208              rarg, rbcp);
4209   __ mov(rbx, rax);  // why?
4210 
4211   // post the breakpoint event
4212   __ get_method(rarg);
4213   __ call_VM(noreg,
4214              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4215              rarg, rbcp);
4216 
4217   // complete the execution of original bytecode
4218   __ dispatch_only_normal(vtos);
4219 }
4220 
4221 //-----------------------------------------------------------------------------
4222 // Exceptions
4223 
4224 void TemplateTable::athrow() {
4225   transition(atos, vtos);
4226   __ null_check(rax);
4227   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4228 }
4229 
4230 //-----------------------------------------------------------------------------
4231 // Synchronization
4232 //
4233 // Note: monitorenter & exit are symmetric routines; which is reflected
4234 //       in the assembly code structure as well
4235 //
4236 // Stack layout:
4237 //
4238 // [expressions  ] <--- rsp               = expression stack top
4239 // ..
4240 // [expressions  ]
4241 // [monitor entry] <--- monitor block top = expression stack bot
4242 // ..
4243 // [monitor entry]
4244 // [frame data   ] <--- monitor block bot
4245 // ...
4246 // [saved rbp    ] <--- rbp
4247 void TemplateTable::monitorenter() {
4248   transition(atos, vtos);
4249 
4250   // check for NULL object
4251   __ null_check(rax);
4252 
4253   const Address monitor_block_top(
4254         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4255   const Address monitor_block_bot(
4256         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4257   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4258 
4259   Label allocated;
4260 
4261   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4262   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4263   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4264 
4265   // initialize entry pointer
4266   __ xorl(rmon, rmon); // points to free slot or NULL
4267 
4268   // find a free slot in the monitor block (result in rmon)
4269   {
4270     Label entry, loop, exit;
4271     __ movptr(rtop, monitor_block_top); // points to current entry,
4272                                         // starting with top-most entry
4273     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4274                                         // of monitor block
4275     __ jmpb(entry);
4276 
4277     __ bind(loop);
4278     // check if current entry is used
4279     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
4280     // if not used then remember entry in rmon
4281     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4282     // check if current entry is for same object
4283     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4284     // if same object then stop searching
4285     __ jccb(Assembler::equal, exit);
4286     // otherwise advance to next entry
4287     __ addptr(rtop, entry_size);
4288     __ bind(entry);
4289     // check if bottom reached
4290     __ cmpptr(rtop, rbot);
4291     // if not at bottom then check this entry
4292     __ jcc(Assembler::notEqual, loop);
4293     __ bind(exit);
4294   }
4295 
4296   __ testptr(rmon, rmon); // check if a slot has been found
4297   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4298 
4299   // allocate one if there's no free slot
4300   {
4301     Label entry, loop;
4302     // 1. compute new pointers          // rsp: old expression stack top
4303     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4304     __ subptr(rsp, entry_size);         // move expression stack top
4305     __ subptr(rmon, entry_size);        // move expression stack bottom
4306     __ mov(rtop, rsp);                  // set start value for copy loop
4307     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4308     __ jmp(entry);
4309     // 2. move expression stack contents
4310     __ bind(loop);
4311     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4312                                                 // word from old location
4313     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4314     __ addptr(rtop, wordSize);                  // advance to next word
4315     __ bind(entry);
4316     __ cmpptr(rtop, rmon);                      // check if bottom reached
4317     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4318                                                 // copy next word
4319   }
4320 
4321   // call run-time routine
4322   // rmon: points to monitor entry
4323   __ bind(allocated);
4324 
4325   // Increment bcp to point to the next bytecode, so exception
4326   // handling for async. exceptions work correctly.
4327   // The object has already been popped from the stack, so the
4328   // expression stack looks correct.
4329   __ increment(rbcp);
4330 
4331   // store object
4332   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4333   __ lock_object(rmon);
4334 
4335   // check to make sure this monitor doesn't cause stack overflow after locking
4336   __ save_bcp();  // in case of exception
4337   __ generate_stack_overflow_check(0);
4338 
4339   // The bcp has already been incremented. Just need to dispatch to
4340   // next instruction.
4341   __ dispatch_next(vtos);
4342 }
4343 
4344 void TemplateTable::monitorexit() {
4345   transition(atos, vtos);
4346 
4347   // check for NULL object
4348   __ null_check(rax);
4349 
4350   const Address monitor_block_top(
4351         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4352   const Address monitor_block_bot(
4353         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4354   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4355 
4356   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4357   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4358 
4359   Label found;
4360 
4361   // find matching slot
4362   {
4363     Label entry, loop;
4364     __ movptr(rtop, monitor_block_top); // points to current entry,
4365                                         // starting with top-most entry
4366     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4367                                         // of monitor block
4368     __ jmpb(entry);
4369 
4370     __ bind(loop);
4371     // check if current entry is for same object
4372     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4373     // if same object then stop searching
4374     __ jcc(Assembler::equal, found);
4375     // otherwise advance to next entry
4376     __ addptr(rtop, entry_size);
4377     __ bind(entry);
4378     // check if bottom reached
4379     __ cmpptr(rtop, rbot);
4380     // if not at bottom then check this entry
4381     __ jcc(Assembler::notEqual, loop);
4382   }
4383 
4384   // error handling. Unlocking was not block-structured
4385   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4386                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4387   __ should_not_reach_here();
4388 
4389   // call run-time routine
4390   __ bind(found);
4391   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4392   __ unlock_object(rtop);
4393   __ pop_ptr(rax); // discard object
4394 }
4395 
4396 // Wide instructions
4397 void TemplateTable::wide() {
4398   transition(vtos, vtos);
4399   __ load_unsigned_byte(rbx, at_bcp(1));
4400   ExternalAddress wtable((address)Interpreter::_wentry_point);
4401   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)), rscratch1);
4402   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4403 }
4404 
4405 // Multi arrays
4406 void TemplateTable::multianewarray() {
4407   transition(vtos, atos);
4408 
4409   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4410   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4411   // last dim is on top of stack; we want address of first one:
4412   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4413   // the latter wordSize to point to the beginning of the array.
4414   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4415   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4416   __ load_unsigned_byte(rbx, at_bcp(3));
4417   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4418 }