1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "interpreter/interp_masm.hpp"
  30 #include "interpreter/templateTable.hpp"
  31 #include "memory/universe.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/objArrayKlass.hpp"
  34 #include "oops/oop.inline.hpp"
  35 #include "prims/methodHandles.hpp"
  36 #include "runtime/frame.inline.hpp"
  37 #include "runtime/safepointMechanism.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "runtime/synchronizer.hpp"
  41 #include "utilities/macros.hpp"
  42 
  43 #define __ _masm->
  44 
  45 // Global Register Names
  46 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  47 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  48 
  49 // Platform-dependent initialization
  50 void TemplateTable::pd_initialize() {
  51   // No x86 specific initialization
  52 }
  53 
  54 // Address Computation: local variables
  55 static inline Address iaddress(int n) {
  56   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  57 }
  58 
  59 static inline Address laddress(int n) {
  60   return iaddress(n + 1);
  61 }
  62 
  63 #ifndef _LP64
  64 static inline Address haddress(int n) {
  65   return iaddress(n + 0);
  66 }
  67 #endif
  68 
  69 static inline Address faddress(int n) {
  70   return iaddress(n);
  71 }
  72 
  73 static inline Address daddress(int n) {
  74   return laddress(n);
  75 }
  76 
  77 static inline Address aaddress(int n) {
  78   return iaddress(n);
  79 }
  80 
  81 static inline Address iaddress(Register r) {
  82   return Address(rlocals, r, Address::times_ptr);
  83 }
  84 
  85 static inline Address laddress(Register r) {
  86   return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
  87 }
  88 
  89 #ifndef _LP64
  90 static inline Address haddress(Register r)       {
  91   return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  92 }
  93 #endif
  94 
  95 static inline Address faddress(Register r) {
  96   return iaddress(r);
  97 }
  98 
  99 static inline Address daddress(Register r) {
 100   return laddress(r);
 101 }
 102 
 103 static inline Address aaddress(Register r) {
 104   return iaddress(r);
 105 }
 106 
 107 
 108 // expression stack
 109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
 110 // data beyond the rsp which is potentially unsafe in an MT environment;
 111 // an interrupt may overwrite that data.)
 112 static inline Address at_rsp   () {
 113   return Address(rsp, 0);
 114 }
 115 
 116 // At top of Java expression stack which may be different than esp().  It
 117 // isn't for category 1 objects.
 118 static inline Address at_tos   () {
 119   return Address(rsp,  Interpreter::expr_offset_in_bytes(0));
 120 }
 121 
 122 static inline Address at_tos_p1() {
 123   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
 124 }
 125 
 126 static inline Address at_tos_p2() {
 127   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
 128 }
 129 
 130 // Condition conversion
 131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 132   switch (cc) {
 133   case TemplateTable::equal        : return Assembler::notEqual;
 134   case TemplateTable::not_equal    : return Assembler::equal;
 135   case TemplateTable::less         : return Assembler::greaterEqual;
 136   case TemplateTable::less_equal   : return Assembler::greater;
 137   case TemplateTable::greater      : return Assembler::lessEqual;
 138   case TemplateTable::greater_equal: return Assembler::less;
 139   }
 140   ShouldNotReachHere();
 141   return Assembler::zero;
 142 }
 143 
 144 
 145 
 146 // Miscelaneous helper routines
 147 // Store an oop (or NULL) at the address described by obj.
 148 // If val == noreg this means store a NULL
 149 
 150 
 151 static void do_oop_store(InterpreterMacroAssembler* _masm,
 152                          Address dst,
 153                          Register val,
 154                          DecoratorSet decorators = 0) {
 155   assert(val == noreg || val == rax, "parameter is just for looks");
 156   __ store_heap_oop(dst, val, rdx, rbx, decorators);
 157 }
 158 
 159 static void do_oop_load(InterpreterMacroAssembler* _masm,
 160                         Address src,
 161                         Register dst,
 162                         DecoratorSet decorators = 0) {
 163   __ load_heap_oop(dst, src, rdx, rbx, decorators);
 164 }
 165 
 166 Address TemplateTable::at_bcp(int offset) {
 167   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 168   return Address(rbcp, offset);
 169 }
 170 
 171 
 172 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 173                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 174                                    int byte_no) {
 175   if (!RewriteBytecodes)  return;
 176   Label L_patch_done;
 177 
 178   switch (bc) {
 179   case Bytecodes::_fast_aputfield:
 180   case Bytecodes::_fast_bputfield:
 181   case Bytecodes::_fast_zputfield:
 182   case Bytecodes::_fast_cputfield:
 183   case Bytecodes::_fast_dputfield:
 184   case Bytecodes::_fast_fputfield:
 185   case Bytecodes::_fast_iputfield:
 186   case Bytecodes::_fast_lputfield:
 187   case Bytecodes::_fast_sputfield:
 188     {
 189       // We skip bytecode quickening for putfield instructions when
 190       // the put_code written to the constant pool cache is zero.
 191       // This is required so that every execution of this instruction
 192       // calls out to InterpreterRuntime::resolve_get_put to do
 193       // additional, required work.
 194       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 195       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 196       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 197       __ movl(bc_reg, bc);
 198       __ cmpl(temp_reg, (int) 0);
 199       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 200     }
 201     break;
 202   default:
 203     assert(byte_no == -1, "sanity");
 204     // the pair bytecodes have already done the load.
 205     if (load_bc_into_bc_reg) {
 206       __ movl(bc_reg, bc);
 207     }
 208   }
 209 
 210   if (JvmtiExport::can_post_breakpoint()) {
 211     Label L_fast_patch;
 212     // if a breakpoint is present we can't rewrite the stream directly
 213     __ movzbl(temp_reg, at_bcp(0));
 214     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 215     __ jcc(Assembler::notEqual, L_fast_patch);
 216     __ get_method(temp_reg);
 217     // Let breakpoint table handling rewrite to quicker bytecode
 218     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 219 #ifndef ASSERT
 220     __ jmpb(L_patch_done);
 221 #else
 222     __ jmp(L_patch_done);
 223 #endif
 224     __ bind(L_fast_patch);
 225   }
 226 
 227 #ifdef ASSERT
 228   Label L_okay;
 229   __ load_unsigned_byte(temp_reg, at_bcp(0));
 230   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 231   __ jcc(Assembler::equal, L_okay);
 232   __ cmpl(temp_reg, bc_reg);
 233   __ jcc(Assembler::equal, L_okay);
 234   __ stop("patching the wrong bytecode");
 235   __ bind(L_okay);
 236 #endif
 237 
 238   // patch bytecode
 239   __ movb(at_bcp(0), bc_reg);
 240   __ bind(L_patch_done);
 241 }
 242 // Individual instructions
 243 
 244 
 245 void TemplateTable::nop() {
 246   transition(vtos, vtos);
 247   // nothing to do
 248 }
 249 
 250 void TemplateTable::shouldnotreachhere() {
 251   transition(vtos, vtos);
 252   __ stop("shouldnotreachhere bytecode");
 253 }
 254 
 255 void TemplateTable::aconst_null() {
 256   transition(vtos, atos);
 257   __ xorl(rax, rax);
 258 }
 259 
 260 void TemplateTable::iconst(int value) {
 261   transition(vtos, itos);
 262   if (value == 0) {
 263     __ xorl(rax, rax);
 264   } else {
 265     __ movl(rax, value);
 266   }
 267 }
 268 
 269 void TemplateTable::lconst(int value) {
 270   transition(vtos, ltos);
 271   if (value == 0) {
 272     __ xorl(rax, rax);
 273   } else {
 274     __ movl(rax, value);
 275   }
 276 #ifndef _LP64
 277   assert(value >= 0, "check this code");
 278   __ xorptr(rdx, rdx);
 279 #endif
 280 }
 281 
 282 
 283 
 284 void TemplateTable::fconst(int value) {
 285   transition(vtos, ftos);
 286   if (UseSSE >= 1) {
 287     static float one = 1.0f, two = 2.0f;
 288     switch (value) {
 289     case 0:
 290       __ xorps(xmm0, xmm0);
 291       break;
 292     case 1:
 293       __ movflt(xmm0, ExternalAddress((address) &one));
 294       break;
 295     case 2:
 296       __ movflt(xmm0, ExternalAddress((address) &two));
 297       break;
 298     default:
 299       ShouldNotReachHere();
 300       break;
 301     }
 302   } else {
 303 #ifdef _LP64
 304     ShouldNotReachHere();
 305 #else
 306            if (value == 0) { __ fldz();
 307     } else if (value == 1) { __ fld1();
 308     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 309     } else                 { ShouldNotReachHere();
 310     }
 311 #endif // _LP64
 312   }
 313 }
 314 
 315 void TemplateTable::dconst(int value) {
 316   transition(vtos, dtos);
 317   if (UseSSE >= 2) {
 318     static double one = 1.0;
 319     switch (value) {
 320     case 0:
 321       __ xorpd(xmm0, xmm0);
 322       break;
 323     case 1:
 324       __ movdbl(xmm0, ExternalAddress((address) &one));
 325       break;
 326     default:
 327       ShouldNotReachHere();
 328       break;
 329     }
 330   } else {
 331 #ifdef _LP64
 332     ShouldNotReachHere();
 333 #else
 334            if (value == 0) { __ fldz();
 335     } else if (value == 1) { __ fld1();
 336     } else                 { ShouldNotReachHere();
 337     }
 338 #endif
 339   }
 340 }
 341 
 342 void TemplateTable::bipush() {
 343   transition(vtos, itos);
 344   __ load_signed_byte(rax, at_bcp(1));
 345 }
 346 
 347 void TemplateTable::sipush() {
 348   transition(vtos, itos);
 349   __ load_unsigned_short(rax, at_bcp(1));
 350   __ bswapl(rax);
 351   __ sarl(rax, 16);
 352 }
 353 
 354 void TemplateTable::ldc(bool wide) {
 355   transition(vtos, vtos);
 356   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 357   Label call_ldc, notFloat, notClass, notInt, Done;
 358 
 359   if (wide) {
 360     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 361   } else {
 362     __ load_unsigned_byte(rbx, at_bcp(1));
 363   }
 364 
 365   __ get_cpool_and_tags(rcx, rax);
 366   const int base_offset = ConstantPool::header_size() * wordSize;
 367   const int tags_offset = Array<u1>::base_offset_in_bytes();
 368 
 369   // get type
 370   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 371 
 372   // unresolved class - get the resolved class
 373   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 374   __ jccb(Assembler::equal, call_ldc);
 375 
 376   // unresolved class in error state - call into runtime to throw the error
 377   // from the first resolution attempt
 378   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 379   __ jccb(Assembler::equal, call_ldc);
 380 
 381   // resolved class - need to call vm to get java mirror of the class
 382   __ cmpl(rdx, JVM_CONSTANT_Class);
 383   __ jcc(Assembler::notEqual, notClass);
 384 
 385   __ bind(call_ldc);
 386 
 387   __ movl(rarg, wide);
 388   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 389 
 390   __ push(atos);
 391   __ jmp(Done);
 392 
 393   __ bind(notClass);
 394   __ cmpl(rdx, JVM_CONSTANT_Float);
 395   __ jccb(Assembler::notEqual, notFloat);
 396 
 397   // ftos
 398   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 399   __ push(ftos);
 400   __ jmp(Done);
 401 
 402   __ bind(notFloat);
 403   __ cmpl(rdx, JVM_CONSTANT_Integer);
 404   __ jccb(Assembler::notEqual, notInt);
 405 
 406   // itos
 407   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 408   __ push(itos);
 409   __ jmp(Done);
 410 
 411   // assume the tag is for condy; if not, the VM runtime will tell us
 412   __ bind(notInt);
 413   condy_helper(Done);
 414 
 415   __ bind(Done);
 416 }
 417 
 418 // Fast path for caching oop constants.
 419 void TemplateTable::fast_aldc(bool wide) {
 420   transition(vtos, atos);
 421 
 422   Register result = rax;
 423   Register tmp = rdx;
 424   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 425   int index_size = wide ? sizeof(u2) : sizeof(u1);
 426 
 427   Label resolved;
 428 
 429   // We are resolved if the resolved reference cache entry contains a
 430   // non-null object (String, MethodType, etc.)
 431   assert_different_registers(result, tmp);
 432   __ get_cache_index_at_bcp(tmp, 1, index_size);
 433   __ load_resolved_reference_at_index(result, tmp);
 434   __ testptr(result, result);
 435   __ jcc(Assembler::notZero, resolved);
 436 
 437   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 438 
 439   // first time invocation - must resolve first
 440   __ movl(rarg, (int)bytecode());
 441   __ call_VM(result, entry, rarg);
 442   __ bind(resolved);
 443 
 444   { // Check for the null sentinel.
 445     // If we just called the VM, it already did the mapping for us,
 446     // but it's harmless to retry.
 447     Label notNull;
 448     ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
 449     __ movptr(tmp, null_sentinel);
 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       __ jcc(Assembler::notEqual, notInt);
 529       // itos
 530       __ movl(rax, field);
 531       __ push(itos);
 532       __ jmp(Done);
 533 
 534       __ bind(notInt);
 535       __ cmpl(flags, ftos);
 536       __ jcc(Assembler::notEqual, notFloat);
 537       // ftos
 538       __ load_float(field);
 539       __ push(ftos);
 540       __ jmp(Done);
 541 
 542       __ bind(notFloat);
 543       __ cmpl(flags, stos);
 544       __ jcc(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       __ jcc(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       __ jcc(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       __ jcc(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       __ jcc(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       __ jcc(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);
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);
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);
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);
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);
1132   // Move superklass into rax
1133   __ load_klass(rax, rdx);
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);
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);
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);
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
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
1557       break;
1558     default:
1559       ShouldNotReachHere();
1560       break;
1561     }
1562   } else {
1563 #ifdef _LP64
1564     ShouldNotReachHere();
1565 #else
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
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
1621       break;
1622     default:
1623       ShouldNotReachHere();
1624       break;
1625     }
1626   } else {
1627 #ifdef _LP64
1628     ShouldNotReachHere();
1629 #else
1630     switch (op) {
1631     case add: __ fadd_d (at_rsp());                break;
1632     case sub: __ fsubr_d(at_rsp());                break;
1633     case mul: {
1634       Label L_strict;
1635       Label L_join;
1636       const Address access_flags      (rcx, Method::access_flags_offset());
1637       __ get_method(rcx);
1638       __ movl(rcx, access_flags);
1639       __ testl(rcx, JVM_ACC_STRICT);
1640       __ jccb(Assembler::notZero, L_strict);
1641       __ fmul_d (at_rsp());
1642       __ jmpb(L_join);
1643       __ bind(L_strict);
1644       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1645       __ fmulp();
1646       __ fmul_d (at_rsp());
1647       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1648       __ fmulp();
1649       __ bind(L_join);
1650       break;
1651     }
1652     case div: {
1653       Label L_strict;
1654       Label L_join;
1655       const Address access_flags      (rcx, Method::access_flags_offset());
1656       __ get_method(rcx);
1657       __ movl(rcx, access_flags);
1658       __ testl(rcx, JVM_ACC_STRICT);
1659       __ jccb(Assembler::notZero, L_strict);
1660       __ fdivr_d(at_rsp());
1661       __ jmp(L_join);
1662       __ bind(L_strict);
1663       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1664       __ fmul_d (at_rsp());
1665       __ fdivrp();
1666       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1667       __ fmulp();
1668       __ bind(L_join);
1669       break;
1670     }
1671     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1672     default : ShouldNotReachHere();
1673     }
1674     __ d2ieee();
1675     // Pop double precision number from rsp.
1676     __ pop(rax);
1677     __ pop(rdx);
1678 #endif
1679   }
1680 }
1681 
1682 void TemplateTable::ineg() {
1683   transition(itos, itos);
1684   __ negl(rax);
1685 }
1686 
1687 void TemplateTable::lneg() {
1688   transition(ltos, ltos);
1689   LP64_ONLY(__ negq(rax));
1690   NOT_LP64(__ lneg(rdx, rax));
1691 }
1692 
1693 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1694 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1695   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1696   // of 128-bits operands for SSE instructions.
1697   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1698   // Store the value to a 128-bits operand.
1699   operand[0] = lo;
1700   operand[1] = hi;
1701   return operand;
1702 }
1703 
1704 // Buffer for 128-bits masks used by SSE instructions.
1705 static jlong float_signflip_pool[2*2];
1706 static jlong double_signflip_pool[2*2];
1707 
1708 void TemplateTable::fneg() {
1709   transition(ftos, ftos);
1710   if (UseSSE >= 1) {
1711     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1712     __ xorps(xmm0, ExternalAddress((address) float_signflip));
1713   } else {
1714     LP64_ONLY(ShouldNotReachHere());
1715     NOT_LP64(__ fchs());
1716   }
1717 }
1718 
1719 void TemplateTable::dneg() {
1720   transition(dtos, dtos);
1721   if (UseSSE >= 2) {
1722     static jlong *double_signflip =
1723       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1724     __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1725   } else {
1726 #ifdef _LP64
1727     ShouldNotReachHere();
1728 #else
1729     __ fchs();
1730 #endif
1731   }
1732 }
1733 
1734 void TemplateTable::iinc() {
1735   transition(vtos, vtos);
1736   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1737   locals_index(rbx);
1738   __ addl(iaddress(rbx), rdx);
1739 }
1740 
1741 void TemplateTable::wide_iinc() {
1742   transition(vtos, vtos);
1743   __ movl(rdx, at_bcp(4)); // get constant
1744   locals_index_wide(rbx);
1745   __ bswapl(rdx); // swap bytes & sign-extend constant
1746   __ sarl(rdx, 16);
1747   __ addl(iaddress(rbx), rdx);
1748   // Note: should probably use only one movl to get both
1749   //       the index and the constant -> fix this
1750 }
1751 
1752 void TemplateTable::convert() {
1753 #ifdef _LP64
1754   // Checking
1755 #ifdef ASSERT
1756   {
1757     TosState tos_in  = ilgl;
1758     TosState tos_out = ilgl;
1759     switch (bytecode()) {
1760     case Bytecodes::_i2l: // fall through
1761     case Bytecodes::_i2f: // fall through
1762     case Bytecodes::_i2d: // fall through
1763     case Bytecodes::_i2b: // fall through
1764     case Bytecodes::_i2c: // fall through
1765     case Bytecodes::_i2s: tos_in = itos; break;
1766     case Bytecodes::_l2i: // fall through
1767     case Bytecodes::_l2f: // fall through
1768     case Bytecodes::_l2d: tos_in = ltos; break;
1769     case Bytecodes::_f2i: // fall through
1770     case Bytecodes::_f2l: // fall through
1771     case Bytecodes::_f2d: tos_in = ftos; break;
1772     case Bytecodes::_d2i: // fall through
1773     case Bytecodes::_d2l: // fall through
1774     case Bytecodes::_d2f: tos_in = dtos; break;
1775     default             : ShouldNotReachHere();
1776     }
1777     switch (bytecode()) {
1778     case Bytecodes::_l2i: // fall through
1779     case Bytecodes::_f2i: // fall through
1780     case Bytecodes::_d2i: // fall through
1781     case Bytecodes::_i2b: // fall through
1782     case Bytecodes::_i2c: // fall through
1783     case Bytecodes::_i2s: tos_out = itos; break;
1784     case Bytecodes::_i2l: // fall through
1785     case Bytecodes::_f2l: // fall through
1786     case Bytecodes::_d2l: tos_out = ltos; break;
1787     case Bytecodes::_i2f: // fall through
1788     case Bytecodes::_l2f: // fall through
1789     case Bytecodes::_d2f: tos_out = ftos; break;
1790     case Bytecodes::_i2d: // fall through
1791     case Bytecodes::_l2d: // fall through
1792     case Bytecodes::_f2d: tos_out = dtos; break;
1793     default             : ShouldNotReachHere();
1794     }
1795     transition(tos_in, tos_out);
1796   }
1797 #endif // ASSERT
1798 
1799   static const int64_t is_nan = 0x8000000000000000L;
1800 
1801   // Conversion
1802   switch (bytecode()) {
1803   case Bytecodes::_i2l:
1804     __ movslq(rax, rax);
1805     break;
1806   case Bytecodes::_i2f:
1807     __ cvtsi2ssl(xmm0, rax);
1808     break;
1809   case Bytecodes::_i2d:
1810     __ cvtsi2sdl(xmm0, rax);
1811     break;
1812   case Bytecodes::_i2b:
1813     __ movsbl(rax, rax);
1814     break;
1815   case Bytecodes::_i2c:
1816     __ movzwl(rax, rax);
1817     break;
1818   case Bytecodes::_i2s:
1819     __ movswl(rax, rax);
1820     break;
1821   case Bytecodes::_l2i:
1822     __ movl(rax, rax);
1823     break;
1824   case Bytecodes::_l2f:
1825     __ cvtsi2ssq(xmm0, rax);
1826     break;
1827   case Bytecodes::_l2d:
1828     __ cvtsi2sdq(xmm0, rax);
1829     break;
1830   case Bytecodes::_f2i:
1831   {
1832     Label L;
1833     __ cvttss2sil(rax, xmm0);
1834     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1835     __ jcc(Assembler::notEqual, L);
1836     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1837     __ bind(L);
1838   }
1839     break;
1840   case Bytecodes::_f2l:
1841   {
1842     Label L;
1843     __ cvttss2siq(rax, xmm0);
1844     // NaN or overflow/underflow?
1845     __ cmp64(rax, ExternalAddress((address) &is_nan));
1846     __ jcc(Assembler::notEqual, L);
1847     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1848     __ bind(L);
1849   }
1850     break;
1851   case Bytecodes::_f2d:
1852     __ cvtss2sd(xmm0, xmm0);
1853     break;
1854   case Bytecodes::_d2i:
1855   {
1856     Label L;
1857     __ cvttsd2sil(rax, xmm0);
1858     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1859     __ jcc(Assembler::notEqual, L);
1860     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1861     __ bind(L);
1862   }
1863     break;
1864   case Bytecodes::_d2l:
1865   {
1866     Label L;
1867     __ cvttsd2siq(rax, xmm0);
1868     // NaN or overflow/underflow?
1869     __ cmp64(rax, ExternalAddress((address) &is_nan));
1870     __ jcc(Assembler::notEqual, L);
1871     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1872     __ bind(L);
1873   }
1874     break;
1875   case Bytecodes::_d2f:
1876     __ cvtsd2ss(xmm0, xmm0);
1877     break;
1878   default:
1879     ShouldNotReachHere();
1880   }
1881 #else
1882   // Checking
1883 #ifdef ASSERT
1884   { TosState tos_in  = ilgl;
1885     TosState tos_out = ilgl;
1886     switch (bytecode()) {
1887       case Bytecodes::_i2l: // fall through
1888       case Bytecodes::_i2f: // fall through
1889       case Bytecodes::_i2d: // fall through
1890       case Bytecodes::_i2b: // fall through
1891       case Bytecodes::_i2c: // fall through
1892       case Bytecodes::_i2s: tos_in = itos; break;
1893       case Bytecodes::_l2i: // fall through
1894       case Bytecodes::_l2f: // fall through
1895       case Bytecodes::_l2d: tos_in = ltos; break;
1896       case Bytecodes::_f2i: // fall through
1897       case Bytecodes::_f2l: // fall through
1898       case Bytecodes::_f2d: tos_in = ftos; break;
1899       case Bytecodes::_d2i: // fall through
1900       case Bytecodes::_d2l: // fall through
1901       case Bytecodes::_d2f: tos_in = dtos; break;
1902       default             : ShouldNotReachHere();
1903     }
1904     switch (bytecode()) {
1905       case Bytecodes::_l2i: // fall through
1906       case Bytecodes::_f2i: // fall through
1907       case Bytecodes::_d2i: // fall through
1908       case Bytecodes::_i2b: // fall through
1909       case Bytecodes::_i2c: // fall through
1910       case Bytecodes::_i2s: tos_out = itos; break;
1911       case Bytecodes::_i2l: // fall through
1912       case Bytecodes::_f2l: // fall through
1913       case Bytecodes::_d2l: tos_out = ltos; break;
1914       case Bytecodes::_i2f: // fall through
1915       case Bytecodes::_l2f: // fall through
1916       case Bytecodes::_d2f: tos_out = ftos; break;
1917       case Bytecodes::_i2d: // fall through
1918       case Bytecodes::_l2d: // fall through
1919       case Bytecodes::_f2d: tos_out = dtos; break;
1920       default             : ShouldNotReachHere();
1921     }
1922     transition(tos_in, tos_out);
1923   }
1924 #endif // ASSERT
1925 
1926   // Conversion
1927   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1928   switch (bytecode()) {
1929     case Bytecodes::_i2l:
1930       __ extend_sign(rdx, rax);
1931       break;
1932     case Bytecodes::_i2f:
1933       if (UseSSE >= 1) {
1934         __ cvtsi2ssl(xmm0, rax);
1935       } else {
1936         __ push(rax);          // store int on tos
1937         __ fild_s(at_rsp());   // load int to ST0
1938         __ f2ieee();           // truncate to float size
1939         __ pop(rcx);           // adjust rsp
1940       }
1941       break;
1942     case Bytecodes::_i2d:
1943       if (UseSSE >= 2) {
1944         __ cvtsi2sdl(xmm0, rax);
1945       } else {
1946       __ push(rax);          // add one slot for d2ieee()
1947       __ push(rax);          // store int on tos
1948       __ fild_s(at_rsp());   // load int to ST0
1949       __ d2ieee();           // truncate to double size
1950       __ pop(rcx);           // adjust rsp
1951       __ pop(rcx);
1952       }
1953       break;
1954     case Bytecodes::_i2b:
1955       __ shll(rax, 24);      // truncate upper 24 bits
1956       __ sarl(rax, 24);      // and sign-extend byte
1957       LP64_ONLY(__ movsbl(rax, rax));
1958       break;
1959     case Bytecodes::_i2c:
1960       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1961       LP64_ONLY(__ movzwl(rax, rax));
1962       break;
1963     case Bytecodes::_i2s:
1964       __ shll(rax, 16);      // truncate upper 16 bits
1965       __ sarl(rax, 16);      // and sign-extend short
1966       LP64_ONLY(__ movswl(rax, rax));
1967       break;
1968     case Bytecodes::_l2i:
1969       /* nothing to do */
1970       break;
1971     case Bytecodes::_l2f:
1972       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1973       // 64-bit long values to floats. On 32-bit platforms it is not possible
1974       // to use that instruction with 64-bit operands, therefore the FPU is
1975       // used to perform the conversion.
1976       __ push(rdx);          // store long on tos
1977       __ push(rax);
1978       __ fild_d(at_rsp());   // load long to ST0
1979       __ f2ieee();           // truncate to float size
1980       __ pop(rcx);           // adjust rsp
1981       __ pop(rcx);
1982       if (UseSSE >= 1) {
1983         __ push_f();
1984         __ pop_f(xmm0);
1985       }
1986       break;
1987     case Bytecodes::_l2d:
1988       // On 32-bit platforms the FPU is used for conversion because on
1989       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1990       // instruction with 64-bit operands.
1991       __ push(rdx);          // store long on tos
1992       __ push(rax);
1993       __ fild_d(at_rsp());   // load long to ST0
1994       __ d2ieee();           // truncate to double size
1995       __ pop(rcx);           // adjust rsp
1996       __ pop(rcx);
1997       if (UseSSE >= 2) {
1998         __ push_d();
1999         __ pop_d(xmm0);
2000       }
2001       break;
2002     case Bytecodes::_f2i:
2003       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2004       // as it returns 0 for any NaN.
2005       if (UseSSE >= 1) {
2006         __ push_f(xmm0);
2007       } else {
2008         __ push(rcx);          // reserve space for argument
2009         __ fstp_s(at_rsp());   // pass float argument on stack
2010       }
2011       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2012       break;
2013     case Bytecodes::_f2l:
2014       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2015       // as it returns 0 for any NaN.
2016       if (UseSSE >= 1) {
2017        __ push_f(xmm0);
2018       } else {
2019         __ push(rcx);          // reserve space for argument
2020         __ fstp_s(at_rsp());   // pass float argument on stack
2021       }
2022       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2023       break;
2024     case Bytecodes::_f2d:
2025       if (UseSSE < 1) {
2026         /* nothing to do */
2027       } else if (UseSSE == 1) {
2028         __ push_f(xmm0);
2029         __ pop_f();
2030       } else { // UseSSE >= 2
2031         __ cvtss2sd(xmm0, xmm0);
2032       }
2033       break;
2034     case Bytecodes::_d2i:
2035       if (UseSSE >= 2) {
2036         __ push_d(xmm0);
2037       } else {
2038         __ push(rcx);          // reserve space for argument
2039         __ push(rcx);
2040         __ fstp_d(at_rsp());   // pass double argument on stack
2041       }
2042       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2043       break;
2044     case Bytecodes::_d2l:
2045       if (UseSSE >= 2) {
2046         __ push_d(xmm0);
2047       } else {
2048         __ push(rcx);          // reserve space for argument
2049         __ push(rcx);
2050         __ fstp_d(at_rsp());   // pass double argument on stack
2051       }
2052       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2053       break;
2054     case Bytecodes::_d2f:
2055       if (UseSSE <= 1) {
2056         __ push(rcx);          // reserve space for f2ieee()
2057         __ f2ieee();           // truncate to float size
2058         __ pop(rcx);           // adjust rsp
2059         if (UseSSE == 1) {
2060           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2061           // the conversion is performed using the FPU in this case.
2062           __ push_f();
2063           __ pop_f(xmm0);
2064         }
2065       } else { // UseSSE >= 2
2066         __ cvtsd2ss(xmm0, xmm0);
2067       }
2068       break;
2069     default             :
2070       ShouldNotReachHere();
2071   }
2072 #endif
2073 }
2074 
2075 void TemplateTable::lcmp() {
2076   transition(ltos, itos);
2077 #ifdef _LP64
2078   Label done;
2079   __ pop_l(rdx);
2080   __ cmpq(rdx, rax);
2081   __ movl(rax, -1);
2082   __ jccb(Assembler::less, done);
2083   __ setb(Assembler::notEqual, rax);
2084   __ movzbl(rax, rax);
2085   __ bind(done);
2086 #else
2087 
2088   // y = rdx:rax
2089   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2090   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2091   __ mov(rax, rcx);
2092 #endif
2093 }
2094 
2095 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2096   if ((is_float && UseSSE >= 1) ||
2097       (!is_float && UseSSE >= 2)) {
2098     Label done;
2099     if (is_float) {
2100       // XXX get rid of pop here, use ... reg, mem32
2101       __ pop_f(xmm1);
2102       __ ucomiss(xmm1, xmm0);
2103     } else {
2104       // XXX get rid of pop here, use ... reg, mem64
2105       __ pop_d(xmm1);
2106       __ ucomisd(xmm1, xmm0);
2107     }
2108     if (unordered_result < 0) {
2109       __ movl(rax, -1);
2110       __ jccb(Assembler::parity, done);
2111       __ jccb(Assembler::below, done);
2112       __ setb(Assembler::notEqual, rdx);
2113       __ movzbl(rax, rdx);
2114     } else {
2115       __ movl(rax, 1);
2116       __ jccb(Assembler::parity, done);
2117       __ jccb(Assembler::above, done);
2118       __ movl(rax, 0);
2119       __ jccb(Assembler::equal, done);
2120       __ decrementl(rax);
2121     }
2122     __ bind(done);
2123   } else {
2124 #ifdef _LP64
2125     ShouldNotReachHere();
2126 #else
2127     if (is_float) {
2128       __ fld_s(at_rsp());
2129     } else {
2130       __ fld_d(at_rsp());
2131       __ pop(rdx);
2132     }
2133     __ pop(rcx);
2134     __ fcmp2int(rax, unordered_result < 0);
2135 #endif // _LP64
2136   }
2137 }
2138 
2139 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2140   __ get_method(rcx); // rcx holds method
2141   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2142                                      // holds bumped taken count
2143 
2144   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2145                              InvocationCounter::counter_offset();
2146   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2147                               InvocationCounter::counter_offset();
2148 
2149   // Load up edx with the branch displacement
2150   if (is_wide) {
2151     __ movl(rdx, at_bcp(1));
2152   } else {
2153     __ load_signed_short(rdx, at_bcp(1));
2154   }
2155   __ bswapl(rdx);
2156 
2157   if (!is_wide) {
2158     __ sarl(rdx, 16);
2159   }
2160   LP64_ONLY(__ movl2ptr(rdx, rdx));
2161 
2162   // Handle all the JSR stuff here, then exit.
2163   // It's much shorter and cleaner than intermingling with the non-JSR
2164   // normal-branch stuff occurring below.
2165   if (is_jsr) {
2166     // Pre-load the next target bytecode into rbx
2167     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2168 
2169     // compute return address as bci in rax
2170     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2171                         in_bytes(ConstMethod::codes_offset())));
2172     __ subptr(rax, Address(rcx, Method::const_offset()));
2173     // Adjust the bcp in r13 by the displacement in rdx
2174     __ addptr(rbcp, rdx);
2175     // jsr returns atos that is not an oop
2176     __ push_i(rax);
2177     __ dispatch_only(vtos, true);
2178     return;
2179   }
2180 
2181   // Normal (non-jsr) branch handling
2182 
2183   // Adjust the bcp in r13 by the displacement in rdx
2184   __ addptr(rbcp, rdx);
2185 
2186   assert(UseLoopCounter || !UseOnStackReplacement,
2187          "on-stack-replacement requires loop counters");
2188   Label backedge_counter_overflow;
2189   Label profile_method;
2190   Label dispatch;
2191   if (UseLoopCounter) {
2192     // increment backedge counter for backward branches
2193     // rax: MDO
2194     // rbx: MDO bumped taken-count
2195     // rcx: method
2196     // rdx: target offset
2197     // r13: target bcp
2198     // r14: locals pointer
2199     __ testl(rdx, rdx);             // check if forward or backward branch
2200     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2201 
2202     // check if MethodCounters exists
2203     Label has_counters;
2204     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2205     __ testptr(rax, rax);
2206     __ jcc(Assembler::notZero, has_counters);
2207     __ push(rdx);
2208     __ push(rcx);
2209     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2210                rcx);
2211     __ pop(rcx);
2212     __ pop(rdx);
2213     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2214     __ testptr(rax, rax);
2215     __ jcc(Assembler::zero, dispatch);
2216     __ bind(has_counters);
2217 
2218     if (TieredCompilation) {
2219       Label no_mdo;
2220       int increment = InvocationCounter::count_increment;
2221       if (ProfileInterpreter) {
2222         // Are we profiling?
2223         __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2224         __ testptr(rbx, rbx);
2225         __ jccb(Assembler::zero, no_mdo);
2226         // Increment the MDO backedge counter
2227         const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2228                                            in_bytes(InvocationCounter::counter_offset()));
2229         const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2230         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, rax, false, Assembler::zero,
2231                                    UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2232         __ jmp(dispatch);
2233       }
2234       __ bind(no_mdo);
2235       // Increment backedge counter in MethodCounters*
2236       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2237       const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2238       __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2239                                  rax, false, Assembler::zero,
2240                                  UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2241   } else { // not TieredCompilation
2242       // increment counter
2243       __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2244       __ movl(rax, Address(rcx, be_offset));        // load backedge counter
2245       __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2246       __ movl(Address(rcx, be_offset), rax);        // store counter
2247 
2248       __ movl(rax, Address(rcx, inv_offset));    // load invocation counter
2249 
2250       __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2251       __ addl(rax, Address(rcx, be_offset));        // add both counters
2252 
2253       if (ProfileInterpreter) {
2254         // Test to see if we should create a method data oop
2255         __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2256         __ jcc(Assembler::less, dispatch);
2257 
2258         // if no method data exists, go to profile method
2259         __ test_method_data_pointer(rax, profile_method);
2260 
2261         if (UseOnStackReplacement) {
2262           // check for overflow against rbx which is the MDO taken count
2263           __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2264           __ jcc(Assembler::below, dispatch);
2265 
2266           // When ProfileInterpreter is on, the backedge_count comes
2267           // from the MethodData*, which value does not get reset on
2268           // the call to frequency_counter_overflow().  To avoid
2269           // excessive calls to the overflow routine while the method is
2270           // being compiled, add a second test to make sure the overflow
2271           // function is called only once every overflow_frequency.
2272           const int overflow_frequency = 1024;
2273           __ andl(rbx, overflow_frequency - 1);
2274           __ jcc(Assembler::zero, backedge_counter_overflow);
2275 
2276         }
2277       } else {
2278         if (UseOnStackReplacement) {
2279           // check for overflow against rax, which is the sum of the
2280           // counters
2281           __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2282           __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2283 
2284         }
2285       }
2286     }
2287     __ bind(dispatch);
2288   }
2289 
2290   // Pre-load the next target bytecode into rbx
2291   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2292 
2293   // continue with the bytecode @ target
2294   // rax: return bci for jsr's, unused otherwise
2295   // rbx: target bytecode
2296   // r13: target bcp
2297   __ dispatch_only(vtos, true);
2298 
2299   if (UseLoopCounter) {
2300     if (ProfileInterpreter) {
2301       // Out-of-line code to allocate method data oop.
2302       __ bind(profile_method);
2303       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2304       __ set_method_data_pointer_for_bcp();
2305       __ jmp(dispatch);
2306     }
2307 
2308     if (UseOnStackReplacement) {
2309       // invocation counter overflow
2310       __ bind(backedge_counter_overflow);
2311       __ negptr(rdx);
2312       __ addptr(rdx, rbcp); // branch bcp
2313       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2314       __ call_VM(noreg,
2315                  CAST_FROM_FN_PTR(address,
2316                                   InterpreterRuntime::frequency_counter_overflow),
2317                  rdx);
2318 
2319       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2320       // rdx: scratch
2321       // r14: locals pointer
2322       // r13: bcp
2323       __ testptr(rax, rax);                        // test result
2324       __ jcc(Assembler::zero, dispatch);         // no osr if null
2325       // nmethod may have been invalidated (VM may block upon call_VM return)
2326       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2327       __ jcc(Assembler::notEqual, dispatch);
2328 
2329       // We have the address of an on stack replacement routine in rax.
2330       // In preparation of invoking it, first we must migrate the locals
2331       // and monitors from off the interpreter frame on the stack.
2332       // Ensure to save the osr nmethod over the migration call,
2333       // it will be preserved in rbx.
2334       __ mov(rbx, rax);
2335 
2336       NOT_LP64(__ get_thread(rcx));
2337 
2338       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2339 
2340       // rax is OSR buffer, move it to expected parameter location
2341       LP64_ONLY(__ mov(j_rarg0, rax));
2342       NOT_LP64(__ mov(rcx, rax));
2343       // We use j_rarg definitions here so that registers don't conflict as parameter
2344       // registers change across platforms as we are in the midst of a calling
2345       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2346 
2347       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2348       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2349 
2350       // pop the interpreter frame
2351       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2352       __ leave();                                // remove frame anchor
2353       __ pop(retaddr);                           // get return address
2354       __ mov(rsp, sender_sp);                   // set sp to sender sp
2355       // Ensure compiled code always sees stack at proper alignment
2356       __ andptr(rsp, -(StackAlignmentInBytes));
2357 
2358       // unlike x86 we need no specialized return from compiled code
2359       // to the interpreter or the call stub.
2360 
2361       // push the return address
2362       __ push(retaddr);
2363 
2364       // and begin the OSR nmethod
2365       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2366     }
2367   }
2368 }
2369 
2370 void TemplateTable::if_0cmp(Condition cc) {
2371   transition(itos, vtos);
2372   // assume branch is more often taken than not (loops use backward branches)
2373   Label not_taken;
2374   __ testl(rax, rax);
2375   __ jcc(j_not(cc), not_taken);
2376   branch(false, false);
2377   __ bind(not_taken);
2378   __ profile_not_taken_branch(rax);
2379 }
2380 
2381 void TemplateTable::if_icmp(Condition cc) {
2382   transition(itos, vtos);
2383   // assume branch is more often taken than not (loops use backward branches)
2384   Label not_taken;
2385   __ pop_i(rdx);
2386   __ cmpl(rdx, rax);
2387   __ jcc(j_not(cc), not_taken);
2388   branch(false, false);
2389   __ bind(not_taken);
2390   __ profile_not_taken_branch(rax);
2391 }
2392 
2393 void TemplateTable::if_nullcmp(Condition cc) {
2394   transition(atos, vtos);
2395   // assume branch is more often taken than not (loops use backward branches)
2396   Label not_taken;
2397   __ testptr(rax, rax);
2398   __ jcc(j_not(cc), not_taken);
2399   branch(false, false);
2400   __ bind(not_taken);
2401   __ profile_not_taken_branch(rax);
2402 }
2403 
2404 void TemplateTable::if_acmp(Condition cc) {
2405   transition(atos, vtos);
2406   // assume branch is more often taken than not (loops use backward branches)
2407   Label not_taken;
2408   __ pop_ptr(rdx);
2409   __ cmpoop(rdx, rax);
2410   __ jcc(j_not(cc), not_taken);
2411   branch(false, false);
2412   __ bind(not_taken);
2413   __ profile_not_taken_branch(rax);
2414 }
2415 
2416 void TemplateTable::ret() {
2417   transition(vtos, vtos);
2418   locals_index(rbx);
2419   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2420   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2421   __ profile_ret(rbx, rcx);
2422   __ get_method(rax);
2423   __ movptr(rbcp, Address(rax, Method::const_offset()));
2424   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2425                       ConstMethod::codes_offset()));
2426   __ dispatch_next(vtos, 0, true);
2427 }
2428 
2429 void TemplateTable::wide_ret() {
2430   transition(vtos, vtos);
2431   locals_index_wide(rbx);
2432   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2433   __ profile_ret(rbx, rcx);
2434   __ get_method(rax);
2435   __ movptr(rbcp, Address(rax, Method::const_offset()));
2436   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2437   __ dispatch_next(vtos, 0, true);
2438 }
2439 
2440 void TemplateTable::tableswitch() {
2441   Label default_case, continue_execution;
2442   transition(itos, vtos);
2443 
2444   // align r13/rsi
2445   __ lea(rbx, at_bcp(BytesPerInt));
2446   __ andptr(rbx, -BytesPerInt);
2447   // load lo & hi
2448   __ movl(rcx, Address(rbx, BytesPerInt));
2449   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2450   __ bswapl(rcx);
2451   __ bswapl(rdx);
2452   // check against lo & hi
2453   __ cmpl(rax, rcx);
2454   __ jcc(Assembler::less, default_case);
2455   __ cmpl(rax, rdx);
2456   __ jcc(Assembler::greater, default_case);
2457   // lookup dispatch offset
2458   __ subl(rax, rcx);
2459   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2460   __ profile_switch_case(rax, rbx, rcx);
2461   // continue execution
2462   __ bind(continue_execution);
2463   __ bswapl(rdx);
2464   LP64_ONLY(__ movl2ptr(rdx, rdx));
2465   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2466   __ addptr(rbcp, rdx);
2467   __ dispatch_only(vtos, true);
2468   // handle default
2469   __ bind(default_case);
2470   __ profile_switch_default(rax);
2471   __ movl(rdx, Address(rbx, 0));
2472   __ jmp(continue_execution);
2473 }
2474 
2475 void TemplateTable::lookupswitch() {
2476   transition(itos, itos);
2477   __ stop("lookupswitch bytecode should have been rewritten");
2478 }
2479 
2480 void TemplateTable::fast_linearswitch() {
2481   transition(itos, vtos);
2482   Label loop_entry, loop, found, continue_execution;
2483   // bswap rax so we can avoid bswapping the table entries
2484   __ bswapl(rax);
2485   // align r13
2486   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2487                                     // this instruction (change offsets
2488                                     // below)
2489   __ andptr(rbx, -BytesPerInt);
2490   // set counter
2491   __ movl(rcx, Address(rbx, BytesPerInt));
2492   __ bswapl(rcx);
2493   __ jmpb(loop_entry);
2494   // table search
2495   __ bind(loop);
2496   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2497   __ jcc(Assembler::equal, found);
2498   __ bind(loop_entry);
2499   __ decrementl(rcx);
2500   __ jcc(Assembler::greaterEqual, loop);
2501   // default case
2502   __ profile_switch_default(rax);
2503   __ movl(rdx, Address(rbx, 0));
2504   __ jmp(continue_execution);
2505   // entry found -> get offset
2506   __ bind(found);
2507   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2508   __ profile_switch_case(rcx, rax, rbx);
2509   // continue execution
2510   __ bind(continue_execution);
2511   __ bswapl(rdx);
2512   __ movl2ptr(rdx, rdx);
2513   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2514   __ addptr(rbcp, rdx);
2515   __ dispatch_only(vtos, true);
2516 }
2517 
2518 void TemplateTable::fast_binaryswitch() {
2519   transition(itos, vtos);
2520   // Implementation using the following core algorithm:
2521   //
2522   // int binary_search(int key, LookupswitchPair* array, int n) {
2523   //   // Binary search according to "Methodik des Programmierens" by
2524   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2525   //   int i = 0;
2526   //   int j = n;
2527   //   while (i+1 < j) {
2528   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2529   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2530   //     // where a stands for the array and assuming that the (inexisting)
2531   //     // element a[n] is infinitely big.
2532   //     int h = (i + j) >> 1;
2533   //     // i < h < j
2534   //     if (key < array[h].fast_match()) {
2535   //       j = h;
2536   //     } else {
2537   //       i = h;
2538   //     }
2539   //   }
2540   //   // R: a[i] <= key < a[i+1] or Q
2541   //   // (i.e., if key is within array, i is the correct index)
2542   //   return i;
2543   // }
2544 
2545   // Register allocation
2546   const Register key   = rax; // already set (tosca)
2547   const Register array = rbx;
2548   const Register i     = rcx;
2549   const Register j     = rdx;
2550   const Register h     = rdi;
2551   const Register temp  = rsi;
2552 
2553   // Find array start
2554   NOT_LP64(__ save_bcp());
2555 
2556   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2557                                           // get rid of this
2558                                           // instruction (change
2559                                           // offsets below)
2560   __ andptr(array, -BytesPerInt);
2561 
2562   // Initialize i & j
2563   __ xorl(i, i);                            // i = 0;
2564   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2565 
2566   // Convert j into native byteordering
2567   __ bswapl(j);
2568 
2569   // And start
2570   Label entry;
2571   __ jmp(entry);
2572 
2573   // binary search loop
2574   {
2575     Label loop;
2576     __ bind(loop);
2577     // int h = (i + j) >> 1;
2578     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2579     __ sarl(h, 1);                               // h = (i + j) >> 1;
2580     // if (key < array[h].fast_match()) {
2581     //   j = h;
2582     // } else {
2583     //   i = h;
2584     // }
2585     // Convert array[h].match to native byte-ordering before compare
2586     __ movl(temp, Address(array, h, Address::times_8));
2587     __ bswapl(temp);
2588     __ cmpl(key, temp);
2589     // j = h if (key <  array[h].fast_match())
2590     __ cmov32(Assembler::less, j, h);
2591     // i = h if (key >= array[h].fast_match())
2592     __ cmov32(Assembler::greaterEqual, i, h);
2593     // while (i+1 < j)
2594     __ bind(entry);
2595     __ leal(h, Address(i, 1)); // i+1
2596     __ cmpl(h, j);             // i+1 < j
2597     __ jcc(Assembler::less, loop);
2598   }
2599 
2600   // end of binary search, result index is i (must check again!)
2601   Label default_case;
2602   // Convert array[i].match to native byte-ordering before compare
2603   __ movl(temp, Address(array, i, Address::times_8));
2604   __ bswapl(temp);
2605   __ cmpl(key, temp);
2606   __ jcc(Assembler::notEqual, default_case);
2607 
2608   // entry found -> j = offset
2609   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2610   __ profile_switch_case(i, key, array);
2611   __ bswapl(j);
2612   LP64_ONLY(__ movslq(j, j));
2613 
2614   NOT_LP64(__ restore_bcp());
2615   NOT_LP64(__ restore_locals());                           // restore rdi
2616 
2617   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2618   __ addptr(rbcp, j);
2619   __ dispatch_only(vtos, true);
2620 
2621   // default case -> j = default offset
2622   __ bind(default_case);
2623   __ profile_switch_default(i);
2624   __ movl(j, Address(array, -2 * BytesPerInt));
2625   __ bswapl(j);
2626   LP64_ONLY(__ movslq(j, j));
2627 
2628   NOT_LP64(__ restore_bcp());
2629   NOT_LP64(__ restore_locals());
2630 
2631   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2632   __ addptr(rbcp, j);
2633   __ dispatch_only(vtos, true);
2634 }
2635 
2636 void TemplateTable::_return(TosState state) {
2637   transition(state, state);
2638 
2639   assert(_desc->calls_vm(),
2640          "inconsistent calls_vm information"); // call in remove_activation
2641 
2642   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2643     assert(state == vtos, "only valid state");
2644     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2645     __ movptr(robj, aaddress(0));
2646     __ load_klass(rdi, robj);
2647     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2648     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2649     Label skip_register_finalizer;
2650     __ jcc(Assembler::zero, skip_register_finalizer);
2651 
2652     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2653 
2654     __ bind(skip_register_finalizer);
2655   }
2656 
2657   if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2658     Label no_safepoint;
2659     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2660 #ifdef _LP64
2661     __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2662 #else
2663     const Register thread = rdi;
2664     __ get_thread(thread);
2665     __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2666 #endif
2667     __ jcc(Assembler::zero, no_safepoint);
2668     __ push(state);
2669     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2670                                     InterpreterRuntime::at_safepoint));
2671     __ pop(state);
2672     __ bind(no_safepoint);
2673   }
2674 
2675   // Narrow result if state is itos but result type is smaller.
2676   // Need to narrow in the return bytecode rather than in generate_return_entry
2677   // since compiled code callers expect the result to already be narrowed.
2678   if (state == itos) {
2679     __ narrow(rax);
2680   }
2681   __ remove_activation(state, rbcp);
2682 
2683   __ jmp(rbcp);
2684 }
2685 
2686 // ----------------------------------------------------------------------------
2687 // Volatile variables demand their effects be made known to all CPU's
2688 // in order.  Store buffers on most chips allow reads & writes to
2689 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2690 // without some kind of memory barrier (i.e., it's not sufficient that
2691 // the interpreter does not reorder volatile references, the hardware
2692 // also must not reorder them).
2693 //
2694 // According to the new Java Memory Model (JMM):
2695 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2696 //     writes act as aquire & release, so:
2697 // (2) A read cannot let unrelated NON-volatile memory refs that
2698 //     happen after the read float up to before the read.  It's OK for
2699 //     non-volatile memory refs that happen before the volatile read to
2700 //     float down below it.
2701 // (3) Similar a volatile write cannot let unrelated NON-volatile
2702 //     memory refs that happen BEFORE the write float down to after the
2703 //     write.  It's OK for non-volatile memory refs that happen after the
2704 //     volatile write to float up before it.
2705 //
2706 // We only put in barriers around volatile refs (they are expensive),
2707 // not _between_ memory refs (that would require us to track the
2708 // flavor of the previous memory refs).  Requirements (2) and (3)
2709 // require some barriers before volatile stores and after volatile
2710 // loads.  These nearly cover requirement (1) but miss the
2711 // volatile-store-volatile-load case.  This final case is placed after
2712 // volatile-stores although it could just as well go before
2713 // volatile-loads.
2714 
2715 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2716   // Helper function to insert a is-volatile test and memory barrier
2717   if(!os::is_MP()) return;    // Not needed on single CPU
2718   __ membar(order_constraint);
2719 }
2720 
2721 void TemplateTable::resolve_cache_and_index(int byte_no,
2722                                             Register Rcache,
2723                                             Register index,
2724                                             size_t index_size) {
2725   const Register temp = rbx;
2726   assert_different_registers(Rcache, index, temp);
2727 
2728   Label resolved;
2729 
2730   Bytecodes::Code code = bytecode();
2731   switch (code) {
2732   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2733   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2734   default: break;
2735   }
2736 
2737   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2738   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2739   __ cmpl(temp, code);  // have we resolved this bytecode?
2740   __ jcc(Assembler::equal, resolved);
2741 
2742   // resolve first time through
2743   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2744   __ movl(temp, code);
2745   __ call_VM(noreg, entry, temp);
2746   // Update registers with resolved info
2747   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2748   __ bind(resolved);
2749 }
2750 
2751 // The cache and index registers must be set before call
2752 void TemplateTable::load_field_cp_cache_entry(Register obj,
2753                                               Register cache,
2754                                               Register index,
2755                                               Register off,
2756                                               Register flags,
2757                                               bool is_static = false) {
2758   assert_different_registers(cache, index, flags, off);
2759 
2760   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2761   // Field offset
2762   __ movptr(off, Address(cache, index, Address::times_ptr,
2763                          in_bytes(cp_base_offset +
2764                                   ConstantPoolCacheEntry::f2_offset())));
2765   // Flags
2766   __ movl(flags, Address(cache, index, Address::times_ptr,
2767                          in_bytes(cp_base_offset +
2768                                   ConstantPoolCacheEntry::flags_offset())));
2769 
2770   // klass overwrite register
2771   if (is_static) {
2772     __ movptr(obj, Address(cache, index, Address::times_ptr,
2773                            in_bytes(cp_base_offset +
2774                                     ConstantPoolCacheEntry::f1_offset())));
2775     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2776     __ movptr(obj, Address(obj, mirror_offset));
2777     __ resolve_oop_handle(obj);
2778   }
2779 }
2780 
2781 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2782                                                Register method,
2783                                                Register itable_index,
2784                                                Register flags,
2785                                                bool is_invokevirtual,
2786                                                bool is_invokevfinal, /*unused*/
2787                                                bool is_invokedynamic) {
2788   // setup registers
2789   const Register cache = rcx;
2790   const Register index = rdx;
2791   assert_different_registers(method, flags);
2792   assert_different_registers(method, cache, index);
2793   assert_different_registers(itable_index, flags);
2794   assert_different_registers(itable_index, cache, index);
2795   // determine constant pool cache field offsets
2796   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2797   const int method_offset = in_bytes(
2798     ConstantPoolCache::base_offset() +
2799       ((byte_no == f2_byte)
2800        ? ConstantPoolCacheEntry::f2_offset()
2801        : ConstantPoolCacheEntry::f1_offset()));
2802   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2803                                     ConstantPoolCacheEntry::flags_offset());
2804   // access constant pool cache fields
2805   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2806                                     ConstantPoolCacheEntry::f2_offset());
2807 
2808   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2809   resolve_cache_and_index(byte_no, cache, index, index_size);
2810     __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2811 
2812   if (itable_index != noreg) {
2813     // pick up itable or appendix index from f2 also:
2814     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2815   }
2816   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2817 }
2818 
2819 // The registers cache and index expected to be set before call.
2820 // Correct values of the cache and index registers are preserved.
2821 void TemplateTable::jvmti_post_field_access(Register cache,
2822                                             Register index,
2823                                             bool is_static,
2824                                             bool has_tos) {
2825   if (JvmtiExport::can_post_field_access()) {
2826     // Check to see if a field access watch has been set before we take
2827     // the time to call into the VM.
2828     Label L1;
2829     assert_different_registers(cache, index, rax);
2830     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2831     __ testl(rax,rax);
2832     __ jcc(Assembler::zero, L1);
2833 
2834     // cache entry pointer
2835     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2836     __ shll(index, LogBytesPerWord);
2837     __ addptr(cache, index);
2838     if (is_static) {
2839       __ xorptr(rax, rax);      // NULL object reference
2840     } else {
2841       __ pop(atos);         // Get the object
2842       __ verify_oop(rax);
2843       __ push(atos);        // Restore stack state
2844     }
2845     // rax,:   object pointer or NULL
2846     // cache: cache entry pointer
2847     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2848                rax, cache);
2849     __ get_cache_and_index_at_bcp(cache, index, 1);
2850     __ bind(L1);
2851   }
2852 }
2853 
2854 void TemplateTable::pop_and_check_object(Register r) {
2855   __ pop_ptr(r);
2856   __ null_check(r);  // for field access must check obj.
2857   __ verify_oop(r);
2858 }
2859 
2860 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2861   transition(vtos, vtos);
2862 
2863   const Register cache = rcx;
2864   const Register index = rdx;
2865   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2866   const Register off   = rbx;
2867   const Register flags = rax;
2868   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2869 
2870   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2871   jvmti_post_field_access(cache, index, is_static, false);
2872   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2873 
2874   if (!is_static) pop_and_check_object(obj);
2875 
2876   const Address field(obj, off, Address::times_1, 0*wordSize);
2877 
2878   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2879 
2880   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2881   // Make sure we don't need to mask edx after the above shift
2882   assert(btos == 0, "change code, btos != 0");
2883 
2884   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2885 
2886   __ jcc(Assembler::notZero, notByte);
2887   // btos
2888   __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2889   __ push(btos);
2890   // Rewrite bytecode to be faster
2891   if (!is_static && rc == may_rewrite) {
2892     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2893   }
2894   __ jmp(Done);
2895 
2896   __ bind(notByte);
2897   __ cmpl(flags, ztos);
2898   __ jcc(Assembler::notEqual, notBool);
2899 
2900   // ztos (same code as btos)
2901   __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2902   __ push(ztos);
2903   // Rewrite bytecode to be faster
2904   if (!is_static && rc == may_rewrite) {
2905     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2906     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2907   }
2908   __ jmp(Done);
2909 
2910   __ bind(notBool);
2911   __ cmpl(flags, atos);
2912   __ jcc(Assembler::notEqual, notObj);
2913   // atos
2914   do_oop_load(_masm, field, rax);
2915   __ push(atos);
2916   if (!is_static && rc == may_rewrite) {
2917     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2918   }
2919   __ jmp(Done);
2920 
2921   __ bind(notObj);
2922   __ cmpl(flags, itos);
2923   __ jcc(Assembler::notEqual, notInt);
2924   // itos
2925   __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2926   __ push(itos);
2927   // Rewrite bytecode to be faster
2928   if (!is_static && rc == may_rewrite) {
2929     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2930   }
2931   __ jmp(Done);
2932 
2933   __ bind(notInt);
2934   __ cmpl(flags, ctos);
2935   __ jcc(Assembler::notEqual, notChar);
2936   // ctos
2937   __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2938   __ push(ctos);
2939   // Rewrite bytecode to be faster
2940   if (!is_static && rc == may_rewrite) {
2941     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2942   }
2943   __ jmp(Done);
2944 
2945   __ bind(notChar);
2946   __ cmpl(flags, stos);
2947   __ jcc(Assembler::notEqual, notShort);
2948   // stos
2949   __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2950   __ push(stos);
2951   // Rewrite bytecode to be faster
2952   if (!is_static && rc == may_rewrite) {
2953     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2954   }
2955   __ jmp(Done);
2956 
2957   __ bind(notShort);
2958   __ cmpl(flags, ltos);
2959   __ jcc(Assembler::notEqual, notLong);
2960   // ltos
2961     // Generate code as if volatile (x86_32).  There just aren't enough registers to
2962     // save that information and this code is faster than the test.
2963   __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2964   __ push(ltos);
2965   // Rewrite bytecode to be faster
2966   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2967   __ jmp(Done);
2968 
2969   __ bind(notLong);
2970   __ cmpl(flags, ftos);
2971   __ jcc(Assembler::notEqual, notFloat);
2972   // ftos
2973 
2974   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2975   __ push(ftos);
2976   // Rewrite bytecode to be faster
2977   if (!is_static && rc == may_rewrite) {
2978     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2979   }
2980   __ jmp(Done);
2981 
2982   __ bind(notFloat);
2983 #ifdef ASSERT
2984   __ cmpl(flags, dtos);
2985   __ jcc(Assembler::notEqual, notDouble);
2986 #endif
2987   // dtos
2988   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
2989   __ push(dtos);
2990   // Rewrite bytecode to be faster
2991   if (!is_static && rc == may_rewrite) {
2992     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2993   }
2994 #ifdef ASSERT
2995   __ jmp(Done);
2996 
2997 
2998   __ bind(notDouble);
2999   __ stop("Bad state");
3000 #endif
3001 
3002   __ bind(Done);
3003   // [jk] not needed currently
3004   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3005   //                                              Assembler::LoadStore));
3006 }
3007 
3008 void TemplateTable::getfield(int byte_no) {
3009   getfield_or_static(byte_no, false);
3010 }
3011 
3012 void TemplateTable::nofast_getfield(int byte_no) {
3013   getfield_or_static(byte_no, false, may_not_rewrite);
3014 }
3015 
3016 void TemplateTable::getstatic(int byte_no) {
3017   getfield_or_static(byte_no, true);
3018 }
3019 
3020 
3021 // The registers cache and index expected to be set before call.
3022 // The function may destroy various registers, just not the cache and index registers.
3023 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3024 
3025   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
3026   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
3027   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
3028   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3029 
3030   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3031 
3032   if (JvmtiExport::can_post_field_modification()) {
3033     // Check to see if a field modification watch has been set before
3034     // we take the time to call into the VM.
3035     Label L1;
3036     assert_different_registers(cache, index, rax);
3037     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3038     __ testl(rax, rax);
3039     __ jcc(Assembler::zero, L1);
3040 
3041     __ get_cache_and_index_at_bcp(robj, RDX, 1);
3042 
3043 
3044     if (is_static) {
3045       // Life is simple.  Null out the object pointer.
3046       __ xorl(RBX, RBX);
3047 
3048     } else {
3049       // Life is harder. The stack holds the value on top, followed by
3050       // the object.  We don't know the size of the value, though; it
3051       // could be one or two words depending on its type. As a result,
3052       // we must find the type to determine where the object is.
3053 #ifndef _LP64
3054       Label two_word, valsize_known;
3055 #endif
3056       __ movl(RCX, Address(robj, RDX,
3057                            Address::times_ptr,
3058                            in_bytes(cp_base_offset +
3059                                      ConstantPoolCacheEntry::flags_offset())));
3060       NOT_LP64(__ mov(rbx, rsp));
3061       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3062 
3063       // Make sure we don't need to mask rcx after the above shift
3064       ConstantPoolCacheEntry::verify_tos_state_shift();
3065 #ifdef _LP64
3066       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3067       __ cmpl(c_rarg3, ltos);
3068       __ cmovptr(Assembler::equal,
3069                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3070       __ cmpl(c_rarg3, dtos);
3071       __ cmovptr(Assembler::equal,
3072                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3073 #else
3074       __ cmpl(rcx, ltos);
3075       __ jccb(Assembler::equal, two_word);
3076       __ cmpl(rcx, dtos);
3077       __ jccb(Assembler::equal, two_word);
3078       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3079       __ jmpb(valsize_known);
3080 
3081       __ bind(two_word);
3082       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3083 
3084       __ bind(valsize_known);
3085       // setup object pointer
3086       __ movptr(rbx, Address(rbx, 0));
3087 #endif
3088     }
3089     // cache entry pointer
3090     __ addptr(robj, in_bytes(cp_base_offset));
3091     __ shll(RDX, LogBytesPerWord);
3092     __ addptr(robj, RDX);
3093     // object (tos)
3094     __ mov(RCX, rsp);
3095     // c_rarg1: object pointer set up above (NULL if static)
3096     // c_rarg2: cache entry pointer
3097     // c_rarg3: jvalue object on the stack
3098     __ call_VM(noreg,
3099                CAST_FROM_FN_PTR(address,
3100                                 InterpreterRuntime::post_field_modification),
3101                RBX, robj, RCX);
3102     __ get_cache_and_index_at_bcp(cache, index, 1);
3103     __ bind(L1);
3104   }
3105 }
3106 
3107 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3108   transition(vtos, vtos);
3109 
3110   const Register cache = rcx;
3111   const Register index = rdx;
3112   const Register obj   = rcx;
3113   const Register off   = rbx;
3114   const Register flags = rax;
3115   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3116 
3117   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3118   jvmti_post_field_mod(cache, index, is_static);
3119   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3120 
3121   // [jk] not needed currently
3122   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3123   //                                              Assembler::StoreStore));
3124 
3125   Label notVolatile, Done;
3126   __ movl(rdx, flags);
3127   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3128   __ andl(rdx, 0x1);
3129 
3130   // field addresses
3131   const Address field(obj, off, Address::times_1, 0*wordSize);
3132   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3133 
3134   Label notByte, notBool, notInt, notShort, notChar,
3135         notLong, notFloat, notObj, notDouble;
3136 
3137   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3138 
3139   assert(btos == 0, "change code, btos != 0");
3140   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3141   __ jcc(Assembler::notZero, notByte);
3142 
3143   // btos
3144   {
3145     __ pop(btos);
3146     if (!is_static) pop_and_check_object(obj);
3147     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3148     if (!is_static && rc == may_rewrite) {
3149       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3150     }
3151     __ jmp(Done);
3152   }
3153 
3154   __ bind(notByte);
3155   __ cmpl(flags, ztos);
3156   __ jcc(Assembler::notEqual, notBool);
3157 
3158   // ztos
3159   {
3160     __ pop(ztos);
3161     if (!is_static) pop_and_check_object(obj);
3162     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3163     if (!is_static && rc == may_rewrite) {
3164       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3165     }
3166     __ jmp(Done);
3167   }
3168 
3169   __ bind(notBool);
3170   __ cmpl(flags, atos);
3171   __ jcc(Assembler::notEqual, notObj);
3172 
3173   // atos
3174   {
3175     __ pop(atos);
3176     if (!is_static) pop_and_check_object(obj);
3177     // Store into the field
3178     do_oop_store(_masm, field, rax);
3179     if (!is_static && rc == may_rewrite) {
3180       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3181     }
3182     __ jmp(Done);
3183   }
3184 
3185   __ bind(notObj);
3186   __ cmpl(flags, itos);
3187   __ jcc(Assembler::notEqual, notInt);
3188 
3189   // itos
3190   {
3191     __ pop(itos);
3192     if (!is_static) pop_and_check_object(obj);
3193     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3194     if (!is_static && rc == may_rewrite) {
3195       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3196     }
3197     __ jmp(Done);
3198   }
3199 
3200   __ bind(notInt);
3201   __ cmpl(flags, ctos);
3202   __ jcc(Assembler::notEqual, notChar);
3203 
3204   // ctos
3205   {
3206     __ pop(ctos);
3207     if (!is_static) pop_and_check_object(obj);
3208     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3209     if (!is_static && rc == may_rewrite) {
3210       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3211     }
3212     __ jmp(Done);
3213   }
3214 
3215   __ bind(notChar);
3216   __ cmpl(flags, stos);
3217   __ jcc(Assembler::notEqual, notShort);
3218 
3219   // stos
3220   {
3221     __ pop(stos);
3222     if (!is_static) pop_and_check_object(obj);
3223     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3224     if (!is_static && rc == may_rewrite) {
3225       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3226     }
3227     __ jmp(Done);
3228   }
3229 
3230   __ bind(notShort);
3231   __ cmpl(flags, ltos);
3232   __ jcc(Assembler::notEqual, notLong);
3233 
3234   // ltos
3235 #ifdef _LP64
3236   {
3237     __ pop(ltos);
3238     if (!is_static) pop_and_check_object(obj);
3239     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos*/, noreg, noreg);
3240     if (!is_static && rc == may_rewrite) {
3241       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3242     }
3243     __ jmp(Done);
3244   }
3245 #else
3246   {
3247     Label notVolatileLong;
3248     __ testl(rdx, rdx);
3249     __ jcc(Assembler::zero, notVolatileLong);
3250 
3251     __ pop(ltos);  // overwrites rdx, do this after testing volatile.
3252     if (!is_static) pop_and_check_object(obj);
3253 
3254     // Replace with real volatile test
3255     __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos */, noreg, noreg);
3256     // volatile_barrier();
3257     volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3258                                                  Assembler::StoreStore));
3259     // Don't rewrite volatile version
3260     __ jmp(notVolatile);
3261 
3262     __ bind(notVolatileLong);
3263 
3264     __ pop(ltos);  // overwrites rdx
3265     if (!is_static) pop_and_check_object(obj);
3266     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg);
3267     // Don't rewrite to _fast_lputfield for potential volatile case.
3268     __ jmp(notVolatile);
3269   }
3270 #endif // _LP64
3271 
3272   __ bind(notLong);
3273   __ cmpl(flags, ftos);
3274   __ jcc(Assembler::notEqual, notFloat);
3275 
3276   // ftos
3277   {
3278     __ pop(ftos);
3279     if (!is_static) pop_and_check_object(obj);
3280     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3281     if (!is_static && rc == may_rewrite) {
3282       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3283     }
3284     __ jmp(Done);
3285   }
3286 
3287   __ bind(notFloat);
3288 #ifdef ASSERT
3289   __ cmpl(flags, dtos);
3290   __ jcc(Assembler::notEqual, notDouble);
3291 #endif
3292 
3293   // dtos
3294   {
3295     __ pop(dtos);
3296     if (!is_static) pop_and_check_object(obj);
3297     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3298     if (!is_static && rc == may_rewrite) {
3299       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3300     }
3301   }
3302 
3303 #ifdef ASSERT
3304   __ jmp(Done);
3305 
3306   __ bind(notDouble);
3307   __ stop("Bad state");
3308 #endif
3309 
3310   __ bind(Done);
3311 
3312   // Check for volatile store
3313   __ testl(rdx, rdx);
3314   __ jcc(Assembler::zero, notVolatile);
3315   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3316                                                Assembler::StoreStore));
3317   __ bind(notVolatile);
3318 }
3319 
3320 void TemplateTable::putfield(int byte_no) {
3321   putfield_or_static(byte_no, false);
3322 }
3323 
3324 void TemplateTable::nofast_putfield(int byte_no) {
3325   putfield_or_static(byte_no, false, may_not_rewrite);
3326 }
3327 
3328 void TemplateTable::putstatic(int byte_no) {
3329   putfield_or_static(byte_no, true);
3330 }
3331 
3332 void TemplateTable::jvmti_post_fast_field_mod() {
3333 
3334   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3335 
3336   if (JvmtiExport::can_post_field_modification()) {
3337     // Check to see if a field modification watch has been set before
3338     // we take the time to call into the VM.
3339     Label L2;
3340     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3341     __ testl(scratch, scratch);
3342     __ jcc(Assembler::zero, L2);
3343     __ pop_ptr(rbx);                  // copy the object pointer from tos
3344     __ verify_oop(rbx);
3345     __ push_ptr(rbx);                 // put the object pointer back on tos
3346     // Save tos values before call_VM() clobbers them. Since we have
3347     // to do it for every data type, we use the saved values as the
3348     // jvalue object.
3349     switch (bytecode()) {          // load values into the jvalue object
3350     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3351     case Bytecodes::_fast_bputfield: // fall through
3352     case Bytecodes::_fast_zputfield: // fall through
3353     case Bytecodes::_fast_sputfield: // fall through
3354     case Bytecodes::_fast_cputfield: // fall through
3355     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3356     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3357     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3358     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3359 
3360     default:
3361       ShouldNotReachHere();
3362     }
3363     __ mov(scratch, rsp);             // points to jvalue on the stack
3364     // access constant pool cache entry
3365     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3366     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3367     __ verify_oop(rbx);
3368     // rbx: object pointer copied above
3369     // c_rarg2: cache entry pointer
3370     // c_rarg3: jvalue object on the stack
3371     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3372     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3373 
3374     switch (bytecode()) {             // restore tos values
3375     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3376     case Bytecodes::_fast_bputfield: // fall through
3377     case Bytecodes::_fast_zputfield: // fall through
3378     case Bytecodes::_fast_sputfield: // fall through
3379     case Bytecodes::_fast_cputfield: // fall through
3380     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3381     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3382     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3383     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3384     default: break;
3385     }
3386     __ bind(L2);
3387   }
3388 }
3389 
3390 void TemplateTable::fast_storefield(TosState state) {
3391   transition(state, vtos);
3392 
3393   ByteSize base = ConstantPoolCache::base_offset();
3394 
3395   jvmti_post_fast_field_mod();
3396 
3397   // access constant pool cache
3398   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3399 
3400   // test for volatile with rdx but rdx is tos register for lputfield.
3401   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3402                        in_bytes(base +
3403                                 ConstantPoolCacheEntry::flags_offset())));
3404 
3405   // replace index with field offset from cache entry
3406   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3407                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3408 
3409   // [jk] not needed currently
3410   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3411   //                                              Assembler::StoreStore));
3412 
3413   Label notVolatile;
3414   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3415   __ andl(rdx, 0x1);
3416 
3417   // Get object from stack
3418   pop_and_check_object(rcx);
3419 
3420   // field address
3421   const Address field(rcx, rbx, Address::times_1);
3422 
3423   // access field
3424   switch (bytecode()) {
3425   case Bytecodes::_fast_aputfield:
3426     do_oop_store(_masm, field, rax);
3427     break;
3428   case Bytecodes::_fast_lputfield:
3429 #ifdef _LP64
3430     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg);
3431 #else
3432   __ stop("should not be rewritten");
3433 #endif
3434     break;
3435   case Bytecodes::_fast_iputfield:
3436     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3437     break;
3438   case Bytecodes::_fast_zputfield:
3439     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3440     break;
3441   case Bytecodes::_fast_bputfield:
3442     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3443     break;
3444   case Bytecodes::_fast_sputfield:
3445     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3446     break;
3447   case Bytecodes::_fast_cputfield:
3448     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3449     break;
3450   case Bytecodes::_fast_fputfield:
3451     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg);
3452     break;
3453   case Bytecodes::_fast_dputfield:
3454     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg);
3455     break;
3456   default:
3457     ShouldNotReachHere();
3458   }
3459 
3460   // Check for volatile store
3461   __ testl(rdx, rdx);
3462   __ jcc(Assembler::zero, notVolatile);
3463   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3464                                                Assembler::StoreStore));
3465   __ bind(notVolatile);
3466 }
3467 
3468 void TemplateTable::fast_accessfield(TosState state) {
3469   transition(atos, state);
3470 
3471   // Do the JVMTI work here to avoid disturbing the register state below
3472   if (JvmtiExport::can_post_field_access()) {
3473     // Check to see if a field access watch has been set before we
3474     // take the time to call into the VM.
3475     Label L1;
3476     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3477     __ testl(rcx, rcx);
3478     __ jcc(Assembler::zero, L1);
3479     // access constant pool cache entry
3480     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3481     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3482     __ verify_oop(rax);
3483     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3484     LP64_ONLY(__ mov(c_rarg1, rax));
3485     // c_rarg1: object pointer copied above
3486     // c_rarg2: cache entry pointer
3487     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3488     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3489     __ pop_ptr(rax); // restore object pointer
3490     __ bind(L1);
3491   }
3492 
3493   // access constant pool cache
3494   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3495   // replace index with field offset from cache entry
3496   // [jk] not needed currently
3497   // if (os::is_MP()) {
3498   //   __ movl(rdx, Address(rcx, rbx, Address::times_8,
3499   //                        in_bytes(ConstantPoolCache::base_offset() +
3500   //                                 ConstantPoolCacheEntry::flags_offset())));
3501   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3502   //   __ andl(rdx, 0x1);
3503   // }
3504   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3505                          in_bytes(ConstantPoolCache::base_offset() +
3506                                   ConstantPoolCacheEntry::f2_offset())));
3507 
3508   // rax: object
3509   __ verify_oop(rax);
3510   __ null_check(rax);
3511   Address field(rax, rbx, Address::times_1);
3512 
3513   // access field
3514   switch (bytecode()) {
3515   case Bytecodes::_fast_agetfield:
3516     do_oop_load(_masm, field, rax);
3517     __ verify_oop(rax);
3518     break;
3519   case Bytecodes::_fast_lgetfield:
3520 #ifdef _LP64
3521     __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3522 #else
3523   __ stop("should not be rewritten");
3524 #endif
3525     break;
3526   case Bytecodes::_fast_igetfield:
3527     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3528     break;
3529   case Bytecodes::_fast_bgetfield:
3530     __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3531     break;
3532   case Bytecodes::_fast_sgetfield:
3533     __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3534     break;
3535   case Bytecodes::_fast_cgetfield:
3536     __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3537     break;
3538   case Bytecodes::_fast_fgetfield:
3539     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3540     break;
3541   case Bytecodes::_fast_dgetfield:
3542     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3543     break;
3544   default:
3545     ShouldNotReachHere();
3546   }
3547   // [jk] not needed currently
3548   // if (os::is_MP()) {
3549   //   Label notVolatile;
3550   //   __ testl(rdx, rdx);
3551   //   __ jcc(Assembler::zero, notVolatile);
3552   //   __ membar(Assembler::LoadLoad);
3553   //   __ bind(notVolatile);
3554   //};
3555 }
3556 
3557 void TemplateTable::fast_xaccess(TosState state) {
3558   transition(vtos, state);
3559 
3560   // get receiver
3561   __ movptr(rax, aaddress(0));
3562   // access constant pool cache
3563   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3564   __ movptr(rbx,
3565             Address(rcx, rdx, Address::times_ptr,
3566                     in_bytes(ConstantPoolCache::base_offset() +
3567                              ConstantPoolCacheEntry::f2_offset())));
3568   // make sure exception is reported in correct bcp range (getfield is
3569   // next instruction)
3570   __ increment(rbcp);
3571   __ null_check(rax);
3572   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3573   switch (state) {
3574   case itos:
3575     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3576     break;
3577   case atos:
3578     do_oop_load(_masm, field, rax);
3579     __ verify_oop(rax);
3580     break;
3581   case ftos:
3582     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3583     break;
3584   default:
3585     ShouldNotReachHere();
3586   }
3587 
3588   // [jk] not needed currently
3589   // if (os::is_MP()) {
3590   //   Label notVolatile;
3591   //   __ movl(rdx, Address(rcx, rdx, Address::times_8,
3592   //                        in_bytes(ConstantPoolCache::base_offset() +
3593   //                                 ConstantPoolCacheEntry::flags_offset())));
3594   //   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3595   //   __ testl(rdx, 0x1);
3596   //   __ jcc(Assembler::zero, notVolatile);
3597   //   __ membar(Assembler::LoadLoad);
3598   //   __ bind(notVolatile);
3599   // }
3600 
3601   __ decrement(rbcp);
3602 }
3603 
3604 //-----------------------------------------------------------------------------
3605 // Calls
3606 
3607 void TemplateTable::count_calls(Register method, Register temp) {
3608   // implemented elsewhere
3609   ShouldNotReachHere();
3610 }
3611 
3612 void TemplateTable::prepare_invoke(int byte_no,
3613                                    Register method,  // linked method (or i-klass)
3614                                    Register index,   // itable index, MethodType, etc.
3615                                    Register recv,    // if caller wants to see it
3616                                    Register flags    // if caller wants to test it
3617                                    ) {
3618   // determine flags
3619   const Bytecodes::Code code = bytecode();
3620   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3621   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3622   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3623   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3624   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3625   const bool load_receiver       = (recv  != noreg);
3626   const bool save_flags          = (flags != noreg);
3627   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3628   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3629   assert(flags == noreg || flags == rdx, "");
3630   assert(recv  == noreg || recv  == rcx, "");
3631 
3632   // setup registers & access constant pool cache
3633   if (recv  == noreg)  recv  = rcx;
3634   if (flags == noreg)  flags = rdx;
3635   assert_different_registers(method, index, recv, flags);
3636 
3637   // save 'interpreter return address'
3638   __ save_bcp();
3639 
3640   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3641 
3642   // maybe push appendix to arguments (just before return address)
3643   if (is_invokedynamic || is_invokehandle) {
3644     Label L_no_push;
3645     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3646     __ jcc(Assembler::zero, L_no_push);
3647     // Push the appendix as a trailing parameter.
3648     // This must be done before we get the receiver,
3649     // since the parameter_size includes it.
3650     __ push(rbx);
3651     __ mov(rbx, index);
3652     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3653     __ load_resolved_reference_at_index(index, rbx);
3654     __ pop(rbx);
3655     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3656     __ bind(L_no_push);
3657   }
3658 
3659   // load receiver if needed (after appendix is pushed so parameter size is correct)
3660   // Note: no return address pushed yet
3661   if (load_receiver) {
3662     __ movl(recv, flags);
3663     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3664     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3665     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3666     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3667     __ movptr(recv, recv_addr);
3668     __ verify_oop(recv);
3669   }
3670 
3671   if (save_flags) {
3672     __ movl(rbcp, flags);
3673   }
3674 
3675   // compute return type
3676   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3677   // Make sure we don't need to mask flags after the above shift
3678   ConstantPoolCacheEntry::verify_tos_state_shift();
3679   // load return address
3680   {
3681     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3682     ExternalAddress table(table_addr);
3683     LP64_ONLY(__ lea(rscratch1, table));
3684     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3685     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3686   }
3687 
3688   // push return address
3689   __ push(flags);
3690 
3691   // Restore flags value from the constant pool cache, and restore rsi
3692   // for later null checks.  r13 is the bytecode pointer
3693   if (save_flags) {
3694     __ movl(flags, rbcp);
3695     __ restore_bcp();
3696   }
3697 }
3698 
3699 void TemplateTable::invokevirtual_helper(Register index,
3700                                          Register recv,
3701                                          Register flags) {
3702   // Uses temporary registers rax, rdx
3703   assert_different_registers(index, recv, rax, rdx);
3704   assert(index == rbx, "");
3705   assert(recv  == rcx, "");
3706 
3707   // Test for an invoke of a final method
3708   Label notFinal;
3709   __ movl(rax, flags);
3710   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3711   __ jcc(Assembler::zero, notFinal);
3712 
3713   const Register method = index;  // method must be rbx
3714   assert(method == rbx,
3715          "Method* must be rbx for interpreter calling convention");
3716 
3717   // do the call - the index is actually the method to call
3718   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3719 
3720   // It's final, need a null check here!
3721   __ null_check(recv);
3722 
3723   // profile this call
3724   __ profile_final_call(rax);
3725   __ profile_arguments_type(rax, method, rbcp, true);
3726 
3727   __ jump_from_interpreted(method, rax);
3728 
3729   __ bind(notFinal);
3730 
3731   // get receiver klass
3732   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3733   __ load_klass(rax, recv);
3734 
3735   // profile this call
3736   __ profile_virtual_call(rax, rlocals, rdx);
3737   // get target Method* & entry point
3738   __ lookup_virtual_method(rax, index, method);
3739   __ profile_called_method(method, rdx, rbcp);
3740 
3741   __ profile_arguments_type(rdx, method, rbcp, true);
3742   __ jump_from_interpreted(method, rdx);
3743 }
3744 
3745 void TemplateTable::invokevirtual(int byte_no) {
3746   transition(vtos, vtos);
3747   assert(byte_no == f2_byte, "use this argument");
3748   prepare_invoke(byte_no,
3749                  rbx,    // method or vtable index
3750                  noreg,  // unused itable index
3751                  rcx, rdx); // recv, flags
3752 
3753   // rbx: index
3754   // rcx: receiver
3755   // rdx: flags
3756 
3757   invokevirtual_helper(rbx, rcx, rdx);
3758 }
3759 
3760 void TemplateTable::invokespecial(int byte_no) {
3761   transition(vtos, vtos);
3762   assert(byte_no == f1_byte, "use this argument");
3763   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3764                  rcx);  // get receiver also for null check
3765   __ verify_oop(rcx);
3766   __ null_check(rcx);
3767   // do the call
3768   __ profile_call(rax);
3769   __ profile_arguments_type(rax, rbx, rbcp, false);
3770   __ jump_from_interpreted(rbx, rax);
3771 }
3772 
3773 void TemplateTable::invokestatic(int byte_no) {
3774   transition(vtos, vtos);
3775   assert(byte_no == f1_byte, "use this argument");
3776   prepare_invoke(byte_no, rbx);  // get f1 Method*
3777   // do the call
3778   __ profile_call(rax);
3779   __ profile_arguments_type(rax, rbx, rbcp, false);
3780   __ jump_from_interpreted(rbx, rax);
3781 }
3782 
3783 
3784 void TemplateTable::fast_invokevfinal(int byte_no) {
3785   transition(vtos, vtos);
3786   assert(byte_no == f2_byte, "use this argument");
3787   __ stop("fast_invokevfinal not used on x86");
3788 }
3789 
3790 
3791 void TemplateTable::invokeinterface(int byte_no) {
3792   transition(vtos, vtos);
3793   assert(byte_no == f1_byte, "use this argument");
3794   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 Method*
3795                  rcx, rdx); // recv, flags
3796 
3797   // rax: reference klass (from f1) if interface method
3798   // rbx: method (from f2)
3799   // rcx: receiver
3800   // rdx: flags
3801 
3802   // First check for Object case, then private interface method,
3803   // then regular interface method.
3804 
3805   // Special case of invokeinterface called for virtual method of
3806   // java.lang.Object.  See cpCache.cpp for details.
3807   Label notObjectMethod;
3808   __ movl(rlocals, rdx);
3809   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3810   __ jcc(Assembler::zero, notObjectMethod);
3811   invokevirtual_helper(rbx, rcx, rdx);
3812   // no return from above
3813   __ bind(notObjectMethod);
3814 
3815   Label no_such_interface; // for receiver subtype check
3816   Register recvKlass; // used for exception processing
3817 
3818   // Check for private method invocation - indicated by vfinal
3819   Label notVFinal;
3820   __ movl(rlocals, rdx);
3821   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3822   __ jcc(Assembler::zero, notVFinal);
3823 
3824   // Get receiver klass into rlocals - also a null check
3825   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3826   __ load_klass(rlocals, rcx);
3827 
3828   Label subtype;
3829   __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3830   // If we get here the typecheck failed
3831   recvKlass = rdx;
3832   __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3833   __ jmp(no_such_interface);
3834 
3835   __ bind(subtype);
3836 
3837   // do the call - rbx is actually the method to call
3838 
3839   __ profile_final_call(rdx);
3840   __ profile_arguments_type(rdx, rbx, rbcp, true);
3841 
3842   __ jump_from_interpreted(rbx, rdx);
3843   // no return from above
3844   __ bind(notVFinal);
3845 
3846   // Get receiver klass into rdx - also a null check
3847   __ restore_locals();  // restore r14
3848   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3849   __ load_klass(rdx, rcx);
3850 
3851   Label no_such_method;
3852 
3853   // Preserve method for throw_AbstractMethodErrorVerbose.
3854   __ mov(rcx, rbx);
3855   // Receiver subtype check against REFC.
3856   // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3857   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3858                              rdx, rax, noreg,
3859                              // outputs: scan temp. reg, scan temp. reg
3860                              rbcp, rlocals,
3861                              no_such_interface,
3862                              /*return_method=*/false);
3863 
3864   // profile this call
3865   __ restore_bcp(); // rbcp was destroyed by receiver type check
3866   __ profile_virtual_call(rdx, rbcp, rlocals);
3867 
3868   // Get declaring interface class from method, and itable index
3869   __ movptr(rax, Address(rbx, Method::const_offset()));
3870   __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3871   __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3872   __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3873   __ subl(rbx, Method::itable_index_max);
3874   __ negl(rbx);
3875 
3876   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3877   __ mov(rlocals, rdx);
3878   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3879                              rlocals, rax, rbx,
3880                              // outputs: method, scan temp. reg
3881                              rbx, rbcp,
3882                              no_such_interface);
3883 
3884   // rbx: Method* to call
3885   // rcx: receiver
3886   // Check for abstract method error
3887   // Note: This should be done more efficiently via a throw_abstract_method_error
3888   //       interpreter entry point and a conditional jump to it in case of a null
3889   //       method.
3890   __ testptr(rbx, rbx);
3891   __ jcc(Assembler::zero, no_such_method);
3892 
3893   __ profile_called_method(rbx, rbcp, rdx);
3894   __ profile_arguments_type(rdx, rbx, rbcp, true);
3895 
3896   // do the call
3897   // rcx: receiver
3898   // rbx,: Method*
3899   __ jump_from_interpreted(rbx, rdx);
3900   __ should_not_reach_here();
3901 
3902   // exception handling code follows...
3903   // note: must restore interpreter registers to canonical
3904   //       state for exception handling to work correctly!
3905 
3906   __ bind(no_such_method);
3907   // throw exception
3908   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3909   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3910   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3911   // Pass arguments for generating a verbose error message.
3912 #ifdef _LP64
3913   recvKlass = c_rarg1;
3914   Register method    = c_rarg2;
3915   if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3916   if (method != rcx)    { __ movq(method, rcx);    }
3917 #else
3918   recvKlass = rdx;
3919   Register method    = rcx;
3920 #endif
3921   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3922              recvKlass, method);
3923   // The call_VM checks for exception, so we should never return here.
3924   __ should_not_reach_here();
3925 
3926   __ bind(no_such_interface);
3927   // throw exception
3928   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3929   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3930   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3931   // Pass arguments for generating a verbose error message.
3932   LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3933   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3934              recvKlass, rax);
3935   // the call_VM checks for exception, so we should never return here.
3936   __ should_not_reach_here();
3937 }
3938 
3939 void TemplateTable::invokehandle(int byte_no) {
3940   transition(vtos, vtos);
3941   assert(byte_no == f1_byte, "use this argument");
3942   const Register rbx_method = rbx;
3943   const Register rax_mtype  = rax;
3944   const Register rcx_recv   = rcx;
3945   const Register rdx_flags  = rdx;
3946 
3947   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3948   __ verify_method_ptr(rbx_method);
3949   __ verify_oop(rcx_recv);
3950   __ null_check(rcx_recv);
3951 
3952   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3953   // rbx: MH.invokeExact_MT method (from f2)
3954 
3955   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3956 
3957   // FIXME: profile the LambdaForm also
3958   __ profile_final_call(rax);
3959   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3960 
3961   __ jump_from_interpreted(rbx_method, rdx);
3962 }
3963 
3964 void TemplateTable::invokedynamic(int byte_no) {
3965   transition(vtos, vtos);
3966   assert(byte_no == f1_byte, "use this argument");
3967 
3968   const Register rbx_method   = rbx;
3969   const Register rax_callsite = rax;
3970 
3971   prepare_invoke(byte_no, rbx_method, rax_callsite);
3972 
3973   // rax: CallSite object (from cpool->resolved_references[f1])
3974   // rbx: MH.linkToCallSite method (from f2)
3975 
3976   // Note:  rax_callsite is already pushed by prepare_invoke
3977 
3978   // %%% should make a type profile for any invokedynamic that takes a ref argument
3979   // profile this call
3980   __ profile_call(rbcp);
3981   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3982 
3983   __ verify_oop(rax_callsite);
3984 
3985   __ jump_from_interpreted(rbx_method, rdx);
3986 }
3987 
3988 //-----------------------------------------------------------------------------
3989 // Allocation
3990 
3991 void TemplateTable::_new() {
3992   transition(vtos, atos);
3993   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3994   Label slow_case;
3995   Label slow_case_no_pop;
3996   Label done;
3997   Label initialize_header;
3998   Label initialize_object;  // including clearing the fields
3999 
4000   __ get_cpool_and_tags(rcx, rax);
4001 
4002   // Make sure the class we're about to instantiate has been resolved.
4003   // This is done before loading InstanceKlass to be consistent with the order
4004   // how Constant Pool is updated (see ConstantPool::klass_at_put)
4005   const int tags_offset = Array<u1>::base_offset_in_bytes();
4006   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4007   __ jcc(Assembler::notEqual, slow_case_no_pop);
4008 
4009   // get InstanceKlass
4010   __ load_resolved_klass_at_index(rcx, rdx, rcx);
4011   __ push(rcx);  // save the contexts of klass for initializing the header
4012 
4013   // make sure klass is initialized & doesn't have finalizer
4014   // make sure klass is fully initialized
4015   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4016   __ jcc(Assembler::notEqual, slow_case);
4017 
4018   // get instance_size in InstanceKlass (scaled to a count of bytes)
4019   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4020   // test to see if it has a finalizer or is malformed in some way
4021   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4022   __ jcc(Assembler::notZero, slow_case);
4023 
4024   // Allocate the instance:
4025   //  If TLAB is enabled:
4026   //    Try to allocate in the TLAB.
4027   //    If fails, go to the slow path.
4028   //  Else If inline contiguous allocations are enabled:
4029   //    Try to allocate in eden.
4030   //    If fails due to heap end, go to slow path.
4031   //
4032   //  If TLAB is enabled OR inline contiguous is enabled:
4033   //    Initialize the allocation.
4034   //    Exit.
4035   //
4036   //  Go to slow path.
4037 
4038   const bool allow_shared_alloc =
4039     Universe::heap()->supports_inline_contig_alloc();
4040 
4041   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4042 #ifndef _LP64
4043   if (UseTLAB || allow_shared_alloc) {
4044     __ get_thread(thread);
4045   }
4046 #endif // _LP64
4047 
4048   if (UseTLAB) {
4049     __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
4050     if (ZeroTLAB) {
4051       // the fields have been already cleared
4052       __ jmp(initialize_header);
4053     } else {
4054       // initialize both the header and fields
4055       __ jmp(initialize_object);
4056     }
4057   } else {
4058     // Allocation in the shared Eden, if allowed.
4059     //
4060     // rdx: instance size in bytes
4061     __ eden_allocate(thread, rax, rdx, 0, rbx, slow_case);
4062   }
4063 
4064   // If UseTLAB or allow_shared_alloc are true, the object is created above and
4065   // there is an initialize need. Otherwise, skip and go to the slow path.
4066   if (UseTLAB || allow_shared_alloc) {
4067     // The object is initialized before the header.  If the object size is
4068     // zero, go directly to the header initialization.
4069     __ bind(initialize_object);
4070     __ decrement(rdx, sizeof(oopDesc));
4071     __ jcc(Assembler::zero, initialize_header);
4072 
4073     // Initialize topmost object field, divide rdx by 8, check if odd and
4074     // test if zero.
4075     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
4076     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4077 
4078     // rdx must have been multiple of 8
4079 #ifdef ASSERT
4080     // make sure rdx was multiple of 8
4081     Label L;
4082     // Ignore partial flag stall after shrl() since it is debug VM
4083     __ jccb(Assembler::carryClear, L);
4084     __ stop("object size is not multiple of 2 - adjust this code");
4085     __ bind(L);
4086     // rdx must be > 0, no extra check needed here
4087 #endif
4088 
4089     // initialize remaining object fields: rdx was a multiple of 8
4090     { Label loop;
4091     __ bind(loop);
4092     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4093     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4094     __ decrement(rdx);
4095     __ jcc(Assembler::notZero, loop);
4096     }
4097 
4098     // initialize object header only.
4099     __ bind(initialize_header);
4100     if (UseBiasedLocking) {
4101       __ pop(rcx);   // get saved klass back in the register.
4102       __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4103       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4104     } else {
4105       __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4106                 (intptr_t)markOopDesc::prototype()); // header
4107       __ pop(rcx);   // get saved klass back in the register.
4108     }
4109 #ifdef _LP64
4110     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4111     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4112 #endif
4113     __ store_klass(rax, rcx);  // klass
4114 
4115     {
4116       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4117       // Trigger dtrace event for fastpath
4118       __ push(atos);
4119       __ call_VM_leaf(
4120            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4121       __ pop(atos);
4122     }
4123 
4124     __ jmp(done);
4125   }
4126 
4127   // slow case
4128   __ bind(slow_case);
4129   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4130   __ bind(slow_case_no_pop);
4131 
4132   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4133   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4134 
4135   __ get_constant_pool(rarg1);
4136   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4137   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4138    __ verify_oop(rax);
4139 
4140   // continue
4141   __ bind(done);
4142 }
4143 
4144 void TemplateTable::newarray() {
4145   transition(itos, atos);
4146   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4147   __ load_unsigned_byte(rarg1, at_bcp(1));
4148   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4149           rarg1, rax);
4150 }
4151 
4152 void TemplateTable::anewarray() {
4153   transition(itos, atos);
4154 
4155   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4156   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4157 
4158   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4159   __ get_constant_pool(rarg1);
4160   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4161           rarg1, rarg2, rax);
4162 }
4163 
4164 void TemplateTable::arraylength() {
4165   transition(atos, itos);
4166   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4167   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4168 }
4169 
4170 void TemplateTable::checkcast() {
4171   transition(atos, atos);
4172   Label done, is_null, ok_is_subtype, quicked, resolved;
4173   __ testptr(rax, rax); // object is in rax
4174   __ jcc(Assembler::zero, is_null);
4175 
4176   // Get cpool & tags index
4177   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4178   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4179   // See if bytecode has already been quicked
4180   __ cmpb(Address(rdx, rbx,
4181                   Address::times_1,
4182                   Array<u1>::base_offset_in_bytes()),
4183           JVM_CONSTANT_Class);
4184   __ jcc(Assembler::equal, quicked);
4185   __ push(atos); // save receiver for result, and for GC
4186   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4187 
4188   // vm_result_2 has metadata result
4189 #ifndef _LP64
4190   // borrow rdi from locals
4191   __ get_thread(rdi);
4192   __ get_vm_result_2(rax, rdi);
4193   __ restore_locals();
4194 #else
4195   __ get_vm_result_2(rax, r15_thread);
4196 #endif
4197 
4198   __ pop_ptr(rdx); // restore receiver
4199   __ jmpb(resolved);
4200 
4201   // Get superklass in rax and subklass in rbx
4202   __ bind(quicked);
4203   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4204   __ load_resolved_klass_at_index(rcx, rbx, rax);
4205 
4206   __ bind(resolved);
4207   __ load_klass(rbx, rdx);
4208 
4209   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4210   // Superklass in rax.  Subklass in rbx.
4211   __ gen_subtype_check(rbx, ok_is_subtype);
4212 
4213   // Come here on failure
4214   __ push_ptr(rdx);
4215   // object is at TOS
4216   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4217 
4218   // Come here on success
4219   __ bind(ok_is_subtype);
4220   __ mov(rax, rdx); // Restore object in rdx
4221 
4222   // Collect counts on whether this check-cast sees NULLs a lot or not.
4223   if (ProfileInterpreter) {
4224     __ jmp(done);
4225     __ bind(is_null);
4226     __ profile_null_seen(rcx);
4227   } else {
4228     __ bind(is_null);   // same as 'done'
4229   }
4230   __ bind(done);
4231 }
4232 
4233 void TemplateTable::instanceof() {
4234   transition(atos, itos);
4235   Label done, is_null, ok_is_subtype, quicked, resolved;
4236   __ testptr(rax, rax);
4237   __ jcc(Assembler::zero, is_null);
4238 
4239   // Get cpool & tags index
4240   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4241   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4242   // See if bytecode has already been quicked
4243   __ cmpb(Address(rdx, rbx,
4244                   Address::times_1,
4245                   Array<u1>::base_offset_in_bytes()),
4246           JVM_CONSTANT_Class);
4247   __ jcc(Assembler::equal, quicked);
4248 
4249   __ push(atos); // save receiver for result, and for GC
4250   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4251   // vm_result_2 has metadata result
4252 
4253 #ifndef _LP64
4254   // borrow rdi from locals
4255   __ get_thread(rdi);
4256   __ get_vm_result_2(rax, rdi);
4257   __ restore_locals();
4258 #else
4259   __ get_vm_result_2(rax, r15_thread);
4260 #endif
4261 
4262   __ pop_ptr(rdx); // restore receiver
4263   __ verify_oop(rdx);
4264   __ load_klass(rdx, rdx);
4265   __ jmpb(resolved);
4266 
4267   // Get superklass in rax and subklass in rdx
4268   __ bind(quicked);
4269   __ load_klass(rdx, rax);
4270   __ load_resolved_klass_at_index(rcx, rbx, rax);
4271 
4272   __ bind(resolved);
4273 
4274   // Generate subtype check.  Blows rcx, rdi
4275   // Superklass in rax.  Subklass in rdx.
4276   __ gen_subtype_check(rdx, ok_is_subtype);
4277 
4278   // Come here on failure
4279   __ xorl(rax, rax);
4280   __ jmpb(done);
4281   // Come here on success
4282   __ bind(ok_is_subtype);
4283   __ movl(rax, 1);
4284 
4285   // Collect counts on whether this test sees NULLs a lot or not.
4286   if (ProfileInterpreter) {
4287     __ jmp(done);
4288     __ bind(is_null);
4289     __ profile_null_seen(rcx);
4290   } else {
4291     __ bind(is_null);   // same as 'done'
4292   }
4293   __ bind(done);
4294   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4295   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4296 }
4297 
4298 
4299 //----------------------------------------------------------------------------------------------------
4300 // Breakpoints
4301 void TemplateTable::_breakpoint() {
4302   // Note: We get here even if we are single stepping..
4303   // jbug insists on setting breakpoints at every bytecode
4304   // even if we are in single step mode.
4305 
4306   transition(vtos, vtos);
4307 
4308   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4309 
4310   // get the unpatched byte code
4311   __ get_method(rarg);
4312   __ call_VM(noreg,
4313              CAST_FROM_FN_PTR(address,
4314                               InterpreterRuntime::get_original_bytecode_at),
4315              rarg, rbcp);
4316   __ mov(rbx, rax);  // why?
4317 
4318   // post the breakpoint event
4319   __ get_method(rarg);
4320   __ call_VM(noreg,
4321              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4322              rarg, rbcp);
4323 
4324   // complete the execution of original bytecode
4325   __ dispatch_only_normal(vtos);
4326 }
4327 
4328 //-----------------------------------------------------------------------------
4329 // Exceptions
4330 
4331 void TemplateTable::athrow() {
4332   transition(atos, vtos);
4333   __ null_check(rax);
4334   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4335 }
4336 
4337 //-----------------------------------------------------------------------------
4338 // Synchronization
4339 //
4340 // Note: monitorenter & exit are symmetric routines; which is reflected
4341 //       in the assembly code structure as well
4342 //
4343 // Stack layout:
4344 //
4345 // [expressions  ] <--- rsp               = expression stack top
4346 // ..
4347 // [expressions  ]
4348 // [monitor entry] <--- monitor block top = expression stack bot
4349 // ..
4350 // [monitor entry]
4351 // [frame data   ] <--- monitor block bot
4352 // ...
4353 // [saved rbp    ] <--- rbp
4354 void TemplateTable::monitorenter() {
4355   transition(atos, vtos);
4356 
4357   // check for NULL object
4358   __ null_check(rax);
4359 
4360   __ resolve_for_write(IS_NOT_NULL, rax);
4361 
4362   const Address monitor_block_top(
4363         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4364   const Address monitor_block_bot(
4365         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4366   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4367 
4368   Label allocated;
4369 
4370   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4371   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4372   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4373 
4374   // initialize entry pointer
4375   __ xorl(rmon, rmon); // points to free slot or NULL
4376 
4377   // find a free slot in the monitor block (result in rmon)
4378   {
4379     Label entry, loop, exit;
4380     __ movptr(rtop, monitor_block_top); // points to current entry,
4381                                         // starting with top-most entry
4382     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4383                                         // of monitor block
4384     __ jmpb(entry);
4385 
4386     __ bind(loop);
4387     // check if current entry is used
4388     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4389     // if not used then remember entry in rmon
4390     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4391     // check if current entry is for same object
4392     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4393     // if same object then stop searching
4394     __ jccb(Assembler::equal, exit);
4395     // otherwise advance to next entry
4396     __ addptr(rtop, entry_size);
4397     __ bind(entry);
4398     // check if bottom reached
4399     __ cmpptr(rtop, rbot);
4400     // if not at bottom then check this entry
4401     __ jcc(Assembler::notEqual, loop);
4402     __ bind(exit);
4403   }
4404 
4405   __ testptr(rmon, rmon); // check if a slot has been found
4406   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4407 
4408   // allocate one if there's no free slot
4409   {
4410     Label entry, loop;
4411     // 1. compute new pointers          // rsp: old expression stack top
4412     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4413     __ subptr(rsp, entry_size);         // move expression stack top
4414     __ subptr(rmon, entry_size);        // move expression stack bottom
4415     __ mov(rtop, rsp);                  // set start value for copy loop
4416     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4417     __ jmp(entry);
4418     // 2. move expression stack contents
4419     __ bind(loop);
4420     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4421                                                 // word from old location
4422     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4423     __ addptr(rtop, wordSize);                  // advance to next word
4424     __ bind(entry);
4425     __ cmpptr(rtop, rmon);                      // check if bottom reached
4426     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4427                                                 // copy next word
4428   }
4429 
4430   // call run-time routine
4431   // rmon: points to monitor entry
4432   __ bind(allocated);
4433 
4434   // Increment bcp to point to the next bytecode, so exception
4435   // handling for async. exceptions work correctly.
4436   // The object has already been poped from the stack, so the
4437   // expression stack looks correct.
4438   __ increment(rbcp);
4439 
4440   // store object
4441   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4442   __ lock_object(rmon);
4443 
4444   // check to make sure this monitor doesn't cause stack overflow after locking
4445   __ save_bcp();  // in case of exception
4446   __ generate_stack_overflow_check(0);
4447 
4448   // The bcp has already been incremented. Just need to dispatch to
4449   // next instruction.
4450   __ dispatch_next(vtos);
4451 }
4452 
4453 void TemplateTable::monitorexit() {
4454   transition(atos, vtos);
4455 
4456   // check for NULL object
4457   __ null_check(rax);
4458 
4459   __ resolve_for_write(IS_NOT_NULL, rax);
4460 
4461   const Address monitor_block_top(
4462         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4463   const Address monitor_block_bot(
4464         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4465   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4466 
4467   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4468   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4469 
4470   Label found;
4471 
4472   // find matching slot
4473   {
4474     Label entry, loop;
4475     __ movptr(rtop, monitor_block_top); // points to current entry,
4476                                         // starting with top-most entry
4477     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4478                                         // of monitor block
4479     __ jmpb(entry);
4480 
4481     __ bind(loop);
4482     // check if current entry is for same object
4483     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4484     // if same object then stop searching
4485     __ jcc(Assembler::equal, found);
4486     // otherwise advance to next entry
4487     __ addptr(rtop, entry_size);
4488     __ bind(entry);
4489     // check if bottom reached
4490     __ cmpptr(rtop, rbot);
4491     // if not at bottom then check this entry
4492     __ jcc(Assembler::notEqual, loop);
4493   }
4494 
4495   // error handling. Unlocking was not block-structured
4496   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4497                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4498   __ should_not_reach_here();
4499 
4500   // call run-time routine
4501   __ bind(found);
4502   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4503   __ unlock_object(rtop);
4504   __ pop_ptr(rax); // discard object
4505 }
4506 
4507 // Wide instructions
4508 void TemplateTable::wide() {
4509   transition(vtos, vtos);
4510   __ load_unsigned_byte(rbx, at_bcp(1));
4511   ExternalAddress wtable((address)Interpreter::_wentry_point);
4512   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4513   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4514 }
4515 
4516 // Multi arrays
4517 void TemplateTable::multianewarray() {
4518   transition(vtos, atos);
4519 
4520   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4521   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4522   // last dim is on top of stack; we want address of first one:
4523   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4524   // the latter wordSize to point to the beginning of the array.
4525   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4526   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4527   __ load_unsigned_byte(rbx, at_bcp(3));
4528   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4529 }