1 /*
   2  * Copyright (c) 1997, 2022, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "compiler/disassembler.hpp"
  28 #include "gc/shared/collectedHeap.hpp"
  29 #include "gc/shared/tlab_globals.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateTable.hpp"
  34 #include "memory/universe.hpp"
  35 #include "oops/methodData.hpp"
  36 #include "oops/objArrayKlass.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "prims/jvmtiExport.hpp"
  39 #include "prims/methodHandles.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/safepointMechanism.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "utilities/macros.hpp"
  46 
  47 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  48 
  49 // Global Register Names
  50 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  51 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  52 
  53 // Address Computation: local variables
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 #ifndef _LP64
  63 static inline Address haddress(int n) {
  64   return iaddress(n + 0);
  65 }
  66 #endif
  67 
  68 static inline Address faddress(int n) {
  69   return iaddress(n);
  70 }
  71 
  72 static inline Address daddress(int n) {
  73   return laddress(n);
  74 }
  75 
  76 static inline Address aaddress(int n) {
  77   return iaddress(n);
  78 }
  79 
  80 static inline Address iaddress(Register r) {
  81   return Address(rlocals, r, Address::times_ptr);
  82 }
  83 
  84 static inline Address laddress(Register r) {
  85   return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
  86 }
  87 
  88 #ifndef _LP64
  89 static inline Address haddress(Register r)       {
  90   return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
  91 }
  92 #endif
  93 
  94 static inline Address faddress(Register r) {
  95   return iaddress(r);
  96 }
  97 
  98 static inline Address daddress(Register r) {
  99   return laddress(r);
 100 }
 101 
 102 static inline Address aaddress(Register r) {
 103   return iaddress(r);
 104 }
 105 
 106 
 107 // expression stack
 108 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
 109 // data beyond the rsp which is potentially unsafe in an MT environment;
 110 // an interrupt may overwrite that data.)
 111 static inline Address at_rsp   () {
 112   return Address(rsp, 0);
 113 }
 114 
 115 // At top of Java expression stack which may be different than esp().  It
 116 // isn't for category 1 objects.
 117 static inline Address at_tos   () {
 118   return Address(rsp,  Interpreter::expr_offset_in_bytes(0));
 119 }
 120 
 121 static inline Address at_tos_p1() {
 122   return Address(rsp,  Interpreter::expr_offset_in_bytes(1));
 123 }
 124 
 125 static inline Address at_tos_p2() {
 126   return Address(rsp,  Interpreter::expr_offset_in_bytes(2));
 127 }
 128 
 129 // Condition conversion
 130 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 131   switch (cc) {
 132   case TemplateTable::equal        : return Assembler::notEqual;
 133   case TemplateTable::not_equal    : return Assembler::equal;
 134   case TemplateTable::less         : return Assembler::greaterEqual;
 135   case TemplateTable::less_equal   : return Assembler::greater;
 136   case TemplateTable::greater      : return Assembler::lessEqual;
 137   case TemplateTable::greater_equal: return Assembler::less;
 138   }
 139   ShouldNotReachHere();
 140   return Assembler::zero;
 141 }
 142 
 143 
 144 
 145 // Miscellaneous helper routines
 146 // Store an oop (or NULL) at the address described by obj.
 147 // If val == noreg this means store a NULL
 148 
 149 
 150 static void do_oop_store(InterpreterMacroAssembler* _masm,
 151                          Address dst,
 152                          Register val,
 153                          DecoratorSet decorators = 0) {
 154   assert(val == noreg || val == rax, "parameter is just for looks");
 155   __ store_heap_oop(dst, val, rdx, rbx, LP64_ONLY(r8) NOT_LP64(rsi), decorators);
 156 }
 157 
 158 static void do_oop_load(InterpreterMacroAssembler* _masm,
 159                         Address src,
 160                         Register dst,
 161                         DecoratorSet decorators = 0) {
 162   __ load_heap_oop(dst, src, rdx, rbx, decorators);
 163 }
 164 
 165 Address TemplateTable::at_bcp(int offset) {
 166   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 167   return Address(rbcp, offset);
 168 }
 169 
 170 
 171 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 172                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 173                                    int byte_no) {
 174   if (!RewriteBytecodes)  return;
 175   Label L_patch_done;
 176 
 177   switch (bc) {
 178   case Bytecodes::_fast_aputfield:
 179   case Bytecodes::_fast_bputfield:
 180   case Bytecodes::_fast_zputfield:
 181   case Bytecodes::_fast_cputfield:
 182   case Bytecodes::_fast_dputfield:
 183   case Bytecodes::_fast_fputfield:
 184   case Bytecodes::_fast_iputfield:
 185   case Bytecodes::_fast_lputfield:
 186   case Bytecodes::_fast_sputfield:
 187     {
 188       // We skip bytecode quickening for putfield instructions when
 189       // the put_code written to the constant pool cache is zero.
 190       // This is required so that every execution of this instruction
 191       // calls out to InterpreterRuntime::resolve_get_put to do
 192       // additional, required work.
 193       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 194       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 195       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 196       __ movl(bc_reg, bc);
 197       __ cmpl(temp_reg, (int) 0);
 198       __ jcc(Assembler::zero, L_patch_done);  // don't patch
 199     }
 200     break;
 201   default:
 202     assert(byte_no == -1, "sanity");
 203     // the pair bytecodes have already done the load.
 204     if (load_bc_into_bc_reg) {
 205       __ movl(bc_reg, bc);
 206     }
 207   }
 208 
 209   if (JvmtiExport::can_post_breakpoint()) {
 210     Label L_fast_patch;
 211     // if a breakpoint is present we can't rewrite the stream directly
 212     __ movzbl(temp_reg, at_bcp(0));
 213     __ cmpl(temp_reg, Bytecodes::_breakpoint);
 214     __ jcc(Assembler::notEqual, L_fast_patch);
 215     __ get_method(temp_reg);
 216     // Let breakpoint table handling rewrite to quicker bytecode
 217     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
 218 #ifndef ASSERT
 219     __ jmpb(L_patch_done);
 220 #else
 221     __ jmp(L_patch_done);
 222 #endif
 223     __ bind(L_fast_patch);
 224   }
 225 
 226 #ifdef ASSERT
 227   Label L_okay;
 228   __ load_unsigned_byte(temp_reg, at_bcp(0));
 229   __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
 230   __ jcc(Assembler::equal, L_okay);
 231   __ cmpl(temp_reg, bc_reg);
 232   __ jcc(Assembler::equal, L_okay);
 233   __ stop("patching the wrong bytecode");
 234   __ bind(L_okay);
 235 #endif
 236 
 237   // patch bytecode
 238   __ movb(at_bcp(0), bc_reg);
 239   __ bind(L_patch_done);
 240 }
 241 // Individual instructions
 242 
 243 
 244 void TemplateTable::nop() {
 245   transition(vtos, vtos);
 246   // nothing to do
 247 }
 248 
 249 void TemplateTable::shouldnotreachhere() {
 250   transition(vtos, vtos);
 251   __ stop("shouldnotreachhere bytecode");
 252 }
 253 
 254 void TemplateTable::aconst_null() {
 255   transition(vtos, atos);
 256   __ xorl(rax, rax);
 257 }
 258 
 259 void TemplateTable::iconst(int value) {
 260   transition(vtos, itos);
 261   if (value == 0) {
 262     __ xorl(rax, rax);
 263   } else {
 264     __ movl(rax, value);
 265   }
 266 }
 267 
 268 void TemplateTable::lconst(int value) {
 269   transition(vtos, ltos);
 270   if (value == 0) {
 271     __ xorl(rax, rax);
 272   } else {
 273     __ movl(rax, value);
 274   }
 275 #ifndef _LP64
 276   assert(value >= 0, "check this code");
 277   __ xorptr(rdx, rdx);
 278 #endif
 279 }
 280 
 281 
 282 
 283 void TemplateTable::fconst(int value) {
 284   transition(vtos, ftos);
 285   if (UseSSE >= 1) {
 286     static float one = 1.0f, two = 2.0f;
 287     switch (value) {
 288     case 0:
 289       __ xorps(xmm0, xmm0);
 290       break;
 291     case 1:
 292       __ movflt(xmm0, ExternalAddress((address) &one));
 293       break;
 294     case 2:
 295       __ movflt(xmm0, ExternalAddress((address) &two));
 296       break;
 297     default:
 298       ShouldNotReachHere();
 299       break;
 300     }
 301   } else {
 302 #ifdef _LP64
 303     ShouldNotReachHere();
 304 #else
 305            if (value == 0) { __ fldz();
 306     } else if (value == 1) { __ fld1();
 307     } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
 308     } else                 { ShouldNotReachHere();
 309     }
 310 #endif // _LP64
 311   }
 312 }
 313 
 314 void TemplateTable::dconst(int value) {
 315   transition(vtos, dtos);
 316   if (UseSSE >= 2) {
 317     static double one = 1.0;
 318     switch (value) {
 319     case 0:
 320       __ xorpd(xmm0, xmm0);
 321       break;
 322     case 1:
 323       __ movdbl(xmm0, ExternalAddress((address) &one));
 324       break;
 325     default:
 326       ShouldNotReachHere();
 327       break;
 328     }
 329   } else {
 330 #ifdef _LP64
 331     ShouldNotReachHere();
 332 #else
 333            if (value == 0) { __ fldz();
 334     } else if (value == 1) { __ fld1();
 335     } else                 { ShouldNotReachHere();
 336     }
 337 #endif
 338   }
 339 }
 340 
 341 void TemplateTable::bipush() {
 342   transition(vtos, itos);
 343   __ load_signed_byte(rax, at_bcp(1));
 344 }
 345 
 346 void TemplateTable::sipush() {
 347   transition(vtos, itos);
 348   __ load_unsigned_short(rax, at_bcp(1));
 349   __ bswapl(rax);
 350   __ sarl(rax, 16);
 351 }
 352 
 353 void TemplateTable::ldc(bool wide) {
 354   transition(vtos, vtos);
 355   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 356   Label call_ldc, notFloat, notClass, notInt, Done;
 357 
 358   if (wide) {
 359     __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 360   } else {
 361     __ load_unsigned_byte(rbx, at_bcp(1));
 362   }
 363 
 364   __ get_cpool_and_tags(rcx, rax);
 365   const int base_offset = ConstantPool::header_size() * wordSize;
 366   const int tags_offset = Array<u1>::base_offset_in_bytes();
 367 
 368   // get type
 369   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 370 
 371   // unresolved class - get the resolved class
 372   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
 373   __ jccb(Assembler::equal, call_ldc);
 374 
 375   // unresolved class in error state - call into runtime to throw the error
 376   // from the first resolution attempt
 377   __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
 378   __ jccb(Assembler::equal, call_ldc);
 379 
 380   // resolved class - need to call vm to get java mirror of the class
 381   __ cmpl(rdx, JVM_CONSTANT_Class);
 382   __ jcc(Assembler::notEqual, notClass);
 383 
 384   __ bind(call_ldc);
 385 
 386   __ movl(rarg, wide);
 387   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
 388 
 389   __ push(atos);
 390   __ jmp(Done);
 391 
 392   __ bind(notClass);
 393   __ cmpl(rdx, JVM_CONSTANT_Float);
 394   __ jccb(Assembler::notEqual, notFloat);
 395 
 396   // ftos
 397   __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
 398   __ push(ftos);
 399   __ jmp(Done);
 400 
 401   __ bind(notFloat);
 402   __ cmpl(rdx, JVM_CONSTANT_Integer);
 403   __ jccb(Assembler::notEqual, notInt);
 404 
 405   // itos
 406   __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
 407   __ push(itos);
 408   __ jmp(Done);
 409 
 410   // assume the tag is for condy; if not, the VM runtime will tell us
 411   __ bind(notInt);
 412   condy_helper(Done);
 413 
 414   __ bind(Done);
 415 }
 416 
 417 // Fast path for caching oop constants.
 418 void TemplateTable::fast_aldc(bool wide) {
 419   transition(vtos, atos);
 420 
 421   Register result = rax;
 422   Register tmp = rdx;
 423   Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 424   int index_size = wide ? sizeof(u2) : sizeof(u1);
 425 
 426   Label resolved;
 427 
 428   // We are resolved if the resolved reference cache entry contains a
 429   // non-null object (String, MethodType, etc.)
 430   assert_different_registers(result, tmp);
 431   __ get_cache_index_at_bcp(tmp, 1, index_size);
 432   __ load_resolved_reference_at_index(result, tmp);
 433   __ testptr(result, result);
 434   __ jcc(Assembler::notZero, resolved);
 435 
 436   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 437 
 438   // first time invocation - must resolve first
 439   __ movl(rarg, (int)bytecode());
 440   __ call_VM(result, entry, rarg);
 441   __ bind(resolved);
 442 
 443   { // Check for the null sentinel.
 444     // If we just called the VM, it already did the mapping for us,
 445     // but it's harmless to retry.
 446     Label notNull;
 447     ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
 448     __ movptr(tmp, null_sentinel);
 449     __ resolve_oop_handle(tmp);
 450     __ cmpoop(tmp, result);
 451     __ jccb(Assembler::notEqual, notNull);
 452     __ xorptr(result, result);  // NULL object reference
 453     __ bind(notNull);
 454   }
 455 
 456   if (VerifyOops) {
 457     __ verify_oop(result);
 458   }
 459 }
 460 
 461 void TemplateTable::ldc2_w() {
 462   transition(vtos, vtos);
 463   Label notDouble, notLong, Done;
 464   __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
 465 
 466   __ get_cpool_and_tags(rcx, rax);
 467   const int base_offset = ConstantPool::header_size() * wordSize;
 468   const int tags_offset = Array<u1>::base_offset_in_bytes();
 469 
 470   // get type
 471   __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
 472   __ cmpl(rdx, JVM_CONSTANT_Double);
 473   __ jccb(Assembler::notEqual, notDouble);
 474 
 475   // dtos
 476   __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
 477   __ push(dtos);
 478 
 479   __ jmp(Done);
 480   __ bind(notDouble);
 481   __ cmpl(rdx, JVM_CONSTANT_Long);
 482   __ jccb(Assembler::notEqual, notLong);
 483 
 484   // ltos
 485   __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
 486   NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
 487   __ push(ltos);
 488   __ jmp(Done);
 489 
 490   __ bind(notLong);
 491   condy_helper(Done);
 492 
 493   __ bind(Done);
 494 }
 495 
 496 void TemplateTable::condy_helper(Label& Done) {
 497   const Register obj = rax;
 498   const Register off = rbx;
 499   const Register flags = rcx;
 500   const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
 501   __ movl(rarg, (int)bytecode());
 502   call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
 503 #ifndef _LP64
 504   // borrow rdi from locals
 505   __ get_thread(rdi);
 506   __ get_vm_result_2(flags, rdi);
 507   __ restore_locals();
 508 #else
 509   __ get_vm_result_2(flags, r15_thread);
 510 #endif
 511   // VMr = obj = base address to find primitive value to push
 512   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 513   __ movl(off, flags);
 514   __ andl(off, ConstantPoolCacheEntry::field_index_mask);
 515   const Address field(obj, off, Address::times_1, 0*wordSize);
 516 
 517   // What sort of thing are we loading?
 518   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
 519   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
 520 
 521   switch (bytecode()) {
 522   case Bytecodes::_ldc:
 523   case Bytecodes::_ldc_w:
 524     {
 525       // tos in (itos, ftos, stos, btos, ctos, ztos)
 526       Label notInt, notFloat, notShort, notByte, notChar, notBool;
 527       __ cmpl(flags, itos);
 528       __ jccb(Assembler::notEqual, notInt);
 529       // itos
 530       __ movl(rax, field);
 531       __ push(itos);
 532       __ jmp(Done);
 533 
 534       __ bind(notInt);
 535       __ cmpl(flags, ftos);
 536       __ jccb(Assembler::notEqual, notFloat);
 537       // ftos
 538       __ load_float(field);
 539       __ push(ftos);
 540       __ jmp(Done);
 541 
 542       __ bind(notFloat);
 543       __ cmpl(flags, stos);
 544       __ jccb(Assembler::notEqual, notShort);
 545       // stos
 546       __ load_signed_short(rax, field);
 547       __ push(stos);
 548       __ jmp(Done);
 549 
 550       __ bind(notShort);
 551       __ cmpl(flags, btos);
 552       __ jccb(Assembler::notEqual, notByte);
 553       // btos
 554       __ load_signed_byte(rax, field);
 555       __ push(btos);
 556       __ jmp(Done);
 557 
 558       __ bind(notByte);
 559       __ cmpl(flags, ctos);
 560       __ jccb(Assembler::notEqual, notChar);
 561       // ctos
 562       __ load_unsigned_short(rax, field);
 563       __ push(ctos);
 564       __ jmp(Done);
 565 
 566       __ bind(notChar);
 567       __ cmpl(flags, ztos);
 568       __ jccb(Assembler::notEqual, notBool);
 569       // ztos
 570       __ load_signed_byte(rax, field);
 571       __ push(ztos);
 572       __ jmp(Done);
 573 
 574       __ bind(notBool);
 575       break;
 576     }
 577 
 578   case Bytecodes::_ldc2_w:
 579     {
 580       Label notLong, notDouble;
 581       __ cmpl(flags, ltos);
 582       __ jccb(Assembler::notEqual, notLong);
 583       // ltos
 584       // Loading high word first because movptr clobbers rax
 585       NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
 586       __ movptr(rax, field);
 587       __ push(ltos);
 588       __ jmp(Done);
 589 
 590       __ bind(notLong);
 591       __ cmpl(flags, dtos);
 592       __ jccb(Assembler::notEqual, notDouble);
 593       // dtos
 594       __ load_double(field);
 595       __ push(dtos);
 596       __ jmp(Done);
 597 
 598       __ bind(notDouble);
 599       break;
 600     }
 601 
 602   default:
 603     ShouldNotReachHere();
 604   }
 605 
 606   __ stop("bad ldc/condy");
 607 }
 608 
 609 void TemplateTable::locals_index(Register reg, int offset) {
 610   __ load_unsigned_byte(reg, at_bcp(offset));
 611   __ negptr(reg);
 612 }
 613 
 614 void TemplateTable::iload() {
 615   iload_internal();
 616 }
 617 
 618 void TemplateTable::nofast_iload() {
 619   iload_internal(may_not_rewrite);
 620 }
 621 
 622 void TemplateTable::iload_internal(RewriteControl rc) {
 623   transition(vtos, itos);
 624   if (RewriteFrequentPairs && rc == may_rewrite) {
 625     Label rewrite, done;
 626     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 627     LP64_ONLY(assert(rbx != bc, "register damaged"));
 628 
 629     // get next byte
 630     __ load_unsigned_byte(rbx,
 631                           at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 632     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 633     // last two iloads in a pair.  Comparing against fast_iload means that
 634     // the next bytecode is neither an iload or a caload, and therefore
 635     // an iload pair.
 636     __ cmpl(rbx, Bytecodes::_iload);
 637     __ jcc(Assembler::equal, done);
 638 
 639     __ cmpl(rbx, Bytecodes::_fast_iload);
 640     __ movl(bc, Bytecodes::_fast_iload2);
 641 
 642     __ jccb(Assembler::equal, rewrite);
 643 
 644     // if _caload, rewrite to fast_icaload
 645     __ cmpl(rbx, Bytecodes::_caload);
 646     __ movl(bc, Bytecodes::_fast_icaload);
 647     __ jccb(Assembler::equal, rewrite);
 648 
 649     // rewrite so iload doesn't check again.
 650     __ movl(bc, Bytecodes::_fast_iload);
 651 
 652     // rewrite
 653     // bc: fast bytecode
 654     __ bind(rewrite);
 655     patch_bytecode(Bytecodes::_iload, bc, rbx, false);
 656     __ bind(done);
 657   }
 658 
 659   // Get the local value into tos
 660   locals_index(rbx);
 661   __ movl(rax, iaddress(rbx));
 662 }
 663 
 664 void TemplateTable::fast_iload2() {
 665   transition(vtos, itos);
 666   locals_index(rbx);
 667   __ movl(rax, iaddress(rbx));
 668   __ push(itos);
 669   locals_index(rbx, 3);
 670   __ movl(rax, iaddress(rbx));
 671 }
 672 
 673 void TemplateTable::fast_iload() {
 674   transition(vtos, itos);
 675   locals_index(rbx);
 676   __ movl(rax, iaddress(rbx));
 677 }
 678 
 679 void TemplateTable::lload() {
 680   transition(vtos, ltos);
 681   locals_index(rbx);
 682   __ movptr(rax, laddress(rbx));
 683   NOT_LP64(__ movl(rdx, haddress(rbx)));
 684 }
 685 
 686 void TemplateTable::fload() {
 687   transition(vtos, ftos);
 688   locals_index(rbx);
 689   __ load_float(faddress(rbx));
 690 }
 691 
 692 void TemplateTable::dload() {
 693   transition(vtos, dtos);
 694   locals_index(rbx);
 695   __ load_double(daddress(rbx));
 696 }
 697 
 698 void TemplateTable::aload() {
 699   transition(vtos, atos);
 700   locals_index(rbx);
 701   __ movptr(rax, aaddress(rbx));
 702 }
 703 
 704 void TemplateTable::locals_index_wide(Register reg) {
 705   __ load_unsigned_short(reg, at_bcp(2));
 706   __ bswapl(reg);
 707   __ shrl(reg, 16);
 708   __ negptr(reg);
 709 }
 710 
 711 void TemplateTable::wide_iload() {
 712   transition(vtos, itos);
 713   locals_index_wide(rbx);
 714   __ movl(rax, iaddress(rbx));
 715 }
 716 
 717 void TemplateTable::wide_lload() {
 718   transition(vtos, ltos);
 719   locals_index_wide(rbx);
 720   __ movptr(rax, laddress(rbx));
 721   NOT_LP64(__ movl(rdx, haddress(rbx)));
 722 }
 723 
 724 void TemplateTable::wide_fload() {
 725   transition(vtos, ftos);
 726   locals_index_wide(rbx);
 727   __ load_float(faddress(rbx));
 728 }
 729 
 730 void TemplateTable::wide_dload() {
 731   transition(vtos, dtos);
 732   locals_index_wide(rbx);
 733   __ load_double(daddress(rbx));
 734 }
 735 
 736 void TemplateTable::wide_aload() {
 737   transition(vtos, atos);
 738   locals_index_wide(rbx);
 739   __ movptr(rax, aaddress(rbx));
 740 }
 741 
 742 void TemplateTable::index_check(Register array, Register index) {
 743   // Pop ptr into array
 744   __ pop_ptr(array);
 745   index_check_without_pop(array, index);
 746 }
 747 
 748 void TemplateTable::index_check_without_pop(Register array, Register index) {
 749   // destroys rbx
 750   // check array
 751   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 752   // sign extend index for use by indexed load
 753   __ movl2ptr(index, index);
 754   // check index
 755   __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
 756   if (index != rbx) {
 757     // ??? convention: move aberrant index into rbx for exception message
 758     assert(rbx != array, "different registers");
 759     __ movl(rbx, index);
 760   }
 761   Label skip;
 762   __ jccb(Assembler::below, skip);
 763   // Pass array to create more detailed exceptions.
 764   __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
 765   __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
 766   __ bind(skip);
 767 }
 768 
 769 void TemplateTable::iaload() {
 770   transition(itos, itos);
 771   // rax: index
 772   // rdx: array
 773   index_check(rdx, rax); // kills rbx
 774   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
 775                     Address(rdx, rax, Address::times_4,
 776                             arrayOopDesc::base_offset_in_bytes(T_INT)),
 777                     noreg, noreg);
 778 }
 779 
 780 void TemplateTable::laload() {
 781   transition(itos, ltos);
 782   // rax: index
 783   // rdx: array
 784   index_check(rdx, rax); // kills rbx
 785   NOT_LP64(__ mov(rbx, rax));
 786   // rbx,: index
 787   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
 788                     Address(rdx, rbx, Address::times_8,
 789                             arrayOopDesc::base_offset_in_bytes(T_LONG)),
 790                     noreg, noreg);
 791 }
 792 
 793 
 794 
 795 void TemplateTable::faload() {
 796   transition(itos, ftos);
 797   // rax: index
 798   // rdx: array
 799   index_check(rdx, rax); // kills rbx
 800   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
 801                     Address(rdx, rax,
 802                             Address::times_4,
 803                             arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
 804                     noreg, noreg);
 805 }
 806 
 807 void TemplateTable::daload() {
 808   transition(itos, dtos);
 809   // rax: index
 810   // rdx: array
 811   index_check(rdx, rax); // kills rbx
 812   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
 813                     Address(rdx, rax,
 814                             Address::times_8,
 815                             arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
 816                     noreg, noreg);
 817 }
 818 
 819 void TemplateTable::aaload() {
 820   transition(itos, atos);
 821   // rax: index
 822   // rdx: array
 823   index_check(rdx, rax); // kills rbx
 824   do_oop_load(_masm,
 825               Address(rdx, rax,
 826                       UseCompressedOops ? Address::times_4 : Address::times_ptr,
 827                       arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
 828               rax,
 829               IS_ARRAY);
 830 }
 831 
 832 void TemplateTable::baload() {
 833   transition(itos, itos);
 834   // rax: index
 835   // rdx: array
 836   index_check(rdx, rax); // kills rbx
 837   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
 838                     Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
 839                     noreg, noreg);
 840 }
 841 
 842 void TemplateTable::caload() {
 843   transition(itos, itos);
 844   // rax: index
 845   // rdx: array
 846   index_check(rdx, rax); // kills rbx
 847   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 848                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 849                     noreg, noreg);
 850 }
 851 
 852 // iload followed by caload frequent pair
 853 void TemplateTable::fast_icaload() {
 854   transition(vtos, itos);
 855   // load index out of locals
 856   locals_index(rbx);
 857   __ movl(rax, iaddress(rbx));
 858 
 859   // rax: index
 860   // rdx: array
 861   index_check(rdx, rax); // kills rbx
 862   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
 863                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
 864                     noreg, noreg);
 865 }
 866 
 867 
 868 void TemplateTable::saload() {
 869   transition(itos, itos);
 870   // rax: index
 871   // rdx: array
 872   index_check(rdx, rax); // kills rbx
 873   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
 874                     Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
 875                     noreg, noreg);
 876 }
 877 
 878 void TemplateTable::iload(int n) {
 879   transition(vtos, itos);
 880   __ movl(rax, iaddress(n));
 881 }
 882 
 883 void TemplateTable::lload(int n) {
 884   transition(vtos, ltos);
 885   __ movptr(rax, laddress(n));
 886   NOT_LP64(__ movptr(rdx, haddress(n)));
 887 }
 888 
 889 void TemplateTable::fload(int n) {
 890   transition(vtos, ftos);
 891   __ load_float(faddress(n));
 892 }
 893 
 894 void TemplateTable::dload(int n) {
 895   transition(vtos, dtos);
 896   __ load_double(daddress(n));
 897 }
 898 
 899 void TemplateTable::aload(int n) {
 900   transition(vtos, atos);
 901   __ movptr(rax, aaddress(n));
 902 }
 903 
 904 void TemplateTable::aload_0() {
 905   aload_0_internal();
 906 }
 907 
 908 void TemplateTable::nofast_aload_0() {
 909   aload_0_internal(may_not_rewrite);
 910 }
 911 
 912 void TemplateTable::aload_0_internal(RewriteControl rc) {
 913   transition(vtos, atos);
 914   // According to bytecode histograms, the pairs:
 915   //
 916   // _aload_0, _fast_igetfield
 917   // _aload_0, _fast_agetfield
 918   // _aload_0, _fast_fgetfield
 919   //
 920   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 921   // _aload_0 bytecode checks if the next bytecode is either
 922   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 923   // rewrites the current bytecode into a pair bytecode; otherwise it
 924   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 925   // the pair check anymore.
 926   //
 927   // Note: If the next bytecode is _getfield, the rewrite must be
 928   //       delayed, otherwise we may miss an opportunity for a pair.
 929   //
 930   // Also rewrite frequent pairs
 931   //   aload_0, aload_1
 932   //   aload_0, iload_1
 933   // These bytecodes with a small amount of code are most profitable
 934   // to rewrite
 935   if (RewriteFrequentPairs && rc == may_rewrite) {
 936     Label rewrite, done;
 937 
 938     const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
 939     LP64_ONLY(assert(rbx != bc, "register damaged"));
 940 
 941     // get next byte
 942     __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 943 
 944     // if _getfield then wait with rewrite
 945     __ cmpl(rbx, Bytecodes::_getfield);
 946     __ jcc(Assembler::equal, done);
 947 
 948     // if _igetfield then rewrite to _fast_iaccess_0
 949     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 950     __ cmpl(rbx, Bytecodes::_fast_igetfield);
 951     __ movl(bc, Bytecodes::_fast_iaccess_0);
 952     __ jccb(Assembler::equal, rewrite);
 953 
 954     // if _agetfield then rewrite to _fast_aaccess_0
 955     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 956     __ cmpl(rbx, Bytecodes::_fast_agetfield);
 957     __ movl(bc, Bytecodes::_fast_aaccess_0);
 958     __ jccb(Assembler::equal, rewrite);
 959 
 960     // if _fgetfield then rewrite to _fast_faccess_0
 961     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 962     __ cmpl(rbx, Bytecodes::_fast_fgetfield);
 963     __ movl(bc, Bytecodes::_fast_faccess_0);
 964     __ jccb(Assembler::equal, rewrite);
 965 
 966     // else rewrite to _fast_aload0
 967     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 968     __ movl(bc, Bytecodes::_fast_aload_0);
 969 
 970     // rewrite
 971     // bc: fast bytecode
 972     __ bind(rewrite);
 973     patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
 974 
 975     __ bind(done);
 976   }
 977 
 978   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 979   aload(0);
 980 }
 981 
 982 void TemplateTable::istore() {
 983   transition(itos, vtos);
 984   locals_index(rbx);
 985   __ movl(iaddress(rbx), rax);
 986 }
 987 
 988 
 989 void TemplateTable::lstore() {
 990   transition(ltos, vtos);
 991   locals_index(rbx);
 992   __ movptr(laddress(rbx), rax);
 993   NOT_LP64(__ movptr(haddress(rbx), rdx));
 994 }
 995 
 996 void TemplateTable::fstore() {
 997   transition(ftos, vtos);
 998   locals_index(rbx);
 999   __ store_float(faddress(rbx));
1000 }
1001 
1002 void TemplateTable::dstore() {
1003   transition(dtos, vtos);
1004   locals_index(rbx);
1005   __ store_double(daddress(rbx));
1006 }
1007 
1008 void TemplateTable::astore() {
1009   transition(vtos, vtos);
1010   __ pop_ptr(rax);
1011   locals_index(rbx);
1012   __ movptr(aaddress(rbx), rax);
1013 }
1014 
1015 void TemplateTable::wide_istore() {
1016   transition(vtos, vtos);
1017   __ pop_i();
1018   locals_index_wide(rbx);
1019   __ movl(iaddress(rbx), rax);
1020 }
1021 
1022 void TemplateTable::wide_lstore() {
1023   transition(vtos, vtos);
1024   NOT_LP64(__ pop_l(rax, rdx));
1025   LP64_ONLY(__ pop_l());
1026   locals_index_wide(rbx);
1027   __ movptr(laddress(rbx), rax);
1028   NOT_LP64(__ movl(haddress(rbx), rdx));
1029 }
1030 
1031 void TemplateTable::wide_fstore() {
1032 #ifdef _LP64
1033   transition(vtos, vtos);
1034   __ pop_f(xmm0);
1035   locals_index_wide(rbx);
1036   __ movflt(faddress(rbx), xmm0);
1037 #else
1038   wide_istore();
1039 #endif
1040 }
1041 
1042 void TemplateTable::wide_dstore() {
1043 #ifdef _LP64
1044   transition(vtos, vtos);
1045   __ pop_d(xmm0);
1046   locals_index_wide(rbx);
1047   __ movdbl(daddress(rbx), xmm0);
1048 #else
1049   wide_lstore();
1050 #endif
1051 }
1052 
1053 void TemplateTable::wide_astore() {
1054   transition(vtos, vtos);
1055   __ pop_ptr(rax);
1056   locals_index_wide(rbx);
1057   __ movptr(aaddress(rbx), rax);
1058 }
1059 
1060 void TemplateTable::iastore() {
1061   transition(itos, vtos);
1062   __ pop_i(rbx);
1063   // rax: value
1064   // rbx: index
1065   // rdx: array
1066   index_check(rdx, rbx); // prefer index in rbx
1067   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1068                      Address(rdx, rbx, Address::times_4,
1069                              arrayOopDesc::base_offset_in_bytes(T_INT)),
1070                      rax, noreg, noreg, noreg);
1071 }
1072 
1073 void TemplateTable::lastore() {
1074   transition(ltos, vtos);
1075   __ pop_i(rbx);
1076   // rax,: low(value)
1077   // rcx: array
1078   // rdx: high(value)
1079   index_check(rcx, rbx);  // prefer index in rbx,
1080   // rbx,: index
1081   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1082                      Address(rcx, rbx, Address::times_8,
1083                              arrayOopDesc::base_offset_in_bytes(T_LONG)),
1084                      noreg /* ltos */, noreg, noreg, noreg);
1085 }
1086 
1087 
1088 void TemplateTable::fastore() {
1089   transition(ftos, vtos);
1090   __ pop_i(rbx);
1091   // value is in UseSSE >= 1 ? xmm0 : ST(0)
1092   // rbx:  index
1093   // rdx:  array
1094   index_check(rdx, rbx); // prefer index in rbx
1095   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1096                      Address(rdx, rbx, Address::times_4,
1097                              arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1098                      noreg /* ftos */, noreg, noreg, noreg);
1099 }
1100 
1101 void TemplateTable::dastore() {
1102   transition(dtos, vtos);
1103   __ pop_i(rbx);
1104   // value is in UseSSE >= 2 ? xmm0 : ST(0)
1105   // rbx:  index
1106   // rdx:  array
1107   index_check(rdx, rbx); // prefer index in rbx
1108   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1109                      Address(rdx, rbx, Address::times_8,
1110                              arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1111                      noreg /* dtos */, noreg, noreg, noreg);
1112 }
1113 
1114 void TemplateTable::aastore() {
1115   Label is_null, ok_is_subtype, done;
1116   transition(vtos, vtos);
1117   // stack: ..., array, index, value
1118   __ movptr(rax, at_tos());    // value
1119   __ movl(rcx, at_tos_p1()); // index
1120   __ movptr(rdx, at_tos_p2()); // array
1121 
1122   Address element_address(rdx, rcx,
1123                           UseCompressedOops? Address::times_4 : Address::times_ptr,
1124                           arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1125 
1126   index_check_without_pop(rdx, rcx);     // kills rbx
1127   __ testptr(rax, rax);
1128   __ jcc(Assembler::zero, is_null);
1129 
1130   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1131   // Move subklass into rbx
1132   __ load_klass(rbx, rax, tmp_load_klass);
1133   // Move superklass into rax
1134   __ load_klass(rax, rdx, tmp_load_klass);
1135   __ movptr(rax, Address(rax,
1136                          ObjArrayKlass::element_klass_offset()));
1137 
1138   // Generate subtype check.  Blows rcx, rdi
1139   // Superklass in rax.  Subklass in rbx.
1140   __ gen_subtype_check(rbx, ok_is_subtype);
1141 
1142   // Come here on failure
1143   // object is at TOS
1144   __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1145 
1146   // Come here on success
1147   __ bind(ok_is_subtype);
1148 
1149   // Get the value we will store
1150   __ movptr(rax, at_tos());
1151   __ movl(rcx, at_tos_p1()); // index
1152   // Now store using the appropriate barrier
1153   do_oop_store(_masm, element_address, rax, IS_ARRAY);
1154   __ jmp(done);
1155 
1156   // Have a NULL in rax, rdx=array, ecx=index.  Store NULL at ary[idx]
1157   __ bind(is_null);
1158   __ profile_null_seen(rbx);
1159 
1160   // Store a NULL
1161   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1162 
1163   // Pop stack arguments
1164   __ bind(done);
1165   __ addptr(rsp, 3 * Interpreter::stackElementSize);
1166 }
1167 
1168 void TemplateTable::bastore() {
1169   transition(itos, vtos);
1170   __ pop_i(rbx);
1171   // rax: value
1172   // rbx: index
1173   // rdx: array
1174   index_check(rdx, rbx); // prefer index in rbx
1175   // Need to check whether array is boolean or byte
1176   // since both types share the bastore bytecode.
1177   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1178   __ load_klass(rcx, rdx, tmp_load_klass);
1179   __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1180   int diffbit = Klass::layout_helper_boolean_diffbit();
1181   __ testl(rcx, diffbit);
1182   Label L_skip;
1183   __ jccb(Assembler::zero, L_skip);
1184   __ andl(rax, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1185   __ bind(L_skip);
1186   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1187                      Address(rdx, rbx,Address::times_1,
1188                              arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1189                      rax, noreg, noreg, noreg);
1190 }
1191 
1192 void TemplateTable::castore() {
1193   transition(itos, vtos);
1194   __ pop_i(rbx);
1195   // rax: value
1196   // rbx: index
1197   // rdx: array
1198   index_check(rdx, rbx);  // prefer index in rbx
1199   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1200                      Address(rdx, rbx, Address::times_2,
1201                              arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1202                      rax, noreg, noreg, noreg);
1203 }
1204 
1205 
1206 void TemplateTable::sastore() {
1207   castore();
1208 }
1209 
1210 void TemplateTable::istore(int n) {
1211   transition(itos, vtos);
1212   __ movl(iaddress(n), rax);
1213 }
1214 
1215 void TemplateTable::lstore(int n) {
1216   transition(ltos, vtos);
1217   __ movptr(laddress(n), rax);
1218   NOT_LP64(__ movptr(haddress(n), rdx));
1219 }
1220 
1221 void TemplateTable::fstore(int n) {
1222   transition(ftos, vtos);
1223   __ store_float(faddress(n));
1224 }
1225 
1226 void TemplateTable::dstore(int n) {
1227   transition(dtos, vtos);
1228   __ store_double(daddress(n));
1229 }
1230 
1231 
1232 void TemplateTable::astore(int n) {
1233   transition(vtos, vtos);
1234   __ pop_ptr(rax);
1235   __ movptr(aaddress(n), rax);
1236 }
1237 
1238 void TemplateTable::pop() {
1239   transition(vtos, vtos);
1240   __ addptr(rsp, Interpreter::stackElementSize);
1241 }
1242 
1243 void TemplateTable::pop2() {
1244   transition(vtos, vtos);
1245   __ addptr(rsp, 2 * Interpreter::stackElementSize);
1246 }
1247 
1248 
1249 void TemplateTable::dup() {
1250   transition(vtos, vtos);
1251   __ load_ptr(0, rax);
1252   __ push_ptr(rax);
1253   // stack: ..., a, a
1254 }
1255 
1256 void TemplateTable::dup_x1() {
1257   transition(vtos, vtos);
1258   // stack: ..., a, b
1259   __ load_ptr( 0, rax);  // load b
1260   __ load_ptr( 1, rcx);  // load a
1261   __ store_ptr(1, rax);  // store b
1262   __ store_ptr(0, rcx);  // store a
1263   __ push_ptr(rax);      // push b
1264   // stack: ..., b, a, b
1265 }
1266 
1267 void TemplateTable::dup_x2() {
1268   transition(vtos, vtos);
1269   // stack: ..., a, b, c
1270   __ load_ptr( 0, rax);  // load c
1271   __ load_ptr( 2, rcx);  // load a
1272   __ store_ptr(2, rax);  // store c in a
1273   __ push_ptr(rax);      // push c
1274   // stack: ..., c, b, c, c
1275   __ load_ptr( 2, rax);  // load b
1276   __ store_ptr(2, rcx);  // store a in b
1277   // stack: ..., c, a, c, c
1278   __ store_ptr(1, rax);  // store b in c
1279   // stack: ..., c, a, b, c
1280 }
1281 
1282 void TemplateTable::dup2() {
1283   transition(vtos, vtos);
1284   // stack: ..., a, b
1285   __ load_ptr(1, rax);  // load a
1286   __ push_ptr(rax);     // push a
1287   __ load_ptr(1, rax);  // load b
1288   __ push_ptr(rax);     // push b
1289   // stack: ..., a, b, a, b
1290 }
1291 
1292 
1293 void TemplateTable::dup2_x1() {
1294   transition(vtos, vtos);
1295   // stack: ..., a, b, c
1296   __ load_ptr( 0, rcx);  // load c
1297   __ load_ptr( 1, rax);  // load b
1298   __ push_ptr(rax);      // push b
1299   __ push_ptr(rcx);      // push c
1300   // stack: ..., a, b, c, b, c
1301   __ store_ptr(3, rcx);  // store c in b
1302   // stack: ..., a, c, c, b, c
1303   __ load_ptr( 4, rcx);  // load a
1304   __ store_ptr(2, rcx);  // store a in 2nd c
1305   // stack: ..., a, c, a, b, c
1306   __ store_ptr(4, rax);  // store b in a
1307   // stack: ..., b, c, a, b, c
1308 }
1309 
1310 void TemplateTable::dup2_x2() {
1311   transition(vtos, vtos);
1312   // stack: ..., a, b, c, d
1313   __ load_ptr( 0, rcx);  // load d
1314   __ load_ptr( 1, rax);  // load c
1315   __ push_ptr(rax);      // push c
1316   __ push_ptr(rcx);      // push d
1317   // stack: ..., a, b, c, d, c, d
1318   __ load_ptr( 4, rax);  // load b
1319   __ store_ptr(2, rax);  // store b in d
1320   __ store_ptr(4, rcx);  // store d in b
1321   // stack: ..., a, d, c, b, c, d
1322   __ load_ptr( 5, rcx);  // load a
1323   __ load_ptr( 3, rax);  // load c
1324   __ store_ptr(3, rcx);  // store a in c
1325   __ store_ptr(5, rax);  // store c in a
1326   // stack: ..., c, d, a, b, c, d
1327 }
1328 
1329 void TemplateTable::swap() {
1330   transition(vtos, vtos);
1331   // stack: ..., a, b
1332   __ load_ptr( 1, rcx);  // load a
1333   __ load_ptr( 0, rax);  // load b
1334   __ store_ptr(0, rcx);  // store a in b
1335   __ store_ptr(1, rax);  // store b in a
1336   // stack: ..., b, a
1337 }
1338 
1339 void TemplateTable::iop2(Operation op) {
1340   transition(itos, itos);
1341   switch (op) {
1342   case add  :                    __ pop_i(rdx); __ addl (rax, rdx); break;
1343   case sub  : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1344   case mul  :                    __ pop_i(rdx); __ imull(rax, rdx); break;
1345   case _and :                    __ pop_i(rdx); __ andl (rax, rdx); break;
1346   case _or  :                    __ pop_i(rdx); __ orl  (rax, rdx); break;
1347   case _xor :                    __ pop_i(rdx); __ xorl (rax, rdx); break;
1348   case shl  : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax);      break;
1349   case shr  : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax);      break;
1350   case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax);      break;
1351   default   : ShouldNotReachHere();
1352   }
1353 }
1354 
1355 void TemplateTable::lop2(Operation op) {
1356   transition(ltos, ltos);
1357 #ifdef _LP64
1358   switch (op) {
1359   case add  :                    __ pop_l(rdx); __ addptr(rax, rdx); break;
1360   case sub  : __ mov(rdx, rax);  __ pop_l(rax); __ subptr(rax, rdx); break;
1361   case _and :                    __ pop_l(rdx); __ andptr(rax, rdx); break;
1362   case _or  :                    __ pop_l(rdx); __ orptr (rax, rdx); break;
1363   case _xor :                    __ pop_l(rdx); __ xorptr(rax, rdx); break;
1364   default   : ShouldNotReachHere();
1365   }
1366 #else
1367   __ pop_l(rbx, rcx);
1368   switch (op) {
1369     case add  : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1370     case sub  : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1371                 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1372     case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1373     case _or  : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1374     case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1375     default   : ShouldNotReachHere();
1376   }
1377 #endif
1378 }
1379 
1380 void TemplateTable::idiv() {
1381   transition(itos, itos);
1382   __ movl(rcx, rax);
1383   __ pop_i(rax);
1384   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1385   //       they are not equal, one could do a normal division (no correction
1386   //       needed), which may speed up this implementation for the common case.
1387   //       (see also JVM spec., p.243 & p.271)
1388   __ corrected_idivl(rcx);
1389 }
1390 
1391 void TemplateTable::irem() {
1392   transition(itos, itos);
1393   __ movl(rcx, rax);
1394   __ pop_i(rax);
1395   // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1396   //       they are not equal, one could do a normal division (no correction
1397   //       needed), which may speed up this implementation for the common case.
1398   //       (see also JVM spec., p.243 & p.271)
1399   __ corrected_idivl(rcx);
1400   __ movl(rax, rdx);
1401 }
1402 
1403 void TemplateTable::lmul() {
1404   transition(ltos, ltos);
1405 #ifdef _LP64
1406   __ pop_l(rdx);
1407   __ imulq(rax, rdx);
1408 #else
1409   __ pop_l(rbx, rcx);
1410   __ push(rcx); __ push(rbx);
1411   __ push(rdx); __ push(rax);
1412   __ lmul(2 * wordSize, 0);
1413   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1414 #endif
1415 }
1416 
1417 void TemplateTable::ldiv() {
1418   transition(ltos, ltos);
1419 #ifdef _LP64
1420   __ mov(rcx, rax);
1421   __ pop_l(rax);
1422   // generate explicit div0 check
1423   __ testq(rcx, rcx);
1424   __ jump_cc(Assembler::zero,
1425              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1426   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1427   //       they are not equal, one could do a normal division (no correction
1428   //       needed), which may speed up this implementation for the common case.
1429   //       (see also JVM spec., p.243 & p.271)
1430   __ corrected_idivq(rcx); // kills rbx
1431 #else
1432   __ pop_l(rbx, rcx);
1433   __ push(rcx); __ push(rbx);
1434   __ push(rdx); __ push(rax);
1435   // check if y = 0
1436   __ orl(rax, rdx);
1437   __ jump_cc(Assembler::zero,
1438              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1439   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1440   __ addptr(rsp, 4 * wordSize);  // take off temporaries
1441 #endif
1442 }
1443 
1444 void TemplateTable::lrem() {
1445   transition(ltos, ltos);
1446 #ifdef _LP64
1447   __ mov(rcx, rax);
1448   __ pop_l(rax);
1449   __ testq(rcx, rcx);
1450   __ jump_cc(Assembler::zero,
1451              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1452   // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1453   //       they are not equal, one could do a normal division (no correction
1454   //       needed), which may speed up this implementation for the common case.
1455   //       (see also JVM spec., p.243 & p.271)
1456   __ corrected_idivq(rcx); // kills rbx
1457   __ mov(rax, rdx);
1458 #else
1459   __ pop_l(rbx, rcx);
1460   __ push(rcx); __ push(rbx);
1461   __ push(rdx); __ push(rax);
1462   // check if y = 0
1463   __ orl(rax, rdx);
1464   __ jump_cc(Assembler::zero,
1465              ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1466   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1467   __ addptr(rsp, 4 * wordSize);
1468 #endif
1469 }
1470 
1471 void TemplateTable::lshl() {
1472   transition(itos, ltos);
1473   __ movl(rcx, rax);                             // get shift count
1474   #ifdef _LP64
1475   __ pop_l(rax);                                 // get shift value
1476   __ shlq(rax);
1477 #else
1478   __ pop_l(rax, rdx);                            // get shift value
1479   __ lshl(rdx, rax);
1480 #endif
1481 }
1482 
1483 void TemplateTable::lshr() {
1484 #ifdef _LP64
1485   transition(itos, ltos);
1486   __ movl(rcx, rax);                             // get shift count
1487   __ pop_l(rax);                                 // get shift value
1488   __ sarq(rax);
1489 #else
1490   transition(itos, ltos);
1491   __ mov(rcx, rax);                              // get shift count
1492   __ pop_l(rax, rdx);                            // get shift value
1493   __ lshr(rdx, rax, true);
1494 #endif
1495 }
1496 
1497 void TemplateTable::lushr() {
1498   transition(itos, ltos);
1499 #ifdef _LP64
1500   __ movl(rcx, rax);                             // get shift count
1501   __ pop_l(rax);                                 // get shift value
1502   __ shrq(rax);
1503 #else
1504   __ mov(rcx, rax);                              // get shift count
1505   __ pop_l(rax, rdx);                            // get shift value
1506   __ lshr(rdx, rax);
1507 #endif
1508 }
1509 
1510 void TemplateTable::fop2(Operation op) {
1511   transition(ftos, ftos);
1512 
1513   if (UseSSE >= 1) {
1514     switch (op) {
1515     case add:
1516       __ addss(xmm0, at_rsp());
1517       __ addptr(rsp, Interpreter::stackElementSize);
1518       break;
1519     case sub:
1520       __ movflt(xmm1, xmm0);
1521       __ pop_f(xmm0);
1522       __ subss(xmm0, xmm1);
1523       break;
1524     case mul:
1525       __ mulss(xmm0, at_rsp());
1526       __ addptr(rsp, Interpreter::stackElementSize);
1527       break;
1528     case div:
1529       __ movflt(xmm1, xmm0);
1530       __ pop_f(xmm0);
1531       __ divss(xmm0, xmm1);
1532       break;
1533     case rem:
1534       // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1535       // modulo operation. The frem method calls the function
1536       // double fmod(double x, double y) in math.h. The documentation of fmod states:
1537       // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1538       // (signalling or quiet) is returned.
1539       //
1540       // On x86_32 platforms the FPU is used to perform the modulo operation. The
1541       // reason is that on 32-bit Windows the sign of modulo operations diverges from
1542       // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1543       // The fprem instruction used on x86_32 is functionally equivalent to
1544       // SharedRuntime::frem in that it returns a NaN.
1545 #ifdef _LP64
1546       __ movflt(xmm1, xmm0);
1547       __ pop_f(xmm0);
1548       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1549 #else
1550       __ push_f(xmm0);
1551       __ pop_f();
1552       __ fld_s(at_rsp());
1553       __ fremr(rax);
1554       __ f2ieee();
1555       __ pop(rax);  // pop second operand off the stack
1556       __ push_f();
1557       __ pop_f(xmm0);
1558 #endif
1559       break;
1560     default:
1561       ShouldNotReachHere();
1562       break;
1563     }
1564   } else {
1565 #ifdef _LP64
1566     ShouldNotReachHere();
1567 #else
1568     switch (op) {
1569     case add: __ fadd_s (at_rsp());                break;
1570     case sub: __ fsubr_s(at_rsp());                break;
1571     case mul: __ fmul_s (at_rsp());                break;
1572     case div: __ fdivr_s(at_rsp());                break;
1573     case rem: __ fld_s  (at_rsp()); __ fremr(rax); break;
1574     default : ShouldNotReachHere();
1575     }
1576     __ f2ieee();
1577     __ pop(rax);  // pop second operand off the stack
1578 #endif // _LP64
1579   }
1580 }
1581 
1582 void TemplateTable::dop2(Operation op) {
1583   transition(dtos, dtos);
1584   if (UseSSE >= 2) {
1585     switch (op) {
1586     case add:
1587       __ addsd(xmm0, at_rsp());
1588       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1589       break;
1590     case sub:
1591       __ movdbl(xmm1, xmm0);
1592       __ pop_d(xmm0);
1593       __ subsd(xmm0, xmm1);
1594       break;
1595     case mul:
1596       __ mulsd(xmm0, at_rsp());
1597       __ addptr(rsp, 2 * Interpreter::stackElementSize);
1598       break;
1599     case div:
1600       __ movdbl(xmm1, xmm0);
1601       __ pop_d(xmm0);
1602       __ divsd(xmm0, xmm1);
1603       break;
1604     case rem:
1605       // Similar to fop2(), the modulo operation is performed using the
1606       // SharedRuntime::drem method (on x86_64 platforms) or using the
1607       // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1608 #ifdef _LP64
1609       __ movdbl(xmm1, xmm0);
1610       __ pop_d(xmm0);
1611       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1612 #else
1613       __ push_d(xmm0);
1614       __ pop_d();
1615       __ fld_d(at_rsp());
1616       __ fremr(rax);
1617       __ d2ieee();
1618       __ pop(rax);
1619       __ pop(rdx);
1620       __ push_d();
1621       __ pop_d(xmm0);
1622 #endif
1623       break;
1624     default:
1625       ShouldNotReachHere();
1626       break;
1627     }
1628   } else {
1629 #ifdef _LP64
1630     ShouldNotReachHere();
1631 #else
1632     switch (op) {
1633     case add: __ fadd_d (at_rsp());                break;
1634     case sub: __ fsubr_d(at_rsp());                break;
1635     case mul: {
1636       // strict semantics
1637       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1638       __ fmulp();
1639       __ fmul_d (at_rsp());
1640       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1641       __ fmulp();
1642       break;
1643     }
1644     case div: {
1645       // strict semantics
1646       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
1647       __ fmul_d (at_rsp());
1648       __ fdivrp();
1649       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
1650       __ fmulp();
1651       break;
1652     }
1653     case rem: __ fld_d  (at_rsp()); __ fremr(rax); break;
1654     default : ShouldNotReachHere();
1655     }
1656     __ d2ieee();
1657     // Pop double precision number from rsp.
1658     __ pop(rax);
1659     __ pop(rdx);
1660 #endif
1661   }
1662 }
1663 
1664 void TemplateTable::ineg() {
1665   transition(itos, itos);
1666   __ negl(rax);
1667 }
1668 
1669 void TemplateTable::lneg() {
1670   transition(ltos, ltos);
1671   LP64_ONLY(__ negq(rax));
1672   NOT_LP64(__ lneg(rdx, rax));
1673 }
1674 
1675 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1676 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1677   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1678   // of 128-bits operands for SSE instructions.
1679   jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1680   // Store the value to a 128-bits operand.
1681   operand[0] = lo;
1682   operand[1] = hi;
1683   return operand;
1684 }
1685 
1686 // Buffer for 128-bits masks used by SSE instructions.
1687 static jlong float_signflip_pool[2*2];
1688 static jlong double_signflip_pool[2*2];
1689 
1690 void TemplateTable::fneg() {
1691   transition(ftos, ftos);
1692   if (UseSSE >= 1) {
1693     static jlong *float_signflip  = double_quadword(&float_signflip_pool[1],  CONST64(0x8000000080000000),  CONST64(0x8000000080000000));
1694     __ xorps(xmm0, ExternalAddress((address) float_signflip));
1695   } else {
1696     LP64_ONLY(ShouldNotReachHere());
1697     NOT_LP64(__ fchs());
1698   }
1699 }
1700 
1701 void TemplateTable::dneg() {
1702   transition(dtos, dtos);
1703   if (UseSSE >= 2) {
1704     static jlong *double_signflip =
1705       double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1706     __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1707   } else {
1708 #ifdef _LP64
1709     ShouldNotReachHere();
1710 #else
1711     __ fchs();
1712 #endif
1713   }
1714 }
1715 
1716 void TemplateTable::iinc() {
1717   transition(vtos, vtos);
1718   __ load_signed_byte(rdx, at_bcp(2)); // get constant
1719   locals_index(rbx);
1720   __ addl(iaddress(rbx), rdx);
1721 }
1722 
1723 void TemplateTable::wide_iinc() {
1724   transition(vtos, vtos);
1725   __ movl(rdx, at_bcp(4)); // get constant
1726   locals_index_wide(rbx);
1727   __ bswapl(rdx); // swap bytes & sign-extend constant
1728   __ sarl(rdx, 16);
1729   __ addl(iaddress(rbx), rdx);
1730   // Note: should probably use only one movl to get both
1731   //       the index and the constant -> fix this
1732 }
1733 
1734 void TemplateTable::convert() {
1735 #ifdef _LP64
1736   // Checking
1737 #ifdef ASSERT
1738   {
1739     TosState tos_in  = ilgl;
1740     TosState tos_out = ilgl;
1741     switch (bytecode()) {
1742     case Bytecodes::_i2l: // fall through
1743     case Bytecodes::_i2f: // fall through
1744     case Bytecodes::_i2d: // fall through
1745     case Bytecodes::_i2b: // fall through
1746     case Bytecodes::_i2c: // fall through
1747     case Bytecodes::_i2s: tos_in = itos; break;
1748     case Bytecodes::_l2i: // fall through
1749     case Bytecodes::_l2f: // fall through
1750     case Bytecodes::_l2d: tos_in = ltos; break;
1751     case Bytecodes::_f2i: // fall through
1752     case Bytecodes::_f2l: // fall through
1753     case Bytecodes::_f2d: tos_in = ftos; break;
1754     case Bytecodes::_d2i: // fall through
1755     case Bytecodes::_d2l: // fall through
1756     case Bytecodes::_d2f: tos_in = dtos; break;
1757     default             : ShouldNotReachHere();
1758     }
1759     switch (bytecode()) {
1760     case Bytecodes::_l2i: // fall through
1761     case Bytecodes::_f2i: // fall through
1762     case Bytecodes::_d2i: // fall through
1763     case Bytecodes::_i2b: // fall through
1764     case Bytecodes::_i2c: // fall through
1765     case Bytecodes::_i2s: tos_out = itos; break;
1766     case Bytecodes::_i2l: // fall through
1767     case Bytecodes::_f2l: // fall through
1768     case Bytecodes::_d2l: tos_out = ltos; break;
1769     case Bytecodes::_i2f: // fall through
1770     case Bytecodes::_l2f: // fall through
1771     case Bytecodes::_d2f: tos_out = ftos; break;
1772     case Bytecodes::_i2d: // fall through
1773     case Bytecodes::_l2d: // fall through
1774     case Bytecodes::_f2d: tos_out = dtos; break;
1775     default             : ShouldNotReachHere();
1776     }
1777     transition(tos_in, tos_out);
1778   }
1779 #endif // ASSERT
1780 
1781   static const int64_t is_nan = 0x8000000000000000L;
1782 
1783   // Conversion
1784   switch (bytecode()) {
1785   case Bytecodes::_i2l:
1786     __ movslq(rax, rax);
1787     break;
1788   case Bytecodes::_i2f:
1789     __ cvtsi2ssl(xmm0, rax);
1790     break;
1791   case Bytecodes::_i2d:
1792     __ cvtsi2sdl(xmm0, rax);
1793     break;
1794   case Bytecodes::_i2b:
1795     __ movsbl(rax, rax);
1796     break;
1797   case Bytecodes::_i2c:
1798     __ movzwl(rax, rax);
1799     break;
1800   case Bytecodes::_i2s:
1801     __ movswl(rax, rax);
1802     break;
1803   case Bytecodes::_l2i:
1804     __ movl(rax, rax);
1805     break;
1806   case Bytecodes::_l2f:
1807     __ cvtsi2ssq(xmm0, rax);
1808     break;
1809   case Bytecodes::_l2d:
1810     __ cvtsi2sdq(xmm0, rax);
1811     break;
1812   case Bytecodes::_f2i:
1813   {
1814     Label L;
1815     __ cvttss2sil(rax, xmm0);
1816     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1817     __ jcc(Assembler::notEqual, L);
1818     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1819     __ bind(L);
1820   }
1821     break;
1822   case Bytecodes::_f2l:
1823   {
1824     Label L;
1825     __ cvttss2siq(rax, xmm0);
1826     // NaN or overflow/underflow?
1827     __ cmp64(rax, ExternalAddress((address) &is_nan));
1828     __ jcc(Assembler::notEqual, L);
1829     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1830     __ bind(L);
1831   }
1832     break;
1833   case Bytecodes::_f2d:
1834     __ cvtss2sd(xmm0, xmm0);
1835     break;
1836   case Bytecodes::_d2i:
1837   {
1838     Label L;
1839     __ cvttsd2sil(rax, xmm0);
1840     __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1841     __ jcc(Assembler::notEqual, L);
1842     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1843     __ bind(L);
1844   }
1845     break;
1846   case Bytecodes::_d2l:
1847   {
1848     Label L;
1849     __ cvttsd2siq(rax, xmm0);
1850     // NaN or overflow/underflow?
1851     __ cmp64(rax, ExternalAddress((address) &is_nan));
1852     __ jcc(Assembler::notEqual, L);
1853     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1854     __ bind(L);
1855   }
1856     break;
1857   case Bytecodes::_d2f:
1858     __ cvtsd2ss(xmm0, xmm0);
1859     break;
1860   default:
1861     ShouldNotReachHere();
1862   }
1863 #else
1864   // Checking
1865 #ifdef ASSERT
1866   { TosState tos_in  = ilgl;
1867     TosState tos_out = ilgl;
1868     switch (bytecode()) {
1869       case Bytecodes::_i2l: // fall through
1870       case Bytecodes::_i2f: // fall through
1871       case Bytecodes::_i2d: // fall through
1872       case Bytecodes::_i2b: // fall through
1873       case Bytecodes::_i2c: // fall through
1874       case Bytecodes::_i2s: tos_in = itos; break;
1875       case Bytecodes::_l2i: // fall through
1876       case Bytecodes::_l2f: // fall through
1877       case Bytecodes::_l2d: tos_in = ltos; break;
1878       case Bytecodes::_f2i: // fall through
1879       case Bytecodes::_f2l: // fall through
1880       case Bytecodes::_f2d: tos_in = ftos; break;
1881       case Bytecodes::_d2i: // fall through
1882       case Bytecodes::_d2l: // fall through
1883       case Bytecodes::_d2f: tos_in = dtos; break;
1884       default             : ShouldNotReachHere();
1885     }
1886     switch (bytecode()) {
1887       case Bytecodes::_l2i: // fall through
1888       case Bytecodes::_f2i: // fall through
1889       case Bytecodes::_d2i: // fall through
1890       case Bytecodes::_i2b: // fall through
1891       case Bytecodes::_i2c: // fall through
1892       case Bytecodes::_i2s: tos_out = itos; break;
1893       case Bytecodes::_i2l: // fall through
1894       case Bytecodes::_f2l: // fall through
1895       case Bytecodes::_d2l: tos_out = ltos; break;
1896       case Bytecodes::_i2f: // fall through
1897       case Bytecodes::_l2f: // fall through
1898       case Bytecodes::_d2f: tos_out = ftos; break;
1899       case Bytecodes::_i2d: // fall through
1900       case Bytecodes::_l2d: // fall through
1901       case Bytecodes::_f2d: tos_out = dtos; break;
1902       default             : ShouldNotReachHere();
1903     }
1904     transition(tos_in, tos_out);
1905   }
1906 #endif // ASSERT
1907 
1908   // Conversion
1909   // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1910   switch (bytecode()) {
1911     case Bytecodes::_i2l:
1912       __ extend_sign(rdx, rax);
1913       break;
1914     case Bytecodes::_i2f:
1915       if (UseSSE >= 1) {
1916         __ cvtsi2ssl(xmm0, rax);
1917       } else {
1918         __ push(rax);          // store int on tos
1919         __ fild_s(at_rsp());   // load int to ST0
1920         __ f2ieee();           // truncate to float size
1921         __ pop(rcx);           // adjust rsp
1922       }
1923       break;
1924     case Bytecodes::_i2d:
1925       if (UseSSE >= 2) {
1926         __ cvtsi2sdl(xmm0, rax);
1927       } else {
1928       __ push(rax);          // add one slot for d2ieee()
1929       __ push(rax);          // store int on tos
1930       __ fild_s(at_rsp());   // load int to ST0
1931       __ d2ieee();           // truncate to double size
1932       __ pop(rcx);           // adjust rsp
1933       __ pop(rcx);
1934       }
1935       break;
1936     case Bytecodes::_i2b:
1937       __ shll(rax, 24);      // truncate upper 24 bits
1938       __ sarl(rax, 24);      // and sign-extend byte
1939       LP64_ONLY(__ movsbl(rax, rax));
1940       break;
1941     case Bytecodes::_i2c:
1942       __ andl(rax, 0xFFFF);  // truncate upper 16 bits
1943       LP64_ONLY(__ movzwl(rax, rax));
1944       break;
1945     case Bytecodes::_i2s:
1946       __ shll(rax, 16);      // truncate upper 16 bits
1947       __ sarl(rax, 16);      // and sign-extend short
1948       LP64_ONLY(__ movswl(rax, rax));
1949       break;
1950     case Bytecodes::_l2i:
1951       /* nothing to do */
1952       break;
1953     case Bytecodes::_l2f:
1954       // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1955       // 64-bit long values to floats. On 32-bit platforms it is not possible
1956       // to use that instruction with 64-bit operands, therefore the FPU is
1957       // used to perform the conversion.
1958       __ push(rdx);          // store long on tos
1959       __ push(rax);
1960       __ fild_d(at_rsp());   // load long to ST0
1961       __ f2ieee();           // truncate to float size
1962       __ pop(rcx);           // adjust rsp
1963       __ pop(rcx);
1964       if (UseSSE >= 1) {
1965         __ push_f();
1966         __ pop_f(xmm0);
1967       }
1968       break;
1969     case Bytecodes::_l2d:
1970       // On 32-bit platforms the FPU is used for conversion because on
1971       // 32-bit platforms it is not not possible to use the cvtsi2sdq
1972       // instruction with 64-bit operands.
1973       __ push(rdx);          // store long on tos
1974       __ push(rax);
1975       __ fild_d(at_rsp());   // load long to ST0
1976       __ d2ieee();           // truncate to double size
1977       __ pop(rcx);           // adjust rsp
1978       __ pop(rcx);
1979       if (UseSSE >= 2) {
1980         __ push_d();
1981         __ pop_d(xmm0);
1982       }
1983       break;
1984     case Bytecodes::_f2i:
1985       // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1986       // as it returns 0 for any NaN.
1987       if (UseSSE >= 1) {
1988         __ push_f(xmm0);
1989       } else {
1990         __ push(rcx);          // reserve space for argument
1991         __ fstp_s(at_rsp());   // pass float argument on stack
1992       }
1993       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1994       break;
1995     case Bytecodes::_f2l:
1996       // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1997       // as it returns 0 for any NaN.
1998       if (UseSSE >= 1) {
1999        __ push_f(xmm0);
2000       } else {
2001         __ push(rcx);          // reserve space for argument
2002         __ fstp_s(at_rsp());   // pass float argument on stack
2003       }
2004       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2005       break;
2006     case Bytecodes::_f2d:
2007       if (UseSSE < 1) {
2008         /* nothing to do */
2009       } else if (UseSSE == 1) {
2010         __ push_f(xmm0);
2011         __ pop_f();
2012       } else { // UseSSE >= 2
2013         __ cvtss2sd(xmm0, xmm0);
2014       }
2015       break;
2016     case Bytecodes::_d2i:
2017       if (UseSSE >= 2) {
2018         __ push_d(xmm0);
2019       } else {
2020         __ push(rcx);          // reserve space for argument
2021         __ push(rcx);
2022         __ fstp_d(at_rsp());   // pass double argument on stack
2023       }
2024       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2025       break;
2026     case Bytecodes::_d2l:
2027       if (UseSSE >= 2) {
2028         __ push_d(xmm0);
2029       } else {
2030         __ push(rcx);          // reserve space for argument
2031         __ push(rcx);
2032         __ fstp_d(at_rsp());   // pass double argument on stack
2033       }
2034       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2035       break;
2036     case Bytecodes::_d2f:
2037       if (UseSSE <= 1) {
2038         __ push(rcx);          // reserve space for f2ieee()
2039         __ f2ieee();           // truncate to float size
2040         __ pop(rcx);           // adjust rsp
2041         if (UseSSE == 1) {
2042           // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2043           // the conversion is performed using the FPU in this case.
2044           __ push_f();
2045           __ pop_f(xmm0);
2046         }
2047       } else { // UseSSE >= 2
2048         __ cvtsd2ss(xmm0, xmm0);
2049       }
2050       break;
2051     default             :
2052       ShouldNotReachHere();
2053   }
2054 #endif
2055 }
2056 
2057 void TemplateTable::lcmp() {
2058   transition(ltos, itos);
2059 #ifdef _LP64
2060   Label done;
2061   __ pop_l(rdx);
2062   __ cmpq(rdx, rax);
2063   __ movl(rax, -1);
2064   __ jccb(Assembler::less, done);
2065   __ setb(Assembler::notEqual, rax);
2066   __ movzbl(rax, rax);
2067   __ bind(done);
2068 #else
2069 
2070   // y = rdx:rax
2071   __ pop_l(rbx, rcx);             // get x = rcx:rbx
2072   __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2073   __ mov(rax, rcx);
2074 #endif
2075 }
2076 
2077 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2078   if ((is_float && UseSSE >= 1) ||
2079       (!is_float && UseSSE >= 2)) {
2080     Label done;
2081     if (is_float) {
2082       // XXX get rid of pop here, use ... reg, mem32
2083       __ pop_f(xmm1);
2084       __ ucomiss(xmm1, xmm0);
2085     } else {
2086       // XXX get rid of pop here, use ... reg, mem64
2087       __ pop_d(xmm1);
2088       __ ucomisd(xmm1, xmm0);
2089     }
2090     if (unordered_result < 0) {
2091       __ movl(rax, -1);
2092       __ jccb(Assembler::parity, done);
2093       __ jccb(Assembler::below, done);
2094       __ setb(Assembler::notEqual, rdx);
2095       __ movzbl(rax, rdx);
2096     } else {
2097       __ movl(rax, 1);
2098       __ jccb(Assembler::parity, done);
2099       __ jccb(Assembler::above, done);
2100       __ movl(rax, 0);
2101       __ jccb(Assembler::equal, done);
2102       __ decrementl(rax);
2103     }
2104     __ bind(done);
2105   } else {
2106 #ifdef _LP64
2107     ShouldNotReachHere();
2108 #else
2109     if (is_float) {
2110       __ fld_s(at_rsp());
2111     } else {
2112       __ fld_d(at_rsp());
2113       __ pop(rdx);
2114     }
2115     __ pop(rcx);
2116     __ fcmp2int(rax, unordered_result < 0);
2117 #endif // _LP64
2118   }
2119 }
2120 
2121 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2122   __ get_method(rcx); // rcx holds method
2123   __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2124                                      // holds bumped taken count
2125 
2126   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2127                              InvocationCounter::counter_offset();
2128   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2129                               InvocationCounter::counter_offset();
2130 
2131   // Load up edx with the branch displacement
2132   if (is_wide) {
2133     __ movl(rdx, at_bcp(1));
2134   } else {
2135     __ load_signed_short(rdx, at_bcp(1));
2136   }
2137   __ bswapl(rdx);
2138 
2139   if (!is_wide) {
2140     __ sarl(rdx, 16);
2141   }
2142   LP64_ONLY(__ movl2ptr(rdx, rdx));
2143 
2144   // Handle all the JSR stuff here, then exit.
2145   // It's much shorter and cleaner than intermingling with the non-JSR
2146   // normal-branch stuff occurring below.
2147   if (is_jsr) {
2148     // Pre-load the next target bytecode into rbx
2149     __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2150 
2151     // compute return address as bci in rax
2152     __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2153                         in_bytes(ConstMethod::codes_offset())));
2154     __ subptr(rax, Address(rcx, Method::const_offset()));
2155     // Adjust the bcp in r13 by the displacement in rdx
2156     __ addptr(rbcp, rdx);
2157     // jsr returns atos that is not an oop
2158     __ push_i(rax);
2159     __ dispatch_only(vtos, true);
2160     return;
2161   }
2162 
2163   // Normal (non-jsr) branch handling
2164 
2165   // Adjust the bcp in r13 by the displacement in rdx
2166   __ addptr(rbcp, rdx);
2167 
2168   assert(UseLoopCounter || !UseOnStackReplacement,
2169          "on-stack-replacement requires loop counters");
2170   Label backedge_counter_overflow;
2171   Label dispatch;
2172   if (UseLoopCounter) {
2173     // increment backedge counter for backward branches
2174     // rax: MDO
2175     // rbx: MDO bumped taken-count
2176     // rcx: method
2177     // rdx: target offset
2178     // r13: target bcp
2179     // r14: locals pointer
2180     __ testl(rdx, rdx);             // check if forward or backward branch
2181     __ jcc(Assembler::positive, dispatch); // count only if backward branch
2182 
2183     // check if MethodCounters exists
2184     Label has_counters;
2185     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2186     __ testptr(rax, rax);
2187     __ jcc(Assembler::notZero, has_counters);
2188     __ push(rdx);
2189     __ push(rcx);
2190     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2191                rcx);
2192     __ pop(rcx);
2193     __ pop(rdx);
2194     __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2195     __ testptr(rax, rax);
2196     __ jcc(Assembler::zero, dispatch);
2197     __ bind(has_counters);
2198 
2199     Label no_mdo;
2200     if (ProfileInterpreter) {
2201       // Are we profiling?
2202       __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2203       __ testptr(rbx, rbx);
2204       __ jccb(Assembler::zero, no_mdo);
2205       // Increment the MDO backedge counter
2206       const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2207           in_bytes(InvocationCounter::counter_offset()));
2208       const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2209       __ increment_mask_and_jump(mdo_backedge_counter, mask, rax,
2210           UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2211       __ jmp(dispatch);
2212     }
2213     __ bind(no_mdo);
2214     // Increment backedge counter in MethodCounters*
2215     __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2216     const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2217     __ increment_mask_and_jump(Address(rcx, be_offset), mask, rax,
2218         UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2219     __ bind(dispatch);
2220   }
2221 
2222   // Pre-load the next target bytecode into rbx
2223   __ load_unsigned_byte(rbx, Address(rbcp, 0));
2224 
2225   // continue with the bytecode @ target
2226   // rax: return bci for jsr's, unused otherwise
2227   // rbx: target bytecode
2228   // r13: target bcp
2229   __ dispatch_only(vtos, true);
2230 
2231   if (UseLoopCounter) {
2232     if (UseOnStackReplacement) {
2233       Label set_mdp;
2234       // invocation counter overflow
2235       __ bind(backedge_counter_overflow);
2236       __ negptr(rdx);
2237       __ addptr(rdx, rbcp); // branch bcp
2238       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2239       __ call_VM(noreg,
2240                  CAST_FROM_FN_PTR(address,
2241                                   InterpreterRuntime::frequency_counter_overflow),
2242                  rdx);
2243 
2244       // rax: osr nmethod (osr ok) or NULL (osr not possible)
2245       // rdx: scratch
2246       // r14: locals pointer
2247       // r13: bcp
2248       __ testptr(rax, rax);                        // test result
2249       __ jcc(Assembler::zero, dispatch);         // no osr if null
2250       // nmethod may have been invalidated (VM may block upon call_VM return)
2251       __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2252       __ jcc(Assembler::notEqual, dispatch);
2253 
2254       // We have the address of an on stack replacement routine in rax.
2255       // In preparation of invoking it, first we must migrate the locals
2256       // and monitors from off the interpreter frame on the stack.
2257       // Ensure to save the osr nmethod over the migration call,
2258       // it will be preserved in rbx.
2259       __ mov(rbx, rax);
2260 
2261       NOT_LP64(__ get_thread(rcx));
2262 
2263       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2264 
2265       // rax is OSR buffer, move it to expected parameter location
2266       LP64_ONLY(__ mov(j_rarg0, rax));
2267       NOT_LP64(__ mov(rcx, rax));
2268       // We use j_rarg definitions here so that registers don't conflict as parameter
2269       // registers change across platforms as we are in the midst of a calling
2270       // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2271 
2272       const Register retaddr   = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2273       const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2274 
2275       // pop the interpreter frame
2276       __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2277       __ leave();                                // remove frame anchor
2278       __ pop(retaddr);                           // get return address
2279       __ mov(rsp, sender_sp);                   // set sp to sender sp
2280       // Ensure compiled code always sees stack at proper alignment
2281       __ andptr(rsp, -(StackAlignmentInBytes));
2282 
2283       // unlike x86 we need no specialized return from compiled code
2284       // to the interpreter or the call stub.
2285 
2286       // push the return address
2287       __ push(retaddr);
2288 
2289       // and begin the OSR nmethod
2290       __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2291     }
2292   }
2293 }
2294 
2295 void TemplateTable::if_0cmp(Condition cc) {
2296   transition(itos, vtos);
2297   // assume branch is more often taken than not (loops use backward branches)
2298   Label not_taken;
2299   __ testl(rax, rax);
2300   __ jcc(j_not(cc), not_taken);
2301   branch(false, false);
2302   __ bind(not_taken);
2303   __ profile_not_taken_branch(rax);
2304 }
2305 
2306 void TemplateTable::if_icmp(Condition cc) {
2307   transition(itos, vtos);
2308   // assume branch is more often taken than not (loops use backward branches)
2309   Label not_taken;
2310   __ pop_i(rdx);
2311   __ cmpl(rdx, rax);
2312   __ jcc(j_not(cc), not_taken);
2313   branch(false, false);
2314   __ bind(not_taken);
2315   __ profile_not_taken_branch(rax);
2316 }
2317 
2318 void TemplateTable::if_nullcmp(Condition cc) {
2319   transition(atos, vtos);
2320   // assume branch is more often taken than not (loops use backward branches)
2321   Label not_taken;
2322   __ testptr(rax, rax);
2323   __ jcc(j_not(cc), not_taken);
2324   branch(false, false);
2325   __ bind(not_taken);
2326   __ profile_not_taken_branch(rax);
2327 }
2328 
2329 void TemplateTable::if_acmp(Condition cc) {
2330   transition(atos, vtos);
2331   // assume branch is more often taken than not (loops use backward branches)
2332   Label not_taken;
2333   __ pop_ptr(rdx);
2334   __ cmpoop(rdx, rax);
2335   __ jcc(j_not(cc), not_taken);
2336   branch(false, false);
2337   __ bind(not_taken);
2338   __ profile_not_taken_branch(rax);
2339 }
2340 
2341 void TemplateTable::ret() {
2342   transition(vtos, vtos);
2343   locals_index(rbx);
2344   LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2345   NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2346   __ profile_ret(rbx, rcx);
2347   __ get_method(rax);
2348   __ movptr(rbcp, Address(rax, Method::const_offset()));
2349   __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2350                       ConstMethod::codes_offset()));
2351   __ dispatch_next(vtos, 0, true);
2352 }
2353 
2354 void TemplateTable::wide_ret() {
2355   transition(vtos, vtos);
2356   locals_index_wide(rbx);
2357   __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2358   __ profile_ret(rbx, rcx);
2359   __ get_method(rax);
2360   __ movptr(rbcp, Address(rax, Method::const_offset()));
2361   __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2362   __ dispatch_next(vtos, 0, true);
2363 }
2364 
2365 void TemplateTable::tableswitch() {
2366   Label default_case, continue_execution;
2367   transition(itos, vtos);
2368 
2369   // align r13/rsi
2370   __ lea(rbx, at_bcp(BytesPerInt));
2371   __ andptr(rbx, -BytesPerInt);
2372   // load lo & hi
2373   __ movl(rcx, Address(rbx, BytesPerInt));
2374   __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2375   __ bswapl(rcx);
2376   __ bswapl(rdx);
2377   // check against lo & hi
2378   __ cmpl(rax, rcx);
2379   __ jcc(Assembler::less, default_case);
2380   __ cmpl(rax, rdx);
2381   __ jcc(Assembler::greater, default_case);
2382   // lookup dispatch offset
2383   __ subl(rax, rcx);
2384   __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2385   __ profile_switch_case(rax, rbx, rcx);
2386   // continue execution
2387   __ bind(continue_execution);
2388   __ bswapl(rdx);
2389   LP64_ONLY(__ movl2ptr(rdx, rdx));
2390   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2391   __ addptr(rbcp, rdx);
2392   __ dispatch_only(vtos, true);
2393   // handle default
2394   __ bind(default_case);
2395   __ profile_switch_default(rax);
2396   __ movl(rdx, Address(rbx, 0));
2397   __ jmp(continue_execution);
2398 }
2399 
2400 void TemplateTable::lookupswitch() {
2401   transition(itos, itos);
2402   __ stop("lookupswitch bytecode should have been rewritten");
2403 }
2404 
2405 void TemplateTable::fast_linearswitch() {
2406   transition(itos, vtos);
2407   Label loop_entry, loop, found, continue_execution;
2408   // bswap rax so we can avoid bswapping the table entries
2409   __ bswapl(rax);
2410   // align r13
2411   __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2412                                     // this instruction (change offsets
2413                                     // below)
2414   __ andptr(rbx, -BytesPerInt);
2415   // set counter
2416   __ movl(rcx, Address(rbx, BytesPerInt));
2417   __ bswapl(rcx);
2418   __ jmpb(loop_entry);
2419   // table search
2420   __ bind(loop);
2421   __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2422   __ jcc(Assembler::equal, found);
2423   __ bind(loop_entry);
2424   __ decrementl(rcx);
2425   __ jcc(Assembler::greaterEqual, loop);
2426   // default case
2427   __ profile_switch_default(rax);
2428   __ movl(rdx, Address(rbx, 0));
2429   __ jmp(continue_execution);
2430   // entry found -> get offset
2431   __ bind(found);
2432   __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2433   __ profile_switch_case(rcx, rax, rbx);
2434   // continue execution
2435   __ bind(continue_execution);
2436   __ bswapl(rdx);
2437   __ movl2ptr(rdx, rdx);
2438   __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2439   __ addptr(rbcp, rdx);
2440   __ dispatch_only(vtos, true);
2441 }
2442 
2443 void TemplateTable::fast_binaryswitch() {
2444   transition(itos, vtos);
2445   // Implementation using the following core algorithm:
2446   //
2447   // int binary_search(int key, LookupswitchPair* array, int n) {
2448   //   // Binary search according to "Methodik des Programmierens" by
2449   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2450   //   int i = 0;
2451   //   int j = n;
2452   //   while (i+1 < j) {
2453   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2454   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2455   //     // where a stands for the array and assuming that the (inexisting)
2456   //     // element a[n] is infinitely big.
2457   //     int h = (i + j) >> 1;
2458   //     // i < h < j
2459   //     if (key < array[h].fast_match()) {
2460   //       j = h;
2461   //     } else {
2462   //       i = h;
2463   //     }
2464   //   }
2465   //   // R: a[i] <= key < a[i+1] or Q
2466   //   // (i.e., if key is within array, i is the correct index)
2467   //   return i;
2468   // }
2469 
2470   // Register allocation
2471   const Register key   = rax; // already set (tosca)
2472   const Register array = rbx;
2473   const Register i     = rcx;
2474   const Register j     = rdx;
2475   const Register h     = rdi;
2476   const Register temp  = rsi;
2477 
2478   // Find array start
2479   NOT_LP64(__ save_bcp());
2480 
2481   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2482                                           // get rid of this
2483                                           // instruction (change
2484                                           // offsets below)
2485   __ andptr(array, -BytesPerInt);
2486 
2487   // Initialize i & j
2488   __ xorl(i, i);                            // i = 0;
2489   __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2490 
2491   // Convert j into native byteordering
2492   __ bswapl(j);
2493 
2494   // And start
2495   Label entry;
2496   __ jmp(entry);
2497 
2498   // binary search loop
2499   {
2500     Label loop;
2501     __ bind(loop);
2502     // int h = (i + j) >> 1;
2503     __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2504     __ sarl(h, 1);                               // h = (i + j) >> 1;
2505     // if (key < array[h].fast_match()) {
2506     //   j = h;
2507     // } else {
2508     //   i = h;
2509     // }
2510     // Convert array[h].match to native byte-ordering before compare
2511     __ movl(temp, Address(array, h, Address::times_8));
2512     __ bswapl(temp);
2513     __ cmpl(key, temp);
2514     // j = h if (key <  array[h].fast_match())
2515     __ cmov32(Assembler::less, j, h);
2516     // i = h if (key >= array[h].fast_match())
2517     __ cmov32(Assembler::greaterEqual, i, h);
2518     // while (i+1 < j)
2519     __ bind(entry);
2520     __ leal(h, Address(i, 1)); // i+1
2521     __ cmpl(h, j);             // i+1 < j
2522     __ jcc(Assembler::less, loop);
2523   }
2524 
2525   // end of binary search, result index is i (must check again!)
2526   Label default_case;
2527   // Convert array[i].match to native byte-ordering before compare
2528   __ movl(temp, Address(array, i, Address::times_8));
2529   __ bswapl(temp);
2530   __ cmpl(key, temp);
2531   __ jcc(Assembler::notEqual, default_case);
2532 
2533   // entry found -> j = offset
2534   __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2535   __ profile_switch_case(i, key, array);
2536   __ bswapl(j);
2537   LP64_ONLY(__ movslq(j, j));
2538 
2539   NOT_LP64(__ restore_bcp());
2540   NOT_LP64(__ restore_locals());                           // restore rdi
2541 
2542   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2543   __ addptr(rbcp, j);
2544   __ dispatch_only(vtos, true);
2545 
2546   // default case -> j = default offset
2547   __ bind(default_case);
2548   __ profile_switch_default(i);
2549   __ movl(j, Address(array, -2 * BytesPerInt));
2550   __ bswapl(j);
2551   LP64_ONLY(__ movslq(j, j));
2552 
2553   NOT_LP64(__ restore_bcp());
2554   NOT_LP64(__ restore_locals());
2555 
2556   __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2557   __ addptr(rbcp, j);
2558   __ dispatch_only(vtos, true);
2559 }
2560 
2561 void TemplateTable::_return(TosState state) {
2562   transition(state, state);
2563 
2564   assert(_desc->calls_vm(),
2565          "inconsistent calls_vm information"); // call in remove_activation
2566 
2567   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2568     assert(state == vtos, "only valid state");
2569     Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2570     __ movptr(robj, aaddress(0));
2571     Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
2572     __ load_klass(rdi, robj, tmp_load_klass);
2573     __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2574     __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2575     Label skip_register_finalizer;
2576     __ jcc(Assembler::zero, skip_register_finalizer);
2577 
2578     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2579 
2580     __ bind(skip_register_finalizer);
2581   }
2582 
2583   if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2584     Label no_safepoint;
2585     NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2586 #ifdef _LP64
2587     __ testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2588 #else
2589     const Register thread = rdi;
2590     __ get_thread(thread);
2591     __ testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2592 #endif
2593     __ jcc(Assembler::zero, no_safepoint);
2594     __ push(state);
2595     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2596                                        InterpreterRuntime::at_safepoint));
2597     __ pop(state);
2598     __ bind(no_safepoint);
2599   }
2600 
2601   // Narrow result if state is itos but result type is smaller.
2602   // Need to narrow in the return bytecode rather than in generate_return_entry
2603   // since compiled code callers expect the result to already be narrowed.
2604   if (state == itos) {
2605     __ narrow(rax);
2606   }
2607   __ remove_activation(state, rbcp);
2608 
2609   __ jmp(rbcp);
2610 }
2611 
2612 // ----------------------------------------------------------------------------
2613 // Volatile variables demand their effects be made known to all CPU's
2614 // in order.  Store buffers on most chips allow reads & writes to
2615 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2616 // without some kind of memory barrier (i.e., it's not sufficient that
2617 // the interpreter does not reorder volatile references, the hardware
2618 // also must not reorder them).
2619 //
2620 // According to the new Java Memory Model (JMM):
2621 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2622 //     writes act as acquire & release, so:
2623 // (2) A read cannot let unrelated NON-volatile memory refs that
2624 //     happen after the read float up to before the read.  It's OK for
2625 //     non-volatile memory refs that happen before the volatile read to
2626 //     float down below it.
2627 // (3) Similar a volatile write cannot let unrelated NON-volatile
2628 //     memory refs that happen BEFORE the write float down to after the
2629 //     write.  It's OK for non-volatile memory refs that happen after the
2630 //     volatile write to float up before it.
2631 //
2632 // We only put in barriers around volatile refs (they are expensive),
2633 // not _between_ memory refs (that would require us to track the
2634 // flavor of the previous memory refs).  Requirements (2) and (3)
2635 // require some barriers before volatile stores and after volatile
2636 // loads.  These nearly cover requirement (1) but miss the
2637 // volatile-store-volatile-load case.  This final case is placed after
2638 // volatile-stores although it could just as well go before
2639 // volatile-loads.
2640 
2641 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2642   // Helper function to insert a is-volatile test and memory barrier
2643   __ membar(order_constraint);
2644 }
2645 
2646 void TemplateTable::resolve_cache_and_index(int byte_no,
2647                                             Register cache,
2648                                             Register index,
2649                                             size_t index_size) {
2650   const Register temp = rbx;
2651   assert_different_registers(cache, index, temp);
2652 
2653   Label L_clinit_barrier_slow;
2654   Label resolved;
2655 
2656   Bytecodes::Code code = bytecode();
2657   switch (code) {
2658   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2659   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2660   default: break;
2661   }
2662 
2663   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2664   __ get_cache_and_index_and_bytecode_at_bcp(cache, index, temp, byte_no, 1, index_size);
2665   __ cmpl(temp, code);  // have we resolved this bytecode?
2666   __ jcc(Assembler::equal, resolved);
2667 
2668   // resolve first time through
2669   // Class initialization barrier slow path lands here as well.
2670   __ bind(L_clinit_barrier_slow);
2671   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2672   __ movl(temp, code);
2673   __ call_VM(noreg, entry, temp);
2674   // Update registers with resolved info
2675   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
2676 
2677   __ bind(resolved);
2678 
2679   // Class initialization barrier for static methods
2680   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2681     const Register method = temp;
2682     const Register klass  = temp;
2683     const Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2684     assert(thread != noreg, "x86_32 not supported");
2685 
2686     __ load_resolved_method_at_index(byte_no, method, cache, index);
2687     __ load_method_holder(klass, method);
2688     __ clinit_barrier(klass, thread, NULL /*L_fast_path*/, &L_clinit_barrier_slow);
2689   }
2690 }
2691 
2692 // The cache and index registers must be set before call
2693 void TemplateTable::load_field_cp_cache_entry(Register obj,
2694                                               Register cache,
2695                                               Register index,
2696                                               Register off,
2697                                               Register flags,
2698                                               bool is_static = false) {
2699   assert_different_registers(cache, index, flags, off);
2700 
2701   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2702   // Field offset
2703   __ movptr(off, Address(cache, index, Address::times_ptr,
2704                          in_bytes(cp_base_offset +
2705                                   ConstantPoolCacheEntry::f2_offset())));
2706   // Flags
2707   __ movl(flags, Address(cache, index, Address::times_ptr,
2708                          in_bytes(cp_base_offset +
2709                                   ConstantPoolCacheEntry::flags_offset())));
2710 
2711   // klass overwrite register
2712   if (is_static) {
2713     __ movptr(obj, Address(cache, index, Address::times_ptr,
2714                            in_bytes(cp_base_offset +
2715                                     ConstantPoolCacheEntry::f1_offset())));
2716     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2717     __ movptr(obj, Address(obj, mirror_offset));
2718     __ resolve_oop_handle(obj);
2719   }
2720 }
2721 
2722 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2723                                                Register method,
2724                                                Register itable_index,
2725                                                Register flags,
2726                                                bool is_invokevirtual,
2727                                                bool is_invokevfinal, /*unused*/
2728                                                bool is_invokedynamic) {
2729   // setup registers
2730   const Register cache = rcx;
2731   const Register index = rdx;
2732   assert_different_registers(method, flags);
2733   assert_different_registers(method, cache, index);
2734   assert_different_registers(itable_index, flags);
2735   assert_different_registers(itable_index, cache, index);
2736   // determine constant pool cache field offsets
2737   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2738   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2739                                     ConstantPoolCacheEntry::flags_offset());
2740   // access constant pool cache fields
2741   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2742                                     ConstantPoolCacheEntry::f2_offset());
2743 
2744   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2745   resolve_cache_and_index(byte_no, cache, index, index_size);
2746   __ load_resolved_method_at_index(byte_no, method, cache, index);
2747 
2748   if (itable_index != noreg) {
2749     // pick up itable or appendix index from f2 also:
2750     __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2751   }
2752   __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2753 }
2754 
2755 // The registers cache and index expected to be set before call.
2756 // Correct values of the cache and index registers are preserved.
2757 void TemplateTable::jvmti_post_field_access(Register cache,
2758                                             Register index,
2759                                             bool is_static,
2760                                             bool has_tos) {
2761   if (JvmtiExport::can_post_field_access()) {
2762     // Check to see if a field access watch has been set before we take
2763     // the time to call into the VM.
2764     Label L1;
2765     assert_different_registers(cache, index, rax);
2766     __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2767     __ testl(rax,rax);
2768     __ jcc(Assembler::zero, L1);
2769 
2770     // cache entry pointer
2771     __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2772     __ shll(index, LogBytesPerWord);
2773     __ addptr(cache, index);
2774     if (is_static) {
2775       __ xorptr(rax, rax);      // NULL object reference
2776     } else {
2777       __ pop(atos);         // Get the object
2778       __ verify_oop(rax);
2779       __ push(atos);        // Restore stack state
2780     }
2781     // rax,:   object pointer or NULL
2782     // cache: cache entry pointer
2783     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2784                rax, cache);
2785     __ get_cache_and_index_at_bcp(cache, index, 1);
2786     __ bind(L1);
2787   }
2788 }
2789 
2790 void TemplateTable::pop_and_check_object(Register r) {
2791   __ pop_ptr(r);
2792   __ null_check(r);  // for field access must check obj.
2793   __ verify_oop(r);
2794 }
2795 
2796 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2797   transition(vtos, vtos);
2798 
2799   const Register cache = rcx;
2800   const Register index = rdx;
2801   const Register obj   = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2802   const Register off   = rbx;
2803   const Register flags = rax;
2804   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2805 
2806   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2807   jvmti_post_field_access(cache, index, is_static, false);
2808   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2809 
2810   if (!is_static) pop_and_check_object(obj);
2811 
2812   const Address field(obj, off, Address::times_1, 0*wordSize);
2813 
2814   Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2815 
2816   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2817   // Make sure we don't need to mask edx after the above shift
2818   assert(btos == 0, "change code, btos != 0");
2819 
2820   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2821 
2822   __ jcc(Assembler::notZero, notByte);
2823   // btos
2824   __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2825   __ push(btos);
2826   // Rewrite bytecode to be faster
2827   if (!is_static && rc == may_rewrite) {
2828     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2829   }
2830   __ jmp(Done);
2831 
2832   __ bind(notByte);
2833   __ cmpl(flags, ztos);
2834   __ jcc(Assembler::notEqual, notBool);
2835 
2836   // ztos (same code as btos)
2837   __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2838   __ push(ztos);
2839   // Rewrite bytecode to be faster
2840   if (!is_static && rc == may_rewrite) {
2841     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2842     patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2843   }
2844   __ jmp(Done);
2845 
2846   __ bind(notBool);
2847   __ cmpl(flags, atos);
2848   __ jcc(Assembler::notEqual, notObj);
2849   // atos
2850   do_oop_load(_masm, field, rax);
2851   __ push(atos);
2852   if (!is_static && rc == may_rewrite) {
2853     patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2854   }
2855   __ jmp(Done);
2856 
2857   __ bind(notObj);
2858   __ cmpl(flags, itos);
2859   __ jcc(Assembler::notEqual, notInt);
2860   // itos
2861   __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2862   __ push(itos);
2863   // Rewrite bytecode to be faster
2864   if (!is_static && rc == may_rewrite) {
2865     patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2866   }
2867   __ jmp(Done);
2868 
2869   __ bind(notInt);
2870   __ cmpl(flags, ctos);
2871   __ jcc(Assembler::notEqual, notChar);
2872   // ctos
2873   __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2874   __ push(ctos);
2875   // Rewrite bytecode to be faster
2876   if (!is_static && rc == may_rewrite) {
2877     patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2878   }
2879   __ jmp(Done);
2880 
2881   __ bind(notChar);
2882   __ cmpl(flags, stos);
2883   __ jcc(Assembler::notEqual, notShort);
2884   // stos
2885   __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2886   __ push(stos);
2887   // Rewrite bytecode to be faster
2888   if (!is_static && rc == may_rewrite) {
2889     patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2890   }
2891   __ jmp(Done);
2892 
2893   __ bind(notShort);
2894   __ cmpl(flags, ltos);
2895   __ jcc(Assembler::notEqual, notLong);
2896   // ltos
2897     // Generate code as if volatile (x86_32).  There just aren't enough registers to
2898     // save that information and this code is faster than the test.
2899   __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2900   __ push(ltos);
2901   // Rewrite bytecode to be faster
2902   LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2903   __ jmp(Done);
2904 
2905   __ bind(notLong);
2906   __ cmpl(flags, ftos);
2907   __ jcc(Assembler::notEqual, notFloat);
2908   // ftos
2909 
2910   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2911   __ push(ftos);
2912   // Rewrite bytecode to be faster
2913   if (!is_static && rc == may_rewrite) {
2914     patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2915   }
2916   __ jmp(Done);
2917 
2918   __ bind(notFloat);
2919 #ifdef ASSERT
2920   Label notDouble;
2921   __ cmpl(flags, dtos);
2922   __ jcc(Assembler::notEqual, notDouble);
2923 #endif
2924   // dtos
2925   // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
2926   __ access_load_at(T_DOUBLE, IN_HEAP | MO_RELAXED, noreg /* dtos */, field, noreg, noreg);
2927   __ push(dtos);
2928   // Rewrite bytecode to be faster
2929   if (!is_static && rc == may_rewrite) {
2930     patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2931   }
2932 #ifdef ASSERT
2933   __ jmp(Done);
2934 
2935   __ bind(notDouble);
2936   __ stop("Bad state");
2937 #endif
2938 
2939   __ bind(Done);
2940   // [jk] not needed currently
2941   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2942   //                                              Assembler::LoadStore));
2943 }
2944 
2945 void TemplateTable::getfield(int byte_no) {
2946   getfield_or_static(byte_no, false);
2947 }
2948 
2949 void TemplateTable::nofast_getfield(int byte_no) {
2950   getfield_or_static(byte_no, false, may_not_rewrite);
2951 }
2952 
2953 void TemplateTable::getstatic(int byte_no) {
2954   getfield_or_static(byte_no, true);
2955 }
2956 
2957 
2958 // The registers cache and index expected to be set before call.
2959 // The function may destroy various registers, just not the cache and index registers.
2960 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2961 
2962   const Register robj = LP64_ONLY(c_rarg2)   NOT_LP64(rax);
2963   const Register RBX  = LP64_ONLY(c_rarg1)   NOT_LP64(rbx);
2964   const Register RCX  = LP64_ONLY(c_rarg3)   NOT_LP64(rcx);
2965   const Register RDX  = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2966 
2967   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2968 
2969   if (JvmtiExport::can_post_field_modification()) {
2970     // Check to see if a field modification watch has been set before
2971     // we take the time to call into the VM.
2972     Label L1;
2973     assert_different_registers(cache, index, rax);
2974     __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2975     __ testl(rax, rax);
2976     __ jcc(Assembler::zero, L1);
2977 
2978     __ get_cache_and_index_at_bcp(robj, RDX, 1);
2979 
2980 
2981     if (is_static) {
2982       // Life is simple.  Null out the object pointer.
2983       __ xorl(RBX, RBX);
2984 
2985     } else {
2986       // Life is harder. The stack holds the value on top, followed by
2987       // the object.  We don't know the size of the value, though; it
2988       // could be one or two words depending on its type. As a result,
2989       // we must find the type to determine where the object is.
2990 #ifndef _LP64
2991       Label two_word, valsize_known;
2992 #endif
2993       __ movl(RCX, Address(robj, RDX,
2994                            Address::times_ptr,
2995                            in_bytes(cp_base_offset +
2996                                      ConstantPoolCacheEntry::flags_offset())));
2997       NOT_LP64(__ mov(rbx, rsp));
2998       __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2999 
3000       // Make sure we don't need to mask rcx after the above shift
3001       ConstantPoolCacheEntry::verify_tos_state_shift();
3002 #ifdef _LP64
3003       __ movptr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
3004       __ cmpl(c_rarg3, ltos);
3005       __ cmovptr(Assembler::equal,
3006                  c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3007       __ cmpl(c_rarg3, dtos);
3008       __ cmovptr(Assembler::equal,
3009                  c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3010 #else
3011       __ cmpl(rcx, ltos);
3012       __ jccb(Assembler::equal, two_word);
3013       __ cmpl(rcx, dtos);
3014       __ jccb(Assembler::equal, two_word);
3015       __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3016       __ jmpb(valsize_known);
3017 
3018       __ bind(two_word);
3019       __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3020 
3021       __ bind(valsize_known);
3022       // setup object pointer
3023       __ movptr(rbx, Address(rbx, 0));
3024 #endif
3025     }
3026     // cache entry pointer
3027     __ addptr(robj, in_bytes(cp_base_offset));
3028     __ shll(RDX, LogBytesPerWord);
3029     __ addptr(robj, RDX);
3030     // object (tos)
3031     __ mov(RCX, rsp);
3032     // c_rarg1: object pointer set up above (NULL if static)
3033     // c_rarg2: cache entry pointer
3034     // c_rarg3: jvalue object on the stack
3035     __ call_VM(noreg,
3036                CAST_FROM_FN_PTR(address,
3037                                 InterpreterRuntime::post_field_modification),
3038                RBX, robj, RCX);
3039     __ get_cache_and_index_at_bcp(cache, index, 1);
3040     __ bind(L1);
3041   }
3042 }
3043 
3044 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3045   transition(vtos, vtos);
3046 
3047   const Register cache = rcx;
3048   const Register index = rdx;
3049   const Register obj   = rcx;
3050   const Register off   = rbx;
3051   const Register flags = rax;
3052 
3053   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3054   jvmti_post_field_mod(cache, index, is_static);
3055   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3056 
3057   // [jk] not needed currently
3058   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3059   //                                              Assembler::StoreStore));
3060 
3061   Label notVolatile, Done;
3062   __ movl(rdx, flags);
3063   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3064   __ andl(rdx, 0x1);
3065 
3066   // Check for volatile store
3067   __ testl(rdx, rdx);
3068   __ jcc(Assembler::zero, notVolatile);
3069 
3070   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3071   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3072                                                Assembler::StoreStore));
3073   __ jmp(Done);
3074   __ bind(notVolatile);
3075 
3076   putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3077 
3078   __ bind(Done);
3079 }
3080 
3081 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3082                                               Register obj, Register off, Register flags) {
3083 
3084   // field addresses
3085   const Address field(obj, off, Address::times_1, 0*wordSize);
3086   NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3087 
3088   Label notByte, notBool, notInt, notShort, notChar,
3089         notLong, notFloat, notObj;
3090   Label Done;
3091 
3092   const Register bc    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3093 
3094   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3095 
3096   assert(btos == 0, "change code, btos != 0");
3097   __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3098   __ jcc(Assembler::notZero, notByte);
3099 
3100   // btos
3101   {
3102     __ pop(btos);
3103     if (!is_static) pop_and_check_object(obj);
3104     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3105     if (!is_static && rc == may_rewrite) {
3106       patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3107     }
3108     __ jmp(Done);
3109   }
3110 
3111   __ bind(notByte);
3112   __ cmpl(flags, ztos);
3113   __ jcc(Assembler::notEqual, notBool);
3114 
3115   // ztos
3116   {
3117     __ pop(ztos);
3118     if (!is_static) pop_and_check_object(obj);
3119     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3120     if (!is_static && rc == may_rewrite) {
3121       patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3122     }
3123     __ jmp(Done);
3124   }
3125 
3126   __ bind(notBool);
3127   __ cmpl(flags, atos);
3128   __ jcc(Assembler::notEqual, notObj);
3129 
3130   // atos
3131   {
3132     __ pop(atos);
3133     if (!is_static) pop_and_check_object(obj);
3134     // Store into the field
3135     do_oop_store(_masm, field, rax);
3136     if (!is_static && rc == may_rewrite) {
3137       patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3138     }
3139     __ jmp(Done);
3140   }
3141 
3142   __ bind(notObj);
3143   __ cmpl(flags, itos);
3144   __ jcc(Assembler::notEqual, notInt);
3145 
3146   // itos
3147   {
3148     __ pop(itos);
3149     if (!is_static) pop_and_check_object(obj);
3150     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3151     if (!is_static && rc == may_rewrite) {
3152       patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3153     }
3154     __ jmp(Done);
3155   }
3156 
3157   __ bind(notInt);
3158   __ cmpl(flags, ctos);
3159   __ jcc(Assembler::notEqual, notChar);
3160 
3161   // ctos
3162   {
3163     __ pop(ctos);
3164     if (!is_static) pop_and_check_object(obj);
3165     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3166     if (!is_static && rc == may_rewrite) {
3167       patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3168     }
3169     __ jmp(Done);
3170   }
3171 
3172   __ bind(notChar);
3173   __ cmpl(flags, stos);
3174   __ jcc(Assembler::notEqual, notShort);
3175 
3176   // stos
3177   {
3178     __ pop(stos);
3179     if (!is_static) pop_and_check_object(obj);
3180     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3181     if (!is_static && rc == may_rewrite) {
3182       patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3183     }
3184     __ jmp(Done);
3185   }
3186 
3187   __ bind(notShort);
3188   __ cmpl(flags, ltos);
3189   __ jcc(Assembler::notEqual, notLong);
3190 
3191   // ltos
3192   {
3193     __ pop(ltos);
3194     if (!is_static) pop_and_check_object(obj);
3195     // MO_RELAXED: generate atomic store for the case of volatile field (important for x86_32)
3196     __ access_store_at(T_LONG, IN_HEAP | MO_RELAXED, field, noreg /* ltos*/, noreg, noreg, noreg);
3197 #ifdef _LP64
3198     if (!is_static && rc == may_rewrite) {
3199       patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3200     }
3201 #endif // _LP64
3202     __ jmp(Done);
3203   }
3204 
3205   __ bind(notLong);
3206   __ cmpl(flags, ftos);
3207   __ jcc(Assembler::notEqual, notFloat);
3208 
3209   // ftos
3210   {
3211     __ pop(ftos);
3212     if (!is_static) pop_and_check_object(obj);
3213     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3214     if (!is_static && rc == may_rewrite) {
3215       patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3216     }
3217     __ jmp(Done);
3218   }
3219 
3220   __ bind(notFloat);
3221 #ifdef ASSERT
3222   Label notDouble;
3223   __ cmpl(flags, dtos);
3224   __ jcc(Assembler::notEqual, notDouble);
3225 #endif
3226 
3227   // dtos
3228   {
3229     __ pop(dtos);
3230     if (!is_static) pop_and_check_object(obj);
3231     // MO_RELAXED: for the case of volatile field, in fact it adds no extra work for the underlying implementation
3232     __ access_store_at(T_DOUBLE, IN_HEAP | MO_RELAXED, field, noreg /* dtos */, noreg, noreg, noreg);
3233     if (!is_static && rc == may_rewrite) {
3234       patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3235     }
3236   }
3237 
3238 #ifdef ASSERT
3239   __ jmp(Done);
3240 
3241   __ bind(notDouble);
3242   __ stop("Bad state");
3243 #endif
3244 
3245   __ bind(Done);
3246 }
3247 
3248 void TemplateTable::putfield(int byte_no) {
3249   putfield_or_static(byte_no, false);
3250 }
3251 
3252 void TemplateTable::nofast_putfield(int byte_no) {
3253   putfield_or_static(byte_no, false, may_not_rewrite);
3254 }
3255 
3256 void TemplateTable::putstatic(int byte_no) {
3257   putfield_or_static(byte_no, true);
3258 }
3259 
3260 void TemplateTable::jvmti_post_fast_field_mod() {
3261 
3262   const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3263 
3264   if (JvmtiExport::can_post_field_modification()) {
3265     // Check to see if a field modification watch has been set before
3266     // we take the time to call into the VM.
3267     Label L2;
3268     __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3269     __ testl(scratch, scratch);
3270     __ jcc(Assembler::zero, L2);
3271     __ pop_ptr(rbx);                  // copy the object pointer from tos
3272     __ verify_oop(rbx);
3273     __ push_ptr(rbx);                 // put the object pointer back on tos
3274     // Save tos values before call_VM() clobbers them. Since we have
3275     // to do it for every data type, we use the saved values as the
3276     // jvalue object.
3277     switch (bytecode()) {          // load values into the jvalue object
3278     case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3279     case Bytecodes::_fast_bputfield: // fall through
3280     case Bytecodes::_fast_zputfield: // fall through
3281     case Bytecodes::_fast_sputfield: // fall through
3282     case Bytecodes::_fast_cputfield: // fall through
3283     case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3284     case Bytecodes::_fast_dputfield: __ push(dtos); break;
3285     case Bytecodes::_fast_fputfield: __ push(ftos); break;
3286     case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3287 
3288     default:
3289       ShouldNotReachHere();
3290     }
3291     __ mov(scratch, rsp);             // points to jvalue on the stack
3292     // access constant pool cache entry
3293     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3294     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3295     __ verify_oop(rbx);
3296     // rbx: object pointer copied above
3297     // c_rarg2: cache entry pointer
3298     // c_rarg3: jvalue object on the stack
3299     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3300     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3301 
3302     switch (bytecode()) {             // restore tos values
3303     case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3304     case Bytecodes::_fast_bputfield: // fall through
3305     case Bytecodes::_fast_zputfield: // fall through
3306     case Bytecodes::_fast_sputfield: // fall through
3307     case Bytecodes::_fast_cputfield: // fall through
3308     case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3309     case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3310     case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3311     case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3312     default: break;
3313     }
3314     __ bind(L2);
3315   }
3316 }
3317 
3318 void TemplateTable::fast_storefield(TosState state) {
3319   transition(state, vtos);
3320 
3321   ByteSize base = ConstantPoolCache::base_offset();
3322 
3323   jvmti_post_fast_field_mod();
3324 
3325   // access constant pool cache
3326   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3327 
3328   // test for volatile with rdx but rdx is tos register for lputfield.
3329   __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3330                        in_bytes(base +
3331                                 ConstantPoolCacheEntry::flags_offset())));
3332 
3333   // replace index with field offset from cache entry
3334   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3335                          in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3336 
3337   // [jk] not needed currently
3338   // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3339   //                                              Assembler::StoreStore));
3340 
3341   Label notVolatile, Done;
3342   __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3343   __ andl(rdx, 0x1);
3344 
3345   // Get object from stack
3346   pop_and_check_object(rcx);
3347 
3348   // field address
3349   const Address field(rcx, rbx, Address::times_1);
3350 
3351   // Check for volatile store
3352   __ testl(rdx, rdx);
3353   __ jcc(Assembler::zero, notVolatile);
3354 
3355   fast_storefield_helper(field, rax);
3356   volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3357                                                Assembler::StoreStore));
3358   __ jmp(Done);
3359   __ bind(notVolatile);
3360 
3361   fast_storefield_helper(field, rax);
3362 
3363   __ bind(Done);
3364 }
3365 
3366 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3367 
3368   // access field
3369   switch (bytecode()) {
3370   case Bytecodes::_fast_aputfield:
3371     do_oop_store(_masm, field, rax);
3372     break;
3373   case Bytecodes::_fast_lputfield:
3374 #ifdef _LP64
3375     __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg, noreg);
3376 #else
3377   __ stop("should not be rewritten");
3378 #endif
3379     break;
3380   case Bytecodes::_fast_iputfield:
3381     __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg, noreg);
3382     break;
3383   case Bytecodes::_fast_zputfield:
3384     __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg, noreg);
3385     break;
3386   case Bytecodes::_fast_bputfield:
3387     __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg, noreg);
3388     break;
3389   case Bytecodes::_fast_sputfield:
3390     __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg, noreg);
3391     break;
3392   case Bytecodes::_fast_cputfield:
3393     __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg, noreg);
3394     break;
3395   case Bytecodes::_fast_fputfield:
3396     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg, noreg);
3397     break;
3398   case Bytecodes::_fast_dputfield:
3399     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg, noreg);
3400     break;
3401   default:
3402     ShouldNotReachHere();
3403   }
3404 }
3405 
3406 void TemplateTable::fast_accessfield(TosState state) {
3407   transition(atos, state);
3408 
3409   // Do the JVMTI work here to avoid disturbing the register state below
3410   if (JvmtiExport::can_post_field_access()) {
3411     // Check to see if a field access watch has been set before we
3412     // take the time to call into the VM.
3413     Label L1;
3414     __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3415     __ testl(rcx, rcx);
3416     __ jcc(Assembler::zero, L1);
3417     // access constant pool cache entry
3418     LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3419     NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3420     __ verify_oop(rax);
3421     __ push_ptr(rax);  // save object pointer before call_VM() clobbers it
3422     LP64_ONLY(__ mov(c_rarg1, rax));
3423     // c_rarg1: object pointer copied above
3424     // c_rarg2: cache entry pointer
3425     LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3426     NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3427     __ pop_ptr(rax); // restore object pointer
3428     __ bind(L1);
3429   }
3430 
3431   // access constant pool cache
3432   __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3433   // replace index with field offset from cache entry
3434   // [jk] not needed currently
3435   // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3436   //                      in_bytes(ConstantPoolCache::base_offset() +
3437   //                               ConstantPoolCacheEntry::flags_offset())));
3438   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3439   // __ andl(rdx, 0x1);
3440   //
3441   __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3442                          in_bytes(ConstantPoolCache::base_offset() +
3443                                   ConstantPoolCacheEntry::f2_offset())));
3444 
3445   // rax: object
3446   __ verify_oop(rax);
3447   __ null_check(rax);
3448   Address field(rax, rbx, Address::times_1);
3449 
3450   // access field
3451   switch (bytecode()) {
3452   case Bytecodes::_fast_agetfield:
3453     do_oop_load(_masm, field, rax);
3454     __ verify_oop(rax);
3455     break;
3456   case Bytecodes::_fast_lgetfield:
3457 #ifdef _LP64
3458     __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3459 #else
3460   __ stop("should not be rewritten");
3461 #endif
3462     break;
3463   case Bytecodes::_fast_igetfield:
3464     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3465     break;
3466   case Bytecodes::_fast_bgetfield:
3467     __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3468     break;
3469   case Bytecodes::_fast_sgetfield:
3470     __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3471     break;
3472   case Bytecodes::_fast_cgetfield:
3473     __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3474     break;
3475   case Bytecodes::_fast_fgetfield:
3476     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3477     break;
3478   case Bytecodes::_fast_dgetfield:
3479     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3480     break;
3481   default:
3482     ShouldNotReachHere();
3483   }
3484   // [jk] not needed currently
3485   //   Label notVolatile;
3486   //   __ testl(rdx, rdx);
3487   //   __ jcc(Assembler::zero, notVolatile);
3488   //   __ membar(Assembler::LoadLoad);
3489   //   __ bind(notVolatile);
3490 }
3491 
3492 void TemplateTable::fast_xaccess(TosState state) {
3493   transition(vtos, state);
3494 
3495   // get receiver
3496   __ movptr(rax, aaddress(0));
3497   // access constant pool cache
3498   __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3499   __ movptr(rbx,
3500             Address(rcx, rdx, Address::times_ptr,
3501                     in_bytes(ConstantPoolCache::base_offset() +
3502                              ConstantPoolCacheEntry::f2_offset())));
3503   // make sure exception is reported in correct bcp range (getfield is
3504   // next instruction)
3505   __ increment(rbcp);
3506   __ null_check(rax);
3507   const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3508   switch (state) {
3509   case itos:
3510     __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3511     break;
3512   case atos:
3513     do_oop_load(_masm, field, rax);
3514     __ verify_oop(rax);
3515     break;
3516   case ftos:
3517     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3518     break;
3519   default:
3520     ShouldNotReachHere();
3521   }
3522 
3523   // [jk] not needed currently
3524   // Label notVolatile;
3525   // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3526   //                      in_bytes(ConstantPoolCache::base_offset() +
3527   //                               ConstantPoolCacheEntry::flags_offset())));
3528   // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3529   // __ testl(rdx, 0x1);
3530   // __ jcc(Assembler::zero, notVolatile);
3531   // __ membar(Assembler::LoadLoad);
3532   // __ bind(notVolatile);
3533 
3534   __ decrement(rbcp);
3535 }
3536 
3537 //-----------------------------------------------------------------------------
3538 // Calls
3539 
3540 void TemplateTable::prepare_invoke(int byte_no,
3541                                    Register method,  // linked method (or i-klass)
3542                                    Register index,   // itable index, MethodType, etc.
3543                                    Register recv,    // if caller wants to see it
3544                                    Register flags    // if caller wants to test it
3545                                    ) {
3546   // determine flags
3547   const Bytecodes::Code code = bytecode();
3548   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3549   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3550   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3551   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3552   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3553   const bool load_receiver       = (recv  != noreg);
3554   const bool save_flags          = (flags != noreg);
3555   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3556   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3557   assert(flags == noreg || flags == rdx, "");
3558   assert(recv  == noreg || recv  == rcx, "");
3559 
3560   // setup registers & access constant pool cache
3561   if (recv  == noreg)  recv  = rcx;
3562   if (flags == noreg)  flags = rdx;
3563   assert_different_registers(method, index, recv, flags);
3564 
3565   // save 'interpreter return address'
3566   __ save_bcp();
3567 
3568   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3569 
3570   // maybe push appendix to arguments (just before return address)
3571   if (is_invokedynamic || is_invokehandle) {
3572     Label L_no_push;
3573     __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3574     __ jcc(Assembler::zero, L_no_push);
3575     // Push the appendix as a trailing parameter.
3576     // This must be done before we get the receiver,
3577     // since the parameter_size includes it.
3578     __ push(rbx);
3579     __ mov(rbx, index);
3580     __ load_resolved_reference_at_index(index, rbx);
3581     __ pop(rbx);
3582     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3583     __ bind(L_no_push);
3584   }
3585 
3586   // load receiver if needed (after appendix is pushed so parameter size is correct)
3587   // Note: no return address pushed yet
3588   if (load_receiver) {
3589     __ movl(recv, flags);
3590     __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3591     const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3592     const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3593     Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3594     __ movptr(recv, recv_addr);
3595     __ verify_oop(recv);
3596   }
3597 
3598   if (save_flags) {
3599     __ movl(rbcp, flags);
3600   }
3601 
3602   // compute return type
3603   __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3604   // Make sure we don't need to mask flags after the above shift
3605   ConstantPoolCacheEntry::verify_tos_state_shift();
3606   // load return address
3607   {
3608     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3609     ExternalAddress table(table_addr);
3610     LP64_ONLY(__ lea(rscratch1, table));
3611     LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3612     NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3613   }
3614 
3615   // push return address
3616   __ push(flags);
3617 
3618   // Restore flags value from the constant pool cache, and restore rsi
3619   // for later null checks.  r13 is the bytecode pointer
3620   if (save_flags) {
3621     __ movl(flags, rbcp);
3622     __ restore_bcp();
3623   }
3624 }
3625 
3626 void TemplateTable::invokevirtual_helper(Register index,
3627                                          Register recv,
3628                                          Register flags) {
3629   // Uses temporary registers rax, rdx
3630   assert_different_registers(index, recv, rax, rdx);
3631   assert(index == rbx, "");
3632   assert(recv  == rcx, "");
3633 
3634   // Test for an invoke of a final method
3635   Label notFinal;
3636   __ movl(rax, flags);
3637   __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3638   __ jcc(Assembler::zero, notFinal);
3639 
3640   const Register method = index;  // method must be rbx
3641   assert(method == rbx,
3642          "Method* must be rbx for interpreter calling convention");
3643 
3644   // do the call - the index is actually the method to call
3645   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3646 
3647   // It's final, need a null check here!
3648   __ null_check(recv);
3649 
3650   // profile this call
3651   __ profile_final_call(rax);
3652   __ profile_arguments_type(rax, method, rbcp, true);
3653 
3654   __ jump_from_interpreted(method, rax);
3655 
3656   __ bind(notFinal);
3657 
3658   // get receiver klass
3659   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3660   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
3661   __ load_klass(rax, recv, tmp_load_klass);
3662 
3663   // profile this call
3664   __ profile_virtual_call(rax, rlocals, rdx);
3665   // get target Method* & entry point
3666   __ lookup_virtual_method(rax, index, method);
3667 
3668   __ profile_arguments_type(rdx, method, rbcp, true);
3669   __ jump_from_interpreted(method, rdx);
3670 }
3671 
3672 void TemplateTable::invokevirtual(int byte_no) {
3673   transition(vtos, vtos);
3674   assert(byte_no == f2_byte, "use this argument");
3675   prepare_invoke(byte_no,
3676                  rbx,    // method or vtable index
3677                  noreg,  // unused itable index
3678                  rcx, rdx); // recv, flags
3679 
3680   // rbx: index
3681   // rcx: receiver
3682   // rdx: flags
3683 
3684   invokevirtual_helper(rbx, rcx, rdx);
3685 }
3686 
3687 void TemplateTable::invokespecial(int byte_no) {
3688   transition(vtos, vtos);
3689   assert(byte_no == f1_byte, "use this argument");
3690   prepare_invoke(byte_no, rbx, noreg,  // get f1 Method*
3691                  rcx);  // get receiver also for null check
3692   __ verify_oop(rcx);
3693   __ null_check(rcx);
3694   // do the call
3695   __ profile_call(rax);
3696   __ profile_arguments_type(rax, rbx, rbcp, false);
3697   __ jump_from_interpreted(rbx, rax);
3698 }
3699 
3700 void TemplateTable::invokestatic(int byte_no) {
3701   transition(vtos, vtos);
3702   assert(byte_no == f1_byte, "use this argument");
3703   prepare_invoke(byte_no, rbx);  // get f1 Method*
3704   // do the call
3705   __ profile_call(rax);
3706   __ profile_arguments_type(rax, rbx, rbcp, false);
3707   __ jump_from_interpreted(rbx, rax);
3708 }
3709 
3710 
3711 void TemplateTable::fast_invokevfinal(int byte_no) {
3712   transition(vtos, vtos);
3713   assert(byte_no == f2_byte, "use this argument");
3714   __ stop("fast_invokevfinal not used on x86");
3715 }
3716 
3717 
3718 void TemplateTable::invokeinterface(int byte_no) {
3719   transition(vtos, vtos);
3720   assert(byte_no == f1_byte, "use this argument");
3721   prepare_invoke(byte_no, rax, rbx,  // get f1 Klass*, f2 Method*
3722                  rcx, rdx); // recv, flags
3723 
3724   // rax: reference klass (from f1) if interface method
3725   // rbx: method (from f2)
3726   // rcx: receiver
3727   // rdx: flags
3728 
3729   // First check for Object case, then private interface method,
3730   // then regular interface method.
3731 
3732   // Special case of invokeinterface called for virtual method of
3733   // java.lang.Object.  See cpCache.cpp for details.
3734   Label notObjectMethod;
3735   __ movl(rlocals, rdx);
3736   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3737   __ jcc(Assembler::zero, notObjectMethod);
3738   invokevirtual_helper(rbx, rcx, rdx);
3739   // no return from above
3740   __ bind(notObjectMethod);
3741 
3742   Label no_such_interface; // for receiver subtype check
3743   Register recvKlass; // used for exception processing
3744 
3745   // Check for private method invocation - indicated by vfinal
3746   Label notVFinal;
3747   __ movl(rlocals, rdx);
3748   __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3749   __ jcc(Assembler::zero, notVFinal);
3750 
3751   // Get receiver klass into rlocals - also a null check
3752   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3753   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
3754   __ load_klass(rlocals, rcx, tmp_load_klass);
3755 
3756   Label subtype;
3757   __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3758   // If we get here the typecheck failed
3759   recvKlass = rdx;
3760   __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3761   __ jmp(no_such_interface);
3762 
3763   __ bind(subtype);
3764 
3765   // do the call - rbx is actually the method to call
3766 
3767   __ profile_final_call(rdx);
3768   __ profile_arguments_type(rdx, rbx, rbcp, true);
3769 
3770   __ jump_from_interpreted(rbx, rdx);
3771   // no return from above
3772   __ bind(notVFinal);
3773 
3774   // Get receiver klass into rdx - also a null check
3775   __ restore_locals();  // restore r14
3776   __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3777   __ load_klass(rdx, rcx, tmp_load_klass);
3778 
3779   Label no_such_method;
3780 
3781   // Preserve method for throw_AbstractMethodErrorVerbose.
3782   __ mov(rcx, rbx);
3783   // Receiver subtype check against REFC.
3784   // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3785   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3786                              rdx, rax, noreg,
3787                              // outputs: scan temp. reg, scan temp. reg
3788                              rbcp, rlocals,
3789                              no_such_interface,
3790                              /*return_method=*/false);
3791 
3792   // profile this call
3793   __ restore_bcp(); // rbcp was destroyed by receiver type check
3794   __ profile_virtual_call(rdx, rbcp, rlocals);
3795 
3796   // Get declaring interface class from method, and itable index
3797   __ load_method_holder(rax, rbx);
3798   __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3799   __ subl(rbx, Method::itable_index_max);
3800   __ negl(rbx);
3801 
3802   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3803   __ mov(rlocals, rdx);
3804   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3805                              rlocals, rax, rbx,
3806                              // outputs: method, scan temp. reg
3807                              rbx, rbcp,
3808                              no_such_interface);
3809 
3810   // rbx: Method* to call
3811   // rcx: receiver
3812   // Check for abstract method error
3813   // Note: This should be done more efficiently via a throw_abstract_method_error
3814   //       interpreter entry point and a conditional jump to it in case of a null
3815   //       method.
3816   __ testptr(rbx, rbx);
3817   __ jcc(Assembler::zero, no_such_method);
3818 
3819   __ profile_arguments_type(rdx, rbx, rbcp, true);
3820 
3821   // do the call
3822   // rcx: receiver
3823   // rbx,: Method*
3824   __ jump_from_interpreted(rbx, rdx);
3825   __ should_not_reach_here();
3826 
3827   // exception handling code follows...
3828   // note: must restore interpreter registers to canonical
3829   //       state for exception handling to work correctly!
3830 
3831   __ bind(no_such_method);
3832   // throw exception
3833   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3834   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3835   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3836   // Pass arguments for generating a verbose error message.
3837 #ifdef _LP64
3838   recvKlass = c_rarg1;
3839   Register method    = c_rarg2;
3840   if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3841   if (method != rcx)    { __ movq(method, rcx);    }
3842 #else
3843   recvKlass = rdx;
3844   Register method    = rcx;
3845 #endif
3846   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3847              recvKlass, method);
3848   // The call_VM checks for exception, so we should never return here.
3849   __ should_not_reach_here();
3850 
3851   __ bind(no_such_interface);
3852   // throw exception
3853   __ pop(rbx);           // pop return address (pushed by prepare_invoke)
3854   __ restore_bcp();      // rbcp must be correct for exception handler   (was destroyed)
3855   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3856   // Pass arguments for generating a verbose error message.
3857   LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3858   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3859              recvKlass, rax);
3860   // the call_VM checks for exception, so we should never return here.
3861   __ should_not_reach_here();
3862 }
3863 
3864 void TemplateTable::invokehandle(int byte_no) {
3865   transition(vtos, vtos);
3866   assert(byte_no == f1_byte, "use this argument");
3867   const Register rbx_method = rbx;
3868   const Register rax_mtype  = rax;
3869   const Register rcx_recv   = rcx;
3870   const Register rdx_flags  = rdx;
3871 
3872   prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3873   __ verify_method_ptr(rbx_method);
3874   __ verify_oop(rcx_recv);
3875   __ null_check(rcx_recv);
3876 
3877   // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3878   // rbx: MH.invokeExact_MT method (from f2)
3879 
3880   // Note:  rax_mtype is already pushed (if necessary) by prepare_invoke
3881 
3882   // FIXME: profile the LambdaForm also
3883   __ profile_final_call(rax);
3884   __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3885 
3886   __ jump_from_interpreted(rbx_method, rdx);
3887 }
3888 
3889 void TemplateTable::invokedynamic(int byte_no) {
3890   transition(vtos, vtos);
3891   assert(byte_no == f1_byte, "use this argument");
3892 
3893   const Register rbx_method   = rbx;
3894   const Register rax_callsite = rax;
3895 
3896   prepare_invoke(byte_no, rbx_method, rax_callsite);
3897 
3898   // rax: CallSite object (from cpool->resolved_references[f1])
3899   // rbx: MH.linkToCallSite method (from f2)
3900 
3901   // Note:  rax_callsite is already pushed by prepare_invoke
3902 
3903   // %%% should make a type profile for any invokedynamic that takes a ref argument
3904   // profile this call
3905   __ profile_call(rbcp);
3906   __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3907 
3908   __ verify_oop(rax_callsite);
3909 
3910   __ jump_from_interpreted(rbx_method, rdx);
3911 }
3912 
3913 //-----------------------------------------------------------------------------
3914 // Allocation
3915 
3916 void TemplateTable::_new() {
3917   transition(vtos, atos);
3918   __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3919   Label slow_case;
3920   Label slow_case_no_pop;
3921   Label done;
3922   Label initialize_header;
3923   Label initialize_object;  // including clearing the fields
3924 
3925   __ get_cpool_and_tags(rcx, rax);
3926 
3927   // Make sure the class we're about to instantiate has been resolved.
3928   // This is done before loading InstanceKlass to be consistent with the order
3929   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3930   const int tags_offset = Array<u1>::base_offset_in_bytes();
3931   __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3932   __ jcc(Assembler::notEqual, slow_case_no_pop);
3933 
3934   // get InstanceKlass
3935   __ load_resolved_klass_at_index(rcx, rcx, rdx);
3936   __ push(rcx);  // save the contexts of klass for initializing the header
3937 
3938   // make sure klass is initialized & doesn't have finalizer
3939   // make sure klass is fully initialized
3940   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3941   __ jcc(Assembler::notEqual, slow_case);
3942 
3943   // get instance_size in InstanceKlass (scaled to a count of bytes)
3944   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3945   // test to see if it has a finalizer or is malformed in some way
3946   __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3947   __ jcc(Assembler::notZero, slow_case);
3948 
3949   // Allocate the instance:
3950   //  If TLAB is enabled:
3951   //    Try to allocate in the TLAB.
3952   //    If fails, go to the slow path.
3953   //  Else If inline contiguous allocations are enabled:
3954   //    Try to allocate in eden.
3955   //    If fails due to heap end, go to slow path.
3956   //
3957   //  If TLAB is enabled OR inline contiguous is enabled:
3958   //    Initialize the allocation.
3959   //    Exit.
3960   //
3961   //  Go to slow path.
3962 
3963   const bool allow_shared_alloc =
3964     Universe::heap()->supports_inline_contig_alloc();
3965 
3966   const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3967 #ifndef _LP64
3968   if (UseTLAB || allow_shared_alloc) {
3969     __ get_thread(thread);
3970   }
3971 #endif // _LP64
3972 
3973   if (UseTLAB) {
3974     __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
3975     if (ZeroTLAB) {
3976       // the fields have been already cleared
3977       __ jmp(initialize_header);
3978     } else {
3979       // initialize both the header and fields
3980       __ jmp(initialize_object);
3981     }
3982   } else {
3983     // Allocation in the shared Eden, if allowed.
3984     //
3985     // rdx: instance size in bytes
3986     __ eden_allocate(thread, rax, rdx, 0, rbx, slow_case);
3987   }
3988 
3989   // If UseTLAB or allow_shared_alloc are true, the object is created above and
3990   // there is an initialize need. Otherwise, skip and go to the slow path.
3991   if (UseTLAB || allow_shared_alloc) {
3992     // The object is initialized before the header.  If the object size is
3993     // zero, go directly to the header initialization.
3994     __ bind(initialize_object);
3995     __ decrement(rdx, sizeof(oopDesc));
3996     __ jcc(Assembler::zero, initialize_header);
3997 
3998     // Initialize topmost object field, divide rdx by 8, check if odd and
3999     // test if zero.
4000     __ xorl(rcx, rcx);    // use zero reg to clear memory (shorter code)
4001     __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4002 
4003     // rdx must have been multiple of 8
4004 #ifdef ASSERT
4005     // make sure rdx was multiple of 8
4006     Label L;
4007     // Ignore partial flag stall after shrl() since it is debug VM
4008     __ jcc(Assembler::carryClear, L);
4009     __ stop("object size is not multiple of 2 - adjust this code");
4010     __ bind(L);
4011     // rdx must be > 0, no extra check needed here
4012 #endif
4013 
4014     // initialize remaining object fields: rdx was a multiple of 8
4015     { Label loop;
4016     __ bind(loop);
4017     __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4018     NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4019     __ decrement(rdx);
4020     __ jcc(Assembler::notZero, loop);
4021     }
4022 
4023     // initialize object header only.
4024     __ bind(initialize_header);
4025     __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
4026               (intptr_t)markWord::prototype().value()); // header
4027     __ pop(rcx);   // get saved klass back in the register.
4028 #ifdef _LP64
4029     __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4030     __ store_klass_gap(rax, rsi);  // zero klass gap for compressed oops
4031 #endif
4032     Register tmp_store_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
4033     __ store_klass(rax, rcx, tmp_store_klass);  // klass
4034 
4035     {
4036       SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4037       // Trigger dtrace event for fastpath
4038       __ push(atos);
4039       __ call_VM_leaf(
4040            CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), rax);
4041       __ pop(atos);
4042     }
4043 
4044     __ jmp(done);
4045   }
4046 
4047   // slow case
4048   __ bind(slow_case);
4049   __ pop(rcx);   // restore stack pointer to what it was when we came in.
4050   __ bind(slow_case_no_pop);
4051 
4052   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4053   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4054 
4055   __ get_constant_pool(rarg1);
4056   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4057   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4058    __ verify_oop(rax);
4059 
4060   // continue
4061   __ bind(done);
4062 }
4063 
4064 void TemplateTable::newarray() {
4065   transition(itos, atos);
4066   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4067   __ load_unsigned_byte(rarg1, at_bcp(1));
4068   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4069           rarg1, rax);
4070 }
4071 
4072 void TemplateTable::anewarray() {
4073   transition(itos, atos);
4074 
4075   Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4076   Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4077 
4078   __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4079   __ get_constant_pool(rarg1);
4080   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4081           rarg1, rarg2, rax);
4082 }
4083 
4084 void TemplateTable::arraylength() {
4085   transition(atos, itos);
4086   __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4087   __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4088 }
4089 
4090 void TemplateTable::checkcast() {
4091   transition(atos, atos);
4092   Label done, is_null, ok_is_subtype, quicked, resolved;
4093   __ testptr(rax, rax); // object is in rax
4094   __ jcc(Assembler::zero, is_null);
4095 
4096   // Get cpool & tags index
4097   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4098   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4099   // See if bytecode has already been quicked
4100   __ cmpb(Address(rdx, rbx,
4101                   Address::times_1,
4102                   Array<u1>::base_offset_in_bytes()),
4103           JVM_CONSTANT_Class);
4104   __ jcc(Assembler::equal, quicked);
4105   __ push(atos); // save receiver for result, and for GC
4106   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4107 
4108   // vm_result_2 has metadata result
4109 #ifndef _LP64
4110   // borrow rdi from locals
4111   __ get_thread(rdi);
4112   __ get_vm_result_2(rax, rdi);
4113   __ restore_locals();
4114 #else
4115   __ get_vm_result_2(rax, r15_thread);
4116 #endif
4117 
4118   __ pop_ptr(rdx); // restore receiver
4119   __ jmpb(resolved);
4120 
4121   // Get superklass in rax and subklass in rbx
4122   __ bind(quicked);
4123   __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4124   __ load_resolved_klass_at_index(rax, rcx, rbx);
4125 
4126   __ bind(resolved);
4127   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
4128   __ load_klass(rbx, rdx, tmp_load_klass);
4129 
4130   // Generate subtype check.  Blows rcx, rdi.  Object in rdx.
4131   // Superklass in rax.  Subklass in rbx.
4132   __ gen_subtype_check(rbx, ok_is_subtype);
4133 
4134   // Come here on failure
4135   __ push_ptr(rdx);
4136   // object is at TOS
4137   __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4138 
4139   // Come here on success
4140   __ bind(ok_is_subtype);
4141   __ mov(rax, rdx); // Restore object in rdx
4142 
4143   // Collect counts on whether this check-cast sees NULLs a lot or not.
4144   if (ProfileInterpreter) {
4145     __ jmp(done);
4146     __ bind(is_null);
4147     __ profile_null_seen(rcx);
4148   } else {
4149     __ bind(is_null);   // same as 'done'
4150   }
4151   __ bind(done);
4152 }
4153 
4154 void TemplateTable::instanceof() {
4155   transition(atos, itos);
4156   Label done, is_null, ok_is_subtype, quicked, resolved;
4157   __ testptr(rax, rax);
4158   __ jcc(Assembler::zero, is_null);
4159 
4160   // Get cpool & tags index
4161   __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4162   __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4163   // See if bytecode has already been quicked
4164   __ cmpb(Address(rdx, rbx,
4165                   Address::times_1,
4166                   Array<u1>::base_offset_in_bytes()),
4167           JVM_CONSTANT_Class);
4168   __ jcc(Assembler::equal, quicked);
4169 
4170   __ push(atos); // save receiver for result, and for GC
4171   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4172   // vm_result_2 has metadata result
4173 
4174 #ifndef _LP64
4175   // borrow rdi from locals
4176   __ get_thread(rdi);
4177   __ get_vm_result_2(rax, rdi);
4178   __ restore_locals();
4179 #else
4180   __ get_vm_result_2(rax, r15_thread);
4181 #endif
4182 
4183   __ pop_ptr(rdx); // restore receiver
4184   __ verify_oop(rdx);
4185   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
4186   __ load_klass(rdx, rdx, tmp_load_klass);
4187   __ jmpb(resolved);
4188 
4189   // Get superklass in rax and subklass in rdx
4190   __ bind(quicked);
4191   __ load_klass(rdx, rax, tmp_load_klass);
4192   __ load_resolved_klass_at_index(rax, rcx, rbx);
4193 
4194   __ bind(resolved);
4195 
4196   // Generate subtype check.  Blows rcx, rdi
4197   // Superklass in rax.  Subklass in rdx.
4198   __ gen_subtype_check(rdx, ok_is_subtype);
4199 
4200   // Come here on failure
4201   __ xorl(rax, rax);
4202   __ jmpb(done);
4203   // Come here on success
4204   __ bind(ok_is_subtype);
4205   __ movl(rax, 1);
4206 
4207   // Collect counts on whether this test sees NULLs a lot or not.
4208   if (ProfileInterpreter) {
4209     __ jmp(done);
4210     __ bind(is_null);
4211     __ profile_null_seen(rcx);
4212   } else {
4213     __ bind(is_null);   // same as 'done'
4214   }
4215   __ bind(done);
4216   // rax = 0: obj == NULL or  obj is not an instanceof the specified klass
4217   // rax = 1: obj != NULL and obj is     an instanceof the specified klass
4218 }
4219 
4220 
4221 //----------------------------------------------------------------------------------------------------
4222 // Breakpoints
4223 void TemplateTable::_breakpoint() {
4224   // Note: We get here even if we are single stepping..
4225   // jbug insists on setting breakpoints at every bytecode
4226   // even if we are in single step mode.
4227 
4228   transition(vtos, vtos);
4229 
4230   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4231 
4232   // get the unpatched byte code
4233   __ get_method(rarg);
4234   __ call_VM(noreg,
4235              CAST_FROM_FN_PTR(address,
4236                               InterpreterRuntime::get_original_bytecode_at),
4237              rarg, rbcp);
4238   __ mov(rbx, rax);  // why?
4239 
4240   // post the breakpoint event
4241   __ get_method(rarg);
4242   __ call_VM(noreg,
4243              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4244              rarg, rbcp);
4245 
4246   // complete the execution of original bytecode
4247   __ dispatch_only_normal(vtos);
4248 }
4249 
4250 //-----------------------------------------------------------------------------
4251 // Exceptions
4252 
4253 void TemplateTable::athrow() {
4254   transition(atos, vtos);
4255   __ null_check(rax);
4256   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4257 }
4258 
4259 //-----------------------------------------------------------------------------
4260 // Synchronization
4261 //
4262 // Note: monitorenter & exit are symmetric routines; which is reflected
4263 //       in the assembly code structure as well
4264 //
4265 // Stack layout:
4266 //
4267 // [expressions  ] <--- rsp               = expression stack top
4268 // ..
4269 // [expressions  ]
4270 // [monitor entry] <--- monitor block top = expression stack bot
4271 // ..
4272 // [monitor entry]
4273 // [frame data   ] <--- monitor block bot
4274 // ...
4275 // [saved rbp    ] <--- rbp
4276 void TemplateTable::monitorenter() {
4277   transition(atos, vtos);
4278 
4279   // check for NULL object
4280   __ null_check(rax);
4281 
4282   const Address monitor_block_top(
4283         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4284   const Address monitor_block_bot(
4285         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4286   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4287 
4288   Label allocated;
4289 
4290   Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4291   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4292   Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4293 
4294   // initialize entry pointer
4295   __ xorl(rmon, rmon); // points to free slot or NULL
4296 
4297   // find a free slot in the monitor block (result in rmon)
4298   {
4299     Label entry, loop, exit;
4300     __ movptr(rtop, monitor_block_top); // points to current entry,
4301                                         // starting with top-most entry
4302     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4303                                         // of monitor block
4304     __ jmpb(entry);
4305 
4306     __ bind(loop);
4307     // check if current entry is used
4308     __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4309     // if not used then remember entry in rmon
4310     __ cmovptr(Assembler::equal, rmon, rtop);   // cmov => cmovptr
4311     // check if current entry is for same object
4312     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4313     // if same object then stop searching
4314     __ jccb(Assembler::equal, exit);
4315     // otherwise advance to next entry
4316     __ addptr(rtop, entry_size);
4317     __ bind(entry);
4318     // check if bottom reached
4319     __ cmpptr(rtop, rbot);
4320     // if not at bottom then check this entry
4321     __ jcc(Assembler::notEqual, loop);
4322     __ bind(exit);
4323   }
4324 
4325   __ testptr(rmon, rmon); // check if a slot has been found
4326   __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4327 
4328   // allocate one if there's no free slot
4329   {
4330     Label entry, loop;
4331     // 1. compute new pointers          // rsp: old expression stack top
4332     __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4333     __ subptr(rsp, entry_size);         // move expression stack top
4334     __ subptr(rmon, entry_size);        // move expression stack bottom
4335     __ mov(rtop, rsp);                  // set start value for copy loop
4336     __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4337     __ jmp(entry);
4338     // 2. move expression stack contents
4339     __ bind(loop);
4340     __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4341                                                 // word from old location
4342     __ movptr(Address(rtop, 0), rbot);          // and store it at new location
4343     __ addptr(rtop, wordSize);                  // advance to next word
4344     __ bind(entry);
4345     __ cmpptr(rtop, rmon);                      // check if bottom reached
4346     __ jcc(Assembler::notEqual, loop);          // if not at bottom then
4347                                                 // copy next word
4348   }
4349 
4350   // call run-time routine
4351   // rmon: points to monitor entry
4352   __ bind(allocated);
4353 
4354   // Increment bcp to point to the next bytecode, so exception
4355   // handling for async. exceptions work correctly.
4356   // The object has already been popped from the stack, so the
4357   // expression stack looks correct.
4358   __ increment(rbcp);
4359 
4360   // store object
4361   __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4362   __ lock_object(rmon);
4363 
4364   // check to make sure this monitor doesn't cause stack overflow after locking
4365   __ save_bcp();  // in case of exception
4366   __ generate_stack_overflow_check(0);
4367 
4368   // The bcp has already been incremented. Just need to dispatch to
4369   // next instruction.
4370   __ dispatch_next(vtos);
4371 }
4372 
4373 void TemplateTable::monitorexit() {
4374   transition(atos, vtos);
4375 
4376   // check for NULL object
4377   __ null_check(rax);
4378 
4379   const Address monitor_block_top(
4380         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4381   const Address monitor_block_bot(
4382         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4383   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4384 
4385   Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4386   Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4387 
4388   Label found;
4389 
4390   // find matching slot
4391   {
4392     Label entry, loop;
4393     __ movptr(rtop, monitor_block_top); // points to current entry,
4394                                         // starting with top-most entry
4395     __ lea(rbot, monitor_block_bot);    // points to word before bottom
4396                                         // of monitor block
4397     __ jmpb(entry);
4398 
4399     __ bind(loop);
4400     // check if current entry is for same object
4401     __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4402     // if same object then stop searching
4403     __ jcc(Assembler::equal, found);
4404     // otherwise advance to next entry
4405     __ addptr(rtop, entry_size);
4406     __ bind(entry);
4407     // check if bottom reached
4408     __ cmpptr(rtop, rbot);
4409     // if not at bottom then check this entry
4410     __ jcc(Assembler::notEqual, loop);
4411   }
4412 
4413   // error handling. Unlocking was not block-structured
4414   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4415                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4416   __ should_not_reach_here();
4417 
4418   // call run-time routine
4419   __ bind(found);
4420   __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4421   __ unlock_object(rtop);
4422   __ pop_ptr(rax); // discard object
4423 }
4424 
4425 // Wide instructions
4426 void TemplateTable::wide() {
4427   transition(vtos, vtos);
4428   __ load_unsigned_byte(rbx, at_bcp(1));
4429   ExternalAddress wtable((address)Interpreter::_wentry_point);
4430   __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4431   // Note: the rbcp increment step is part of the individual wide bytecode implementations
4432 }
4433 
4434 // Multi arrays
4435 void TemplateTable::multianewarray() {
4436   transition(vtos, atos);
4437 
4438   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4439   __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4440   // last dim is on top of stack; we want address of first one:
4441   // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4442   // the latter wordSize to point to the beginning of the array.
4443   __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4444   call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4445   __ load_unsigned_byte(rbx, at_bcp(3));
4446   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));  // get rid of counts
4447 }