1 /*
   2  * Copyright (c) 1997, 2011, 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 "assembler_x86.inline.hpp"
  27 #include "gc_interface/collectedHeap.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "memory/cardTableModRefBS.hpp"
  30 #include "memory/resourceArea.hpp"
  31 #include "prims/methodHandles.hpp"
  32 #include "runtime/biasedLocking.hpp"
  33 #include "runtime/interfaceSupport.hpp"
  34 #include "runtime/objectMonitor.hpp"
  35 #include "runtime/os.hpp"
  36 #include "runtime/sharedRuntime.hpp"
  37 #include "runtime/stubRoutines.hpp"
  38 #ifndef SERIALGC
  39 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
  40 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
  41 #include "gc_implementation/g1/heapRegion.hpp"
  42 #endif
  43 
  44 // Implementation of AddressLiteral
  45 
  46 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) {
  47   _is_lval = false;
  48   _target = target;
  49   switch (rtype) {
  50   case relocInfo::oop_type:
  51     // Oops are a special case. Normally they would be their own section
  52     // but in cases like icBuffer they are literals in the code stream that
  53     // we don't have a section for. We use none so that we get a literal address
  54     // which is always patchable.
  55     break;
  56   case relocInfo::external_word_type:
  57     _rspec = external_word_Relocation::spec(target);
  58     break;
  59   case relocInfo::internal_word_type:
  60     _rspec = internal_word_Relocation::spec(target);
  61     break;
  62   case relocInfo::opt_virtual_call_type:
  63     _rspec = opt_virtual_call_Relocation::spec();
  64     break;
  65   case relocInfo::static_call_type:
  66     _rspec = static_call_Relocation::spec();
  67     break;
  68   case relocInfo::runtime_call_type:
  69     _rspec = runtime_call_Relocation::spec();
  70     break;
  71   case relocInfo::poll_type:
  72   case relocInfo::poll_return_type:
  73     _rspec = Relocation::spec_simple(rtype);
  74     break;
  75   case relocInfo::none:
  76     break;
  77   default:
  78     ShouldNotReachHere();
  79     break;
  80   }
  81 }
  82 
  83 // Implementation of Address
  84 
  85 #ifdef _LP64
  86 
  87 Address Address::make_array(ArrayAddress adr) {
  88   // Not implementable on 64bit machines
  89   // Should have been handled higher up the call chain.
  90   ShouldNotReachHere();
  91   return Address();
  92 }
  93 
  94 // exceedingly dangerous constructor
  95 Address::Address(int disp, address loc, relocInfo::relocType rtype) {
  96   _base  = noreg;
  97   _index = noreg;
  98   _scale = no_scale;
  99   _disp  = disp;
 100   switch (rtype) {
 101     case relocInfo::external_word_type:
 102       _rspec = external_word_Relocation::spec(loc);
 103       break;
 104     case relocInfo::internal_word_type:
 105       _rspec = internal_word_Relocation::spec(loc);
 106       break;
 107     case relocInfo::runtime_call_type:
 108       // HMM
 109       _rspec = runtime_call_Relocation::spec();
 110       break;
 111     case relocInfo::poll_type:
 112     case relocInfo::poll_return_type:
 113       _rspec = Relocation::spec_simple(rtype);
 114       break;
 115     case relocInfo::none:
 116       break;
 117     default:
 118       ShouldNotReachHere();
 119   }
 120 }
 121 #else // LP64
 122 
 123 Address Address::make_array(ArrayAddress adr) {
 124   AddressLiteral base = adr.base();
 125   Address index = adr.index();
 126   assert(index._disp == 0, "must not have disp"); // maybe it can?
 127   Address array(index._base, index._index, index._scale, (intptr_t) base.target());
 128   array._rspec = base._rspec;
 129   return array;
 130 }
 131 
 132 // exceedingly dangerous constructor
 133 Address::Address(address loc, RelocationHolder spec) {
 134   _base  = noreg;
 135   _index = noreg;
 136   _scale = no_scale;
 137   _disp  = (intptr_t) loc;
 138   _rspec = spec;
 139 }
 140 
 141 #endif // _LP64
 142 
 143 
 144 
 145 // Convert the raw encoding form into the form expected by the constructor for
 146 // Address.  An index of 4 (rsp) corresponds to having no index, so convert
 147 // that to noreg for the Address constructor.
 148 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) {
 149   RelocationHolder rspec;
 150   if (disp_is_oop) {
 151     rspec = Relocation::spec_simple(relocInfo::oop_type);
 152   }
 153   bool valid_index = index != rsp->encoding();
 154   if (valid_index) {
 155     Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp));
 156     madr._rspec = rspec;
 157     return madr;
 158   } else {
 159     Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp));
 160     madr._rspec = rspec;
 161     return madr;
 162   }
 163 }
 164 
 165 // Implementation of Assembler
 166 
 167 int AbstractAssembler::code_fill_byte() {
 168   return (u_char)'\xF4'; // hlt
 169 }
 170 
 171 // make this go away someday
 172 void Assembler::emit_data(jint data, relocInfo::relocType rtype, int format) {
 173   if (rtype == relocInfo::none)
 174         emit_long(data);
 175   else  emit_data(data, Relocation::spec_simple(rtype), format);
 176 }
 177 
 178 void Assembler::emit_data(jint data, RelocationHolder const& rspec, int format) {
 179   assert(imm_operand == 0, "default format must be immediate in this file");
 180   assert(inst_mark() != NULL, "must be inside InstructionMark");
 181   if (rspec.type() !=  relocInfo::none) {
 182     #ifdef ASSERT
 183       check_relocation(rspec, format);
 184     #endif
 185     // Do not use AbstractAssembler::relocate, which is not intended for
 186     // embedded words.  Instead, relocate to the enclosing instruction.
 187 
 188     // hack. call32 is too wide for mask so use disp32
 189     if (format == call32_operand)
 190       code_section()->relocate(inst_mark(), rspec, disp32_operand);
 191     else
 192       code_section()->relocate(inst_mark(), rspec, format);
 193   }
 194   emit_long(data);
 195 }
 196 
 197 static int encode(Register r) {
 198   int enc = r->encoding();
 199   if (enc >= 8) {
 200     enc -= 8;
 201   }
 202   return enc;
 203 }
 204 
 205 static int encode(XMMRegister r) {
 206   int enc = r->encoding();
 207   if (enc >= 8) {
 208     enc -= 8;
 209   }
 210   return enc;
 211 }
 212 
 213 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) {
 214   assert(dst->has_byte_register(), "must have byte register");
 215   assert(isByte(op1) && isByte(op2), "wrong opcode");
 216   assert(isByte(imm8), "not a byte");
 217   assert((op1 & 0x01) == 0, "should be 8bit operation");
 218   emit_byte(op1);
 219   emit_byte(op2 | encode(dst));
 220   emit_byte(imm8);
 221 }
 222 
 223 
 224 void Assembler::emit_arith(int op1, int op2, Register dst, int32_t imm32) {
 225   assert(isByte(op1) && isByte(op2), "wrong opcode");
 226   assert((op1 & 0x01) == 1, "should be 32bit operation");
 227   assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
 228   if (is8bit(imm32)) {
 229     emit_byte(op1 | 0x02); // set sign bit
 230     emit_byte(op2 | encode(dst));
 231     emit_byte(imm32 & 0xFF);
 232   } else {
 233     emit_byte(op1);
 234     emit_byte(op2 | encode(dst));
 235     emit_long(imm32);
 236   }
 237 }
 238 
 239 // immediate-to-memory forms
 240 void Assembler::emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32) {
 241   assert((op1 & 0x01) == 1, "should be 32bit operation");
 242   assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
 243   if (is8bit(imm32)) {
 244     emit_byte(op1 | 0x02); // set sign bit
 245     emit_operand(rm, adr, 1);
 246     emit_byte(imm32 & 0xFF);
 247   } else {
 248     emit_byte(op1);
 249     emit_operand(rm, adr, 4);
 250     emit_long(imm32);
 251   }
 252 }
 253 
 254 void Assembler::emit_arith(int op1, int op2, Register dst, jobject obj) {
 255   LP64_ONLY(ShouldNotReachHere());
 256   assert(isByte(op1) && isByte(op2), "wrong opcode");
 257   assert((op1 & 0x01) == 1, "should be 32bit operation");
 258   assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
 259   InstructionMark im(this);
 260   emit_byte(op1);
 261   emit_byte(op2 | encode(dst));
 262   emit_data((intptr_t)obj, relocInfo::oop_type, 0);
 263 }
 264 
 265 
 266 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) {
 267   assert(isByte(op1) && isByte(op2), "wrong opcode");
 268   emit_byte(op1);
 269   emit_byte(op2 | encode(dst) << 3 | encode(src));
 270 }
 271 
 272 
 273 void Assembler::emit_operand(Register reg, Register base, Register index,
 274                              Address::ScaleFactor scale, int disp,
 275                              RelocationHolder const& rspec,
 276                              int rip_relative_correction) {
 277   relocInfo::relocType rtype = (relocInfo::relocType) rspec.type();
 278 
 279   // Encode the registers as needed in the fields they are used in
 280 
 281   int regenc = encode(reg) << 3;
 282   int indexenc = index->is_valid() ? encode(index) << 3 : 0;
 283   int baseenc = base->is_valid() ? encode(base) : 0;
 284 
 285   if (base->is_valid()) {
 286     if (index->is_valid()) {
 287       assert(scale != Address::no_scale, "inconsistent address");
 288       // [base + index*scale + disp]
 289       if (disp == 0 && rtype == relocInfo::none  &&
 290           base != rbp LP64_ONLY(&& base != r13)) {
 291         // [base + index*scale]
 292         // [00 reg 100][ss index base]
 293         assert(index != rsp, "illegal addressing mode");
 294         emit_byte(0x04 | regenc);
 295         emit_byte(scale << 6 | indexenc | baseenc);
 296       } else if (is8bit(disp) && rtype == relocInfo::none) {
 297         // [base + index*scale + imm8]
 298         // [01 reg 100][ss index base] imm8
 299         assert(index != rsp, "illegal addressing mode");
 300         emit_byte(0x44 | regenc);
 301         emit_byte(scale << 6 | indexenc | baseenc);
 302         emit_byte(disp & 0xFF);
 303       } else {
 304         // [base + index*scale + disp32]
 305         // [10 reg 100][ss index base] disp32
 306         assert(index != rsp, "illegal addressing mode");
 307         emit_byte(0x84 | regenc);
 308         emit_byte(scale << 6 | indexenc | baseenc);
 309         emit_data(disp, rspec, disp32_operand);
 310       }
 311     } else if (base == rsp LP64_ONLY(|| base == r12)) {
 312       // [rsp + disp]
 313       if (disp == 0 && rtype == relocInfo::none) {
 314         // [rsp]
 315         // [00 reg 100][00 100 100]
 316         emit_byte(0x04 | regenc);
 317         emit_byte(0x24);
 318       } else if (is8bit(disp) && rtype == relocInfo::none) {
 319         // [rsp + imm8]
 320         // [01 reg 100][00 100 100] disp8
 321         emit_byte(0x44 | regenc);
 322         emit_byte(0x24);
 323         emit_byte(disp & 0xFF);
 324       } else {
 325         // [rsp + imm32]
 326         // [10 reg 100][00 100 100] disp32
 327         emit_byte(0x84 | regenc);
 328         emit_byte(0x24);
 329         emit_data(disp, rspec, disp32_operand);
 330       }
 331     } else {
 332       // [base + disp]
 333       assert(base != rsp LP64_ONLY(&& base != r12), "illegal addressing mode");
 334       if (disp == 0 && rtype == relocInfo::none &&
 335           base != rbp LP64_ONLY(&& base != r13)) {
 336         // [base]
 337         // [00 reg base]
 338         emit_byte(0x00 | regenc | baseenc);
 339       } else if (is8bit(disp) && rtype == relocInfo::none) {
 340         // [base + disp8]
 341         // [01 reg base] disp8
 342         emit_byte(0x40 | regenc | baseenc);
 343         emit_byte(disp & 0xFF);
 344       } else {
 345         // [base + disp32]
 346         // [10 reg base] disp32
 347         emit_byte(0x80 | regenc | baseenc);
 348         emit_data(disp, rspec, disp32_operand);
 349       }
 350     }
 351   } else {
 352     if (index->is_valid()) {
 353       assert(scale != Address::no_scale, "inconsistent address");
 354       // [index*scale + disp]
 355       // [00 reg 100][ss index 101] disp32
 356       assert(index != rsp, "illegal addressing mode");
 357       emit_byte(0x04 | regenc);
 358       emit_byte(scale << 6 | indexenc | 0x05);
 359       emit_data(disp, rspec, disp32_operand);
 360     } else if (rtype != relocInfo::none ) {
 361       // [disp] (64bit) RIP-RELATIVE (32bit) abs
 362       // [00 000 101] disp32
 363 
 364       emit_byte(0x05 | regenc);
 365       // Note that the RIP-rel. correction applies to the generated
 366       // disp field, but _not_ to the target address in the rspec.
 367 
 368       // disp was created by converting the target address minus the pc
 369       // at the start of the instruction. That needs more correction here.
 370       // intptr_t disp = target - next_ip;
 371       assert(inst_mark() != NULL, "must be inside InstructionMark");
 372       address next_ip = pc() + sizeof(int32_t) + rip_relative_correction;
 373       int64_t adjusted = disp;
 374       // Do rip-rel adjustment for 64bit
 375       LP64_ONLY(adjusted -=  (next_ip - inst_mark()));
 376       assert(is_simm32(adjusted),
 377              "must be 32bit offset (RIP relative address)");
 378       emit_data((int32_t) adjusted, rspec, disp32_operand);
 379 
 380     } else {
 381       // 32bit never did this, did everything as the rip-rel/disp code above
 382       // [disp] ABSOLUTE
 383       // [00 reg 100][00 100 101] disp32
 384       emit_byte(0x04 | regenc);
 385       emit_byte(0x25);
 386       emit_data(disp, rspec, disp32_operand);
 387     }
 388   }
 389 }
 390 
 391 void Assembler::emit_operand(XMMRegister reg, Register base, Register index,
 392                              Address::ScaleFactor scale, int disp,
 393                              RelocationHolder const& rspec) {
 394   emit_operand((Register)reg, base, index, scale, disp, rspec);
 395 }
 396 
 397 // Secret local extension to Assembler::WhichOperand:
 398 #define end_pc_operand (_WhichOperand_limit)
 399 
 400 address Assembler::locate_operand(address inst, WhichOperand which) {
 401   // Decode the given instruction, and return the address of
 402   // an embedded 32-bit operand word.
 403 
 404   // If "which" is disp32_operand, selects the displacement portion
 405   // of an effective address specifier.
 406   // If "which" is imm64_operand, selects the trailing immediate constant.
 407   // If "which" is call32_operand, selects the displacement of a call or jump.
 408   // Caller is responsible for ensuring that there is such an operand,
 409   // and that it is 32/64 bits wide.
 410 
 411   // If "which" is end_pc_operand, find the end of the instruction.
 412 
 413   address ip = inst;
 414   bool is_64bit = false;
 415 
 416   debug_only(bool has_disp32 = false);
 417   int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn
 418 
 419   again_after_prefix:
 420   switch (0xFF & *ip++) {
 421 
 422   // These convenience macros generate groups of "case" labels for the switch.
 423 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3
 424 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \
 425              case (x)+4: case (x)+5: case (x)+6: case (x)+7
 426 #define REP16(x) REP8((x)+0): \
 427               case REP8((x)+8)
 428 
 429   case CS_segment:
 430   case SS_segment:
 431   case DS_segment:
 432   case ES_segment:
 433   case FS_segment:
 434   case GS_segment:
 435     // Seems dubious
 436     LP64_ONLY(assert(false, "shouldn't have that prefix"));
 437     assert(ip == inst+1, "only one prefix allowed");
 438     goto again_after_prefix;
 439 
 440   case 0x67:
 441   case REX:
 442   case REX_B:
 443   case REX_X:
 444   case REX_XB:
 445   case REX_R:
 446   case REX_RB:
 447   case REX_RX:
 448   case REX_RXB:
 449     NOT_LP64(assert(false, "64bit prefixes"));
 450     goto again_after_prefix;
 451 
 452   case REX_W:
 453   case REX_WB:
 454   case REX_WX:
 455   case REX_WXB:
 456   case REX_WR:
 457   case REX_WRB:
 458   case REX_WRX:
 459   case REX_WRXB:
 460     NOT_LP64(assert(false, "64bit prefixes"));
 461     is_64bit = true;
 462     goto again_after_prefix;
 463 
 464   case 0xFF: // pushq a; decl a; incl a; call a; jmp a
 465   case 0x88: // movb a, r
 466   case 0x89: // movl a, r
 467   case 0x8A: // movb r, a
 468   case 0x8B: // movl r, a
 469   case 0x8F: // popl a
 470     debug_only(has_disp32 = true);
 471     break;
 472 
 473   case 0x68: // pushq #32
 474     if (which == end_pc_operand) {
 475       return ip + 4;
 476     }
 477     assert(which == imm_operand && !is_64bit, "pushl has no disp32 or 64bit immediate");
 478     return ip;                  // not produced by emit_operand
 479 
 480   case 0x66: // movw ... (size prefix)
 481     again_after_size_prefix2:
 482     switch (0xFF & *ip++) {
 483     case REX:
 484     case REX_B:
 485     case REX_X:
 486     case REX_XB:
 487     case REX_R:
 488     case REX_RB:
 489     case REX_RX:
 490     case REX_RXB:
 491     case REX_W:
 492     case REX_WB:
 493     case REX_WX:
 494     case REX_WXB:
 495     case REX_WR:
 496     case REX_WRB:
 497     case REX_WRX:
 498     case REX_WRXB:
 499       NOT_LP64(assert(false, "64bit prefix found"));
 500       goto again_after_size_prefix2;
 501     case 0x8B: // movw r, a
 502     case 0x89: // movw a, r
 503       debug_only(has_disp32 = true);
 504       break;
 505     case 0xC7: // movw a, #16
 506       debug_only(has_disp32 = true);
 507       tail_size = 2;  // the imm16
 508       break;
 509     case 0x0F: // several SSE/SSE2 variants
 510       ip--;    // reparse the 0x0F
 511       goto again_after_prefix;
 512     default:
 513       ShouldNotReachHere();
 514     }
 515     break;
 516 
 517   case REP8(0xB8): // movl/q r, #32/#64(oop?)
 518     if (which == end_pc_operand)  return ip + (is_64bit ? 8 : 4);
 519     // these asserts are somewhat nonsensical
 520 #ifndef _LP64
 521     assert(which == imm_operand || which == disp32_operand, "");
 522 #else
 523     assert((which == call32_operand || which == imm_operand) && is_64bit ||
 524            which == narrow_oop_operand && !is_64bit, "");
 525 #endif // _LP64
 526     return ip;
 527 
 528   case 0x69: // imul r, a, #32
 529   case 0xC7: // movl a, #32(oop?)
 530     tail_size = 4;
 531     debug_only(has_disp32 = true); // has both kinds of operands!
 532     break;
 533 
 534   case 0x0F: // movx..., etc.
 535     switch (0xFF & *ip++) {
 536     case 0x12: // movlps
 537     case 0x28: // movaps
 538     case 0x2E: // ucomiss
 539     case 0x2F: // comiss
 540     case 0x54: // andps
 541     case 0x55: // andnps
 542     case 0x56: // orps
 543     case 0x57: // xorps
 544     case 0x6E: // movd
 545     case 0x7E: // movd
 546     case 0xAE: // ldmxcsr   a
 547       // 64bit side says it these have both operands but that doesn't
 548       // appear to be true
 549       debug_only(has_disp32 = true);
 550       break;
 551 
 552     case 0xAD: // shrd r, a, %cl
 553     case 0xAF: // imul r, a
 554     case 0xBE: // movsbl r, a (movsxb)
 555     case 0xBF: // movswl r, a (movsxw)
 556     case 0xB6: // movzbl r, a (movzxb)
 557     case 0xB7: // movzwl r, a (movzxw)
 558     case REP16(0x40): // cmovl cc, r, a
 559     case 0xB0: // cmpxchgb
 560     case 0xB1: // cmpxchg
 561     case 0xC1: // xaddl
 562     case 0xC7: // cmpxchg8
 563     case REP16(0x90): // setcc a
 564       debug_only(has_disp32 = true);
 565       // fall out of the switch to decode the address
 566       break;
 567 
 568     case 0xAC: // shrd r, a, #8
 569       debug_only(has_disp32 = true);
 570       tail_size = 1;  // the imm8
 571       break;
 572 
 573     case REP16(0x80): // jcc rdisp32
 574       if (which == end_pc_operand)  return ip + 4;
 575       assert(which == call32_operand, "jcc has no disp32 or imm");
 576       return ip;
 577     default:
 578       ShouldNotReachHere();
 579     }
 580     break;
 581 
 582   case 0x81: // addl a, #32; addl r, #32
 583     // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
 584     // on 32bit in the case of cmpl, the imm might be an oop
 585     tail_size = 4;
 586     debug_only(has_disp32 = true); // has both kinds of operands!
 587     break;
 588 
 589   case 0x83: // addl a, #8; addl r, #8
 590     // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
 591     debug_only(has_disp32 = true); // has both kinds of operands!
 592     tail_size = 1;
 593     break;
 594 
 595   case 0x9B:
 596     switch (0xFF & *ip++) {
 597     case 0xD9: // fnstcw a
 598       debug_only(has_disp32 = true);
 599       break;
 600     default:
 601       ShouldNotReachHere();
 602     }
 603     break;
 604 
 605   case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a
 606   case REP4(0x10): // adc...
 607   case REP4(0x20): // and...
 608   case REP4(0x30): // xor...
 609   case REP4(0x08): // or...
 610   case REP4(0x18): // sbb...
 611   case REP4(0x28): // sub...
 612   case 0xF7: // mull a
 613   case 0x8D: // lea r, a
 614   case 0x87: // xchg r, a
 615   case REP4(0x38): // cmp...
 616   case 0x85: // test r, a
 617     debug_only(has_disp32 = true); // has both kinds of operands!
 618     break;
 619 
 620   case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8
 621   case 0xC6: // movb a, #8
 622   case 0x80: // cmpb a, #8
 623   case 0x6B: // imul r, a, #8
 624     debug_only(has_disp32 = true); // has both kinds of operands!
 625     tail_size = 1; // the imm8
 626     break;
 627 
 628   case 0xE8: // call rdisp32
 629   case 0xE9: // jmp  rdisp32
 630     if (which == end_pc_operand)  return ip + 4;
 631     assert(which == call32_operand, "call has no disp32 or imm");
 632     return ip;
 633 
 634   case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1
 635   case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl
 636   case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a
 637   case 0xDD: // fld_d a; fst_d a; fstp_d a
 638   case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a
 639   case 0xDF: // fild_d a; fistp_d a
 640   case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a
 641   case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a
 642   case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a
 643     debug_only(has_disp32 = true);
 644     break;
 645 
 646   case 0xF0:                    // Lock
 647     assert(os::is_MP(), "only on MP");
 648     goto again_after_prefix;
 649 
 650   case 0xF3:                    // For SSE
 651   case 0xF2:                    // For SSE2
 652     switch (0xFF & *ip++) {
 653     case REX:
 654     case REX_B:
 655     case REX_X:
 656     case REX_XB:
 657     case REX_R:
 658     case REX_RB:
 659     case REX_RX:
 660     case REX_RXB:
 661     case REX_W:
 662     case REX_WB:
 663     case REX_WX:
 664     case REX_WXB:
 665     case REX_WR:
 666     case REX_WRB:
 667     case REX_WRX:
 668     case REX_WRXB:
 669       NOT_LP64(assert(false, "found 64bit prefix"));
 670       ip++;
 671     default:
 672       ip++;
 673     }
 674     debug_only(has_disp32 = true); // has both kinds of operands!
 675     break;
 676 
 677   default:
 678     ShouldNotReachHere();
 679 
 680 #undef REP8
 681 #undef REP16
 682   }
 683 
 684   assert(which != call32_operand, "instruction is not a call, jmp, or jcc");
 685 #ifdef _LP64
 686   assert(which != imm_operand, "instruction is not a movq reg, imm64");
 687 #else
 688   // assert(which != imm_operand || has_imm32, "instruction has no imm32 field");
 689   assert(which != imm_operand || has_disp32, "instruction has no imm32 field");
 690 #endif // LP64
 691   assert(which != disp32_operand || has_disp32, "instruction has no disp32 field");
 692 
 693   // parse the output of emit_operand
 694   int op2 = 0xFF & *ip++;
 695   int base = op2 & 0x07;
 696   int op3 = -1;
 697   const int b100 = 4;
 698   const int b101 = 5;
 699   if (base == b100 && (op2 >> 6) != 3) {
 700     op3 = 0xFF & *ip++;
 701     base = op3 & 0x07;   // refetch the base
 702   }
 703   // now ip points at the disp (if any)
 704 
 705   switch (op2 >> 6) {
 706   case 0:
 707     // [00 reg  100][ss index base]
 708     // [00 reg  100][00   100  esp]
 709     // [00 reg base]
 710     // [00 reg  100][ss index  101][disp32]
 711     // [00 reg  101]               [disp32]
 712 
 713     if (base == b101) {
 714       if (which == disp32_operand)
 715         return ip;              // caller wants the disp32
 716       ip += 4;                  // skip the disp32
 717     }
 718     break;
 719 
 720   case 1:
 721     // [01 reg  100][ss index base][disp8]
 722     // [01 reg  100][00   100  esp][disp8]
 723     // [01 reg base]               [disp8]
 724     ip += 1;                    // skip the disp8
 725     break;
 726 
 727   case 2:
 728     // [10 reg  100][ss index base][disp32]
 729     // [10 reg  100][00   100  esp][disp32]
 730     // [10 reg base]               [disp32]
 731     if (which == disp32_operand)
 732       return ip;                // caller wants the disp32
 733     ip += 4;                    // skip the disp32
 734     break;
 735 
 736   case 3:
 737     // [11 reg base]  (not a memory addressing mode)
 738     break;
 739   }
 740 
 741   if (which == end_pc_operand) {
 742     return ip + tail_size;
 743   }
 744 
 745 #ifdef _LP64
 746   assert(which == narrow_oop_operand && !is_64bit, "instruction is not a movl adr, imm32");
 747 #else
 748   assert(which == imm_operand, "instruction has only an imm field");
 749 #endif // LP64
 750   return ip;
 751 }
 752 
 753 address Assembler::locate_next_instruction(address inst) {
 754   // Secretly share code with locate_operand:
 755   return locate_operand(inst, end_pc_operand);
 756 }
 757 
 758 
 759 #ifdef ASSERT
 760 void Assembler::check_relocation(RelocationHolder const& rspec, int format) {
 761   address inst = inst_mark();
 762   assert(inst != NULL && inst < pc(), "must point to beginning of instruction");
 763   address opnd;
 764 
 765   Relocation* r = rspec.reloc();
 766   if (r->type() == relocInfo::none) {
 767     return;
 768   } else if (r->is_call() || format == call32_operand) {
 769     // assert(format == imm32_operand, "cannot specify a nonzero format");
 770     opnd = locate_operand(inst, call32_operand);
 771   } else if (r->is_data()) {
 772     assert(format == imm_operand || format == disp32_operand
 773            LP64_ONLY(|| format == narrow_oop_operand), "format ok");
 774     opnd = locate_operand(inst, (WhichOperand)format);
 775   } else {
 776     assert(format == imm_operand, "cannot specify a format");
 777     return;
 778   }
 779   assert(opnd == pc(), "must put operand where relocs can find it");
 780 }
 781 #endif // ASSERT
 782 
 783 void Assembler::emit_operand32(Register reg, Address adr) {
 784   assert(reg->encoding() < 8, "no extended registers");
 785   assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
 786   emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
 787                adr._rspec);
 788 }
 789 
 790 void Assembler::emit_operand(Register reg, Address adr,
 791                              int rip_relative_correction) {
 792   emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
 793                adr._rspec,
 794                rip_relative_correction);
 795 }
 796 
 797 void Assembler::emit_operand(XMMRegister reg, Address adr) {
 798   emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
 799                adr._rspec);
 800 }
 801 
 802 // MMX operations
 803 void Assembler::emit_operand(MMXRegister reg, Address adr) {
 804   assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
 805   emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
 806 }
 807 
 808 // work around gcc (3.2.1-7a) bug
 809 void Assembler::emit_operand(Address adr, MMXRegister reg) {
 810   assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
 811   emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
 812 }
 813 
 814 
 815 void Assembler::emit_farith(int b1, int b2, int i) {
 816   assert(isByte(b1) && isByte(b2), "wrong opcode");
 817   assert(0 <= i &&  i < 8, "illegal stack offset");
 818   emit_byte(b1);
 819   emit_byte(b2 + i);
 820 }
 821 
 822 
 823 // Now the Assembler instructions (identical for 32/64 bits)
 824 
 825 void Assembler::adcl(Address dst, int32_t imm32) {
 826   InstructionMark im(this);
 827   prefix(dst);
 828   emit_arith_operand(0x81, rdx, dst, imm32);
 829 }
 830 
 831 void Assembler::adcl(Address dst, Register src) {
 832   InstructionMark im(this);
 833   prefix(dst, src);
 834   emit_byte(0x11);
 835   emit_operand(src, dst);
 836 }
 837 
 838 void Assembler::adcl(Register dst, int32_t imm32) {
 839   prefix(dst);
 840   emit_arith(0x81, 0xD0, dst, imm32);
 841 }
 842 
 843 void Assembler::adcl(Register dst, Address src) {
 844   InstructionMark im(this);
 845   prefix(src, dst);
 846   emit_byte(0x13);
 847   emit_operand(dst, src);
 848 }
 849 
 850 void Assembler::adcl(Register dst, Register src) {
 851   (void) prefix_and_encode(dst->encoding(), src->encoding());
 852   emit_arith(0x13, 0xC0, dst, src);
 853 }
 854 
 855 void Assembler::addl(Address dst, int32_t imm32) {
 856   InstructionMark im(this);
 857   prefix(dst);
 858   emit_arith_operand(0x81, rax, dst, imm32);
 859 }
 860 
 861 void Assembler::addl(Address dst, Register src) {
 862   InstructionMark im(this);
 863   prefix(dst, src);
 864   emit_byte(0x01);
 865   emit_operand(src, dst);
 866 }
 867 
 868 void Assembler::addl(Register dst, int32_t imm32) {
 869   prefix(dst);
 870   emit_arith(0x81, 0xC0, dst, imm32);
 871 }
 872 
 873 void Assembler::addl(Register dst, Address src) {
 874   InstructionMark im(this);
 875   prefix(src, dst);
 876   emit_byte(0x03);
 877   emit_operand(dst, src);
 878 }
 879 
 880 void Assembler::addl(Register dst, Register src) {
 881   (void) prefix_and_encode(dst->encoding(), src->encoding());
 882   emit_arith(0x03, 0xC0, dst, src);
 883 }
 884 
 885 void Assembler::addr_nop_4() {
 886   // 4 bytes: NOP DWORD PTR [EAX+0]
 887   emit_byte(0x0F);
 888   emit_byte(0x1F);
 889   emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc);
 890   emit_byte(0);    // 8-bits offset (1 byte)
 891 }
 892 
 893 void Assembler::addr_nop_5() {
 894   // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset
 895   emit_byte(0x0F);
 896   emit_byte(0x1F);
 897   emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4);
 898   emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
 899   emit_byte(0);    // 8-bits offset (1 byte)
 900 }
 901 
 902 void Assembler::addr_nop_7() {
 903   // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset
 904   emit_byte(0x0F);
 905   emit_byte(0x1F);
 906   emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc);
 907   emit_long(0);    // 32-bits offset (4 bytes)
 908 }
 909 
 910 void Assembler::addr_nop_8() {
 911   // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset
 912   emit_byte(0x0F);
 913   emit_byte(0x1F);
 914   emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4);
 915   emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
 916   emit_long(0);    // 32-bits offset (4 bytes)
 917 }
 918 
 919 void Assembler::addsd(XMMRegister dst, XMMRegister src) {
 920   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
 921   emit_byte(0xF2);
 922   int encode = prefix_and_encode(dst->encoding(), src->encoding());
 923   emit_byte(0x0F);
 924   emit_byte(0x58);
 925   emit_byte(0xC0 | encode);
 926 }
 927 
 928 void Assembler::addsd(XMMRegister dst, Address src) {
 929   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
 930   InstructionMark im(this);
 931   emit_byte(0xF2);
 932   prefix(src, dst);
 933   emit_byte(0x0F);
 934   emit_byte(0x58);
 935   emit_operand(dst, src);
 936 }
 937 
 938 void Assembler::addss(XMMRegister dst, XMMRegister src) {
 939   NOT_LP64(assert(VM_Version::supports_sse(), ""));
 940   emit_byte(0xF3);
 941   int encode = prefix_and_encode(dst->encoding(), src->encoding());
 942   emit_byte(0x0F);
 943   emit_byte(0x58);
 944   emit_byte(0xC0 | encode);
 945 }
 946 
 947 void Assembler::addss(XMMRegister dst, Address src) {
 948   NOT_LP64(assert(VM_Version::supports_sse(), ""));
 949   InstructionMark im(this);
 950   emit_byte(0xF3);
 951   prefix(src, dst);
 952   emit_byte(0x0F);
 953   emit_byte(0x58);
 954   emit_operand(dst, src);
 955 }
 956 
 957 void Assembler::andl(Register dst, int32_t imm32) {
 958   prefix(dst);
 959   emit_arith(0x81, 0xE0, dst, imm32);
 960 }
 961 
 962 void Assembler::andl(Register dst, Address src) {
 963   InstructionMark im(this);
 964   prefix(src, dst);
 965   emit_byte(0x23);
 966   emit_operand(dst, src);
 967 }
 968 
 969 void Assembler::andl(Register dst, Register src) {
 970   (void) prefix_and_encode(dst->encoding(), src->encoding());
 971   emit_arith(0x23, 0xC0, dst, src);
 972 }
 973 
 974 void Assembler::andpd(XMMRegister dst, Address src) {
 975   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
 976   InstructionMark im(this);
 977   emit_byte(0x66);
 978   prefix(src, dst);
 979   emit_byte(0x0F);
 980   emit_byte(0x54);
 981   emit_operand(dst, src);
 982 }
 983 
 984 void Assembler::bsfl(Register dst, Register src) {
 985   int encode = prefix_and_encode(dst->encoding(), src->encoding());
 986   emit_byte(0x0F);
 987   emit_byte(0xBC);
 988   emit_byte(0xC0 | encode);
 989 }
 990 
 991 void Assembler::bsrl(Register dst, Register src) {
 992   assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT");
 993   int encode = prefix_and_encode(dst->encoding(), src->encoding());
 994   emit_byte(0x0F);
 995   emit_byte(0xBD);
 996   emit_byte(0xC0 | encode);
 997 }
 998 
 999 void Assembler::bswapl(Register reg) { // bswap
1000   int encode = prefix_and_encode(reg->encoding());
1001   emit_byte(0x0F);
1002   emit_byte(0xC8 | encode);
1003 }
1004 
1005 void Assembler::call(Label& L, relocInfo::relocType rtype) {
1006   // suspect disp32 is always good
1007   int operand = LP64_ONLY(disp32_operand) NOT_LP64(imm_operand);
1008 
1009   if (L.is_bound()) {
1010     const int long_size = 5;
1011     int offs = (int)( target(L) - pc() );
1012     assert(offs <= 0, "assembler error");
1013     InstructionMark im(this);
1014     // 1110 1000 #32-bit disp
1015     emit_byte(0xE8);
1016     emit_data(offs - long_size, rtype, operand);
1017   } else {
1018     InstructionMark im(this);
1019     // 1110 1000 #32-bit disp
1020     L.add_patch_at(code(), locator());
1021 
1022     emit_byte(0xE8);
1023     emit_data(int(0), rtype, operand);
1024   }
1025 }
1026 
1027 void Assembler::call(Register dst) {
1028   // This was originally using a 32bit register encoding
1029   // and surely we want 64bit!
1030   // this is a 32bit encoding but in 64bit mode the default
1031   // operand size is 64bit so there is no need for the
1032   // wide prefix. So prefix only happens if we use the
1033   // new registers. Much like push/pop.
1034   int x = offset();
1035   // this may be true but dbx disassembles it as if it
1036   // were 32bits...
1037   // int encode = prefix_and_encode(dst->encoding());
1038   // if (offset() != x) assert(dst->encoding() >= 8, "what?");
1039   int encode = prefixq_and_encode(dst->encoding());
1040 
1041   emit_byte(0xFF);
1042   emit_byte(0xD0 | encode);
1043 }
1044 
1045 
1046 void Assembler::call(Address adr) {
1047   InstructionMark im(this);
1048   prefix(adr);
1049   emit_byte(0xFF);
1050   emit_operand(rdx, adr);
1051 }
1052 
1053 void Assembler::call_literal(address entry, RelocationHolder const& rspec) {
1054   assert(entry != NULL, "call most probably wrong");
1055   InstructionMark im(this);
1056   emit_byte(0xE8);
1057   intptr_t disp = entry - (_code_pos + sizeof(int32_t));
1058   assert(is_simm32(disp), "must be 32bit offset (call2)");
1059   // Technically, should use call32_operand, but this format is
1060   // implied by the fact that we're emitting a call instruction.
1061 
1062   int operand = LP64_ONLY(disp32_operand) NOT_LP64(call32_operand);
1063   emit_data((int) disp, rspec, operand);
1064 }
1065 
1066 void Assembler::cdql() {
1067   emit_byte(0x99);
1068 }
1069 
1070 void Assembler::cmovl(Condition cc, Register dst, Register src) {
1071   NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction"));
1072   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1073   emit_byte(0x0F);
1074   emit_byte(0x40 | cc);
1075   emit_byte(0xC0 | encode);
1076 }
1077 
1078 
1079 void Assembler::cmovl(Condition cc, Register dst, Address src) {
1080   NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction"));
1081   prefix(src, dst);
1082   emit_byte(0x0F);
1083   emit_byte(0x40 | cc);
1084   emit_operand(dst, src);
1085 }
1086 
1087 void Assembler::cmpb(Address dst, int imm8) {
1088   InstructionMark im(this);
1089   prefix(dst);
1090   emit_byte(0x80);
1091   emit_operand(rdi, dst, 1);
1092   emit_byte(imm8);
1093 }
1094 
1095 void Assembler::cmpl(Address dst, int32_t imm32) {
1096   InstructionMark im(this);
1097   prefix(dst);
1098   emit_byte(0x81);
1099   emit_operand(rdi, dst, 4);
1100   emit_long(imm32);
1101 }
1102 
1103 void Assembler::cmpl(Register dst, int32_t imm32) {
1104   prefix(dst);
1105   emit_arith(0x81, 0xF8, dst, imm32);
1106 }
1107 
1108 void Assembler::cmpl(Register dst, Register src) {
1109   (void) prefix_and_encode(dst->encoding(), src->encoding());
1110   emit_arith(0x3B, 0xC0, dst, src);
1111 }
1112 
1113 
1114 void Assembler::cmpl(Register dst, Address  src) {
1115   InstructionMark im(this);
1116   prefix(src, dst);
1117   emit_byte(0x3B);
1118   emit_operand(dst, src);
1119 }
1120 
1121 void Assembler::cmpw(Address dst, int imm16) {
1122   InstructionMark im(this);
1123   assert(!dst.base_needs_rex() && !dst.index_needs_rex(), "no extended registers");
1124   emit_byte(0x66);
1125   emit_byte(0x81);
1126   emit_operand(rdi, dst, 2);
1127   emit_word(imm16);
1128 }
1129 
1130 // The 32-bit cmpxchg compares the value at adr with the contents of rax,
1131 // and stores reg into adr if so; otherwise, the value at adr is loaded into rax,.
1132 // The ZF is set if the compared values were equal, and cleared otherwise.
1133 void Assembler::cmpxchgl(Register reg, Address adr) { // cmpxchg
1134   if (Atomics & 2) {
1135      // caveat: no instructionmark, so this isn't relocatable.
1136      // Emit a synthetic, non-atomic, CAS equivalent.
1137      // Beware.  The synthetic form sets all ICCs, not just ZF.
1138      // cmpxchg r,[m] is equivalent to rax, = CAS (m, rax, r)
1139      cmpl(rax, adr);
1140      movl(rax, adr);
1141      if (reg != rax) {
1142         Label L ;
1143         jcc(Assembler::notEqual, L);
1144         movl(adr, reg);
1145         bind(L);
1146      }
1147   } else {
1148      InstructionMark im(this);
1149      prefix(adr, reg);
1150      emit_byte(0x0F);
1151      emit_byte(0xB1);
1152      emit_operand(reg, adr);
1153   }
1154 }
1155 
1156 void Assembler::comisd(XMMRegister dst, Address src) {
1157   // NOTE: dbx seems to decode this as comiss even though the
1158   // 0x66 is there. Strangly ucomisd comes out correct
1159   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1160   emit_byte(0x66);
1161   comiss(dst, src);
1162 }
1163 
1164 void Assembler::comiss(XMMRegister dst, Address src) {
1165   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1166 
1167   InstructionMark im(this);
1168   prefix(src, dst);
1169   emit_byte(0x0F);
1170   emit_byte(0x2F);
1171   emit_operand(dst, src);
1172 }
1173 
1174 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) {
1175   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1176   emit_byte(0xF3);
1177   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1178   emit_byte(0x0F);
1179   emit_byte(0xE6);
1180   emit_byte(0xC0 | encode);
1181 }
1182 
1183 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) {
1184   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1185   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1186   emit_byte(0x0F);
1187   emit_byte(0x5B);
1188   emit_byte(0xC0 | encode);
1189 }
1190 
1191 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) {
1192   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1193   emit_byte(0xF2);
1194   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1195   emit_byte(0x0F);
1196   emit_byte(0x5A);
1197   emit_byte(0xC0 | encode);
1198 }
1199 
1200 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) {
1201   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1202   emit_byte(0xF2);
1203   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1204   emit_byte(0x0F);
1205   emit_byte(0x2A);
1206   emit_byte(0xC0 | encode);
1207 }
1208 
1209 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) {
1210   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1211   emit_byte(0xF3);
1212   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1213   emit_byte(0x0F);
1214   emit_byte(0x2A);
1215   emit_byte(0xC0 | encode);
1216 }
1217 
1218 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) {
1219   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1220   emit_byte(0xF3);
1221   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1222   emit_byte(0x0F);
1223   emit_byte(0x5A);
1224   emit_byte(0xC0 | encode);
1225 }
1226 
1227 void Assembler::cvttsd2sil(Register dst, XMMRegister src) {
1228   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1229   emit_byte(0xF2);
1230   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1231   emit_byte(0x0F);
1232   emit_byte(0x2C);
1233   emit_byte(0xC0 | encode);
1234 }
1235 
1236 void Assembler::cvttss2sil(Register dst, XMMRegister src) {
1237   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1238   emit_byte(0xF3);
1239   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1240   emit_byte(0x0F);
1241   emit_byte(0x2C);
1242   emit_byte(0xC0 | encode);
1243 }
1244 
1245 void Assembler::decl(Address dst) {
1246   // Don't use it directly. Use MacroAssembler::decrement() instead.
1247   InstructionMark im(this);
1248   prefix(dst);
1249   emit_byte(0xFF);
1250   emit_operand(rcx, dst);
1251 }
1252 
1253 void Assembler::divsd(XMMRegister dst, Address src) {
1254   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1255   InstructionMark im(this);
1256   emit_byte(0xF2);
1257   prefix(src, dst);
1258   emit_byte(0x0F);
1259   emit_byte(0x5E);
1260   emit_operand(dst, src);
1261 }
1262 
1263 void Assembler::divsd(XMMRegister dst, XMMRegister src) {
1264   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1265   emit_byte(0xF2);
1266   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1267   emit_byte(0x0F);
1268   emit_byte(0x5E);
1269   emit_byte(0xC0 | encode);
1270 }
1271 
1272 void Assembler::divss(XMMRegister dst, Address src) {
1273   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1274   InstructionMark im(this);
1275   emit_byte(0xF3);
1276   prefix(src, dst);
1277   emit_byte(0x0F);
1278   emit_byte(0x5E);
1279   emit_operand(dst, src);
1280 }
1281 
1282 void Assembler::divss(XMMRegister dst, XMMRegister src) {
1283   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1284   emit_byte(0xF3);
1285   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1286   emit_byte(0x0F);
1287   emit_byte(0x5E);
1288   emit_byte(0xC0 | encode);
1289 }
1290 
1291 void Assembler::emms() {
1292   NOT_LP64(assert(VM_Version::supports_mmx(), ""));
1293   emit_byte(0x0F);
1294   emit_byte(0x77);
1295 }
1296 
1297 void Assembler::hlt() {
1298   emit_byte(0xF4);
1299 }
1300 
1301 void Assembler::idivl(Register src) {
1302   int encode = prefix_and_encode(src->encoding());
1303   emit_byte(0xF7);
1304   emit_byte(0xF8 | encode);
1305 }
1306 
1307 void Assembler::divl(Register src) { // Unsigned
1308   int encode = prefix_and_encode(src->encoding());
1309   emit_byte(0xF7);
1310   emit_byte(0xF0 | encode);
1311 }
1312 
1313 void Assembler::imull(Register dst, Register src) {
1314   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1315   emit_byte(0x0F);
1316   emit_byte(0xAF);
1317   emit_byte(0xC0 | encode);
1318 }
1319 
1320 
1321 void Assembler::imull(Register dst, Register src, int value) {
1322   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1323   if (is8bit(value)) {
1324     emit_byte(0x6B);
1325     emit_byte(0xC0 | encode);
1326     emit_byte(value & 0xFF);
1327   } else {
1328     emit_byte(0x69);
1329     emit_byte(0xC0 | encode);
1330     emit_long(value);
1331   }
1332 }
1333 
1334 void Assembler::incl(Address dst) {
1335   // Don't use it directly. Use MacroAssembler::increment() instead.
1336   InstructionMark im(this);
1337   prefix(dst);
1338   emit_byte(0xFF);
1339   emit_operand(rax, dst);
1340 }
1341 
1342 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) {
1343   InstructionMark im(this);
1344   relocate(rtype);
1345   assert((0 <= cc) && (cc < 16), "illegal cc");
1346   if (L.is_bound()) {
1347     address dst = target(L);
1348     assert(dst != NULL, "jcc most probably wrong");
1349 
1350     const int short_size = 2;
1351     const int long_size = 6;
1352     intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos;
1353     if (rtype == relocInfo::none && is8bit(offs - short_size)) {
1354       // 0111 tttn #8-bit disp
1355       emit_byte(0x70 | cc);
1356       emit_byte((offs - short_size) & 0xFF);
1357     } else {
1358       // 0000 1111 1000 tttn #32-bit disp
1359       assert(is_simm32(offs - long_size),
1360              "must be 32bit offset (call4)");
1361       emit_byte(0x0F);
1362       emit_byte(0x80 | cc);
1363       emit_long(offs - long_size);
1364     }
1365   } else {
1366     // Note: could eliminate cond. jumps to this jump if condition
1367     //       is the same however, seems to be rather unlikely case.
1368     // Note: use jccb() if label to be bound is very close to get
1369     //       an 8-bit displacement
1370     L.add_patch_at(code(), locator());
1371     emit_byte(0x0F);
1372     emit_byte(0x80 | cc);
1373     emit_long(0);
1374   }
1375 }
1376 
1377 void Assembler::jccb(Condition cc, Label& L) {
1378   if (L.is_bound()) {
1379     const int short_size = 2;
1380     address entry = target(L);
1381     assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)),
1382            "Dispacement too large for a short jmp");
1383     intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos;
1384     // 0111 tttn #8-bit disp
1385     emit_byte(0x70 | cc);
1386     emit_byte((offs - short_size) & 0xFF);
1387   } else {
1388     InstructionMark im(this);
1389     L.add_patch_at(code(), locator());
1390     emit_byte(0x70 | cc);
1391     emit_byte(0);
1392   }
1393 }
1394 
1395 void Assembler::jmp(Address adr) {
1396   InstructionMark im(this);
1397   prefix(adr);
1398   emit_byte(0xFF);
1399   emit_operand(rsp, adr);
1400 }
1401 
1402 void Assembler::jmp(Label& L, relocInfo::relocType rtype) {
1403   if (L.is_bound()) {
1404     address entry = target(L);
1405     assert(entry != NULL, "jmp most probably wrong");
1406     InstructionMark im(this);
1407     const int short_size = 2;
1408     const int long_size = 5;
1409     intptr_t offs = entry - _code_pos;
1410     if (rtype == relocInfo::none && is8bit(offs - short_size)) {
1411       emit_byte(0xEB);
1412       emit_byte((offs - short_size) & 0xFF);
1413     } else {
1414       emit_byte(0xE9);
1415       emit_long(offs - long_size);
1416     }
1417   } else {
1418     // By default, forward jumps are always 32-bit displacements, since
1419     // we can't yet know where the label will be bound.  If you're sure that
1420     // the forward jump will not run beyond 256 bytes, use jmpb to
1421     // force an 8-bit displacement.
1422     InstructionMark im(this);
1423     relocate(rtype);
1424     L.add_patch_at(code(), locator());
1425     emit_byte(0xE9);
1426     emit_long(0);
1427   }
1428 }
1429 
1430 void Assembler::jmp(Register entry) {
1431   int encode = prefix_and_encode(entry->encoding());
1432   emit_byte(0xFF);
1433   emit_byte(0xE0 | encode);
1434 }
1435 
1436 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) {
1437   InstructionMark im(this);
1438   emit_byte(0xE9);
1439   assert(dest != NULL, "must have a target");
1440   intptr_t disp = dest - (_code_pos + sizeof(int32_t));
1441   assert(is_simm32(disp), "must be 32bit offset (jmp)");
1442   emit_data(disp, rspec.reloc(), call32_operand);
1443 }
1444 
1445 void Assembler::jmpb(Label& L) {
1446   if (L.is_bound()) {
1447     const int short_size = 2;
1448     address entry = target(L);
1449     assert(is8bit((entry - _code_pos) + short_size),
1450            "Dispacement too large for a short jmp");
1451     assert(entry != NULL, "jmp most probably wrong");
1452     intptr_t offs = entry - _code_pos;
1453     emit_byte(0xEB);
1454     emit_byte((offs - short_size) & 0xFF);
1455   } else {
1456     InstructionMark im(this);
1457     L.add_patch_at(code(), locator());
1458     emit_byte(0xEB);
1459     emit_byte(0);
1460   }
1461 }
1462 
1463 void Assembler::ldmxcsr( Address src) {
1464   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1465   InstructionMark im(this);
1466   prefix(src);
1467   emit_byte(0x0F);
1468   emit_byte(0xAE);
1469   emit_operand(as_Register(2), src);
1470 }
1471 
1472 void Assembler::leal(Register dst, Address src) {
1473   InstructionMark im(this);
1474 #ifdef _LP64
1475   emit_byte(0x67); // addr32
1476   prefix(src, dst);
1477 #endif // LP64
1478   emit_byte(0x8D);
1479   emit_operand(dst, src);
1480 }
1481 
1482 void Assembler::lock() {
1483   if (Atomics & 1) {
1484      // Emit either nothing, a NOP, or a NOP: prefix
1485      emit_byte(0x90) ;
1486   } else {
1487      emit_byte(0xF0);
1488   }
1489 }
1490 
1491 void Assembler::lzcntl(Register dst, Register src) {
1492   assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR");
1493   emit_byte(0xF3);
1494   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1495   emit_byte(0x0F);
1496   emit_byte(0xBD);
1497   emit_byte(0xC0 | encode);
1498 }
1499 
1500 // Emit mfence instruction
1501 void Assembler::mfence() {
1502   NOT_LP64(assert(VM_Version::supports_sse2(), "unsupported");)
1503   emit_byte( 0x0F );
1504   emit_byte( 0xAE );
1505   emit_byte( 0xF0 );
1506 }
1507 
1508 void Assembler::mov(Register dst, Register src) {
1509   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
1510 }
1511 
1512 void Assembler::movapd(XMMRegister dst, XMMRegister src) {
1513   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1514   int dstenc = dst->encoding();
1515   int srcenc = src->encoding();
1516   emit_byte(0x66);
1517   if (dstenc < 8) {
1518     if (srcenc >= 8) {
1519       prefix(REX_B);
1520       srcenc -= 8;
1521     }
1522   } else {
1523     if (srcenc < 8) {
1524       prefix(REX_R);
1525     } else {
1526       prefix(REX_RB);
1527       srcenc -= 8;
1528     }
1529     dstenc -= 8;
1530   }
1531   emit_byte(0x0F);
1532   emit_byte(0x28);
1533   emit_byte(0xC0 | dstenc << 3 | srcenc);
1534 }
1535 
1536 void Assembler::movaps(XMMRegister dst, XMMRegister src) {
1537   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1538   int dstenc = dst->encoding();
1539   int srcenc = src->encoding();
1540   if (dstenc < 8) {
1541     if (srcenc >= 8) {
1542       prefix(REX_B);
1543       srcenc -= 8;
1544     }
1545   } else {
1546     if (srcenc < 8) {
1547       prefix(REX_R);
1548     } else {
1549       prefix(REX_RB);
1550       srcenc -= 8;
1551     }
1552     dstenc -= 8;
1553   }
1554   emit_byte(0x0F);
1555   emit_byte(0x28);
1556   emit_byte(0xC0 | dstenc << 3 | srcenc);
1557 }
1558 
1559 void Assembler::movb(Register dst, Address src) {
1560   NOT_LP64(assert(dst->has_byte_register(), "must have byte register"));
1561   InstructionMark im(this);
1562   prefix(src, dst, true);
1563   emit_byte(0x8A);
1564   emit_operand(dst, src);
1565 }
1566 
1567 
1568 void Assembler::movb(Address dst, int imm8) {
1569   InstructionMark im(this);
1570    prefix(dst);
1571   emit_byte(0xC6);
1572   emit_operand(rax, dst, 1);
1573   emit_byte(imm8);
1574 }
1575 
1576 
1577 void Assembler::movb(Address dst, Register src) {
1578   assert(src->has_byte_register(), "must have byte register");
1579   InstructionMark im(this);
1580   prefix(dst, src, true);
1581   emit_byte(0x88);
1582   emit_operand(src, dst);
1583 }
1584 
1585 void Assembler::movdl(XMMRegister dst, Register src) {
1586   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1587   emit_byte(0x66);
1588   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1589   emit_byte(0x0F);
1590   emit_byte(0x6E);
1591   emit_byte(0xC0 | encode);
1592 }
1593 
1594 void Assembler::movdl(Register dst, XMMRegister src) {
1595   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1596   emit_byte(0x66);
1597   // swap src/dst to get correct prefix
1598   int encode = prefix_and_encode(src->encoding(), dst->encoding());
1599   emit_byte(0x0F);
1600   emit_byte(0x7E);
1601   emit_byte(0xC0 | encode);
1602 }
1603 
1604 void Assembler::movdl(XMMRegister dst, Address src) {
1605   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1606   InstructionMark im(this);
1607   emit_byte(0x66);
1608   prefix(src, dst);
1609   emit_byte(0x0F);
1610   emit_byte(0x6E);
1611   emit_operand(dst, src);
1612 }
1613 
1614 
1615 void Assembler::movdqa(XMMRegister dst, Address src) {
1616   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1617   InstructionMark im(this);
1618   emit_byte(0x66);
1619   prefix(src, dst);
1620   emit_byte(0x0F);
1621   emit_byte(0x6F);
1622   emit_operand(dst, src);
1623 }
1624 
1625 void Assembler::movdqa(XMMRegister dst, XMMRegister src) {
1626   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1627   emit_byte(0x66);
1628   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1629   emit_byte(0x0F);
1630   emit_byte(0x6F);
1631   emit_byte(0xC0 | encode);
1632 }
1633 
1634 void Assembler::movdqa(Address dst, XMMRegister src) {
1635   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1636   InstructionMark im(this);
1637   emit_byte(0x66);
1638   prefix(dst, src);
1639   emit_byte(0x0F);
1640   emit_byte(0x7F);
1641   emit_operand(src, dst);
1642 }
1643 
1644 void Assembler::movdqu(XMMRegister dst, Address src) {
1645   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1646   InstructionMark im(this);
1647   emit_byte(0xF3);
1648   prefix(src, dst);
1649   emit_byte(0x0F);
1650   emit_byte(0x6F);
1651   emit_operand(dst, src);
1652 }
1653 
1654 void Assembler::movdqu(XMMRegister dst, XMMRegister src) {
1655   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1656   emit_byte(0xF3);
1657   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1658   emit_byte(0x0F);
1659   emit_byte(0x6F);
1660   emit_byte(0xC0 | encode);
1661 }
1662 
1663 void Assembler::movdqu(Address dst, XMMRegister src) {
1664   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1665   InstructionMark im(this);
1666   emit_byte(0xF3);
1667   prefix(dst, src);
1668   emit_byte(0x0F);
1669   emit_byte(0x7F);
1670   emit_operand(src, dst);
1671 }
1672 
1673 // Uses zero extension on 64bit
1674 
1675 void Assembler::movl(Register dst, int32_t imm32) {
1676   int encode = prefix_and_encode(dst->encoding());
1677   emit_byte(0xB8 | encode);
1678   emit_long(imm32);
1679 }
1680 
1681 void Assembler::movl(Register dst, Register src) {
1682   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1683   emit_byte(0x8B);
1684   emit_byte(0xC0 | encode);
1685 }
1686 
1687 void Assembler::movl(Register dst, Address src) {
1688   InstructionMark im(this);
1689   prefix(src, dst);
1690   emit_byte(0x8B);
1691   emit_operand(dst, src);
1692 }
1693 
1694 void Assembler::movl(Address dst, int32_t imm32) {
1695   InstructionMark im(this);
1696   prefix(dst);
1697   emit_byte(0xC7);
1698   emit_operand(rax, dst, 4);
1699   emit_long(imm32);
1700 }
1701 
1702 void Assembler::movl(Address dst, Register src) {
1703   InstructionMark im(this);
1704   prefix(dst, src);
1705   emit_byte(0x89);
1706   emit_operand(src, dst);
1707 }
1708 
1709 // New cpus require to use movsd and movss to avoid partial register stall
1710 // when loading from memory. But for old Opteron use movlpd instead of movsd.
1711 // The selection is done in MacroAssembler::movdbl() and movflt().
1712 void Assembler::movlpd(XMMRegister dst, Address src) {
1713   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1714   InstructionMark im(this);
1715   emit_byte(0x66);
1716   prefix(src, dst);
1717   emit_byte(0x0F);
1718   emit_byte(0x12);
1719   emit_operand(dst, src);
1720 }
1721 
1722 void Assembler::movq( MMXRegister dst, Address src ) {
1723   assert( VM_Version::supports_mmx(), "" );
1724   emit_byte(0x0F);
1725   emit_byte(0x6F);
1726   emit_operand(dst, src);
1727 }
1728 
1729 void Assembler::movq( Address dst, MMXRegister src ) {
1730   assert( VM_Version::supports_mmx(), "" );
1731   emit_byte(0x0F);
1732   emit_byte(0x7F);
1733   // workaround gcc (3.2.1-7a) bug
1734   // In that version of gcc with only an emit_operand(MMX, Address)
1735   // gcc will tail jump and try and reverse the parameters completely
1736   // obliterating dst in the process. By having a version available
1737   // that doesn't need to swap the args at the tail jump the bug is
1738   // avoided.
1739   emit_operand(dst, src);
1740 }
1741 
1742 void Assembler::movq(XMMRegister dst, Address src) {
1743   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1744   InstructionMark im(this);
1745   emit_byte(0xF3);
1746   prefix(src, dst);
1747   emit_byte(0x0F);
1748   emit_byte(0x7E);
1749   emit_operand(dst, src);
1750 }
1751 
1752 void Assembler::movq(Address dst, XMMRegister src) {
1753   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1754   InstructionMark im(this);
1755   emit_byte(0x66);
1756   prefix(dst, src);
1757   emit_byte(0x0F);
1758   emit_byte(0xD6);
1759   emit_operand(src, dst);
1760 }
1761 
1762 void Assembler::movsbl(Register dst, Address src) { // movsxb
1763   InstructionMark im(this);
1764   prefix(src, dst);
1765   emit_byte(0x0F);
1766   emit_byte(0xBE);
1767   emit_operand(dst, src);
1768 }
1769 
1770 void Assembler::movsbl(Register dst, Register src) { // movsxb
1771   NOT_LP64(assert(src->has_byte_register(), "must have byte register"));
1772   int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1773   emit_byte(0x0F);
1774   emit_byte(0xBE);
1775   emit_byte(0xC0 | encode);
1776 }
1777 
1778 void Assembler::movsd(XMMRegister dst, XMMRegister src) {
1779   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1780   emit_byte(0xF2);
1781   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1782   emit_byte(0x0F);
1783   emit_byte(0x10);
1784   emit_byte(0xC0 | encode);
1785 }
1786 
1787 void Assembler::movsd(XMMRegister dst, Address src) {
1788   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1789   InstructionMark im(this);
1790   emit_byte(0xF2);
1791   prefix(src, dst);
1792   emit_byte(0x0F);
1793   emit_byte(0x10);
1794   emit_operand(dst, src);
1795 }
1796 
1797 void Assembler::movsd(Address dst, XMMRegister src) {
1798   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1799   InstructionMark im(this);
1800   emit_byte(0xF2);
1801   prefix(dst, src);
1802   emit_byte(0x0F);
1803   emit_byte(0x11);
1804   emit_operand(src, dst);
1805 }
1806 
1807 void Assembler::movss(XMMRegister dst, XMMRegister src) {
1808   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1809   emit_byte(0xF3);
1810   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1811   emit_byte(0x0F);
1812   emit_byte(0x10);
1813   emit_byte(0xC0 | encode);
1814 }
1815 
1816 void Assembler::movss(XMMRegister dst, Address src) {
1817   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1818   InstructionMark im(this);
1819   emit_byte(0xF3);
1820   prefix(src, dst);
1821   emit_byte(0x0F);
1822   emit_byte(0x10);
1823   emit_operand(dst, src);
1824 }
1825 
1826 void Assembler::movss(Address dst, XMMRegister src) {
1827   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1828   InstructionMark im(this);
1829   emit_byte(0xF3);
1830   prefix(dst, src);
1831   emit_byte(0x0F);
1832   emit_byte(0x11);
1833   emit_operand(src, dst);
1834 }
1835 
1836 void Assembler::movswl(Register dst, Address src) { // movsxw
1837   InstructionMark im(this);
1838   prefix(src, dst);
1839   emit_byte(0x0F);
1840   emit_byte(0xBF);
1841   emit_operand(dst, src);
1842 }
1843 
1844 void Assembler::movswl(Register dst, Register src) { // movsxw
1845   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1846   emit_byte(0x0F);
1847   emit_byte(0xBF);
1848   emit_byte(0xC0 | encode);
1849 }
1850 
1851 void Assembler::movw(Address dst, int imm16) {
1852   InstructionMark im(this);
1853 
1854   emit_byte(0x66); // switch to 16-bit mode
1855   prefix(dst);
1856   emit_byte(0xC7);
1857   emit_operand(rax, dst, 2);
1858   emit_word(imm16);
1859 }
1860 
1861 void Assembler::movw(Register dst, Address src) {
1862   InstructionMark im(this);
1863   emit_byte(0x66);
1864   prefix(src, dst);
1865   emit_byte(0x8B);
1866   emit_operand(dst, src);
1867 }
1868 
1869 void Assembler::movw(Address dst, Register src) {
1870   InstructionMark im(this);
1871   emit_byte(0x66);
1872   prefix(dst, src);
1873   emit_byte(0x89);
1874   emit_operand(src, dst);
1875 }
1876 
1877 void Assembler::movzbl(Register dst, Address src) { // movzxb
1878   InstructionMark im(this);
1879   prefix(src, dst);
1880   emit_byte(0x0F);
1881   emit_byte(0xB6);
1882   emit_operand(dst, src);
1883 }
1884 
1885 void Assembler::movzbl(Register dst, Register src) { // movzxb
1886   NOT_LP64(assert(src->has_byte_register(), "must have byte register"));
1887   int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1888   emit_byte(0x0F);
1889   emit_byte(0xB6);
1890   emit_byte(0xC0 | encode);
1891 }
1892 
1893 void Assembler::movzwl(Register dst, Address src) { // movzxw
1894   InstructionMark im(this);
1895   prefix(src, dst);
1896   emit_byte(0x0F);
1897   emit_byte(0xB7);
1898   emit_operand(dst, src);
1899 }
1900 
1901 void Assembler::movzwl(Register dst, Register src) { // movzxw
1902   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1903   emit_byte(0x0F);
1904   emit_byte(0xB7);
1905   emit_byte(0xC0 | encode);
1906 }
1907 
1908 void Assembler::mull(Address src) {
1909   InstructionMark im(this);
1910   prefix(src);
1911   emit_byte(0xF7);
1912   emit_operand(rsp, src);
1913 }
1914 
1915 void Assembler::mull(Register src) {
1916   int encode = prefix_and_encode(src->encoding());
1917   emit_byte(0xF7);
1918   emit_byte(0xE0 | encode);
1919 }
1920 
1921 void Assembler::mulsd(XMMRegister dst, Address src) {
1922   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1923   InstructionMark im(this);
1924   emit_byte(0xF2);
1925   prefix(src, dst);
1926   emit_byte(0x0F);
1927   emit_byte(0x59);
1928   emit_operand(dst, src);
1929 }
1930 
1931 void Assembler::mulsd(XMMRegister dst, XMMRegister src) {
1932   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1933   emit_byte(0xF2);
1934   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1935   emit_byte(0x0F);
1936   emit_byte(0x59);
1937   emit_byte(0xC0 | encode);
1938 }
1939 
1940 void Assembler::mulss(XMMRegister dst, Address src) {
1941   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1942   InstructionMark im(this);
1943   emit_byte(0xF3);
1944   prefix(src, dst);
1945   emit_byte(0x0F);
1946   emit_byte(0x59);
1947   emit_operand(dst, src);
1948 }
1949 
1950 void Assembler::mulss(XMMRegister dst, XMMRegister src) {
1951   NOT_LP64(assert(VM_Version::supports_sse(), ""));
1952   emit_byte(0xF3);
1953   int encode = prefix_and_encode(dst->encoding(), src->encoding());
1954   emit_byte(0x0F);
1955   emit_byte(0x59);
1956   emit_byte(0xC0 | encode);
1957 }
1958 
1959 void Assembler::negl(Register dst) {
1960   int encode = prefix_and_encode(dst->encoding());
1961   emit_byte(0xF7);
1962   emit_byte(0xD8 | encode);
1963 }
1964 
1965 void Assembler::nop(int i) {
1966 #ifdef ASSERT
1967   assert(i > 0, " ");
1968   // The fancy nops aren't currently recognized by debuggers making it a
1969   // pain to disassemble code while debugging. If asserts are on clearly
1970   // speed is not an issue so simply use the single byte traditional nop
1971   // to do alignment.
1972 
1973   for (; i > 0 ; i--) emit_byte(0x90);
1974   return;
1975 
1976 #endif // ASSERT
1977 
1978   if (UseAddressNop && VM_Version::is_intel()) {
1979     //
1980     // Using multi-bytes nops "0x0F 0x1F [address]" for Intel
1981     //  1: 0x90
1982     //  2: 0x66 0x90
1983     //  3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
1984     //  4: 0x0F 0x1F 0x40 0x00
1985     //  5: 0x0F 0x1F 0x44 0x00 0x00
1986     //  6: 0x66 0x0F 0x1F 0x44 0x00 0x00
1987     //  7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
1988     //  8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1989     //  9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1990     // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1991     // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1992 
1993     // The rest coding is Intel specific - don't use consecutive address nops
1994 
1995     // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1996     // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1997     // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1998     // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1999 
2000     while(i >= 15) {
2001       // For Intel don't generate consecutive addess nops (mix with regular nops)
2002       i -= 15;
2003       emit_byte(0x66);   // size prefix
2004       emit_byte(0x66);   // size prefix
2005       emit_byte(0x66);   // size prefix
2006       addr_nop_8();
2007       emit_byte(0x66);   // size prefix
2008       emit_byte(0x66);   // size prefix
2009       emit_byte(0x66);   // size prefix
2010       emit_byte(0x90);   // nop
2011     }
2012     switch (i) {
2013       case 14:
2014         emit_byte(0x66); // size prefix
2015       case 13:
2016         emit_byte(0x66); // size prefix
2017       case 12:
2018         addr_nop_8();
2019         emit_byte(0x66); // size prefix
2020         emit_byte(0x66); // size prefix
2021         emit_byte(0x66); // size prefix
2022         emit_byte(0x90); // nop
2023         break;
2024       case 11:
2025         emit_byte(0x66); // size prefix
2026       case 10:
2027         emit_byte(0x66); // size prefix
2028       case 9:
2029         emit_byte(0x66); // size prefix
2030       case 8:
2031         addr_nop_8();
2032         break;
2033       case 7:
2034         addr_nop_7();
2035         break;
2036       case 6:
2037         emit_byte(0x66); // size prefix
2038       case 5:
2039         addr_nop_5();
2040         break;
2041       case 4:
2042         addr_nop_4();
2043         break;
2044       case 3:
2045         // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2046         emit_byte(0x66); // size prefix
2047       case 2:
2048         emit_byte(0x66); // size prefix
2049       case 1:
2050         emit_byte(0x90); // nop
2051         break;
2052       default:
2053         assert(i == 0, " ");
2054     }
2055     return;
2056   }
2057   if (UseAddressNop && VM_Version::is_amd()) {
2058     //
2059     // Using multi-bytes nops "0x0F 0x1F [address]" for AMD.
2060     //  1: 0x90
2061     //  2: 0x66 0x90
2062     //  3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
2063     //  4: 0x0F 0x1F 0x40 0x00
2064     //  5: 0x0F 0x1F 0x44 0x00 0x00
2065     //  6: 0x66 0x0F 0x1F 0x44 0x00 0x00
2066     //  7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2067     //  8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2068     //  9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2069     // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2070     // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2071 
2072     // The rest coding is AMD specific - use consecutive address nops
2073 
2074     // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2075     // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2076     // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2077     // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2078     // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2079     //     Size prefixes (0x66) are added for larger sizes
2080 
2081     while(i >= 22) {
2082       i -= 11;
2083       emit_byte(0x66); // size prefix
2084       emit_byte(0x66); // size prefix
2085       emit_byte(0x66); // size prefix
2086       addr_nop_8();
2087     }
2088     // Generate first nop for size between 21-12
2089     switch (i) {
2090       case 21:
2091         i -= 1;
2092         emit_byte(0x66); // size prefix
2093       case 20:
2094       case 19:
2095         i -= 1;
2096         emit_byte(0x66); // size prefix
2097       case 18:
2098       case 17:
2099         i -= 1;
2100         emit_byte(0x66); // size prefix
2101       case 16:
2102       case 15:
2103         i -= 8;
2104         addr_nop_8();
2105         break;
2106       case 14:
2107       case 13:
2108         i -= 7;
2109         addr_nop_7();
2110         break;
2111       case 12:
2112         i -= 6;
2113         emit_byte(0x66); // size prefix
2114         addr_nop_5();
2115         break;
2116       default:
2117         assert(i < 12, " ");
2118     }
2119 
2120     // Generate second nop for size between 11-1
2121     switch (i) {
2122       case 11:
2123         emit_byte(0x66); // size prefix
2124       case 10:
2125         emit_byte(0x66); // size prefix
2126       case 9:
2127         emit_byte(0x66); // size prefix
2128       case 8:
2129         addr_nop_8();
2130         break;
2131       case 7:
2132         addr_nop_7();
2133         break;
2134       case 6:
2135         emit_byte(0x66); // size prefix
2136       case 5:
2137         addr_nop_5();
2138         break;
2139       case 4:
2140         addr_nop_4();
2141         break;
2142       case 3:
2143         // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2144         emit_byte(0x66); // size prefix
2145       case 2:
2146         emit_byte(0x66); // size prefix
2147       case 1:
2148         emit_byte(0x90); // nop
2149         break;
2150       default:
2151         assert(i == 0, " ");
2152     }
2153     return;
2154   }
2155 
2156   // Using nops with size prefixes "0x66 0x90".
2157   // From AMD Optimization Guide:
2158   //  1: 0x90
2159   //  2: 0x66 0x90
2160   //  3: 0x66 0x66 0x90
2161   //  4: 0x66 0x66 0x66 0x90
2162   //  5: 0x66 0x66 0x90 0x66 0x90
2163   //  6: 0x66 0x66 0x90 0x66 0x66 0x90
2164   //  7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90
2165   //  8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90
2166   //  9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2167   // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2168   //
2169   while(i > 12) {
2170     i -= 4;
2171     emit_byte(0x66); // size prefix
2172     emit_byte(0x66);
2173     emit_byte(0x66);
2174     emit_byte(0x90); // nop
2175   }
2176   // 1 - 12 nops
2177   if(i > 8) {
2178     if(i > 9) {
2179       i -= 1;
2180       emit_byte(0x66);
2181     }
2182     i -= 3;
2183     emit_byte(0x66);
2184     emit_byte(0x66);
2185     emit_byte(0x90);
2186   }
2187   // 1 - 8 nops
2188   if(i > 4) {
2189     if(i > 6) {
2190       i -= 1;
2191       emit_byte(0x66);
2192     }
2193     i -= 3;
2194     emit_byte(0x66);
2195     emit_byte(0x66);
2196     emit_byte(0x90);
2197   }
2198   switch (i) {
2199     case 4:
2200       emit_byte(0x66);
2201     case 3:
2202       emit_byte(0x66);
2203     case 2:
2204       emit_byte(0x66);
2205     case 1:
2206       emit_byte(0x90);
2207       break;
2208     default:
2209       assert(i == 0, " ");
2210   }
2211 }
2212 
2213 void Assembler::notl(Register dst) {
2214   int encode = prefix_and_encode(dst->encoding());
2215   emit_byte(0xF7);
2216   emit_byte(0xD0 | encode );
2217 }
2218 
2219 void Assembler::orl(Address dst, int32_t imm32) {
2220   InstructionMark im(this);
2221   prefix(dst);
2222   emit_arith_operand(0x81, rcx, dst, imm32);
2223 }
2224 
2225 void Assembler::orl(Register dst, int32_t imm32) {
2226   prefix(dst);
2227   emit_arith(0x81, 0xC8, dst, imm32);
2228 }
2229 
2230 void Assembler::orl(Register dst, Address src) {
2231   InstructionMark im(this);
2232   prefix(src, dst);
2233   emit_byte(0x0B);
2234   emit_operand(dst, src);
2235 }
2236 
2237 void Assembler::orl(Register dst, Register src) {
2238   (void) prefix_and_encode(dst->encoding(), src->encoding());
2239   emit_arith(0x0B, 0xC0, dst, src);
2240 }
2241 
2242 void Assembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
2243   assert(VM_Version::supports_sse4_2(), "");
2244 
2245   InstructionMark im(this);
2246   emit_byte(0x66);
2247   prefix(src, dst);
2248   emit_byte(0x0F);
2249   emit_byte(0x3A);
2250   emit_byte(0x61);
2251   emit_operand(dst, src);
2252   emit_byte(imm8);
2253 }
2254 
2255 void Assembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
2256   assert(VM_Version::supports_sse4_2(), "");
2257 
2258   emit_byte(0x66);
2259   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2260   emit_byte(0x0F);
2261   emit_byte(0x3A);
2262   emit_byte(0x61);
2263   emit_byte(0xC0 | encode);
2264   emit_byte(imm8);
2265 }
2266 
2267 // generic
2268 void Assembler::pop(Register dst) {
2269   int encode = prefix_and_encode(dst->encoding());
2270   emit_byte(0x58 | encode);
2271 }
2272 
2273 void Assembler::popcntl(Register dst, Address src) {
2274   assert(VM_Version::supports_popcnt(), "must support");
2275   InstructionMark im(this);
2276   emit_byte(0xF3);
2277   prefix(src, dst);
2278   emit_byte(0x0F);
2279   emit_byte(0xB8);
2280   emit_operand(dst, src);
2281 }
2282 
2283 void Assembler::popcntl(Register dst, Register src) {
2284   assert(VM_Version::supports_popcnt(), "must support");
2285   emit_byte(0xF3);
2286   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2287   emit_byte(0x0F);
2288   emit_byte(0xB8);
2289   emit_byte(0xC0 | encode);
2290 }
2291 
2292 void Assembler::popf() {
2293   emit_byte(0x9D);
2294 }
2295 
2296 #ifndef _LP64 // no 32bit push/pop on amd64
2297 void Assembler::popl(Address dst) {
2298   // NOTE: this will adjust stack by 8byte on 64bits
2299   InstructionMark im(this);
2300   prefix(dst);
2301   emit_byte(0x8F);
2302   emit_operand(rax, dst);
2303 }
2304 #endif
2305 
2306 void Assembler::prefetch_prefix(Address src) {
2307   prefix(src);
2308   emit_byte(0x0F);
2309 }
2310 
2311 void Assembler::prefetchnta(Address src) {
2312   NOT_LP64(assert(VM_Version::supports_sse2(), "must support"));
2313   InstructionMark im(this);
2314   prefetch_prefix(src);
2315   emit_byte(0x18);
2316   emit_operand(rax, src); // 0, src
2317 }
2318 
2319 void Assembler::prefetchr(Address src) {
2320   NOT_LP64(assert(VM_Version::supports_3dnow_prefetch(), "must support"));
2321   InstructionMark im(this);
2322   prefetch_prefix(src);
2323   emit_byte(0x0D);
2324   emit_operand(rax, src); // 0, src
2325 }
2326 
2327 void Assembler::prefetcht0(Address src) {
2328   NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2329   InstructionMark im(this);
2330   prefetch_prefix(src);
2331   emit_byte(0x18);
2332   emit_operand(rcx, src); // 1, src
2333 }
2334 
2335 void Assembler::prefetcht1(Address src) {
2336   NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2337   InstructionMark im(this);
2338   prefetch_prefix(src);
2339   emit_byte(0x18);
2340   emit_operand(rdx, src); // 2, src
2341 }
2342 
2343 void Assembler::prefetcht2(Address src) {
2344   NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2345   InstructionMark im(this);
2346   prefetch_prefix(src);
2347   emit_byte(0x18);
2348   emit_operand(rbx, src); // 3, src
2349 }
2350 
2351 void Assembler::prefetchw(Address src) {
2352   NOT_LP64(assert(VM_Version::supports_3dnow_prefetch(), "must support"));
2353   InstructionMark im(this);
2354   prefetch_prefix(src);
2355   emit_byte(0x0D);
2356   emit_operand(rcx, src); // 1, src
2357 }
2358 
2359 void Assembler::prefix(Prefix p) {
2360   a_byte(p);
2361 }
2362 
2363 void Assembler::por(XMMRegister dst, XMMRegister src) {
2364   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2365 
2366   emit_byte(0x66);
2367   int  encode = prefix_and_encode(dst->encoding(), src->encoding());
2368   emit_byte(0x0F);
2369 
2370   emit_byte(0xEB);
2371   emit_byte(0xC0 | encode);
2372 }
2373 
2374 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) {
2375   assert(isByte(mode), "invalid value");
2376   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2377 
2378   emit_byte(0x66);
2379   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2380   emit_byte(0x0F);
2381   emit_byte(0x70);
2382   emit_byte(0xC0 | encode);
2383   emit_byte(mode & 0xFF);
2384 
2385 }
2386 
2387 void Assembler::pshufd(XMMRegister dst, Address src, int mode) {
2388   assert(isByte(mode), "invalid value");
2389   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2390 
2391   InstructionMark im(this);
2392   emit_byte(0x66);
2393   prefix(src, dst);
2394   emit_byte(0x0F);
2395   emit_byte(0x70);
2396   emit_operand(dst, src);
2397   emit_byte(mode & 0xFF);
2398 }
2399 
2400 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
2401   assert(isByte(mode), "invalid value");
2402   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2403 
2404   emit_byte(0xF2);
2405   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2406   emit_byte(0x0F);
2407   emit_byte(0x70);
2408   emit_byte(0xC0 | encode);
2409   emit_byte(mode & 0xFF);
2410 }
2411 
2412 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) {
2413   assert(isByte(mode), "invalid value");
2414   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2415 
2416   InstructionMark im(this);
2417   emit_byte(0xF2);
2418   prefix(src, dst); // QQ new
2419   emit_byte(0x0F);
2420   emit_byte(0x70);
2421   emit_operand(dst, src);
2422   emit_byte(mode & 0xFF);
2423 }
2424 
2425 void Assembler::psrlq(XMMRegister dst, int shift) {
2426   // Shift 64 bit value logically right by specified number of bits.
2427   // HMM Table D-1 says sse2 or mmx.
2428   // Do not confuse it with psrldq SSE2 instruction which
2429   // shifts 128 bit value in xmm register by number of bytes.
2430   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2431 
2432   int encode = prefixq_and_encode(xmm2->encoding(), dst->encoding());
2433   emit_byte(0x66);
2434   emit_byte(0x0F);
2435   emit_byte(0x73);
2436   emit_byte(0xC0 | encode);
2437   emit_byte(shift);
2438 }
2439 
2440 void Assembler::psrldq(XMMRegister dst, int shift) {
2441   // Shift 128 bit value in xmm register by number of bytes.
2442   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2443 
2444   int encode = prefixq_and_encode(xmm3->encoding(), dst->encoding());
2445   emit_byte(0x66);
2446   emit_byte(0x0F);
2447   emit_byte(0x73);
2448   emit_byte(0xC0 | encode);
2449   emit_byte(shift);
2450 }
2451 
2452 void Assembler::ptest(XMMRegister dst, Address src) {
2453   assert(VM_Version::supports_sse4_1(), "");
2454 
2455   InstructionMark im(this);
2456   emit_byte(0x66);
2457   prefix(src, dst);
2458   emit_byte(0x0F);
2459   emit_byte(0x38);
2460   emit_byte(0x17);
2461   emit_operand(dst, src);
2462 }
2463 
2464 void Assembler::ptest(XMMRegister dst, XMMRegister src) {
2465   assert(VM_Version::supports_sse4_1(), "");
2466 
2467   emit_byte(0x66);
2468   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2469   emit_byte(0x0F);
2470   emit_byte(0x38);
2471   emit_byte(0x17);
2472   emit_byte(0xC0 | encode);
2473 }
2474 
2475 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) {
2476   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2477   emit_byte(0x66);
2478   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2479   emit_byte(0x0F);
2480   emit_byte(0x60);
2481   emit_byte(0xC0 | encode);
2482 }
2483 
2484 void Assembler::push(int32_t imm32) {
2485   // in 64bits we push 64bits onto the stack but only
2486   // take a 32bit immediate
2487   emit_byte(0x68);
2488   emit_long(imm32);
2489 }
2490 
2491 void Assembler::push(Register src) {
2492   int encode = prefix_and_encode(src->encoding());
2493 
2494   emit_byte(0x50 | encode);
2495 }
2496 
2497 void Assembler::pushf() {
2498   emit_byte(0x9C);
2499 }
2500 
2501 #ifndef _LP64 // no 32bit push/pop on amd64
2502 void Assembler::pushl(Address src) {
2503   // Note this will push 64bit on 64bit
2504   InstructionMark im(this);
2505   prefix(src);
2506   emit_byte(0xFF);
2507   emit_operand(rsi, src);
2508 }
2509 #endif
2510 
2511 void Assembler::pxor(XMMRegister dst, Address src) {
2512   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2513   InstructionMark im(this);
2514   emit_byte(0x66);
2515   prefix(src, dst);
2516   emit_byte(0x0F);
2517   emit_byte(0xEF);
2518   emit_operand(dst, src);
2519 }
2520 
2521 void Assembler::pxor(XMMRegister dst, XMMRegister src) {
2522   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2523   InstructionMark im(this);
2524   emit_byte(0x66);
2525   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2526   emit_byte(0x0F);
2527   emit_byte(0xEF);
2528   emit_byte(0xC0 | encode);
2529 }
2530 
2531 void Assembler::rcll(Register dst, int imm8) {
2532   assert(isShiftCount(imm8), "illegal shift count");
2533   int encode = prefix_and_encode(dst->encoding());
2534   if (imm8 == 1) {
2535     emit_byte(0xD1);
2536     emit_byte(0xD0 | encode);
2537   } else {
2538     emit_byte(0xC1);
2539     emit_byte(0xD0 | encode);
2540     emit_byte(imm8);
2541   }
2542 }
2543 
2544 // copies data from [esi] to [edi] using rcx pointer sized words
2545 // generic
2546 void Assembler::rep_mov() {
2547   emit_byte(0xF3);
2548   // MOVSQ
2549   LP64_ONLY(prefix(REX_W));
2550   emit_byte(0xA5);
2551 }
2552 
2553 // sets rcx pointer sized words with rax, value at [edi]
2554 // generic
2555 void Assembler::rep_set() { // rep_set
2556   emit_byte(0xF3);
2557   // STOSQ
2558   LP64_ONLY(prefix(REX_W));
2559   emit_byte(0xAB);
2560 }
2561 
2562 // scans rcx pointer sized words at [edi] for occurance of rax,
2563 // generic
2564 void Assembler::repne_scan() { // repne_scan
2565   emit_byte(0xF2);
2566   // SCASQ
2567   LP64_ONLY(prefix(REX_W));
2568   emit_byte(0xAF);
2569 }
2570 
2571 #ifdef _LP64
2572 // scans rcx 4 byte words at [edi] for occurance of rax,
2573 // generic
2574 void Assembler::repne_scanl() { // repne_scan
2575   emit_byte(0xF2);
2576   // SCASL
2577   emit_byte(0xAF);
2578 }
2579 #endif
2580 
2581 void Assembler::ret(int imm16) {
2582   if (imm16 == 0) {
2583     emit_byte(0xC3);
2584   } else {
2585     emit_byte(0xC2);
2586     emit_word(imm16);
2587   }
2588 }
2589 
2590 void Assembler::sahf() {
2591 #ifdef _LP64
2592   // Not supported in 64bit mode
2593   ShouldNotReachHere();
2594 #endif
2595   emit_byte(0x9E);
2596 }
2597 
2598 void Assembler::sarl(Register dst, int imm8) {
2599   int encode = prefix_and_encode(dst->encoding());
2600   assert(isShiftCount(imm8), "illegal shift count");
2601   if (imm8 == 1) {
2602     emit_byte(0xD1);
2603     emit_byte(0xF8 | encode);
2604   } else {
2605     emit_byte(0xC1);
2606     emit_byte(0xF8 | encode);
2607     emit_byte(imm8);
2608   }
2609 }
2610 
2611 void Assembler::sarl(Register dst) {
2612   int encode = prefix_and_encode(dst->encoding());
2613   emit_byte(0xD3);
2614   emit_byte(0xF8 | encode);
2615 }
2616 
2617 void Assembler::sbbl(Address dst, int32_t imm32) {
2618   InstructionMark im(this);
2619   prefix(dst);
2620   emit_arith_operand(0x81, rbx, dst, imm32);
2621 }
2622 
2623 void Assembler::sbbl(Register dst, int32_t imm32) {
2624   prefix(dst);
2625   emit_arith(0x81, 0xD8, dst, imm32);
2626 }
2627 
2628 
2629 void Assembler::sbbl(Register dst, Address src) {
2630   InstructionMark im(this);
2631   prefix(src, dst);
2632   emit_byte(0x1B);
2633   emit_operand(dst, src);
2634 }
2635 
2636 void Assembler::sbbl(Register dst, Register src) {
2637   (void) prefix_and_encode(dst->encoding(), src->encoding());
2638   emit_arith(0x1B, 0xC0, dst, src);
2639 }
2640 
2641 void Assembler::setb(Condition cc, Register dst) {
2642   assert(0 <= cc && cc < 16, "illegal cc");
2643   int encode = prefix_and_encode(dst->encoding(), true);
2644   emit_byte(0x0F);
2645   emit_byte(0x90 | cc);
2646   emit_byte(0xC0 | encode);
2647 }
2648 
2649 void Assembler::shll(Register dst, int imm8) {
2650   assert(isShiftCount(imm8), "illegal shift count");
2651   int encode = prefix_and_encode(dst->encoding());
2652   if (imm8 == 1 ) {
2653     emit_byte(0xD1);
2654     emit_byte(0xE0 | encode);
2655   } else {
2656     emit_byte(0xC1);
2657     emit_byte(0xE0 | encode);
2658     emit_byte(imm8);
2659   }
2660 }
2661 
2662 void Assembler::shll(Register dst) {
2663   int encode = prefix_and_encode(dst->encoding());
2664   emit_byte(0xD3);
2665   emit_byte(0xE0 | encode);
2666 }
2667 
2668 void Assembler::shrl(Register dst, int imm8) {
2669   assert(isShiftCount(imm8), "illegal shift count");
2670   int encode = prefix_and_encode(dst->encoding());
2671   emit_byte(0xC1);
2672   emit_byte(0xE8 | encode);
2673   emit_byte(imm8);
2674 }
2675 
2676 void Assembler::shrl(Register dst) {
2677   int encode = prefix_and_encode(dst->encoding());
2678   emit_byte(0xD3);
2679   emit_byte(0xE8 | encode);
2680 }
2681 
2682 // copies a single word from [esi] to [edi]
2683 void Assembler::smovl() {
2684   emit_byte(0xA5);
2685 }
2686 
2687 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) {
2688   // HMM Table D-1 says sse2
2689   // NOT_LP64(assert(VM_Version::supports_sse(), ""));
2690   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2691   emit_byte(0xF2);
2692   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2693   emit_byte(0x0F);
2694   emit_byte(0x51);
2695   emit_byte(0xC0 | encode);
2696 }
2697 
2698 void Assembler::sqrtsd(XMMRegister dst, Address src) {
2699   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2700   InstructionMark im(this);
2701   emit_byte(0xF2);
2702   prefix(src, dst);
2703   emit_byte(0x0F);
2704   emit_byte(0x51);
2705   emit_operand(dst, src);
2706 }
2707 
2708 void Assembler::sqrtss(XMMRegister dst, XMMRegister src) {
2709   // HMM Table D-1 says sse2
2710   // NOT_LP64(assert(VM_Version::supports_sse(), ""));
2711   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2712   emit_byte(0xF3);
2713   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2714   emit_byte(0x0F);
2715   emit_byte(0x51);
2716   emit_byte(0xC0 | encode);
2717 }
2718 
2719 void Assembler::sqrtss(XMMRegister dst, Address src) {
2720   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2721   InstructionMark im(this);
2722   emit_byte(0xF3);
2723   prefix(src, dst);
2724   emit_byte(0x0F);
2725   emit_byte(0x51);
2726   emit_operand(dst, src);
2727 }
2728 
2729 void Assembler::stmxcsr( Address dst) {
2730   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2731   InstructionMark im(this);
2732   prefix(dst);
2733   emit_byte(0x0F);
2734   emit_byte(0xAE);
2735   emit_operand(as_Register(3), dst);
2736 }
2737 
2738 void Assembler::subl(Address dst, int32_t imm32) {
2739   InstructionMark im(this);
2740   prefix(dst);
2741   emit_arith_operand(0x81, rbp, dst, imm32);
2742 }
2743 
2744 void Assembler::subl(Address dst, Register src) {
2745   InstructionMark im(this);
2746   prefix(dst, src);
2747   emit_byte(0x29);
2748   emit_operand(src, dst);
2749 }
2750 
2751 void Assembler::subl(Register dst, int32_t imm32) {
2752   prefix(dst);
2753   emit_arith(0x81, 0xE8, dst, imm32);
2754 }
2755 
2756 void Assembler::subl(Register dst, Address src) {
2757   InstructionMark im(this);
2758   prefix(src, dst);
2759   emit_byte(0x2B);
2760   emit_operand(dst, src);
2761 }
2762 
2763 void Assembler::subl(Register dst, Register src) {
2764   (void) prefix_and_encode(dst->encoding(), src->encoding());
2765   emit_arith(0x2B, 0xC0, dst, src);
2766 }
2767 
2768 void Assembler::subsd(XMMRegister dst, XMMRegister src) {
2769   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2770   emit_byte(0xF2);
2771   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2772   emit_byte(0x0F);
2773   emit_byte(0x5C);
2774   emit_byte(0xC0 | encode);
2775 }
2776 
2777 void Assembler::subsd(XMMRegister dst, Address src) {
2778   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2779   InstructionMark im(this);
2780   emit_byte(0xF2);
2781   prefix(src, dst);
2782   emit_byte(0x0F);
2783   emit_byte(0x5C);
2784   emit_operand(dst, src);
2785 }
2786 
2787 void Assembler::subss(XMMRegister dst, XMMRegister src) {
2788   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2789   emit_byte(0xF3);
2790   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2791   emit_byte(0x0F);
2792   emit_byte(0x5C);
2793   emit_byte(0xC0 | encode);
2794 }
2795 
2796 void Assembler::subss(XMMRegister dst, Address src) {
2797   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2798   InstructionMark im(this);
2799   emit_byte(0xF3);
2800   prefix(src, dst);
2801   emit_byte(0x0F);
2802   emit_byte(0x5C);
2803   emit_operand(dst, src);
2804 }
2805 
2806 void Assembler::testb(Register dst, int imm8) {
2807   NOT_LP64(assert(dst->has_byte_register(), "must have byte register"));
2808   (void) prefix_and_encode(dst->encoding(), true);
2809   emit_arith_b(0xF6, 0xC0, dst, imm8);
2810 }
2811 
2812 void Assembler::testl(Register dst, int32_t imm32) {
2813   // not using emit_arith because test
2814   // doesn't support sign-extension of
2815   // 8bit operands
2816   int encode = dst->encoding();
2817   if (encode == 0) {
2818     emit_byte(0xA9);
2819   } else {
2820     encode = prefix_and_encode(encode);
2821     emit_byte(0xF7);
2822     emit_byte(0xC0 | encode);
2823   }
2824   emit_long(imm32);
2825 }
2826 
2827 void Assembler::testl(Register dst, Register src) {
2828   (void) prefix_and_encode(dst->encoding(), src->encoding());
2829   emit_arith(0x85, 0xC0, dst, src);
2830 }
2831 
2832 void Assembler::testl(Register dst, Address  src) {
2833   InstructionMark im(this);
2834   prefix(src, dst);
2835   emit_byte(0x85);
2836   emit_operand(dst, src);
2837 }
2838 
2839 void Assembler::ucomisd(XMMRegister dst, Address src) {
2840   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2841   emit_byte(0x66);
2842   ucomiss(dst, src);
2843 }
2844 
2845 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) {
2846   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2847   emit_byte(0x66);
2848   ucomiss(dst, src);
2849 }
2850 
2851 void Assembler::ucomiss(XMMRegister dst, Address src) {
2852   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2853 
2854   InstructionMark im(this);
2855   prefix(src, dst);
2856   emit_byte(0x0F);
2857   emit_byte(0x2E);
2858   emit_operand(dst, src);
2859 }
2860 
2861 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) {
2862   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2863   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2864   emit_byte(0x0F);
2865   emit_byte(0x2E);
2866   emit_byte(0xC0 | encode);
2867 }
2868 
2869 
2870 void Assembler::xaddl(Address dst, Register src) {
2871   InstructionMark im(this);
2872   prefix(dst, src);
2873   emit_byte(0x0F);
2874   emit_byte(0xC1);
2875   emit_operand(src, dst);
2876 }
2877 
2878 void Assembler::xchgl(Register dst, Address src) { // xchg
2879   InstructionMark im(this);
2880   prefix(src, dst);
2881   emit_byte(0x87);
2882   emit_operand(dst, src);
2883 }
2884 
2885 void Assembler::xchgl(Register dst, Register src) {
2886   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2887   emit_byte(0x87);
2888   emit_byte(0xc0 | encode);
2889 }
2890 
2891 void Assembler::xorl(Register dst, int32_t imm32) {
2892   prefix(dst);
2893   emit_arith(0x81, 0xF0, dst, imm32);
2894 }
2895 
2896 void Assembler::xorl(Register dst, Address src) {
2897   InstructionMark im(this);
2898   prefix(src, dst);
2899   emit_byte(0x33);
2900   emit_operand(dst, src);
2901 }
2902 
2903 void Assembler::xorl(Register dst, Register src) {
2904   (void) prefix_and_encode(dst->encoding(), src->encoding());
2905   emit_arith(0x33, 0xC0, dst, src);
2906 }
2907 
2908 void Assembler::xorpd(XMMRegister dst, XMMRegister src) {
2909   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2910   emit_byte(0x66);
2911   xorps(dst, src);
2912 }
2913 
2914 void Assembler::xorpd(XMMRegister dst, Address src) {
2915   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2916   InstructionMark im(this);
2917   emit_byte(0x66);
2918   prefix(src, dst);
2919   emit_byte(0x0F);
2920   emit_byte(0x57);
2921   emit_operand(dst, src);
2922 }
2923 
2924 
2925 void Assembler::xorps(XMMRegister dst, XMMRegister src) {
2926   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2927   int encode = prefix_and_encode(dst->encoding(), src->encoding());
2928   emit_byte(0x0F);
2929   emit_byte(0x57);
2930   emit_byte(0xC0 | encode);
2931 }
2932 
2933 void Assembler::xorps(XMMRegister dst, Address src) {
2934   NOT_LP64(assert(VM_Version::supports_sse(), ""));
2935   InstructionMark im(this);
2936   prefix(src, dst);
2937   emit_byte(0x0F);
2938   emit_byte(0x57);
2939   emit_operand(dst, src);
2940 }
2941 
2942 #ifndef _LP64
2943 // 32bit only pieces of the assembler
2944 
2945 void Assembler::cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec) {
2946   // NO PREFIX AS NEVER 64BIT
2947   InstructionMark im(this);
2948   emit_byte(0x81);
2949   emit_byte(0xF8 | src1->encoding());
2950   emit_data(imm32, rspec, 0);
2951 }
2952 
2953 void Assembler::cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec) {
2954   // NO PREFIX AS NEVER 64BIT (not even 32bit versions of 64bit regs
2955   InstructionMark im(this);
2956   emit_byte(0x81);
2957   emit_operand(rdi, src1);
2958   emit_data(imm32, rspec, 0);
2959 }
2960 
2961 // The 64-bit (32bit platform) cmpxchg compares the value at adr with the contents of rdx:rax,
2962 // and stores rcx:rbx into adr if so; otherwise, the value at adr is loaded
2963 // into rdx:rax.  The ZF is set if the compared values were equal, and cleared otherwise.
2964 void Assembler::cmpxchg8(Address adr) {
2965   InstructionMark im(this);
2966   emit_byte(0x0F);
2967   emit_byte(0xc7);
2968   emit_operand(rcx, adr);
2969 }
2970 
2971 void Assembler::decl(Register dst) {
2972   // Don't use it directly. Use MacroAssembler::decrementl() instead.
2973  emit_byte(0x48 | dst->encoding());
2974 }
2975 
2976 #endif // _LP64
2977 
2978 // 64bit typically doesn't use the x87 but needs to for the trig funcs
2979 
2980 void Assembler::fabs() {
2981   emit_byte(0xD9);
2982   emit_byte(0xE1);
2983 }
2984 
2985 void Assembler::fadd(int i) {
2986   emit_farith(0xD8, 0xC0, i);
2987 }
2988 
2989 void Assembler::fadd_d(Address src) {
2990   InstructionMark im(this);
2991   emit_byte(0xDC);
2992   emit_operand32(rax, src);
2993 }
2994 
2995 void Assembler::fadd_s(Address src) {
2996   InstructionMark im(this);
2997   emit_byte(0xD8);
2998   emit_operand32(rax, src);
2999 }
3000 
3001 void Assembler::fadda(int i) {
3002   emit_farith(0xDC, 0xC0, i);
3003 }
3004 
3005 void Assembler::faddp(int i) {
3006   emit_farith(0xDE, 0xC0, i);
3007 }
3008 
3009 void Assembler::fchs() {
3010   emit_byte(0xD9);
3011   emit_byte(0xE0);
3012 }
3013 
3014 void Assembler::fcom(int i) {
3015   emit_farith(0xD8, 0xD0, i);
3016 }
3017 
3018 void Assembler::fcomp(int i) {
3019   emit_farith(0xD8, 0xD8, i);
3020 }
3021 
3022 void Assembler::fcomp_d(Address src) {
3023   InstructionMark im(this);
3024   emit_byte(0xDC);
3025   emit_operand32(rbx, src);
3026 }
3027 
3028 void Assembler::fcomp_s(Address src) {
3029   InstructionMark im(this);
3030   emit_byte(0xD8);
3031   emit_operand32(rbx, src);
3032 }
3033 
3034 void Assembler::fcompp() {
3035   emit_byte(0xDE);
3036   emit_byte(0xD9);
3037 }
3038 
3039 void Assembler::fcos() {
3040   emit_byte(0xD9);
3041   emit_byte(0xFF);
3042 }
3043 
3044 void Assembler::fdecstp() {
3045   emit_byte(0xD9);
3046   emit_byte(0xF6);
3047 }
3048 
3049 void Assembler::fdiv(int i) {
3050   emit_farith(0xD8, 0xF0, i);
3051 }
3052 
3053 void Assembler::fdiv_d(Address src) {
3054   InstructionMark im(this);
3055   emit_byte(0xDC);
3056   emit_operand32(rsi, src);
3057 }
3058 
3059 void Assembler::fdiv_s(Address src) {
3060   InstructionMark im(this);
3061   emit_byte(0xD8);
3062   emit_operand32(rsi, src);
3063 }
3064 
3065 void Assembler::fdiva(int i) {
3066   emit_farith(0xDC, 0xF8, i);
3067 }
3068 
3069 // Note: The Intel manual (Pentium Processor User's Manual, Vol.3, 1994)
3070 //       is erroneous for some of the floating-point instructions below.
3071 
3072 void Assembler::fdivp(int i) {
3073   emit_farith(0xDE, 0xF8, i);                    // ST(0) <- ST(0) / ST(1) and pop (Intel manual wrong)
3074 }
3075 
3076 void Assembler::fdivr(int i) {
3077   emit_farith(0xD8, 0xF8, i);
3078 }
3079 
3080 void Assembler::fdivr_d(Address src) {
3081   InstructionMark im(this);
3082   emit_byte(0xDC);
3083   emit_operand32(rdi, src);
3084 }
3085 
3086 void Assembler::fdivr_s(Address src) {
3087   InstructionMark im(this);
3088   emit_byte(0xD8);
3089   emit_operand32(rdi, src);
3090 }
3091 
3092 void Assembler::fdivra(int i) {
3093   emit_farith(0xDC, 0xF0, i);
3094 }
3095 
3096 void Assembler::fdivrp(int i) {
3097   emit_farith(0xDE, 0xF0, i);                    // ST(0) <- ST(1) / ST(0) and pop (Intel manual wrong)
3098 }
3099 
3100 void Assembler::ffree(int i) {
3101   emit_farith(0xDD, 0xC0, i);
3102 }
3103 
3104 void Assembler::fild_d(Address adr) {
3105   InstructionMark im(this);
3106   emit_byte(0xDF);
3107   emit_operand32(rbp, adr);
3108 }
3109 
3110 void Assembler::fild_s(Address adr) {
3111   InstructionMark im(this);
3112   emit_byte(0xDB);
3113   emit_operand32(rax, adr);
3114 }
3115 
3116 void Assembler::fincstp() {
3117   emit_byte(0xD9);
3118   emit_byte(0xF7);
3119 }
3120 
3121 void Assembler::finit() {
3122   emit_byte(0x9B);
3123   emit_byte(0xDB);
3124   emit_byte(0xE3);
3125 }
3126 
3127 void Assembler::fist_s(Address adr) {
3128   InstructionMark im(this);
3129   emit_byte(0xDB);
3130   emit_operand32(rdx, adr);
3131 }
3132 
3133 void Assembler::fistp_d(Address adr) {
3134   InstructionMark im(this);
3135   emit_byte(0xDF);
3136   emit_operand32(rdi, adr);
3137 }
3138 
3139 void Assembler::fistp_s(Address adr) {
3140   InstructionMark im(this);
3141   emit_byte(0xDB);
3142   emit_operand32(rbx, adr);
3143 }
3144 
3145 void Assembler::fld1() {
3146   emit_byte(0xD9);
3147   emit_byte(0xE8);
3148 }
3149 
3150 void Assembler::fld_d(Address adr) {
3151   InstructionMark im(this);
3152   emit_byte(0xDD);
3153   emit_operand32(rax, adr);
3154 }
3155 
3156 void Assembler::fld_s(Address adr) {
3157   InstructionMark im(this);
3158   emit_byte(0xD9);
3159   emit_operand32(rax, adr);
3160 }
3161 
3162 
3163 void Assembler::fld_s(int index) {
3164   emit_farith(0xD9, 0xC0, index);
3165 }
3166 
3167 void Assembler::fld_x(Address adr) {
3168   InstructionMark im(this);
3169   emit_byte(0xDB);
3170   emit_operand32(rbp, adr);
3171 }
3172 
3173 void Assembler::fldcw(Address src) {
3174   InstructionMark im(this);
3175   emit_byte(0xd9);
3176   emit_operand32(rbp, src);
3177 }
3178 
3179 void Assembler::fldenv(Address src) {
3180   InstructionMark im(this);
3181   emit_byte(0xD9);
3182   emit_operand32(rsp, src);
3183 }
3184 
3185 void Assembler::fldlg2() {
3186   emit_byte(0xD9);
3187   emit_byte(0xEC);
3188 }
3189 
3190 void Assembler::fldln2() {
3191   emit_byte(0xD9);
3192   emit_byte(0xED);
3193 }
3194 
3195 void Assembler::fldz() {
3196   emit_byte(0xD9);
3197   emit_byte(0xEE);
3198 }
3199 
3200 void Assembler::flog() {
3201   fldln2();
3202   fxch();
3203   fyl2x();
3204 }
3205 
3206 void Assembler::flog10() {
3207   fldlg2();
3208   fxch();
3209   fyl2x();
3210 }
3211 
3212 void Assembler::fmul(int i) {
3213   emit_farith(0xD8, 0xC8, i);
3214 }
3215 
3216 void Assembler::fmul_d(Address src) {
3217   InstructionMark im(this);
3218   emit_byte(0xDC);
3219   emit_operand32(rcx, src);
3220 }
3221 
3222 void Assembler::fmul_s(Address src) {
3223   InstructionMark im(this);
3224   emit_byte(0xD8);
3225   emit_operand32(rcx, src);
3226 }
3227 
3228 void Assembler::fmula(int i) {
3229   emit_farith(0xDC, 0xC8, i);
3230 }
3231 
3232 void Assembler::fmulp(int i) {
3233   emit_farith(0xDE, 0xC8, i);
3234 }
3235 
3236 void Assembler::fnsave(Address dst) {
3237   InstructionMark im(this);
3238   emit_byte(0xDD);
3239   emit_operand32(rsi, dst);
3240 }
3241 
3242 void Assembler::fnstcw(Address src) {
3243   InstructionMark im(this);
3244   emit_byte(0x9B);
3245   emit_byte(0xD9);
3246   emit_operand32(rdi, src);
3247 }
3248 
3249 void Assembler::fnstsw_ax() {
3250   emit_byte(0xdF);
3251   emit_byte(0xE0);
3252 }
3253 
3254 void Assembler::fprem() {
3255   emit_byte(0xD9);
3256   emit_byte(0xF8);
3257 }
3258 
3259 void Assembler::fprem1() {
3260   emit_byte(0xD9);
3261   emit_byte(0xF5);
3262 }
3263 
3264 void Assembler::frstor(Address src) {
3265   InstructionMark im(this);
3266   emit_byte(0xDD);
3267   emit_operand32(rsp, src);
3268 }
3269 
3270 void Assembler::fsin() {
3271   emit_byte(0xD9);
3272   emit_byte(0xFE);
3273 }
3274 
3275 void Assembler::fsqrt() {
3276   emit_byte(0xD9);
3277   emit_byte(0xFA);
3278 }
3279 
3280 void Assembler::fst_d(Address adr) {
3281   InstructionMark im(this);
3282   emit_byte(0xDD);
3283   emit_operand32(rdx, adr);
3284 }
3285 
3286 void Assembler::fst_s(Address adr) {
3287   InstructionMark im(this);
3288   emit_byte(0xD9);
3289   emit_operand32(rdx, adr);
3290 }
3291 
3292 void Assembler::fstp_d(Address adr) {
3293   InstructionMark im(this);
3294   emit_byte(0xDD);
3295   emit_operand32(rbx, adr);
3296 }
3297 
3298 void Assembler::fstp_d(int index) {
3299   emit_farith(0xDD, 0xD8, index);
3300 }
3301 
3302 void Assembler::fstp_s(Address adr) {
3303   InstructionMark im(this);
3304   emit_byte(0xD9);
3305   emit_operand32(rbx, adr);
3306 }
3307 
3308 void Assembler::fstp_x(Address adr) {
3309   InstructionMark im(this);
3310   emit_byte(0xDB);
3311   emit_operand32(rdi, adr);
3312 }
3313 
3314 void Assembler::fsub(int i) {
3315   emit_farith(0xD8, 0xE0, i);
3316 }
3317 
3318 void Assembler::fsub_d(Address src) {
3319   InstructionMark im(this);
3320   emit_byte(0xDC);
3321   emit_operand32(rsp, src);
3322 }
3323 
3324 void Assembler::fsub_s(Address src) {
3325   InstructionMark im(this);
3326   emit_byte(0xD8);
3327   emit_operand32(rsp, src);
3328 }
3329 
3330 void Assembler::fsuba(int i) {
3331   emit_farith(0xDC, 0xE8, i);
3332 }
3333 
3334 void Assembler::fsubp(int i) {
3335   emit_farith(0xDE, 0xE8, i);                    // ST(0) <- ST(0) - ST(1) and pop (Intel manual wrong)
3336 }
3337 
3338 void Assembler::fsubr(int i) {
3339   emit_farith(0xD8, 0xE8, i);
3340 }
3341 
3342 void Assembler::fsubr_d(Address src) {
3343   InstructionMark im(this);
3344   emit_byte(0xDC);
3345   emit_operand32(rbp, src);
3346 }
3347 
3348 void Assembler::fsubr_s(Address src) {
3349   InstructionMark im(this);
3350   emit_byte(0xD8);
3351   emit_operand32(rbp, src);
3352 }
3353 
3354 void Assembler::fsubra(int i) {
3355   emit_farith(0xDC, 0xE0, i);
3356 }
3357 
3358 void Assembler::fsubrp(int i) {
3359   emit_farith(0xDE, 0xE0, i);                    // ST(0) <- ST(1) - ST(0) and pop (Intel manual wrong)
3360 }
3361 
3362 void Assembler::ftan() {
3363   emit_byte(0xD9);
3364   emit_byte(0xF2);
3365   emit_byte(0xDD);
3366   emit_byte(0xD8);
3367 }
3368 
3369 void Assembler::ftst() {
3370   emit_byte(0xD9);
3371   emit_byte(0xE4);
3372 }
3373 
3374 void Assembler::fucomi(int i) {
3375   // make sure the instruction is supported (introduced for P6, together with cmov)
3376   guarantee(VM_Version::supports_cmov(), "illegal instruction");
3377   emit_farith(0xDB, 0xE8, i);
3378 }
3379 
3380 void Assembler::fucomip(int i) {
3381   // make sure the instruction is supported (introduced for P6, together with cmov)
3382   guarantee(VM_Version::supports_cmov(), "illegal instruction");
3383   emit_farith(0xDF, 0xE8, i);
3384 }
3385 
3386 void Assembler::fwait() {
3387   emit_byte(0x9B);
3388 }
3389 
3390 void Assembler::fxch(int i) {
3391   emit_farith(0xD9, 0xC8, i);
3392 }
3393 
3394 void Assembler::fyl2x() {
3395   emit_byte(0xD9);
3396   emit_byte(0xF1);
3397 }
3398 
3399 
3400 #ifndef _LP64
3401 
3402 void Assembler::incl(Register dst) {
3403   // Don't use it directly. Use MacroAssembler::incrementl() instead.
3404  emit_byte(0x40 | dst->encoding());
3405 }
3406 
3407 void Assembler::lea(Register dst, Address src) {
3408   leal(dst, src);
3409 }
3410 
3411 void Assembler::mov_literal32(Address dst, int32_t imm32,  RelocationHolder const& rspec) {
3412   InstructionMark im(this);
3413   emit_byte(0xC7);
3414   emit_operand(rax, dst);
3415   emit_data((int)imm32, rspec, 0);
3416 }
3417 
3418 void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) {
3419   InstructionMark im(this);
3420   int encode = prefix_and_encode(dst->encoding());
3421   emit_byte(0xB8 | encode);
3422   emit_data((int)imm32, rspec, 0);
3423 }
3424 
3425 void Assembler::popa() { // 32bit
3426   emit_byte(0x61);
3427 }
3428 
3429 void Assembler::push_literal32(int32_t imm32, RelocationHolder const& rspec) {
3430   InstructionMark im(this);
3431   emit_byte(0x68);
3432   emit_data(imm32, rspec, 0);
3433 }
3434 
3435 void Assembler::pusha() { // 32bit
3436   emit_byte(0x60);
3437 }
3438 
3439 void Assembler::set_byte_if_not_zero(Register dst) {
3440   emit_byte(0x0F);
3441   emit_byte(0x95);
3442   emit_byte(0xE0 | dst->encoding());
3443 }
3444 
3445 void Assembler::shldl(Register dst, Register src) {
3446   emit_byte(0x0F);
3447   emit_byte(0xA5);
3448   emit_byte(0xC0 | src->encoding() << 3 | dst->encoding());
3449 }
3450 
3451 void Assembler::shrdl(Register dst, Register src) {
3452   emit_byte(0x0F);
3453   emit_byte(0xAD);
3454   emit_byte(0xC0 | src->encoding() << 3 | dst->encoding());
3455 }
3456 
3457 #else // LP64
3458 
3459 void Assembler::set_byte_if_not_zero(Register dst) {
3460   int enc = prefix_and_encode(dst->encoding(), true);
3461   emit_byte(0x0F);
3462   emit_byte(0x95);
3463   emit_byte(0xE0 | enc);
3464 }
3465 
3466 // 64bit only pieces of the assembler
3467 // This should only be used by 64bit instructions that can use rip-relative
3468 // it cannot be used by instructions that want an immediate value.
3469 
3470 bool Assembler::reachable(AddressLiteral adr) {
3471   int64_t disp;
3472   // None will force a 64bit literal to the code stream. Likely a placeholder
3473   // for something that will be patched later and we need to certain it will
3474   // always be reachable.
3475   if (adr.reloc() == relocInfo::none) {
3476     return false;
3477   }
3478   if (adr.reloc() == relocInfo::internal_word_type) {
3479     // This should be rip relative and easily reachable.
3480     return true;
3481   }
3482   if (adr.reloc() == relocInfo::virtual_call_type ||
3483       adr.reloc() == relocInfo::opt_virtual_call_type ||
3484       adr.reloc() == relocInfo::static_call_type ||
3485       adr.reloc() == relocInfo::static_stub_type ) {
3486     // This should be rip relative within the code cache and easily
3487     // reachable until we get huge code caches. (At which point
3488     // ic code is going to have issues).
3489     return true;
3490   }
3491   if (adr.reloc() != relocInfo::external_word_type &&
3492       adr.reloc() != relocInfo::poll_return_type &&  // these are really external_word but need special
3493       adr.reloc() != relocInfo::poll_type &&         // relocs to identify them
3494       adr.reloc() != relocInfo::runtime_call_type ) {
3495     return false;
3496   }
3497 
3498   // Stress the correction code
3499   if (ForceUnreachable) {
3500     // Must be runtimecall reloc, see if it is in the codecache
3501     // Flipping stuff in the codecache to be unreachable causes issues
3502     // with things like inline caches where the additional instructions
3503     // are not handled.
3504     if (CodeCache::find_blob(adr._target) == NULL) {
3505       return false;
3506     }
3507   }
3508   // For external_word_type/runtime_call_type if it is reachable from where we
3509   // are now (possibly a temp buffer) and where we might end up
3510   // anywhere in the codeCache then we are always reachable.
3511   // This would have to change if we ever save/restore shared code
3512   // to be more pessimistic.
3513   disp = (int64_t)adr._target - ((int64_t)CodeCache::low_bound() + sizeof(int));
3514   if (!is_simm32(disp)) return false;
3515   disp = (int64_t)adr._target - ((int64_t)CodeCache::high_bound() + sizeof(int));
3516   if (!is_simm32(disp)) return false;
3517 
3518   disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int));
3519 
3520   // Because rip relative is a disp + address_of_next_instruction and we
3521   // don't know the value of address_of_next_instruction we apply a fudge factor
3522   // to make sure we will be ok no matter the size of the instruction we get placed into.
3523   // We don't have to fudge the checks above here because they are already worst case.
3524 
3525   // 12 == override/rex byte, opcode byte, rm byte, sib byte, a 4-byte disp , 4-byte literal
3526   // + 4 because better safe than sorry.
3527   const int fudge = 12 + 4;
3528   if (disp < 0) {
3529     disp -= fudge;
3530   } else {
3531     disp += fudge;
3532   }
3533   return is_simm32(disp);
3534 }
3535 
3536 // Check if the polling page is not reachable from the code cache using rip-relative
3537 // addressing.
3538 bool Assembler::is_polling_page_far() {
3539   intptr_t addr = (intptr_t)os::get_polling_page();
3540   return !is_simm32(addr - (intptr_t)CodeCache::low_bound()) ||
3541          !is_simm32(addr - (intptr_t)CodeCache::high_bound());
3542 }
3543 
3544 void Assembler::emit_data64(jlong data,
3545                             relocInfo::relocType rtype,
3546                             int format) {
3547   if (rtype == relocInfo::none) {
3548     emit_long64(data);
3549   } else {
3550     emit_data64(data, Relocation::spec_simple(rtype), format);
3551   }
3552 }
3553 
3554 void Assembler::emit_data64(jlong data,
3555                             RelocationHolder const& rspec,
3556                             int format) {
3557   assert(imm_operand == 0, "default format must be immediate in this file");
3558   assert(imm_operand == format, "must be immediate");
3559   assert(inst_mark() != NULL, "must be inside InstructionMark");
3560   // Do not use AbstractAssembler::relocate, which is not intended for
3561   // embedded words.  Instead, relocate to the enclosing instruction.
3562   code_section()->relocate(inst_mark(), rspec, format);
3563 #ifdef ASSERT
3564   check_relocation(rspec, format);
3565 #endif
3566   emit_long64(data);
3567 }
3568 
3569 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) {
3570   if (reg_enc >= 8) {
3571     prefix(REX_B);
3572     reg_enc -= 8;
3573   } else if (byteinst && reg_enc >= 4) {
3574     prefix(REX);
3575   }
3576   return reg_enc;
3577 }
3578 
3579 int Assembler::prefixq_and_encode(int reg_enc) {
3580   if (reg_enc < 8) {
3581     prefix(REX_W);
3582   } else {
3583     prefix(REX_WB);
3584     reg_enc -= 8;
3585   }
3586   return reg_enc;
3587 }
3588 
3589 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) {
3590   if (dst_enc < 8) {
3591     if (src_enc >= 8) {
3592       prefix(REX_B);
3593       src_enc -= 8;
3594     } else if (byteinst && src_enc >= 4) {
3595       prefix(REX);
3596     }
3597   } else {
3598     if (src_enc < 8) {
3599       prefix(REX_R);
3600     } else {
3601       prefix(REX_RB);
3602       src_enc -= 8;
3603     }
3604     dst_enc -= 8;
3605   }
3606   return dst_enc << 3 | src_enc;
3607 }
3608 
3609 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) {
3610   if (dst_enc < 8) {
3611     if (src_enc < 8) {
3612       prefix(REX_W);
3613     } else {
3614       prefix(REX_WB);
3615       src_enc -= 8;
3616     }
3617   } else {
3618     if (src_enc < 8) {
3619       prefix(REX_WR);
3620     } else {
3621       prefix(REX_WRB);
3622       src_enc -= 8;
3623     }
3624     dst_enc -= 8;
3625   }
3626   return dst_enc << 3 | src_enc;
3627 }
3628 
3629 void Assembler::prefix(Register reg) {
3630   if (reg->encoding() >= 8) {
3631     prefix(REX_B);
3632   }
3633 }
3634 
3635 void Assembler::prefix(Address adr) {
3636   if (adr.base_needs_rex()) {
3637     if (adr.index_needs_rex()) {
3638       prefix(REX_XB);
3639     } else {
3640       prefix(REX_B);
3641     }
3642   } else {
3643     if (adr.index_needs_rex()) {
3644       prefix(REX_X);
3645     }
3646   }
3647 }
3648 
3649 void Assembler::prefixq(Address adr) {
3650   if (adr.base_needs_rex()) {
3651     if (adr.index_needs_rex()) {
3652       prefix(REX_WXB);
3653     } else {
3654       prefix(REX_WB);
3655     }
3656   } else {
3657     if (adr.index_needs_rex()) {
3658       prefix(REX_WX);
3659     } else {
3660       prefix(REX_W);
3661     }
3662   }
3663 }
3664 
3665 
3666 void Assembler::prefix(Address adr, Register reg, bool byteinst) {
3667   if (reg->encoding() < 8) {
3668     if (adr.base_needs_rex()) {
3669       if (adr.index_needs_rex()) {
3670         prefix(REX_XB);
3671       } else {
3672         prefix(REX_B);
3673       }
3674     } else {
3675       if (adr.index_needs_rex()) {
3676         prefix(REX_X);
3677       } else if (reg->encoding() >= 4 ) {
3678         prefix(REX);
3679       }
3680     }
3681   } else {
3682     if (adr.base_needs_rex()) {
3683       if (adr.index_needs_rex()) {
3684         prefix(REX_RXB);
3685       } else {
3686         prefix(REX_RB);
3687       }
3688     } else {
3689       if (adr.index_needs_rex()) {
3690         prefix(REX_RX);
3691       } else {
3692         prefix(REX_R);
3693       }
3694     }
3695   }
3696 }
3697 
3698 void Assembler::prefixq(Address adr, Register src) {
3699   if (src->encoding() < 8) {
3700     if (adr.base_needs_rex()) {
3701       if (adr.index_needs_rex()) {
3702         prefix(REX_WXB);
3703       } else {
3704         prefix(REX_WB);
3705       }
3706     } else {
3707       if (adr.index_needs_rex()) {
3708         prefix(REX_WX);
3709       } else {
3710         prefix(REX_W);
3711       }
3712     }
3713   } else {
3714     if (adr.base_needs_rex()) {
3715       if (adr.index_needs_rex()) {
3716         prefix(REX_WRXB);
3717       } else {
3718         prefix(REX_WRB);
3719       }
3720     } else {
3721       if (adr.index_needs_rex()) {
3722         prefix(REX_WRX);
3723       } else {
3724         prefix(REX_WR);
3725       }
3726     }
3727   }
3728 }
3729 
3730 void Assembler::prefix(Address adr, XMMRegister reg) {
3731   if (reg->encoding() < 8) {
3732     if (adr.base_needs_rex()) {
3733       if (adr.index_needs_rex()) {
3734         prefix(REX_XB);
3735       } else {
3736         prefix(REX_B);
3737       }
3738     } else {
3739       if (adr.index_needs_rex()) {
3740         prefix(REX_X);
3741       }
3742     }
3743   } else {
3744     if (adr.base_needs_rex()) {
3745       if (adr.index_needs_rex()) {
3746         prefix(REX_RXB);
3747       } else {
3748         prefix(REX_RB);
3749       }
3750     } else {
3751       if (adr.index_needs_rex()) {
3752         prefix(REX_RX);
3753       } else {
3754         prefix(REX_R);
3755       }
3756     }
3757   }
3758 }
3759 
3760 void Assembler::adcq(Register dst, int32_t imm32) {
3761   (void) prefixq_and_encode(dst->encoding());
3762   emit_arith(0x81, 0xD0, dst, imm32);
3763 }
3764 
3765 void Assembler::adcq(Register dst, Address src) {
3766   InstructionMark im(this);
3767   prefixq(src, dst);
3768   emit_byte(0x13);
3769   emit_operand(dst, src);
3770 }
3771 
3772 void Assembler::adcq(Register dst, Register src) {
3773   (int) prefixq_and_encode(dst->encoding(), src->encoding());
3774   emit_arith(0x13, 0xC0, dst, src);
3775 }
3776 
3777 void Assembler::addq(Address dst, int32_t imm32) {
3778   InstructionMark im(this);
3779   prefixq(dst);
3780   emit_arith_operand(0x81, rax, dst,imm32);
3781 }
3782 
3783 void Assembler::addq(Address dst, Register src) {
3784   InstructionMark im(this);
3785   prefixq(dst, src);
3786   emit_byte(0x01);
3787   emit_operand(src, dst);
3788 }
3789 
3790 void Assembler::addq(Register dst, int32_t imm32) {
3791   (void) prefixq_and_encode(dst->encoding());
3792   emit_arith(0x81, 0xC0, dst, imm32);
3793 }
3794 
3795 void Assembler::addq(Register dst, Address src) {
3796   InstructionMark im(this);
3797   prefixq(src, dst);
3798   emit_byte(0x03);
3799   emit_operand(dst, src);
3800 }
3801 
3802 void Assembler::addq(Register dst, Register src) {
3803   (void) prefixq_and_encode(dst->encoding(), src->encoding());
3804   emit_arith(0x03, 0xC0, dst, src);
3805 }
3806 
3807 void Assembler::andq(Register dst, int32_t imm32) {
3808   (void) prefixq_and_encode(dst->encoding());
3809   emit_arith(0x81, 0xE0, dst, imm32);
3810 }
3811 
3812 void Assembler::andq(Register dst, Address src) {
3813   InstructionMark im(this);
3814   prefixq(src, dst);
3815   emit_byte(0x23);
3816   emit_operand(dst, src);
3817 }
3818 
3819 void Assembler::andq(Register dst, Register src) {
3820   (int) prefixq_and_encode(dst->encoding(), src->encoding());
3821   emit_arith(0x23, 0xC0, dst, src);
3822 }
3823 
3824 void Assembler::bsfq(Register dst, Register src) {
3825   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3826   emit_byte(0x0F);
3827   emit_byte(0xBC);
3828   emit_byte(0xC0 | encode);
3829 }
3830 
3831 void Assembler::bsrq(Register dst, Register src) {
3832   assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT");
3833   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3834   emit_byte(0x0F);
3835   emit_byte(0xBD);
3836   emit_byte(0xC0 | encode);
3837 }
3838 
3839 void Assembler::bswapq(Register reg) {
3840   int encode = prefixq_and_encode(reg->encoding());
3841   emit_byte(0x0F);
3842   emit_byte(0xC8 | encode);
3843 }
3844 
3845 void Assembler::cdqq() {
3846   prefix(REX_W);
3847   emit_byte(0x99);
3848 }
3849 
3850 void Assembler::clflush(Address adr) {
3851   prefix(adr);
3852   emit_byte(0x0F);
3853   emit_byte(0xAE);
3854   emit_operand(rdi, adr);
3855 }
3856 
3857 void Assembler::cmovq(Condition cc, Register dst, Register src) {
3858   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3859   emit_byte(0x0F);
3860   emit_byte(0x40 | cc);
3861   emit_byte(0xC0 | encode);
3862 }
3863 
3864 void Assembler::cmovq(Condition cc, Register dst, Address src) {
3865   InstructionMark im(this);
3866   prefixq(src, dst);
3867   emit_byte(0x0F);
3868   emit_byte(0x40 | cc);
3869   emit_operand(dst, src);
3870 }
3871 
3872 void Assembler::cmpq(Address dst, int32_t imm32) {
3873   InstructionMark im(this);
3874   prefixq(dst);
3875   emit_byte(0x81);
3876   emit_operand(rdi, dst, 4);
3877   emit_long(imm32);
3878 }
3879 
3880 void Assembler::cmpq(Register dst, int32_t imm32) {
3881   (void) prefixq_and_encode(dst->encoding());
3882   emit_arith(0x81, 0xF8, dst, imm32);
3883 }
3884 
3885 void Assembler::cmpq(Address dst, Register src) {
3886   InstructionMark im(this);
3887   prefixq(dst, src);
3888   emit_byte(0x3B);
3889   emit_operand(src, dst);
3890 }
3891 
3892 void Assembler::cmpq(Register dst, Register src) {
3893   (void) prefixq_and_encode(dst->encoding(), src->encoding());
3894   emit_arith(0x3B, 0xC0, dst, src);
3895 }
3896 
3897 void Assembler::cmpq(Register dst, Address  src) {
3898   InstructionMark im(this);
3899   prefixq(src, dst);
3900   emit_byte(0x3B);
3901   emit_operand(dst, src);
3902 }
3903 
3904 void Assembler::cmpxchgq(Register reg, Address adr) {
3905   InstructionMark im(this);
3906   prefixq(adr, reg);
3907   emit_byte(0x0F);
3908   emit_byte(0xB1);
3909   emit_operand(reg, adr);
3910 }
3911 
3912 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) {
3913   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
3914   emit_byte(0xF2);
3915   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3916   emit_byte(0x0F);
3917   emit_byte(0x2A);
3918   emit_byte(0xC0 | encode);
3919 }
3920 
3921 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) {
3922   NOT_LP64(assert(VM_Version::supports_sse(), ""));
3923   emit_byte(0xF3);
3924   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3925   emit_byte(0x0F);
3926   emit_byte(0x2A);
3927   emit_byte(0xC0 | encode);
3928 }
3929 
3930 void Assembler::cvttsd2siq(Register dst, XMMRegister src) {
3931   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
3932   emit_byte(0xF2);
3933   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3934   emit_byte(0x0F);
3935   emit_byte(0x2C);
3936   emit_byte(0xC0 | encode);
3937 }
3938 
3939 void Assembler::cvttss2siq(Register dst, XMMRegister src) {
3940   NOT_LP64(assert(VM_Version::supports_sse(), ""));
3941   emit_byte(0xF3);
3942   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3943   emit_byte(0x0F);
3944   emit_byte(0x2C);
3945   emit_byte(0xC0 | encode);
3946 }
3947 
3948 void Assembler::decl(Register dst) {
3949   // Don't use it directly. Use MacroAssembler::decrementl() instead.
3950   // Use two-byte form (one-byte form is a REX prefix in 64-bit mode)
3951   int encode = prefix_and_encode(dst->encoding());
3952   emit_byte(0xFF);
3953   emit_byte(0xC8 | encode);
3954 }
3955 
3956 void Assembler::decq(Register dst) {
3957   // Don't use it directly. Use MacroAssembler::decrementq() instead.
3958   // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
3959   int encode = prefixq_and_encode(dst->encoding());
3960   emit_byte(0xFF);
3961   emit_byte(0xC8 | encode);
3962 }
3963 
3964 void Assembler::decq(Address dst) {
3965   // Don't use it directly. Use MacroAssembler::decrementq() instead.
3966   InstructionMark im(this);
3967   prefixq(dst);
3968   emit_byte(0xFF);
3969   emit_operand(rcx, dst);
3970 }
3971 
3972 void Assembler::fxrstor(Address src) {
3973   prefixq(src);
3974   emit_byte(0x0F);
3975   emit_byte(0xAE);
3976   emit_operand(as_Register(1), src);
3977 }
3978 
3979 void Assembler::fxsave(Address dst) {
3980   prefixq(dst);
3981   emit_byte(0x0F);
3982   emit_byte(0xAE);
3983   emit_operand(as_Register(0), dst);
3984 }
3985 
3986 void Assembler::idivq(Register src) {
3987   int encode = prefixq_and_encode(src->encoding());
3988   emit_byte(0xF7);
3989   emit_byte(0xF8 | encode);
3990 }
3991 
3992 void Assembler::imulq(Register dst, Register src) {
3993   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3994   emit_byte(0x0F);
3995   emit_byte(0xAF);
3996   emit_byte(0xC0 | encode);
3997 }
3998 
3999 void Assembler::imulq(Register dst, Register src, int value) {
4000   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4001   if (is8bit(value)) {
4002     emit_byte(0x6B);
4003     emit_byte(0xC0 | encode);
4004     emit_byte(value & 0xFF);
4005   } else {
4006     emit_byte(0x69);
4007     emit_byte(0xC0 | encode);
4008     emit_long(value);
4009   }
4010 }
4011 
4012 void Assembler::incl(Register dst) {
4013   // Don't use it directly. Use MacroAssembler::incrementl() instead.
4014   // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
4015   int encode = prefix_and_encode(dst->encoding());
4016   emit_byte(0xFF);
4017   emit_byte(0xC0 | encode);
4018 }
4019 
4020 void Assembler::incq(Register dst) {
4021   // Don't use it directly. Use MacroAssembler::incrementq() instead.
4022   // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
4023   int encode = prefixq_and_encode(dst->encoding());
4024   emit_byte(0xFF);
4025   emit_byte(0xC0 | encode);
4026 }
4027 
4028 void Assembler::incq(Address dst) {
4029   // Don't use it directly. Use MacroAssembler::incrementq() instead.
4030   InstructionMark im(this);
4031   prefixq(dst);
4032   emit_byte(0xFF);
4033   emit_operand(rax, dst);
4034 }
4035 
4036 void Assembler::lea(Register dst, Address src) {
4037   leaq(dst, src);
4038 }
4039 
4040 void Assembler::leaq(Register dst, Address src) {
4041   InstructionMark im(this);
4042   prefixq(src, dst);
4043   emit_byte(0x8D);
4044   emit_operand(dst, src);
4045 }
4046 
4047 void Assembler::mov64(Register dst, int64_t imm64) {
4048   InstructionMark im(this);
4049   int encode = prefixq_and_encode(dst->encoding());
4050   emit_byte(0xB8 | encode);
4051   emit_long64(imm64);
4052 }
4053 
4054 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) {
4055   InstructionMark im(this);
4056   int encode = prefixq_and_encode(dst->encoding());
4057   emit_byte(0xB8 | encode);
4058   emit_data64(imm64, rspec);
4059 }
4060 
4061 void Assembler::mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec) {
4062   InstructionMark im(this);
4063   int encode = prefix_and_encode(dst->encoding());
4064   emit_byte(0xB8 | encode);
4065   emit_data((int)imm32, rspec, narrow_oop_operand);
4066 }
4067 
4068 void Assembler::mov_narrow_oop(Address dst, int32_t imm32,  RelocationHolder const& rspec) {
4069   InstructionMark im(this);
4070   prefix(dst);
4071   emit_byte(0xC7);
4072   emit_operand(rax, dst, 4);
4073   emit_data((int)imm32, rspec, narrow_oop_operand);
4074 }
4075 
4076 void Assembler::cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec) {
4077   InstructionMark im(this);
4078   int encode = prefix_and_encode(src1->encoding());
4079   emit_byte(0x81);
4080   emit_byte(0xF8 | encode);
4081   emit_data((int)imm32, rspec, narrow_oop_operand);
4082 }
4083 
4084 void Assembler::cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec) {
4085   InstructionMark im(this);
4086   prefix(src1);
4087   emit_byte(0x81);
4088   emit_operand(rax, src1, 4);
4089   emit_data((int)imm32, rspec, narrow_oop_operand);
4090 }
4091 
4092 void Assembler::lzcntq(Register dst, Register src) {
4093   assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR");
4094   emit_byte(0xF3);
4095   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4096   emit_byte(0x0F);
4097   emit_byte(0xBD);
4098   emit_byte(0xC0 | encode);
4099 }
4100 
4101 void Assembler::movdq(XMMRegister dst, Register src) {
4102   // table D-1 says MMX/SSE2
4103   NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), ""));
4104   emit_byte(0x66);
4105   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4106   emit_byte(0x0F);
4107   emit_byte(0x6E);
4108   emit_byte(0xC0 | encode);
4109 }
4110 
4111 void Assembler::movdq(Register dst, XMMRegister src) {
4112   // table D-1 says MMX/SSE2
4113   NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), ""));
4114   emit_byte(0x66);
4115   // swap src/dst to get correct prefix
4116   int encode = prefixq_and_encode(src->encoding(), dst->encoding());
4117   emit_byte(0x0F);
4118   emit_byte(0x7E);
4119   emit_byte(0xC0 | encode);
4120 }
4121 
4122 void Assembler::movq(Register dst, Register src) {
4123   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4124   emit_byte(0x8B);
4125   emit_byte(0xC0 | encode);
4126 }
4127 
4128 void Assembler::movq(Register dst, Address src) {
4129   InstructionMark im(this);
4130   prefixq(src, dst);
4131   emit_byte(0x8B);
4132   emit_operand(dst, src);
4133 }
4134 
4135 void Assembler::movq(Address dst, Register src) {
4136   InstructionMark im(this);
4137   prefixq(dst, src);
4138   emit_byte(0x89);
4139   emit_operand(src, dst);
4140 }
4141 
4142 void Assembler::movsbq(Register dst, Address src) {
4143   InstructionMark im(this);
4144   prefixq(src, dst);
4145   emit_byte(0x0F);
4146   emit_byte(0xBE);
4147   emit_operand(dst, src);
4148 }
4149 
4150 void Assembler::movsbq(Register dst, Register src) {
4151   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4152   emit_byte(0x0F);
4153   emit_byte(0xBE);
4154   emit_byte(0xC0 | encode);
4155 }
4156 
4157 void Assembler::movslq(Register dst, int32_t imm32) {
4158   // dbx shows movslq(rcx, 3) as movq     $0x0000000049000000,(%rbx)
4159   // and movslq(r8, 3); as movl     $0x0000000048000000,(%rbx)
4160   // as a result we shouldn't use until tested at runtime...
4161   ShouldNotReachHere();
4162   InstructionMark im(this);
4163   int encode = prefixq_and_encode(dst->encoding());
4164   emit_byte(0xC7 | encode);
4165   emit_long(imm32);
4166 }
4167 
4168 void Assembler::movslq(Address dst, int32_t imm32) {
4169   assert(is_simm32(imm32), "lost bits");
4170   InstructionMark im(this);
4171   prefixq(dst);
4172   emit_byte(0xC7);
4173   emit_operand(rax, dst, 4);
4174   emit_long(imm32);
4175 }
4176 
4177 void Assembler::movslq(Register dst, Address src) {
4178   InstructionMark im(this);
4179   prefixq(src, dst);
4180   emit_byte(0x63);
4181   emit_operand(dst, src);
4182 }
4183 
4184 void Assembler::movslq(Register dst, Register src) {
4185   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4186   emit_byte(0x63);
4187   emit_byte(0xC0 | encode);
4188 }
4189 
4190 void Assembler::movswq(Register dst, Address src) {
4191   InstructionMark im(this);
4192   prefixq(src, dst);
4193   emit_byte(0x0F);
4194   emit_byte(0xBF);
4195   emit_operand(dst, src);
4196 }
4197 
4198 void Assembler::movswq(Register dst, Register src) {
4199   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4200   emit_byte(0x0F);
4201   emit_byte(0xBF);
4202   emit_byte(0xC0 | encode);
4203 }
4204 
4205 void Assembler::movzbq(Register dst, Address src) {
4206   InstructionMark im(this);
4207   prefixq(src, dst);
4208   emit_byte(0x0F);
4209   emit_byte(0xB6);
4210   emit_operand(dst, src);
4211 }
4212 
4213 void Assembler::movzbq(Register dst, Register src) {
4214   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4215   emit_byte(0x0F);
4216   emit_byte(0xB6);
4217   emit_byte(0xC0 | encode);
4218 }
4219 
4220 void Assembler::movzwq(Register dst, Address src) {
4221   InstructionMark im(this);
4222   prefixq(src, dst);
4223   emit_byte(0x0F);
4224   emit_byte(0xB7);
4225   emit_operand(dst, src);
4226 }
4227 
4228 void Assembler::movzwq(Register dst, Register src) {
4229   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4230   emit_byte(0x0F);
4231   emit_byte(0xB7);
4232   emit_byte(0xC0 | encode);
4233 }
4234 
4235 void Assembler::negq(Register dst) {
4236   int encode = prefixq_and_encode(dst->encoding());
4237   emit_byte(0xF7);
4238   emit_byte(0xD8 | encode);
4239 }
4240 
4241 void Assembler::notq(Register dst) {
4242   int encode = prefixq_and_encode(dst->encoding());
4243   emit_byte(0xF7);
4244   emit_byte(0xD0 | encode);
4245 }
4246 
4247 void Assembler::orq(Address dst, int32_t imm32) {
4248   InstructionMark im(this);
4249   prefixq(dst);
4250   emit_byte(0x81);
4251   emit_operand(rcx, dst, 4);
4252   emit_long(imm32);
4253 }
4254 
4255 void Assembler::orq(Register dst, int32_t imm32) {
4256   (void) prefixq_and_encode(dst->encoding());
4257   emit_arith(0x81, 0xC8, dst, imm32);
4258 }
4259 
4260 void Assembler::orq(Register dst, Address src) {
4261   InstructionMark im(this);
4262   prefixq(src, dst);
4263   emit_byte(0x0B);
4264   emit_operand(dst, src);
4265 }
4266 
4267 void Assembler::orq(Register dst, Register src) {
4268   (void) prefixq_and_encode(dst->encoding(), src->encoding());
4269   emit_arith(0x0B, 0xC0, dst, src);
4270 }
4271 
4272 void Assembler::popa() { // 64bit
4273   movq(r15, Address(rsp, 0));
4274   movq(r14, Address(rsp, wordSize));
4275   movq(r13, Address(rsp, 2 * wordSize));
4276   movq(r12, Address(rsp, 3 * wordSize));
4277   movq(r11, Address(rsp, 4 * wordSize));
4278   movq(r10, Address(rsp, 5 * wordSize));
4279   movq(r9,  Address(rsp, 6 * wordSize));
4280   movq(r8,  Address(rsp, 7 * wordSize));
4281   movq(rdi, Address(rsp, 8 * wordSize));
4282   movq(rsi, Address(rsp, 9 * wordSize));
4283   movq(rbp, Address(rsp, 10 * wordSize));
4284   // skip rsp
4285   movq(rbx, Address(rsp, 12 * wordSize));
4286   movq(rdx, Address(rsp, 13 * wordSize));
4287   movq(rcx, Address(rsp, 14 * wordSize));
4288   movq(rax, Address(rsp, 15 * wordSize));
4289 
4290   addq(rsp, 16 * wordSize);
4291 }
4292 
4293 void Assembler::popcntq(Register dst, Address src) {
4294   assert(VM_Version::supports_popcnt(), "must support");
4295   InstructionMark im(this);
4296   emit_byte(0xF3);
4297   prefixq(src, dst);
4298   emit_byte(0x0F);
4299   emit_byte(0xB8);
4300   emit_operand(dst, src);
4301 }
4302 
4303 void Assembler::popcntq(Register dst, Register src) {
4304   assert(VM_Version::supports_popcnt(), "must support");
4305   emit_byte(0xF3);
4306   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4307   emit_byte(0x0F);
4308   emit_byte(0xB8);
4309   emit_byte(0xC0 | encode);
4310 }
4311 
4312 void Assembler::popq(Address dst) {
4313   InstructionMark im(this);
4314   prefixq(dst);
4315   emit_byte(0x8F);
4316   emit_operand(rax, dst);
4317 }
4318 
4319 void Assembler::pusha() { // 64bit
4320   // we have to store original rsp.  ABI says that 128 bytes
4321   // below rsp are local scratch.
4322   movq(Address(rsp, -5 * wordSize), rsp);
4323 
4324   subq(rsp, 16 * wordSize);
4325 
4326   movq(Address(rsp, 15 * wordSize), rax);
4327   movq(Address(rsp, 14 * wordSize), rcx);
4328   movq(Address(rsp, 13 * wordSize), rdx);
4329   movq(Address(rsp, 12 * wordSize), rbx);
4330   // skip rsp
4331   movq(Address(rsp, 10 * wordSize), rbp);
4332   movq(Address(rsp, 9 * wordSize), rsi);
4333   movq(Address(rsp, 8 * wordSize), rdi);
4334   movq(Address(rsp, 7 * wordSize), r8);
4335   movq(Address(rsp, 6 * wordSize), r9);
4336   movq(Address(rsp, 5 * wordSize), r10);
4337   movq(Address(rsp, 4 * wordSize), r11);
4338   movq(Address(rsp, 3 * wordSize), r12);
4339   movq(Address(rsp, 2 * wordSize), r13);
4340   movq(Address(rsp, wordSize), r14);
4341   movq(Address(rsp, 0), r15);
4342 }
4343 
4344 void Assembler::pushq(Address src) {
4345   InstructionMark im(this);
4346   prefixq(src);
4347   emit_byte(0xFF);
4348   emit_operand(rsi, src);
4349 }
4350 
4351 void Assembler::rclq(Register dst, int imm8) {
4352   assert(isShiftCount(imm8 >> 1), "illegal shift count");
4353   int encode = prefixq_and_encode(dst->encoding());
4354   if (imm8 == 1) {
4355     emit_byte(0xD1);
4356     emit_byte(0xD0 | encode);
4357   } else {
4358     emit_byte(0xC1);
4359     emit_byte(0xD0 | encode);
4360     emit_byte(imm8);
4361   }
4362 }
4363 void Assembler::sarq(Register dst, int imm8) {
4364   assert(isShiftCount(imm8 >> 1), "illegal shift count");
4365   int encode = prefixq_and_encode(dst->encoding());
4366   if (imm8 == 1) {
4367     emit_byte(0xD1);
4368     emit_byte(0xF8 | encode);
4369   } else {
4370     emit_byte(0xC1);
4371     emit_byte(0xF8 | encode);
4372     emit_byte(imm8);
4373   }
4374 }
4375 
4376 void Assembler::sarq(Register dst) {
4377   int encode = prefixq_and_encode(dst->encoding());
4378   emit_byte(0xD3);
4379   emit_byte(0xF8 | encode);
4380 }
4381 
4382 void Assembler::sbbq(Address dst, int32_t imm32) {
4383   InstructionMark im(this);
4384   prefixq(dst);
4385   emit_arith_operand(0x81, rbx, dst, imm32);
4386 }
4387 
4388 void Assembler::sbbq(Register dst, int32_t imm32) {
4389   (void) prefixq_and_encode(dst->encoding());
4390   emit_arith(0x81, 0xD8, dst, imm32);
4391 }
4392 
4393 void Assembler::sbbq(Register dst, Address src) {
4394   InstructionMark im(this);
4395   prefixq(src, dst);
4396   emit_byte(0x1B);
4397   emit_operand(dst, src);
4398 }
4399 
4400 void Assembler::sbbq(Register dst, Register src) {
4401   (void) prefixq_and_encode(dst->encoding(), src->encoding());
4402   emit_arith(0x1B, 0xC0, dst, src);
4403 }
4404 
4405 void Assembler::shlq(Register dst, int imm8) {
4406   assert(isShiftCount(imm8 >> 1), "illegal shift count");
4407   int encode = prefixq_and_encode(dst->encoding());
4408   if (imm8 == 1) {
4409     emit_byte(0xD1);
4410     emit_byte(0xE0 | encode);
4411   } else {
4412     emit_byte(0xC1);
4413     emit_byte(0xE0 | encode);
4414     emit_byte(imm8);
4415   }
4416 }
4417 
4418 void Assembler::shlq(Register dst) {
4419   int encode = prefixq_and_encode(dst->encoding());
4420   emit_byte(0xD3);
4421   emit_byte(0xE0 | encode);
4422 }
4423 
4424 void Assembler::shrq(Register dst, int imm8) {
4425   assert(isShiftCount(imm8 >> 1), "illegal shift count");
4426   int encode = prefixq_and_encode(dst->encoding());
4427   emit_byte(0xC1);
4428   emit_byte(0xE8 | encode);
4429   emit_byte(imm8);
4430 }
4431 
4432 void Assembler::shrq(Register dst) {
4433   int encode = prefixq_and_encode(dst->encoding());
4434   emit_byte(0xD3);
4435   emit_byte(0xE8 | encode);
4436 }
4437 
4438 void Assembler::subq(Address dst, int32_t imm32) {
4439   InstructionMark im(this);
4440   prefixq(dst);
4441   emit_arith_operand(0x81, rbp, dst, imm32);
4442 }
4443 
4444 void Assembler::subq(Address dst, Register src) {
4445   InstructionMark im(this);
4446   prefixq(dst, src);
4447   emit_byte(0x29);
4448   emit_operand(src, dst);
4449 }
4450 
4451 void Assembler::subq(Register dst, int32_t imm32) {
4452   (void) prefixq_and_encode(dst->encoding());
4453   emit_arith(0x81, 0xE8, dst, imm32);
4454 }
4455 
4456 void Assembler::subq(Register dst, Address src) {
4457   InstructionMark im(this);
4458   prefixq(src, dst);
4459   emit_byte(0x2B);
4460   emit_operand(dst, src);
4461 }
4462 
4463 void Assembler::subq(Register dst, Register src) {
4464   (void) prefixq_and_encode(dst->encoding(), src->encoding());
4465   emit_arith(0x2B, 0xC0, dst, src);
4466 }
4467 
4468 void Assembler::testq(Register dst, int32_t imm32) {
4469   // not using emit_arith because test
4470   // doesn't support sign-extension of
4471   // 8bit operands
4472   int encode = dst->encoding();
4473   if (encode == 0) {
4474     prefix(REX_W);
4475     emit_byte(0xA9);
4476   } else {
4477     encode = prefixq_and_encode(encode);
4478     emit_byte(0xF7);
4479     emit_byte(0xC0 | encode);
4480   }
4481   emit_long(imm32);
4482 }
4483 
4484 void Assembler::testq(Register dst, Register src) {
4485   (void) prefixq_and_encode(dst->encoding(), src->encoding());
4486   emit_arith(0x85, 0xC0, dst, src);
4487 }
4488 
4489 void Assembler::xaddq(Address dst, Register src) {
4490   InstructionMark im(this);
4491   prefixq(dst, src);
4492   emit_byte(0x0F);
4493   emit_byte(0xC1);
4494   emit_operand(src, dst);
4495 }
4496 
4497 void Assembler::xchgq(Register dst, Address src) {
4498   InstructionMark im(this);
4499   prefixq(src, dst);
4500   emit_byte(0x87);
4501   emit_operand(dst, src);
4502 }
4503 
4504 void Assembler::xchgq(Register dst, Register src) {
4505   int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4506   emit_byte(0x87);
4507   emit_byte(0xc0 | encode);
4508 }
4509 
4510 void Assembler::xorq(Register dst, Register src) {
4511   (void) prefixq_and_encode(dst->encoding(), src->encoding());
4512   emit_arith(0x33, 0xC0, dst, src);
4513 }
4514 
4515 void Assembler::xorq(Register dst, Address src) {
4516   InstructionMark im(this);
4517   prefixq(src, dst);
4518   emit_byte(0x33);
4519   emit_operand(dst, src);
4520 }
4521 
4522 #endif // !LP64
4523 
4524 static Assembler::Condition reverse[] = {
4525     Assembler::noOverflow     /* overflow      = 0x0 */ ,
4526     Assembler::overflow       /* noOverflow    = 0x1 */ ,
4527     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
4528     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
4529     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
4530     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
4531     Assembler::above          /* belowEqual    = 0x6 */ ,
4532     Assembler::belowEqual     /* above         = 0x7 */ ,
4533     Assembler::positive       /* negative      = 0x8 */ ,
4534     Assembler::negative       /* positive      = 0x9 */ ,
4535     Assembler::noParity       /* parity        = 0xa */ ,
4536     Assembler::parity         /* noParity      = 0xb */ ,
4537     Assembler::greaterEqual   /* less          = 0xc */ ,
4538     Assembler::less           /* greaterEqual  = 0xd */ ,
4539     Assembler::greater        /* lessEqual     = 0xe */ ,
4540     Assembler::lessEqual      /* greater       = 0xf, */
4541 
4542 };
4543 
4544 
4545 // Implementation of MacroAssembler
4546 
4547 // First all the versions that have distinct versions depending on 32/64 bit
4548 // Unless the difference is trivial (1 line or so).
4549 
4550 #ifndef _LP64
4551 
4552 // 32bit versions
4553 
4554 Address MacroAssembler::as_Address(AddressLiteral adr) {
4555   return Address(adr.target(), adr.rspec());
4556 }
4557 
4558 Address MacroAssembler::as_Address(ArrayAddress adr) {
4559   return Address::make_array(adr);
4560 }
4561 
4562 int MacroAssembler::biased_locking_enter(Register lock_reg,
4563                                          Register obj_reg,
4564                                          Register swap_reg,
4565                                          Register tmp_reg,
4566                                          bool swap_reg_contains_mark,
4567                                          Label& done,
4568                                          Label* slow_case,
4569                                          BiasedLockingCounters* counters) {
4570   assert(UseBiasedLocking, "why call this otherwise?");
4571   assert(swap_reg == rax, "swap_reg must be rax, for cmpxchg");
4572   assert_different_registers(lock_reg, obj_reg, swap_reg);
4573 
4574   if (PrintBiasedLockingStatistics && counters == NULL)
4575     counters = BiasedLocking::counters();
4576 
4577   bool need_tmp_reg = false;
4578   if (tmp_reg == noreg) {
4579     need_tmp_reg = true;
4580     tmp_reg = lock_reg;
4581   } else {
4582     assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
4583   }
4584   assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
4585   Address mark_addr      (obj_reg, oopDesc::mark_offset_in_bytes());
4586   Address klass_addr     (obj_reg, oopDesc::klass_offset_in_bytes());
4587   Address saved_mark_addr(lock_reg, 0);
4588 
4589   // Biased locking
4590   // See whether the lock is currently biased toward our thread and
4591   // whether the epoch is still valid
4592   // Note that the runtime guarantees sufficient alignment of JavaThread
4593   // pointers to allow age to be placed into low bits
4594   // First check to see whether biasing is even enabled for this object
4595   Label cas_label;
4596   int null_check_offset = -1;
4597   if (!swap_reg_contains_mark) {
4598     null_check_offset = offset();
4599     movl(swap_reg, mark_addr);
4600   }
4601   if (need_tmp_reg) {
4602     push(tmp_reg);
4603   }
4604   movl(tmp_reg, swap_reg);
4605   andl(tmp_reg, markOopDesc::biased_lock_mask_in_place);
4606   cmpl(tmp_reg, markOopDesc::biased_lock_pattern);
4607   if (need_tmp_reg) {
4608     pop(tmp_reg);
4609   }
4610   jcc(Assembler::notEqual, cas_label);
4611   // The bias pattern is present in the object's header. Need to check
4612   // whether the bias owner and the epoch are both still current.
4613   // Note that because there is no current thread register on x86 we
4614   // need to store off the mark word we read out of the object to
4615   // avoid reloading it and needing to recheck invariants below. This
4616   // store is unfortunate but it makes the overall code shorter and
4617   // simpler.
4618   movl(saved_mark_addr, swap_reg);
4619   if (need_tmp_reg) {
4620     push(tmp_reg);
4621   }
4622   get_thread(tmp_reg);
4623   xorl(swap_reg, tmp_reg);
4624   if (swap_reg_contains_mark) {
4625     null_check_offset = offset();
4626   }
4627   movl(tmp_reg, klass_addr);
4628   xorl(swap_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4629   andl(swap_reg, ~((int) markOopDesc::age_mask_in_place));
4630   if (need_tmp_reg) {
4631     pop(tmp_reg);
4632   }
4633   if (counters != NULL) {
4634     cond_inc32(Assembler::zero,
4635                ExternalAddress((address)counters->biased_lock_entry_count_addr()));
4636   }
4637   jcc(Assembler::equal, done);
4638 
4639   Label try_revoke_bias;
4640   Label try_rebias;
4641 
4642   // At this point we know that the header has the bias pattern and
4643   // that we are not the bias owner in the current epoch. We need to
4644   // figure out more details about the state of the header in order to
4645   // know what operations can be legally performed on the object's
4646   // header.
4647 
4648   // If the low three bits in the xor result aren't clear, that means
4649   // the prototype header is no longer biased and we have to revoke
4650   // the bias on this object.
4651   testl(swap_reg, markOopDesc::biased_lock_mask_in_place);
4652   jcc(Assembler::notZero, try_revoke_bias);
4653 
4654   // Biasing is still enabled for this data type. See whether the
4655   // epoch of the current bias is still valid, meaning that the epoch
4656   // bits of the mark word are equal to the epoch bits of the
4657   // prototype header. (Note that the prototype header's epoch bits
4658   // only change at a safepoint.) If not, attempt to rebias the object
4659   // toward the current thread. Note that we must be absolutely sure
4660   // that the current epoch is invalid in order to do this because
4661   // otherwise the manipulations it performs on the mark word are
4662   // illegal.
4663   testl(swap_reg, markOopDesc::epoch_mask_in_place);
4664   jcc(Assembler::notZero, try_rebias);
4665 
4666   // The epoch of the current bias is still valid but we know nothing
4667   // about the owner; it might be set or it might be clear. Try to
4668   // acquire the bias of the object using an atomic operation. If this
4669   // fails we will go in to the runtime to revoke the object's bias.
4670   // Note that we first construct the presumed unbiased header so we
4671   // don't accidentally blow away another thread's valid bias.
4672   movl(swap_reg, saved_mark_addr);
4673   andl(swap_reg,
4674        markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
4675   if (need_tmp_reg) {
4676     push(tmp_reg);
4677   }
4678   get_thread(tmp_reg);
4679   orl(tmp_reg, swap_reg);
4680   if (os::is_MP()) {
4681     lock();
4682   }
4683   cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4684   if (need_tmp_reg) {
4685     pop(tmp_reg);
4686   }
4687   // If the biasing toward our thread failed, this means that
4688   // another thread succeeded in biasing it toward itself and we
4689   // need to revoke that bias. The revocation will occur in the
4690   // interpreter runtime in the slow case.
4691   if (counters != NULL) {
4692     cond_inc32(Assembler::zero,
4693                ExternalAddress((address)counters->anonymously_biased_lock_entry_count_addr()));
4694   }
4695   if (slow_case != NULL) {
4696     jcc(Assembler::notZero, *slow_case);
4697   }
4698   jmp(done);
4699 
4700   bind(try_rebias);
4701   // At this point we know the epoch has expired, meaning that the
4702   // current "bias owner", if any, is actually invalid. Under these
4703   // circumstances _only_, we are allowed to use the current header's
4704   // value as the comparison value when doing the cas to acquire the
4705   // bias in the current epoch. In other words, we allow transfer of
4706   // the bias from one thread to another directly in this situation.
4707   //
4708   // FIXME: due to a lack of registers we currently blow away the age
4709   // bits in this situation. Should attempt to preserve them.
4710   if (need_tmp_reg) {
4711     push(tmp_reg);
4712   }
4713   get_thread(tmp_reg);
4714   movl(swap_reg, klass_addr);
4715   orl(tmp_reg, Address(swap_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4716   movl(swap_reg, saved_mark_addr);
4717   if (os::is_MP()) {
4718     lock();
4719   }
4720   cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4721   if (need_tmp_reg) {
4722     pop(tmp_reg);
4723   }
4724   // If the biasing toward our thread failed, then another thread
4725   // succeeded in biasing it toward itself and we need to revoke that
4726   // bias. The revocation will occur in the runtime in the slow case.
4727   if (counters != NULL) {
4728     cond_inc32(Assembler::zero,
4729                ExternalAddress((address)counters->rebiased_lock_entry_count_addr()));
4730   }
4731   if (slow_case != NULL) {
4732     jcc(Assembler::notZero, *slow_case);
4733   }
4734   jmp(done);
4735 
4736   bind(try_revoke_bias);
4737   // The prototype mark in the klass doesn't have the bias bit set any
4738   // more, indicating that objects of this data type are not supposed
4739   // to be biased any more. We are going to try to reset the mark of
4740   // this object to the prototype value and fall through to the
4741   // CAS-based locking scheme. Note that if our CAS fails, it means
4742   // that another thread raced us for the privilege of revoking the
4743   // bias of this particular object, so it's okay to continue in the
4744   // normal locking code.
4745   //
4746   // FIXME: due to a lack of registers we currently blow away the age
4747   // bits in this situation. Should attempt to preserve them.
4748   movl(swap_reg, saved_mark_addr);
4749   if (need_tmp_reg) {
4750     push(tmp_reg);
4751   }
4752   movl(tmp_reg, klass_addr);
4753   movl(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4754   if (os::is_MP()) {
4755     lock();
4756   }
4757   cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4758   if (need_tmp_reg) {
4759     pop(tmp_reg);
4760   }
4761   // Fall through to the normal CAS-based lock, because no matter what
4762   // the result of the above CAS, some thread must have succeeded in
4763   // removing the bias bit from the object's header.
4764   if (counters != NULL) {
4765     cond_inc32(Assembler::zero,
4766                ExternalAddress((address)counters->revoked_lock_entry_count_addr()));
4767   }
4768 
4769   bind(cas_label);
4770 
4771   return null_check_offset;
4772 }
4773 void MacroAssembler::call_VM_leaf_base(address entry_point,
4774                                        int number_of_arguments) {
4775   call(RuntimeAddress(entry_point));
4776   increment(rsp, number_of_arguments * wordSize);
4777 }
4778 
4779 void MacroAssembler::cmpoop(Address src1, jobject obj) {
4780   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
4781 }
4782 
4783 void MacroAssembler::cmpoop(Register src1, jobject obj) {
4784   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
4785 }
4786 
4787 void MacroAssembler::extend_sign(Register hi, Register lo) {
4788   // According to Intel Doc. AP-526, "Integer Divide", p.18.
4789   if (VM_Version::is_P6() && hi == rdx && lo == rax) {
4790     cdql();
4791   } else {
4792     movl(hi, lo);
4793     sarl(hi, 31);
4794   }
4795 }
4796 
4797 void MacroAssembler::fat_nop() {
4798   // A 5 byte nop that is safe for patching (see patch_verified_entry)
4799   emit_byte(0x26); // es:
4800   emit_byte(0x2e); // cs:
4801   emit_byte(0x64); // fs:
4802   emit_byte(0x65); // gs:
4803   emit_byte(0x90);
4804 }
4805 
4806 void MacroAssembler::jC2(Register tmp, Label& L) {
4807   // set parity bit if FPU flag C2 is set (via rax)
4808   save_rax(tmp);
4809   fwait(); fnstsw_ax();
4810   sahf();
4811   restore_rax(tmp);
4812   // branch
4813   jcc(Assembler::parity, L);
4814 }
4815 
4816 void MacroAssembler::jnC2(Register tmp, Label& L) {
4817   // set parity bit if FPU flag C2 is set (via rax)
4818   save_rax(tmp);
4819   fwait(); fnstsw_ax();
4820   sahf();
4821   restore_rax(tmp);
4822   // branch
4823   jcc(Assembler::noParity, L);
4824 }
4825 
4826 // 32bit can do a case table jump in one instruction but we no longer allow the base
4827 // to be installed in the Address class
4828 void MacroAssembler::jump(ArrayAddress entry) {
4829   jmp(as_Address(entry));
4830 }
4831 
4832 // Note: y_lo will be destroyed
4833 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
4834   // Long compare for Java (semantics as described in JVM spec.)
4835   Label high, low, done;
4836 
4837   cmpl(x_hi, y_hi);
4838   jcc(Assembler::less, low);
4839   jcc(Assembler::greater, high);
4840   // x_hi is the return register
4841   xorl(x_hi, x_hi);
4842   cmpl(x_lo, y_lo);
4843   jcc(Assembler::below, low);
4844   jcc(Assembler::equal, done);
4845 
4846   bind(high);
4847   xorl(x_hi, x_hi);
4848   increment(x_hi);
4849   jmp(done);
4850 
4851   bind(low);
4852   xorl(x_hi, x_hi);
4853   decrementl(x_hi);
4854 
4855   bind(done);
4856 }
4857 
4858 void MacroAssembler::lea(Register dst, AddressLiteral src) {
4859     mov_literal32(dst, (int32_t)src.target(), src.rspec());
4860 }
4861 
4862 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
4863   // leal(dst, as_Address(adr));
4864   // see note in movl as to why we must use a move
4865   mov_literal32(dst, (int32_t) adr.target(), adr.rspec());
4866 }
4867 
4868 void MacroAssembler::leave() {
4869   mov(rsp, rbp);
4870   pop(rbp);
4871 }
4872 
4873 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) {
4874   // Multiplication of two Java long values stored on the stack
4875   // as illustrated below. Result is in rdx:rax.
4876   //
4877   // rsp ---> [  ??  ] \               \
4878   //            ....    | y_rsp_offset  |
4879   //          [ y_lo ] /  (in bytes)    | x_rsp_offset
4880   //          [ y_hi ]                  | (in bytes)
4881   //            ....                    |
4882   //          [ x_lo ]                 /
4883   //          [ x_hi ]
4884   //            ....
4885   //
4886   // Basic idea: lo(result) = lo(x_lo * y_lo)
4887   //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
4888   Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset);
4889   Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset);
4890   Label quick;
4891   // load x_hi, y_hi and check if quick
4892   // multiplication is possible
4893   movl(rbx, x_hi);
4894   movl(rcx, y_hi);
4895   movl(rax, rbx);
4896   orl(rbx, rcx);                                 // rbx, = 0 <=> x_hi = 0 and y_hi = 0
4897   jcc(Assembler::zero, quick);                   // if rbx, = 0 do quick multiply
4898   // do full multiplication
4899   // 1st step
4900   mull(y_lo);                                    // x_hi * y_lo
4901   movl(rbx, rax);                                // save lo(x_hi * y_lo) in rbx,
4902   // 2nd step
4903   movl(rax, x_lo);
4904   mull(rcx);                                     // x_lo * y_hi
4905   addl(rbx, rax);                                // add lo(x_lo * y_hi) to rbx,
4906   // 3rd step
4907   bind(quick);                                   // note: rbx, = 0 if quick multiply!
4908   movl(rax, x_lo);
4909   mull(y_lo);                                    // x_lo * y_lo
4910   addl(rdx, rbx);                                // correct hi(x_lo * y_lo)
4911 }
4912 
4913 void MacroAssembler::lneg(Register hi, Register lo) {
4914   negl(lo);
4915   adcl(hi, 0);
4916   negl(hi);
4917 }
4918 
4919 void MacroAssembler::lshl(Register hi, Register lo) {
4920   // Java shift left long support (semantics as described in JVM spec., p.305)
4921   // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n))
4922   // shift value is in rcx !
4923   assert(hi != rcx, "must not use rcx");
4924   assert(lo != rcx, "must not use rcx");
4925   const Register s = rcx;                        // shift count
4926   const int      n = BitsPerWord;
4927   Label L;
4928   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
4929   cmpl(s, n);                                    // if (s < n)
4930   jcc(Assembler::less, L);                       // else (s >= n)
4931   movl(hi, lo);                                  // x := x << n
4932   xorl(lo, lo);
4933   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
4934   bind(L);                                       // s (mod n) < n
4935   shldl(hi, lo);                                 // x := x << s
4936   shll(lo);
4937 }
4938 
4939 
4940 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) {
4941   // Java shift right long support (semantics as described in JVM spec., p.306 & p.310)
4942   // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n))
4943   assert(hi != rcx, "must not use rcx");
4944   assert(lo != rcx, "must not use rcx");
4945   const Register s = rcx;                        // shift count
4946   const int      n = BitsPerWord;
4947   Label L;
4948   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
4949   cmpl(s, n);                                    // if (s < n)
4950   jcc(Assembler::less, L);                       // else (s >= n)
4951   movl(lo, hi);                                  // x := x >> n
4952   if (sign_extension) sarl(hi, 31);
4953   else                xorl(hi, hi);
4954   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
4955   bind(L);                                       // s (mod n) < n
4956   shrdl(lo, hi);                                 // x := x >> s
4957   if (sign_extension) sarl(hi);
4958   else                shrl(hi);
4959 }
4960 
4961 void MacroAssembler::movoop(Register dst, jobject obj) {
4962   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
4963 }
4964 
4965 void MacroAssembler::movoop(Address dst, jobject obj) {
4966   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
4967 }
4968 
4969 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
4970   if (src.is_lval()) {
4971     mov_literal32(dst, (intptr_t)src.target(), src.rspec());
4972   } else {
4973     movl(dst, as_Address(src));
4974   }
4975 }
4976 
4977 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
4978   movl(as_Address(dst), src);
4979 }
4980 
4981 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
4982   movl(dst, as_Address(src));
4983 }
4984 
4985 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
4986 void MacroAssembler::movptr(Address dst, intptr_t src) {
4987   movl(dst, src);
4988 }
4989 
4990 
4991 void MacroAssembler::pop_callee_saved_registers() {
4992   pop(rcx);
4993   pop(rdx);
4994   pop(rdi);
4995   pop(rsi);
4996 }
4997 
4998 void MacroAssembler::pop_fTOS() {
4999   fld_d(Address(rsp, 0));
5000   addl(rsp, 2 * wordSize);
5001 }
5002 
5003 void MacroAssembler::push_callee_saved_registers() {
5004   push(rsi);
5005   push(rdi);
5006   push(rdx);
5007   push(rcx);
5008 }
5009 
5010 void MacroAssembler::push_fTOS() {
5011   subl(rsp, 2 * wordSize);
5012   fstp_d(Address(rsp, 0));
5013 }
5014 
5015 
5016 void MacroAssembler::pushoop(jobject obj) {
5017   push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
5018 }
5019 
5020 
5021 void MacroAssembler::pushptr(AddressLiteral src) {
5022   if (src.is_lval()) {
5023     push_literal32((int32_t)src.target(), src.rspec());
5024   } else {
5025     pushl(as_Address(src));
5026   }
5027 }
5028 
5029 void MacroAssembler::set_word_if_not_zero(Register dst) {
5030   xorl(dst, dst);
5031   set_byte_if_not_zero(dst);
5032 }
5033 
5034 static void pass_arg0(MacroAssembler* masm, Register arg) {
5035   masm->push(arg);
5036 }
5037 
5038 static void pass_arg1(MacroAssembler* masm, Register arg) {
5039   masm->push(arg);
5040 }
5041 
5042 static void pass_arg2(MacroAssembler* masm, Register arg) {
5043   masm->push(arg);
5044 }
5045 
5046 static void pass_arg3(MacroAssembler* masm, Register arg) {
5047   masm->push(arg);
5048 }
5049 
5050 #ifndef PRODUCT
5051 extern "C" void findpc(intptr_t x);
5052 #endif
5053 
5054 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) {
5055   // In order to get locks to work, we need to fake a in_VM state
5056   JavaThread* thread = JavaThread::current();
5057   JavaThreadState saved_state = thread->thread_state();
5058   thread->set_thread_state(_thread_in_vm);
5059   if (ShowMessageBoxOnError) {
5060     JavaThread* thread = JavaThread::current();
5061     JavaThreadState saved_state = thread->thread_state();
5062     thread->set_thread_state(_thread_in_vm);
5063     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
5064       ttyLocker ttyl;
5065       BytecodeCounter::print();
5066     }
5067     // To see where a verify_oop failed, get $ebx+40/X for this frame.
5068     // This is the value of eip which points to where verify_oop will return.
5069     if (os::message_box(msg, "Execution stopped, print registers?")) {
5070       ttyLocker ttyl;
5071       tty->print_cr("eip = 0x%08x", eip);
5072 #ifndef PRODUCT
5073       if ((WizardMode || Verbose) && PrintMiscellaneous) {
5074         tty->cr();
5075         findpc(eip);
5076         tty->cr();
5077       }
5078 #endif
5079       tty->print_cr("rax = 0x%08x", rax);
5080       tty->print_cr("rbx = 0x%08x", rbx);
5081       tty->print_cr("rcx = 0x%08x", rcx);
5082       tty->print_cr("rdx = 0x%08x", rdx);
5083       tty->print_cr("rdi = 0x%08x", rdi);
5084       tty->print_cr("rsi = 0x%08x", rsi);
5085       tty->print_cr("rbp = 0x%08x", rbp);
5086       tty->print_cr("rsp = 0x%08x", rsp);
5087       BREAKPOINT;
5088       assert(false, "start up GDB");
5089     }
5090   } else {
5091     ttyLocker ttyl;
5092     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
5093     assert(false, "DEBUG MESSAGE");
5094   }
5095   ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
5096 }
5097 
5098 void MacroAssembler::stop(const char* msg) {
5099   ExternalAddress message((address)msg);
5100   // push address of message
5101   pushptr(message.addr());
5102   { Label L; call(L, relocInfo::none); bind(L); }     // push eip
5103   pusha();                                           // push registers
5104   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
5105   hlt();
5106 }
5107 
5108 void MacroAssembler::warn(const char* msg) {
5109   push_CPU_state();
5110 
5111   ExternalAddress message((address) msg);
5112   // push address of message
5113   pushptr(message.addr());
5114 
5115   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
5116   addl(rsp, wordSize);       // discard argument
5117   pop_CPU_state();
5118 }
5119 
5120 #else // _LP64
5121 
5122 // 64 bit versions
5123 
5124 Address MacroAssembler::as_Address(AddressLiteral adr) {
5125   // amd64 always does this as a pc-rel
5126   // we can be absolute or disp based on the instruction type
5127   // jmp/call are displacements others are absolute
5128   assert(!adr.is_lval(), "must be rval");
5129   assert(reachable(adr), "must be");
5130   return Address((int32_t)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc());
5131 
5132 }
5133 
5134 Address MacroAssembler::as_Address(ArrayAddress adr) {
5135   AddressLiteral base = adr.base();
5136   lea(rscratch1, base);
5137   Address index = adr.index();
5138   assert(index._disp == 0, "must not have disp"); // maybe it can?
5139   Address array(rscratch1, index._index, index._scale, index._disp);
5140   return array;
5141 }
5142 
5143 int MacroAssembler::biased_locking_enter(Register lock_reg,
5144                                          Register obj_reg,
5145                                          Register swap_reg,
5146                                          Register tmp_reg,
5147                                          bool swap_reg_contains_mark,
5148                                          Label& done,
5149                                          Label* slow_case,
5150                                          BiasedLockingCounters* counters) {
5151   assert(UseBiasedLocking, "why call this otherwise?");
5152   assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq");
5153   assert(tmp_reg != noreg, "tmp_reg must be supplied");
5154   assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
5155   assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
5156   Address mark_addr      (obj_reg, oopDesc::mark_offset_in_bytes());
5157   Address saved_mark_addr(lock_reg, 0);
5158 
5159   if (PrintBiasedLockingStatistics && counters == NULL)
5160     counters = BiasedLocking::counters();
5161 
5162   // Biased locking
5163   // See whether the lock is currently biased toward our thread and
5164   // whether the epoch is still valid
5165   // Note that the runtime guarantees sufficient alignment of JavaThread
5166   // pointers to allow age to be placed into low bits
5167   // First check to see whether biasing is even enabled for this object
5168   Label cas_label;
5169   int null_check_offset = -1;
5170   if (!swap_reg_contains_mark) {
5171     null_check_offset = offset();
5172     movq(swap_reg, mark_addr);
5173   }
5174   movq(tmp_reg, swap_reg);
5175   andq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5176   cmpq(tmp_reg, markOopDesc::biased_lock_pattern);
5177   jcc(Assembler::notEqual, cas_label);
5178   // The bias pattern is present in the object's header. Need to check
5179   // whether the bias owner and the epoch are both still current.
5180   load_prototype_header(tmp_reg, obj_reg);
5181   orq(tmp_reg, r15_thread);
5182   xorq(tmp_reg, swap_reg);
5183   andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place));
5184   if (counters != NULL) {
5185     cond_inc32(Assembler::zero,
5186                ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5187   }
5188   jcc(Assembler::equal, done);
5189 
5190   Label try_revoke_bias;
5191   Label try_rebias;
5192 
5193   // At this point we know that the header has the bias pattern and
5194   // that we are not the bias owner in the current epoch. We need to
5195   // figure out more details about the state of the header in order to
5196   // know what operations can be legally performed on the object's
5197   // header.
5198 
5199   // If the low three bits in the xor result aren't clear, that means
5200   // the prototype header is no longer biased and we have to revoke
5201   // the bias on this object.
5202   testq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5203   jcc(Assembler::notZero, try_revoke_bias);
5204 
5205   // Biasing is still enabled for this data type. See whether the
5206   // epoch of the current bias is still valid, meaning that the epoch
5207   // bits of the mark word are equal to the epoch bits of the
5208   // prototype header. (Note that the prototype header's epoch bits
5209   // only change at a safepoint.) If not, attempt to rebias the object
5210   // toward the current thread. Note that we must be absolutely sure
5211   // that the current epoch is invalid in order to do this because
5212   // otherwise the manipulations it performs on the mark word are
5213   // illegal.
5214   testq(tmp_reg, markOopDesc::epoch_mask_in_place);
5215   jcc(Assembler::notZero, try_rebias);
5216 
5217   // The epoch of the current bias is still valid but we know nothing
5218   // about the owner; it might be set or it might be clear. Try to
5219   // acquire the bias of the object using an atomic operation. If this
5220   // fails we will go in to the runtime to revoke the object's bias.
5221   // Note that we first construct the presumed unbiased header so we
5222   // don't accidentally blow away another thread's valid bias.
5223   andq(swap_reg,
5224        markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
5225   movq(tmp_reg, swap_reg);
5226   orq(tmp_reg, r15_thread);
5227   if (os::is_MP()) {
5228     lock();
5229   }
5230   cmpxchgq(tmp_reg, Address(obj_reg, 0));
5231   // If the biasing toward our thread failed, this means that
5232   // another thread succeeded in biasing it toward itself and we
5233   // need to revoke that bias. The revocation will occur in the
5234   // interpreter runtime in the slow case.
5235   if (counters != NULL) {
5236     cond_inc32(Assembler::zero,
5237                ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5238   }
5239   if (slow_case != NULL) {
5240     jcc(Assembler::notZero, *slow_case);
5241   }
5242   jmp(done);
5243 
5244   bind(try_rebias);
5245   // At this point we know the epoch has expired, meaning that the
5246   // current "bias owner", if any, is actually invalid. Under these
5247   // circumstances _only_, we are allowed to use the current header's
5248   // value as the comparison value when doing the cas to acquire the
5249   // bias in the current epoch. In other words, we allow transfer of
5250   // the bias from one thread to another directly in this situation.
5251   //
5252   // FIXME: due to a lack of registers we currently blow away the age
5253   // bits in this situation. Should attempt to preserve them.
5254   load_prototype_header(tmp_reg, obj_reg);
5255   orq(tmp_reg, r15_thread);
5256   if (os::is_MP()) {
5257     lock();
5258   }
5259   cmpxchgq(tmp_reg, Address(obj_reg, 0));
5260   // If the biasing toward our thread failed, then another thread
5261   // succeeded in biasing it toward itself and we need to revoke that
5262   // bias. The revocation will occur in the runtime in the slow case.
5263   if (counters != NULL) {
5264     cond_inc32(Assembler::zero,
5265                ExternalAddress((address) counters->rebiased_lock_entry_count_addr()));
5266   }
5267   if (slow_case != NULL) {
5268     jcc(Assembler::notZero, *slow_case);
5269   }
5270   jmp(done);
5271 
5272   bind(try_revoke_bias);
5273   // The prototype mark in the klass doesn't have the bias bit set any
5274   // more, indicating that objects of this data type are not supposed
5275   // to be biased any more. We are going to try to reset the mark of
5276   // this object to the prototype value and fall through to the
5277   // CAS-based locking scheme. Note that if our CAS fails, it means
5278   // that another thread raced us for the privilege of revoking the
5279   // bias of this particular object, so it's okay to continue in the
5280   // normal locking code.
5281   //
5282   // FIXME: due to a lack of registers we currently blow away the age
5283   // bits in this situation. Should attempt to preserve them.
5284   load_prototype_header(tmp_reg, obj_reg);
5285   if (os::is_MP()) {
5286     lock();
5287   }
5288   cmpxchgq(tmp_reg, Address(obj_reg, 0));
5289   // Fall through to the normal CAS-based lock, because no matter what
5290   // the result of the above CAS, some thread must have succeeded in
5291   // removing the bias bit from the object's header.
5292   if (counters != NULL) {
5293     cond_inc32(Assembler::zero,
5294                ExternalAddress((address) counters->revoked_lock_entry_count_addr()));
5295   }
5296 
5297   bind(cas_label);
5298 
5299   return null_check_offset;
5300 }
5301 
5302 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
5303   Label L, E;
5304 
5305 #ifdef _WIN64
5306   // Windows always allocates space for it's register args
5307   assert(num_args <= 4, "only register arguments supported");
5308   subq(rsp,  frame::arg_reg_save_area_bytes);
5309 #endif
5310 
5311   // Align stack if necessary
5312   testl(rsp, 15);
5313   jcc(Assembler::zero, L);
5314 
5315   subq(rsp, 8);
5316   {
5317     call(RuntimeAddress(entry_point));
5318   }
5319   addq(rsp, 8);
5320   jmp(E);
5321 
5322   bind(L);
5323   {
5324     call(RuntimeAddress(entry_point));
5325   }
5326 
5327   bind(E);
5328 
5329 #ifdef _WIN64
5330   // restore stack pointer
5331   addq(rsp, frame::arg_reg_save_area_bytes);
5332 #endif
5333 
5334 }
5335 
5336 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) {
5337   assert(!src2.is_lval(), "should use cmpptr");
5338 
5339   if (reachable(src2)) {
5340     cmpq(src1, as_Address(src2));
5341   } else {
5342     lea(rscratch1, src2);
5343     Assembler::cmpq(src1, Address(rscratch1, 0));
5344   }
5345 }
5346 
5347 int MacroAssembler::corrected_idivq(Register reg) {
5348   // Full implementation of Java ldiv and lrem; checks for special
5349   // case as described in JVM spec., p.243 & p.271.  The function
5350   // returns the (pc) offset of the idivl instruction - may be needed
5351   // for implicit exceptions.
5352   //
5353   //         normal case                           special case
5354   //
5355   // input : rax: dividend                         min_long
5356   //         reg: divisor   (may not be eax/edx)   -1
5357   //
5358   // output: rax: quotient  (= rax idiv reg)       min_long
5359   //         rdx: remainder (= rax irem reg)       0
5360   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
5361   static const int64_t min_long = 0x8000000000000000;
5362   Label normal_case, special_case;
5363 
5364   // check for special case
5365   cmp64(rax, ExternalAddress((address) &min_long));
5366   jcc(Assembler::notEqual, normal_case);
5367   xorl(rdx, rdx); // prepare rdx for possible special case (where
5368                   // remainder = 0)
5369   cmpq(reg, -1);
5370   jcc(Assembler::equal, special_case);
5371 
5372   // handle normal case
5373   bind(normal_case);
5374   cdqq();
5375   int idivq_offset = offset();
5376   idivq(reg);
5377 
5378   // normal and special case exit
5379   bind(special_case);
5380 
5381   return idivq_offset;
5382 }
5383 
5384 void MacroAssembler::decrementq(Register reg, int value) {
5385   if (value == min_jint) { subq(reg, value); return; }
5386   if (value <  0) { incrementq(reg, -value); return; }
5387   if (value == 0) {                        ; return; }
5388   if (value == 1 && UseIncDec) { decq(reg) ; return; }
5389   /* else */      { subq(reg, value)       ; return; }
5390 }
5391 
5392 void MacroAssembler::decrementq(Address dst, int value) {
5393   if (value == min_jint) { subq(dst, value); return; }
5394   if (value <  0) { incrementq(dst, -value); return; }
5395   if (value == 0) {                        ; return; }
5396   if (value == 1 && UseIncDec) { decq(dst) ; return; }
5397   /* else */      { subq(dst, value)       ; return; }
5398 }
5399 
5400 void MacroAssembler::fat_nop() {
5401   // A 5 byte nop that is safe for patching (see patch_verified_entry)
5402   // Recommened sequence from 'Software Optimization Guide for the AMD
5403   // Hammer Processor'
5404   emit_byte(0x66);
5405   emit_byte(0x66);
5406   emit_byte(0x90);
5407   emit_byte(0x66);
5408   emit_byte(0x90);
5409 }
5410 
5411 void MacroAssembler::incrementq(Register reg, int value) {
5412   if (value == min_jint) { addq(reg, value); return; }
5413   if (value <  0) { decrementq(reg, -value); return; }
5414   if (value == 0) {                        ; return; }
5415   if (value == 1 && UseIncDec) { incq(reg) ; return; }
5416   /* else */      { addq(reg, value)       ; return; }
5417 }
5418 
5419 void MacroAssembler::incrementq(Address dst, int value) {
5420   if (value == min_jint) { addq(dst, value); return; }
5421   if (value <  0) { decrementq(dst, -value); return; }
5422   if (value == 0) {                        ; return; }
5423   if (value == 1 && UseIncDec) { incq(dst) ; return; }
5424   /* else */      { addq(dst, value)       ; return; }
5425 }
5426 
5427 // 32bit can do a case table jump in one instruction but we no longer allow the base
5428 // to be installed in the Address class
5429 void MacroAssembler::jump(ArrayAddress entry) {
5430   lea(rscratch1, entry.base());
5431   Address dispatch = entry.index();
5432   assert(dispatch._base == noreg, "must be");
5433   dispatch._base = rscratch1;
5434   jmp(dispatch);
5435 }
5436 
5437 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
5438   ShouldNotReachHere(); // 64bit doesn't use two regs
5439   cmpq(x_lo, y_lo);
5440 }
5441 
5442 void MacroAssembler::lea(Register dst, AddressLiteral src) {
5443     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
5444 }
5445 
5446 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
5447   mov_literal64(rscratch1, (intptr_t)adr.target(), adr.rspec());
5448   movptr(dst, rscratch1);
5449 }
5450 
5451 void MacroAssembler::leave() {
5452   // %%% is this really better? Why not on 32bit too?
5453   emit_byte(0xC9); // LEAVE
5454 }
5455 
5456 void MacroAssembler::lneg(Register hi, Register lo) {
5457   ShouldNotReachHere(); // 64bit doesn't use two regs
5458   negq(lo);
5459 }
5460 
5461 void MacroAssembler::movoop(Register dst, jobject obj) {
5462   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
5463 }
5464 
5465 void MacroAssembler::movoop(Address dst, jobject obj) {
5466   mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate());
5467   movq(dst, rscratch1);
5468 }
5469 
5470 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
5471   if (src.is_lval()) {
5472     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
5473   } else {
5474     if (reachable(src)) {
5475       movq(dst, as_Address(src));
5476     } else {
5477       lea(rscratch1, src);
5478       movq(dst, Address(rscratch1,0));
5479     }
5480   }
5481 }
5482 
5483 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
5484   movq(as_Address(dst), src);
5485 }
5486 
5487 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
5488   movq(dst, as_Address(src));
5489 }
5490 
5491 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
5492 void MacroAssembler::movptr(Address dst, intptr_t src) {
5493   mov64(rscratch1, src);
5494   movq(dst, rscratch1);
5495 }
5496 
5497 // These are mostly for initializing NULL
5498 void MacroAssembler::movptr(Address dst, int32_t src) {
5499   movslq(dst, src);
5500 }
5501 
5502 void MacroAssembler::movptr(Register dst, int32_t src) {
5503   mov64(dst, (intptr_t)src);
5504 }
5505 
5506 void MacroAssembler::pushoop(jobject obj) {
5507   movoop(rscratch1, obj);
5508   push(rscratch1);
5509 }
5510 
5511 void MacroAssembler::pushptr(AddressLiteral src) {
5512   lea(rscratch1, src);
5513   if (src.is_lval()) {
5514     push(rscratch1);
5515   } else {
5516     pushq(Address(rscratch1, 0));
5517   }
5518 }
5519 
5520 void MacroAssembler::reset_last_Java_frame(bool clear_fp,
5521                                            bool clear_pc) {
5522   // we must set sp to zero to clear frame
5523   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
5524   // must clear fp, so that compiled frames are not confused; it is
5525   // possible that we need it only for debugging
5526   if (clear_fp) {
5527     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
5528   }
5529 
5530   if (clear_pc) {
5531     movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
5532   }
5533 }
5534 
5535 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
5536                                          Register last_java_fp,
5537                                          address  last_java_pc) {
5538   // determine last_java_sp register
5539   if (!last_java_sp->is_valid()) {
5540     last_java_sp = rsp;
5541   }
5542 
5543   // last_java_fp is optional
5544   if (last_java_fp->is_valid()) {
5545     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()),
5546            last_java_fp);
5547   }
5548 
5549   // last_java_pc is optional
5550   if (last_java_pc != NULL) {
5551     Address java_pc(r15_thread,
5552                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
5553     lea(rscratch1, InternalAddress(last_java_pc));
5554     movptr(java_pc, rscratch1);
5555   }
5556 
5557   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
5558 }
5559 
5560 static void pass_arg0(MacroAssembler* masm, Register arg) {
5561   if (c_rarg0 != arg ) {
5562     masm->mov(c_rarg0, arg);
5563   }
5564 }
5565 
5566 static void pass_arg1(MacroAssembler* masm, Register arg) {
5567   if (c_rarg1 != arg ) {
5568     masm->mov(c_rarg1, arg);
5569   }
5570 }
5571 
5572 static void pass_arg2(MacroAssembler* masm, Register arg) {
5573   if (c_rarg2 != arg ) {
5574     masm->mov(c_rarg2, arg);
5575   }
5576 }
5577 
5578 static void pass_arg3(MacroAssembler* masm, Register arg) {
5579   if (c_rarg3 != arg ) {
5580     masm->mov(c_rarg3, arg);
5581   }
5582 }
5583 
5584 void MacroAssembler::stop(const char* msg) {
5585   address rip = pc();
5586   pusha(); // get regs on stack
5587   lea(c_rarg0, ExternalAddress((address) msg));
5588   lea(c_rarg1, InternalAddress(rip));
5589   movq(c_rarg2, rsp); // pass pointer to regs array
5590   andq(rsp, -16); // align stack as required by ABI
5591   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
5592   hlt();
5593 }
5594 
5595 void MacroAssembler::warn(const char* msg) {
5596   push(rsp);
5597   andq(rsp, -16);     // align stack as required by push_CPU_state and call
5598 
5599   push_CPU_state();   // keeps alignment at 16 bytes
5600   lea(c_rarg0, ExternalAddress((address) msg));
5601   call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0);
5602   pop_CPU_state();
5603   pop(rsp);
5604 }
5605 
5606 #ifndef PRODUCT
5607 extern "C" void findpc(intptr_t x);
5608 #endif
5609 
5610 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
5611   // In order to get locks to work, we need to fake a in_VM state
5612   if (ShowMessageBoxOnError ) {
5613     JavaThread* thread = JavaThread::current();
5614     JavaThreadState saved_state = thread->thread_state();
5615     thread->set_thread_state(_thread_in_vm);
5616 #ifndef PRODUCT
5617     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
5618       ttyLocker ttyl;
5619       BytecodeCounter::print();
5620     }
5621 #endif
5622     // To see where a verify_oop failed, get $ebx+40/X for this frame.
5623     // XXX correct this offset for amd64
5624     // This is the value of eip which points to where verify_oop will return.
5625     if (os::message_box(msg, "Execution stopped, print registers?")) {
5626       ttyLocker ttyl;
5627       tty->print_cr("rip = 0x%016lx", pc);
5628 #ifndef PRODUCT
5629       tty->cr();
5630       findpc(pc);
5631       tty->cr();
5632 #endif
5633       tty->print_cr("rax = 0x%016lx", regs[15]);
5634       tty->print_cr("rbx = 0x%016lx", regs[12]);
5635       tty->print_cr("rcx = 0x%016lx", regs[14]);
5636       tty->print_cr("rdx = 0x%016lx", regs[13]);
5637       tty->print_cr("rdi = 0x%016lx", regs[8]);
5638       tty->print_cr("rsi = 0x%016lx", regs[9]);
5639       tty->print_cr("rbp = 0x%016lx", regs[10]);
5640       tty->print_cr("rsp = 0x%016lx", regs[11]);
5641       tty->print_cr("r8  = 0x%016lx", regs[7]);
5642       tty->print_cr("r9  = 0x%016lx", regs[6]);
5643       tty->print_cr("r10 = 0x%016lx", regs[5]);
5644       tty->print_cr("r11 = 0x%016lx", regs[4]);
5645       tty->print_cr("r12 = 0x%016lx", regs[3]);
5646       tty->print_cr("r13 = 0x%016lx", regs[2]);
5647       tty->print_cr("r14 = 0x%016lx", regs[1]);
5648       tty->print_cr("r15 = 0x%016lx", regs[0]);
5649       BREAKPOINT;
5650     }
5651     ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
5652   } else {
5653     ttyLocker ttyl;
5654     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n",
5655                     msg);
5656   }
5657 }
5658 
5659 #endif // _LP64
5660 
5661 // Now versions that are common to 32/64 bit
5662 
5663 void MacroAssembler::addptr(Register dst, int32_t imm32) {
5664   LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32));
5665 }
5666 
5667 void MacroAssembler::addptr(Register dst, Register src) {
5668   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
5669 }
5670 
5671 void MacroAssembler::addptr(Address dst, Register src) {
5672   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
5673 }
5674 
5675 void MacroAssembler::align(int modulus) {
5676   if (offset() % modulus != 0) {
5677     nop(modulus - (offset() % modulus));
5678   }
5679 }
5680 
5681 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) {
5682   if (reachable(src)) {
5683     andpd(dst, as_Address(src));
5684   } else {
5685     lea(rscratch1, src);
5686     andpd(dst, Address(rscratch1, 0));
5687   }
5688 }
5689 
5690 void MacroAssembler::andptr(Register dst, int32_t imm32) {
5691   LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32));
5692 }
5693 
5694 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) {
5695   pushf();
5696   if (os::is_MP())
5697     lock();
5698   incrementl(counter_addr);
5699   popf();
5700 }
5701 
5702 // Writes to stack successive pages until offset reached to check for
5703 // stack overflow + shadow pages.  This clobbers tmp.
5704 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5705   movptr(tmp, rsp);
5706   // Bang stack for total size given plus shadow page size.
5707   // Bang one page at a time because large size can bang beyond yellow and
5708   // red zones.
5709   Label loop;
5710   bind(loop);
5711   movl(Address(tmp, (-os::vm_page_size())), size );
5712   subptr(tmp, os::vm_page_size());
5713   subl(size, os::vm_page_size());
5714   jcc(Assembler::greater, loop);
5715 
5716   // Bang down shadow pages too.
5717   // The -1 because we already subtracted 1 page.
5718   for (int i = 0; i< StackShadowPages-1; i++) {
5719     // this could be any sized move but this is can be a debugging crumb
5720     // so the bigger the better.
5721     movptr(Address(tmp, (-i*os::vm_page_size())), size );
5722   }
5723 }
5724 
5725 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) {
5726   assert(UseBiasedLocking, "why call this otherwise?");
5727 
5728   // Check for biased locking unlock case, which is a no-op
5729   // Note: we do not have to check the thread ID for two reasons.
5730   // First, the interpreter checks for IllegalMonitorStateException at
5731   // a higher level. Second, if the bias was revoked while we held the
5732   // lock, the object could not be rebiased toward another thread, so
5733   // the bias bit would be clear.
5734   movptr(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
5735   andptr(temp_reg, markOopDesc::biased_lock_mask_in_place);
5736   cmpptr(temp_reg, markOopDesc::biased_lock_pattern);
5737   jcc(Assembler::equal, done);
5738 }
5739 
5740 void MacroAssembler::c2bool(Register x) {
5741   // implements x == 0 ? 0 : 1
5742   // note: must only look at least-significant byte of x
5743   //       since C-style booleans are stored in one byte
5744   //       only! (was bug)
5745   andl(x, 0xFF);
5746   setb(Assembler::notZero, x);
5747 }
5748 
5749 // Wouldn't need if AddressLiteral version had new name
5750 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
5751   Assembler::call(L, rtype);
5752 }
5753 
5754 void MacroAssembler::call(Register entry) {
5755   Assembler::call(entry);
5756 }
5757 
5758 void MacroAssembler::call(AddressLiteral entry) {
5759   if (reachable(entry)) {
5760     Assembler::call_literal(entry.target(), entry.rspec());
5761   } else {
5762     lea(rscratch1, entry);
5763     Assembler::call(rscratch1);
5764   }
5765 }
5766 
5767 // Implementation of call_VM versions
5768 
5769 void MacroAssembler::call_VM(Register oop_result,
5770                              address entry_point,
5771                              bool check_exceptions) {
5772   Label C, E;
5773   call(C, relocInfo::none);
5774   jmp(E);
5775 
5776   bind(C);
5777   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
5778   ret(0);
5779 
5780   bind(E);
5781 }
5782 
5783 void MacroAssembler::call_VM(Register oop_result,
5784                              address entry_point,
5785                              Register arg_1,
5786                              bool check_exceptions) {
5787   Label C, E;
5788   call(C, relocInfo::none);
5789   jmp(E);
5790 
5791   bind(C);
5792   pass_arg1(this, arg_1);
5793   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
5794   ret(0);
5795 
5796   bind(E);
5797 }
5798 
5799 void MacroAssembler::call_VM(Register oop_result,
5800                              address entry_point,
5801                              Register arg_1,
5802                              Register arg_2,
5803                              bool check_exceptions) {
5804   Label C, E;
5805   call(C, relocInfo::none);
5806   jmp(E);
5807 
5808   bind(C);
5809 
5810   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5811 
5812   pass_arg2(this, arg_2);
5813   pass_arg1(this, arg_1);
5814   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
5815   ret(0);
5816 
5817   bind(E);
5818 }
5819 
5820 void MacroAssembler::call_VM(Register oop_result,
5821                              address entry_point,
5822                              Register arg_1,
5823                              Register arg_2,
5824                              Register arg_3,
5825                              bool check_exceptions) {
5826   Label C, E;
5827   call(C, relocInfo::none);
5828   jmp(E);
5829 
5830   bind(C);
5831 
5832   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
5833   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
5834   pass_arg3(this, arg_3);
5835 
5836   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5837   pass_arg2(this, arg_2);
5838 
5839   pass_arg1(this, arg_1);
5840   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
5841   ret(0);
5842 
5843   bind(E);
5844 }
5845 
5846 void MacroAssembler::call_VM(Register oop_result,
5847                              Register last_java_sp,
5848                              address entry_point,
5849                              int number_of_arguments,
5850                              bool check_exceptions) {
5851   Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
5852   call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
5853 }
5854 
5855 void MacroAssembler::call_VM(Register oop_result,
5856                              Register last_java_sp,
5857                              address entry_point,
5858                              Register arg_1,
5859                              bool check_exceptions) {
5860   pass_arg1(this, arg_1);
5861   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
5862 }
5863 
5864 void MacroAssembler::call_VM(Register oop_result,
5865                              Register last_java_sp,
5866                              address entry_point,
5867                              Register arg_1,
5868                              Register arg_2,
5869                              bool check_exceptions) {
5870 
5871   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5872   pass_arg2(this, arg_2);
5873   pass_arg1(this, arg_1);
5874   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
5875 }
5876 
5877 void MacroAssembler::call_VM(Register oop_result,
5878                              Register last_java_sp,
5879                              address entry_point,
5880                              Register arg_1,
5881                              Register arg_2,
5882                              Register arg_3,
5883                              bool check_exceptions) {
5884   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
5885   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
5886   pass_arg3(this, arg_3);
5887   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5888   pass_arg2(this, arg_2);
5889   pass_arg1(this, arg_1);
5890   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
5891 }
5892 
5893 void MacroAssembler::call_VM_base(Register oop_result,
5894                                   Register java_thread,
5895                                   Register last_java_sp,
5896                                   address  entry_point,
5897                                   int      number_of_arguments,
5898                                   bool     check_exceptions) {
5899   // determine java_thread register
5900   if (!java_thread->is_valid()) {
5901 #ifdef _LP64
5902     java_thread = r15_thread;
5903 #else
5904     java_thread = rdi;
5905     get_thread(java_thread);
5906 #endif // LP64
5907   }
5908   // determine last_java_sp register
5909   if (!last_java_sp->is_valid()) {
5910     last_java_sp = rsp;
5911   }
5912   // debugging support
5913   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
5914   LP64_ONLY(assert(java_thread == r15_thread, "unexpected register"));
5915 #ifdef ASSERT
5916   LP64_ONLY(if (UseCompressedOops) verify_heapbase("call_VM_base");)
5917 #endif // ASSERT
5918 
5919   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
5920   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
5921 
5922   // push java thread (becomes first argument of C function)
5923 
5924   NOT_LP64(push(java_thread); number_of_arguments++);
5925   LP64_ONLY(mov(c_rarg0, r15_thread));
5926 
5927   // set last Java frame before call
5928   assert(last_java_sp != rbp, "can't use ebp/rbp");
5929 
5930   // Only interpreter should have to set fp
5931   set_last_Java_frame(java_thread, last_java_sp, rbp, NULL);
5932 
5933   // do the call, remove parameters
5934   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
5935 
5936   // restore the thread (cannot use the pushed argument since arguments
5937   // may be overwritten by C code generated by an optimizing compiler);
5938   // however can use the register value directly if it is callee saved.
5939   if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) {
5940     // rdi & rsi (also r15) are callee saved -> nothing to do
5941 #ifdef ASSERT
5942     guarantee(java_thread != rax, "change this code");
5943     push(rax);
5944     { Label L;
5945       get_thread(rax);
5946       cmpptr(java_thread, rax);
5947       jcc(Assembler::equal, L);
5948       stop("MacroAssembler::call_VM_base: rdi not callee saved?");
5949       bind(L);
5950     }
5951     pop(rax);
5952 #endif
5953   } else {
5954     get_thread(java_thread);
5955   }
5956   // reset last Java frame
5957   // Only interpreter should have to clear fp
5958   reset_last_Java_frame(java_thread, true, false);
5959 
5960 #ifndef CC_INTERP
5961    // C++ interp handles this in the interpreter
5962   check_and_handle_popframe(java_thread);
5963   check_and_handle_earlyret(java_thread);
5964 #endif /* CC_INTERP */
5965 
5966   if (check_exceptions) {
5967     // check for pending exceptions (java_thread is set upon return)
5968     cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
5969 #ifndef _LP64
5970     jump_cc(Assembler::notEqual,
5971             RuntimeAddress(StubRoutines::forward_exception_entry()));
5972 #else
5973     // This used to conditionally jump to forward_exception however it is
5974     // possible if we relocate that the branch will not reach. So we must jump
5975     // around so we can always reach
5976 
5977     Label ok;
5978     jcc(Assembler::equal, ok);
5979     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5980     bind(ok);
5981 #endif // LP64
5982   }
5983 
5984   // get oop result if there is one and reset the value in the thread
5985   if (oop_result->is_valid()) {
5986     movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
5987     movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD);
5988     verify_oop(oop_result, "broken oop in call_VM_base");
5989   }
5990 }
5991 
5992 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
5993 
5994   // Calculate the value for last_Java_sp
5995   // somewhat subtle. call_VM does an intermediate call
5996   // which places a return address on the stack just under the
5997   // stack pointer as the user finsihed with it. This allows
5998   // use to retrieve last_Java_pc from last_Java_sp[-1].
5999   // On 32bit we then have to push additional args on the stack to accomplish
6000   // the actual requested call. On 64bit call_VM only can use register args
6001   // so the only extra space is the return address that call_VM created.
6002   // This hopefully explains the calculations here.
6003 
6004 #ifdef _LP64
6005   // We've pushed one address, correct last_Java_sp
6006   lea(rax, Address(rsp, wordSize));
6007 #else
6008   lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize));
6009 #endif // LP64
6010 
6011   call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions);
6012 
6013 }
6014 
6015 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
6016   call_VM_leaf_base(entry_point, number_of_arguments);
6017 }
6018 
6019 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
6020   pass_arg0(this, arg_0);
6021   call_VM_leaf(entry_point, 1);
6022 }
6023 
6024 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
6025 
6026   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
6027   pass_arg1(this, arg_1);
6028   pass_arg0(this, arg_0);
6029   call_VM_leaf(entry_point, 2);
6030 }
6031 
6032 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
6033   LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
6034   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
6035   pass_arg2(this, arg_2);
6036   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
6037   pass_arg1(this, arg_1);
6038   pass_arg0(this, arg_0);
6039   call_VM_leaf(entry_point, 3);
6040 }
6041 
6042 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
6043 }
6044 
6045 void MacroAssembler::check_and_handle_popframe(Register java_thread) {
6046 }
6047 
6048 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) {
6049   if (reachable(src1)) {
6050     cmpl(as_Address(src1), imm);
6051   } else {
6052     lea(rscratch1, src1);
6053     cmpl(Address(rscratch1, 0), imm);
6054   }
6055 }
6056 
6057 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) {
6058   assert(!src2.is_lval(), "use cmpptr");
6059   if (reachable(src2)) {
6060     cmpl(src1, as_Address(src2));
6061   } else {
6062     lea(rscratch1, src2);
6063     cmpl(src1, Address(rscratch1, 0));
6064   }
6065 }
6066 
6067 void MacroAssembler::cmp32(Register src1, int32_t imm) {
6068   Assembler::cmpl(src1, imm);
6069 }
6070 
6071 void MacroAssembler::cmp32(Register src1, Address src2) {
6072   Assembler::cmpl(src1, src2);
6073 }
6074 
6075 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
6076   ucomisd(opr1, opr2);
6077 
6078   Label L;
6079   if (unordered_is_less) {
6080     movl(dst, -1);
6081     jcc(Assembler::parity, L);
6082     jcc(Assembler::below , L);
6083     movl(dst, 0);
6084     jcc(Assembler::equal , L);
6085     increment(dst);
6086   } else { // unordered is greater
6087     movl(dst, 1);
6088     jcc(Assembler::parity, L);
6089     jcc(Assembler::above , L);
6090     movl(dst, 0);
6091     jcc(Assembler::equal , L);
6092     decrementl(dst);
6093   }
6094   bind(L);
6095 }
6096 
6097 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
6098   ucomiss(opr1, opr2);
6099 
6100   Label L;
6101   if (unordered_is_less) {
6102     movl(dst, -1);
6103     jcc(Assembler::parity, L);
6104     jcc(Assembler::below , L);
6105     movl(dst, 0);
6106     jcc(Assembler::equal , L);
6107     increment(dst);
6108   } else { // unordered is greater
6109     movl(dst, 1);
6110     jcc(Assembler::parity, L);
6111     jcc(Assembler::above , L);
6112     movl(dst, 0);
6113     jcc(Assembler::equal , L);
6114     decrementl(dst);
6115   }
6116   bind(L);
6117 }
6118 
6119 
6120 void MacroAssembler::cmp8(AddressLiteral src1, int imm) {
6121   if (reachable(src1)) {
6122     cmpb(as_Address(src1), imm);
6123   } else {
6124     lea(rscratch1, src1);
6125     cmpb(Address(rscratch1, 0), imm);
6126   }
6127 }
6128 
6129 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) {
6130 #ifdef _LP64
6131   if (src2.is_lval()) {
6132     movptr(rscratch1, src2);
6133     Assembler::cmpq(src1, rscratch1);
6134   } else if (reachable(src2)) {
6135     cmpq(src1, as_Address(src2));
6136   } else {
6137     lea(rscratch1, src2);
6138     Assembler::cmpq(src1, Address(rscratch1, 0));
6139   }
6140 #else
6141   if (src2.is_lval()) {
6142     cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
6143   } else {
6144     cmpl(src1, as_Address(src2));
6145   }
6146 #endif // _LP64
6147 }
6148 
6149 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) {
6150   assert(src2.is_lval(), "not a mem-mem compare");
6151 #ifdef _LP64
6152   // moves src2's literal address
6153   movptr(rscratch1, src2);
6154   Assembler::cmpq(src1, rscratch1);
6155 #else
6156   cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
6157 #endif // _LP64
6158 }
6159 
6160 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr) {
6161   if (reachable(adr)) {
6162     if (os::is_MP())
6163       lock();
6164     cmpxchgptr(reg, as_Address(adr));
6165   } else {
6166     lea(rscratch1, adr);
6167     if (os::is_MP())
6168       lock();
6169     cmpxchgptr(reg, Address(rscratch1, 0));
6170   }
6171 }
6172 
6173 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
6174   LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr));
6175 }
6176 
6177 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) {
6178   if (reachable(src)) {
6179     comisd(dst, as_Address(src));
6180   } else {
6181     lea(rscratch1, src);
6182     comisd(dst, Address(rscratch1, 0));
6183   }
6184 }
6185 
6186 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) {
6187   if (reachable(src)) {
6188     comiss(dst, as_Address(src));
6189   } else {
6190     lea(rscratch1, src);
6191     comiss(dst, Address(rscratch1, 0));
6192   }
6193 }
6194 
6195 
6196 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) {
6197   Condition negated_cond = negate_condition(cond);
6198   Label L;
6199   jcc(negated_cond, L);
6200   atomic_incl(counter_addr);
6201   bind(L);
6202 }
6203 
6204 int MacroAssembler::corrected_idivl(Register reg) {
6205   // Full implementation of Java idiv and irem; checks for
6206   // special case as described in JVM spec., p.243 & p.271.
6207   // The function returns the (pc) offset of the idivl
6208   // instruction - may be needed for implicit exceptions.
6209   //
6210   //         normal case                           special case
6211   //
6212   // input : rax,: dividend                         min_int
6213   //         reg: divisor   (may not be rax,/rdx)   -1
6214   //
6215   // output: rax,: quotient  (= rax, idiv reg)       min_int
6216   //         rdx: remainder (= rax, irem reg)       0
6217   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
6218   const int min_int = 0x80000000;
6219   Label normal_case, special_case;
6220 
6221   // check for special case
6222   cmpl(rax, min_int);
6223   jcc(Assembler::notEqual, normal_case);
6224   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
6225   cmpl(reg, -1);
6226   jcc(Assembler::equal, special_case);
6227 
6228   // handle normal case
6229   bind(normal_case);
6230   cdql();
6231   int idivl_offset = offset();
6232   idivl(reg);
6233 
6234   // normal and special case exit
6235   bind(special_case);
6236 
6237   return idivl_offset;
6238 }
6239 
6240 
6241 
6242 void MacroAssembler::decrementl(Register reg, int value) {
6243   if (value == min_jint) {subl(reg, value) ; return; }
6244   if (value <  0) { incrementl(reg, -value); return; }
6245   if (value == 0) {                        ; return; }
6246   if (value == 1 && UseIncDec) { decl(reg) ; return; }
6247   /* else */      { subl(reg, value)       ; return; }
6248 }
6249 
6250 void MacroAssembler::decrementl(Address dst, int value) {
6251   if (value == min_jint) {subl(dst, value) ; return; }
6252   if (value <  0) { incrementl(dst, -value); return; }
6253   if (value == 0) {                        ; return; }
6254   if (value == 1 && UseIncDec) { decl(dst) ; return; }
6255   /* else */      { subl(dst, value)       ; return; }
6256 }
6257 
6258 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
6259   assert (shift_value > 0, "illegal shift value");
6260   Label _is_positive;
6261   testl (reg, reg);
6262   jcc (Assembler::positive, _is_positive);
6263   int offset = (1 << shift_value) - 1 ;
6264 
6265   if (offset == 1) {
6266     incrementl(reg);
6267   } else {
6268     addl(reg, offset);
6269   }
6270 
6271   bind (_is_positive);
6272   sarl(reg, shift_value);
6273 }
6274 
6275 // !defined(COMPILER2) is because of stupid core builds
6276 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2)
6277 void MacroAssembler::empty_FPU_stack() {
6278   if (VM_Version::supports_mmx()) {
6279     emms();
6280   } else {
6281     for (int i = 8; i-- > 0; ) ffree(i);
6282   }
6283 }
6284 #endif // !LP64 || C1 || !C2
6285 
6286 
6287 // Defines obj, preserves var_size_in_bytes
6288 void MacroAssembler::eden_allocate(Register obj,
6289                                    Register var_size_in_bytes,
6290                                    int con_size_in_bytes,
6291                                    Register t1,
6292                                    Label& slow_case) {
6293   assert(obj == rax, "obj must be in rax, for cmpxchg");
6294   assert_different_registers(obj, var_size_in_bytes, t1);
6295   if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
6296     jmp(slow_case);
6297   } else {
6298     Register end = t1;
6299     Label retry;
6300     bind(retry);
6301     ExternalAddress heap_top((address) Universe::heap()->top_addr());
6302     movptr(obj, heap_top);
6303     if (var_size_in_bytes == noreg) {
6304       lea(end, Address(obj, con_size_in_bytes));
6305     } else {
6306       lea(end, Address(obj, var_size_in_bytes, Address::times_1));
6307     }
6308     // if end < obj then we wrapped around => object too long => slow case
6309     cmpptr(end, obj);
6310     jcc(Assembler::below, slow_case);
6311     cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr()));
6312     jcc(Assembler::above, slow_case);
6313     // Compare obj with the top addr, and if still equal, store the new top addr in
6314     // end at the address of the top addr pointer. Sets ZF if was equal, and clears
6315     // it otherwise. Use lock prefix for atomicity on MPs.
6316     locked_cmpxchgptr(end, heap_top);
6317     jcc(Assembler::notEqual, retry);
6318   }
6319 }
6320 
6321 void MacroAssembler::enter() {
6322   push(rbp);
6323   mov(rbp, rsp);
6324 }
6325 
6326 void MacroAssembler::fcmp(Register tmp) {
6327   fcmp(tmp, 1, true, true);
6328 }
6329 
6330 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) {
6331   assert(!pop_right || pop_left, "usage error");
6332   if (VM_Version::supports_cmov()) {
6333     assert(tmp == noreg, "unneeded temp");
6334     if (pop_left) {
6335       fucomip(index);
6336     } else {
6337       fucomi(index);
6338     }
6339     if (pop_right) {
6340       fpop();
6341     }
6342   } else {
6343     assert(tmp != noreg, "need temp");
6344     if (pop_left) {
6345       if (pop_right) {
6346         fcompp();
6347       } else {
6348         fcomp(index);
6349       }
6350     } else {
6351       fcom(index);
6352     }
6353     // convert FPU condition into eflags condition via rax,
6354     save_rax(tmp);
6355     fwait(); fnstsw_ax();
6356     sahf();
6357     restore_rax(tmp);
6358   }
6359   // condition codes set as follows:
6360   //
6361   // CF (corresponds to C0) if x < y
6362   // PF (corresponds to C2) if unordered
6363   // ZF (corresponds to C3) if x = y
6364 }
6365 
6366 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) {
6367   fcmp2int(dst, unordered_is_less, 1, true, true);
6368 }
6369 
6370 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) {
6371   fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right);
6372   Label L;
6373   if (unordered_is_less) {
6374     movl(dst, -1);
6375     jcc(Assembler::parity, L);
6376     jcc(Assembler::below , L);
6377     movl(dst, 0);
6378     jcc(Assembler::equal , L);
6379     increment(dst);
6380   } else { // unordered is greater
6381     movl(dst, 1);
6382     jcc(Assembler::parity, L);
6383     jcc(Assembler::above , L);
6384     movl(dst, 0);
6385     jcc(Assembler::equal , L);
6386     decrementl(dst);
6387   }
6388   bind(L);
6389 }
6390 
6391 void MacroAssembler::fld_d(AddressLiteral src) {
6392   fld_d(as_Address(src));
6393 }
6394 
6395 void MacroAssembler::fld_s(AddressLiteral src) {
6396   fld_s(as_Address(src));
6397 }
6398 
6399 void MacroAssembler::fld_x(AddressLiteral src) {
6400   Assembler::fld_x(as_Address(src));
6401 }
6402 
6403 void MacroAssembler::fldcw(AddressLiteral src) {
6404   Assembler::fldcw(as_Address(src));
6405 }
6406 
6407 void MacroAssembler::fpop() {
6408   ffree();
6409   fincstp();
6410 }
6411 
6412 void MacroAssembler::fremr(Register tmp) {
6413   save_rax(tmp);
6414   { Label L;
6415     bind(L);
6416     fprem();
6417     fwait(); fnstsw_ax();
6418 #ifdef _LP64
6419     testl(rax, 0x400);
6420     jcc(Assembler::notEqual, L);
6421 #else
6422     sahf();
6423     jcc(Assembler::parity, L);
6424 #endif // _LP64
6425   }
6426   restore_rax(tmp);
6427   // Result is in ST0.
6428   // Note: fxch & fpop to get rid of ST1
6429   // (otherwise FPU stack could overflow eventually)
6430   fxch(1);
6431   fpop();
6432 }
6433 
6434 
6435 void MacroAssembler::incrementl(AddressLiteral dst) {
6436   if (reachable(dst)) {
6437     incrementl(as_Address(dst));
6438   } else {
6439     lea(rscratch1, dst);
6440     incrementl(Address(rscratch1, 0));
6441   }
6442 }
6443 
6444 void MacroAssembler::incrementl(ArrayAddress dst) {
6445   incrementl(as_Address(dst));
6446 }
6447 
6448 void MacroAssembler::incrementl(Register reg, int value) {
6449   if (value == min_jint) {addl(reg, value) ; return; }
6450   if (value <  0) { decrementl(reg, -value); return; }
6451   if (value == 0) {                        ; return; }
6452   if (value == 1 && UseIncDec) { incl(reg) ; return; }
6453   /* else */      { addl(reg, value)       ; return; }
6454 }
6455 
6456 void MacroAssembler::incrementl(Address dst, int value) {
6457   if (value == min_jint) {addl(dst, value) ; return; }
6458   if (value <  0) { decrementl(dst, -value); return; }
6459   if (value == 0) {                        ; return; }
6460   if (value == 1 && UseIncDec) { incl(dst) ; return; }
6461   /* else */      { addl(dst, value)       ; return; }
6462 }
6463 
6464 void MacroAssembler::jump(AddressLiteral dst) {
6465   if (reachable(dst)) {
6466     jmp_literal(dst.target(), dst.rspec());
6467   } else {
6468     lea(rscratch1, dst);
6469     jmp(rscratch1);
6470   }
6471 }
6472 
6473 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) {
6474   if (reachable(dst)) {
6475     InstructionMark im(this);
6476     relocate(dst.reloc());
6477     const int short_size = 2;
6478     const int long_size = 6;
6479     int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos);
6480     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
6481       // 0111 tttn #8-bit disp
6482       emit_byte(0x70 | cc);
6483       emit_byte((offs - short_size) & 0xFF);
6484     } else {
6485       // 0000 1111 1000 tttn #32-bit disp
6486       emit_byte(0x0F);
6487       emit_byte(0x80 | cc);
6488       emit_long(offs - long_size);
6489     }
6490   } else {
6491 #ifdef ASSERT
6492     warning("reversing conditional branch");
6493 #endif /* ASSERT */
6494     Label skip;
6495     jccb(reverse[cc], skip);
6496     lea(rscratch1, dst);
6497     Assembler::jmp(rscratch1);
6498     bind(skip);
6499   }
6500 }
6501 
6502 void MacroAssembler::ldmxcsr(AddressLiteral src) {
6503   if (reachable(src)) {
6504     Assembler::ldmxcsr(as_Address(src));
6505   } else {
6506     lea(rscratch1, src);
6507     Assembler::ldmxcsr(Address(rscratch1, 0));
6508   }
6509 }
6510 
6511 int MacroAssembler::load_signed_byte(Register dst, Address src) {
6512   int off;
6513   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6514     off = offset();
6515     movsbl(dst, src); // movsxb
6516   } else {
6517     off = load_unsigned_byte(dst, src);
6518     shll(dst, 24);
6519     sarl(dst, 24);
6520   }
6521   return off;
6522 }
6523 
6524 // Note: load_signed_short used to be called load_signed_word.
6525 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
6526 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
6527 // The term "word" in HotSpot means a 32- or 64-bit machine word.
6528 int MacroAssembler::load_signed_short(Register dst, Address src) {
6529   int off;
6530   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6531     // This is dubious to me since it seems safe to do a signed 16 => 64 bit
6532     // version but this is what 64bit has always done. This seems to imply
6533     // that users are only using 32bits worth.
6534     off = offset();
6535     movswl(dst, src); // movsxw
6536   } else {
6537     off = load_unsigned_short(dst, src);
6538     shll(dst, 16);
6539     sarl(dst, 16);
6540   }
6541   return off;
6542 }
6543 
6544 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
6545   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
6546   // and "3.9 Partial Register Penalties", p. 22).
6547   int off;
6548   if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) {
6549     off = offset();
6550     movzbl(dst, src); // movzxb
6551   } else {
6552     xorl(dst, dst);
6553     off = offset();
6554     movb(dst, src);
6555   }
6556   return off;
6557 }
6558 
6559 // Note: load_unsigned_short used to be called load_unsigned_word.
6560 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
6561   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
6562   // and "3.9 Partial Register Penalties", p. 22).
6563   int off;
6564   if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) {
6565     off = offset();
6566     movzwl(dst, src); // movzxw
6567   } else {
6568     xorl(dst, dst);
6569     off = offset();
6570     movw(dst, src);
6571   }
6572   return off;
6573 }
6574 
6575 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
6576   switch (size_in_bytes) {
6577 #ifndef _LP64
6578   case  8:
6579     assert(dst2 != noreg, "second dest register required");
6580     movl(dst,  src);
6581     movl(dst2, src.plus_disp(BytesPerInt));
6582     break;
6583 #else
6584   case  8:  movq(dst, src); break;
6585 #endif
6586   case  4:  movl(dst, src); break;
6587   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
6588   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
6589   default:  ShouldNotReachHere();
6590   }
6591 }
6592 
6593 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
6594   switch (size_in_bytes) {
6595 #ifndef _LP64
6596   case  8:
6597     assert(src2 != noreg, "second source register required");
6598     movl(dst,                        src);
6599     movl(dst.plus_disp(BytesPerInt), src2);
6600     break;
6601 #else
6602   case  8:  movq(dst, src); break;
6603 #endif
6604   case  4:  movl(dst, src); break;
6605   case  2:  movw(dst, src); break;
6606   case  1:  movb(dst, src); break;
6607   default:  ShouldNotReachHere();
6608   }
6609 }
6610 
6611 void MacroAssembler::mov32(AddressLiteral dst, Register src) {
6612   if (reachable(dst)) {
6613     movl(as_Address(dst), src);
6614   } else {
6615     lea(rscratch1, dst);
6616     movl(Address(rscratch1, 0), src);
6617   }
6618 }
6619 
6620 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
6621   if (reachable(src)) {
6622     movl(dst, as_Address(src));
6623   } else {
6624     lea(rscratch1, src);
6625     movl(dst, Address(rscratch1, 0));
6626   }
6627 }
6628 
6629 // C++ bool manipulation
6630 
6631 void MacroAssembler::movbool(Register dst, Address src) {
6632   if(sizeof(bool) == 1)
6633     movb(dst, src);
6634   else if(sizeof(bool) == 2)
6635     movw(dst, src);
6636   else if(sizeof(bool) == 4)
6637     movl(dst, src);
6638   else
6639     // unsupported
6640     ShouldNotReachHere();
6641 }
6642 
6643 void MacroAssembler::movbool(Address dst, bool boolconst) {
6644   if(sizeof(bool) == 1)
6645     movb(dst, (int) boolconst);
6646   else if(sizeof(bool) == 2)
6647     movw(dst, (int) boolconst);
6648   else if(sizeof(bool) == 4)
6649     movl(dst, (int) boolconst);
6650   else
6651     // unsupported
6652     ShouldNotReachHere();
6653 }
6654 
6655 void MacroAssembler::movbool(Address dst, Register src) {
6656   if(sizeof(bool) == 1)
6657     movb(dst, src);
6658   else if(sizeof(bool) == 2)
6659     movw(dst, src);
6660   else if(sizeof(bool) == 4)
6661     movl(dst, src);
6662   else
6663     // unsupported
6664     ShouldNotReachHere();
6665 }
6666 
6667 void MacroAssembler::movbyte(ArrayAddress dst, int src) {
6668   movb(as_Address(dst), src);
6669 }
6670 
6671 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) {
6672   if (reachable(src)) {
6673     if (UseXmmLoadAndClearUpper) {
6674       movsd (dst, as_Address(src));
6675     } else {
6676       movlpd(dst, as_Address(src));
6677     }
6678   } else {
6679     lea(rscratch1, src);
6680     if (UseXmmLoadAndClearUpper) {
6681       movsd (dst, Address(rscratch1, 0));
6682     } else {
6683       movlpd(dst, Address(rscratch1, 0));
6684     }
6685   }
6686 }
6687 
6688 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) {
6689   if (reachable(src)) {
6690     movss(dst, as_Address(src));
6691   } else {
6692     lea(rscratch1, src);
6693     movss(dst, Address(rscratch1, 0));
6694   }
6695 }
6696 
6697 void MacroAssembler::movptr(Register dst, Register src) {
6698   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6699 }
6700 
6701 void MacroAssembler::movptr(Register dst, Address src) {
6702   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6703 }
6704 
6705 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
6706 void MacroAssembler::movptr(Register dst, intptr_t src) {
6707   LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src));
6708 }
6709 
6710 void MacroAssembler::movptr(Address dst, Register src) {
6711   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6712 }
6713 
6714 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) {
6715   if (reachable(src)) {
6716     movss(dst, as_Address(src));
6717   } else {
6718     lea(rscratch1, src);
6719     movss(dst, Address(rscratch1, 0));
6720   }
6721 }
6722 
6723 void MacroAssembler::null_check(Register reg, int offset) {
6724   if (needs_explicit_null_check(offset)) {
6725     // provoke OS NULL exception if reg = NULL by
6726     // accessing M[reg] w/o changing any (non-CC) registers
6727     // NOTE: cmpl is plenty here to provoke a segv
6728     cmpptr(rax, Address(reg, 0));
6729     // Note: should probably use testl(rax, Address(reg, 0));
6730     //       may be shorter code (however, this version of
6731     //       testl needs to be implemented first)
6732   } else {
6733     // nothing to do, (later) access of M[reg + offset]
6734     // will provoke OS NULL exception if reg = NULL
6735   }
6736 }
6737 
6738 void MacroAssembler::os_breakpoint() {
6739   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
6740   // (e.g., MSVC can't call ps() otherwise)
6741   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
6742 }
6743 
6744 void MacroAssembler::pop_CPU_state() {
6745   pop_FPU_state();
6746   pop_IU_state();
6747 }
6748 
6749 void MacroAssembler::pop_FPU_state() {
6750   NOT_LP64(frstor(Address(rsp, 0));)
6751   LP64_ONLY(fxrstor(Address(rsp, 0));)
6752   addptr(rsp, FPUStateSizeInWords * wordSize);
6753 }
6754 
6755 void MacroAssembler::pop_IU_state() {
6756   popa();
6757   LP64_ONLY(addq(rsp, 8));
6758   popf();
6759 }
6760 
6761 // Save Integer and Float state
6762 // Warning: Stack must be 16 byte aligned (64bit)
6763 void MacroAssembler::push_CPU_state() {
6764   push_IU_state();
6765   push_FPU_state();
6766 }
6767 
6768 void MacroAssembler::push_FPU_state() {
6769   subptr(rsp, FPUStateSizeInWords * wordSize);
6770 #ifndef _LP64
6771   fnsave(Address(rsp, 0));
6772   fwait();
6773 #else
6774   fxsave(Address(rsp, 0));
6775 #endif // LP64
6776 }
6777 
6778 void MacroAssembler::push_IU_state() {
6779   // Push flags first because pusha kills them
6780   pushf();
6781   // Make sure rsp stays 16-byte aligned
6782   LP64_ONLY(subq(rsp, 8));
6783   pusha();
6784 }
6785 
6786 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) {
6787   // determine java_thread register
6788   if (!java_thread->is_valid()) {
6789     java_thread = rdi;
6790     get_thread(java_thread);
6791   }
6792   // we must set sp to zero to clear frame
6793   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
6794   if (clear_fp) {
6795     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
6796   }
6797 
6798   if (clear_pc)
6799     movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
6800 
6801 }
6802 
6803 void MacroAssembler::restore_rax(Register tmp) {
6804   if (tmp == noreg) pop(rax);
6805   else if (tmp != rax) mov(rax, tmp);
6806 }
6807 
6808 void MacroAssembler::round_to(Register reg, int modulus) {
6809   addptr(reg, modulus - 1);
6810   andptr(reg, -modulus);
6811 }
6812 
6813 void MacroAssembler::save_rax(Register tmp) {
6814   if (tmp == noreg) push(rax);
6815   else if (tmp != rax) mov(tmp, rax);
6816 }
6817 
6818 // Write serialization page so VM thread can do a pseudo remote membar.
6819 // We use the current thread pointer to calculate a thread specific
6820 // offset to write to within the page. This minimizes bus traffic
6821 // due to cache line collision.
6822 void MacroAssembler::serialize_memory(Register thread, Register tmp) {
6823   movl(tmp, thread);
6824   shrl(tmp, os::get_serialize_page_shift_count());
6825   andl(tmp, (os::vm_page_size() - sizeof(int)));
6826 
6827   Address index(noreg, tmp, Address::times_1);
6828   ExternalAddress page(os::get_memory_serialize_page());
6829 
6830   // Size of store must match masking code above
6831   movl(as_Address(ArrayAddress(page, index)), tmp);
6832 }
6833 
6834 // Calls to C land
6835 //
6836 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
6837 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
6838 // has to be reset to 0. This is required to allow proper stack traversal.
6839 void MacroAssembler::set_last_Java_frame(Register java_thread,
6840                                          Register last_java_sp,
6841                                          Register last_java_fp,
6842                                          address  last_java_pc) {
6843   // determine java_thread register
6844   if (!java_thread->is_valid()) {
6845     java_thread = rdi;
6846     get_thread(java_thread);
6847   }
6848   // determine last_java_sp register
6849   if (!last_java_sp->is_valid()) {
6850     last_java_sp = rsp;
6851   }
6852 
6853   // last_java_fp is optional
6854 
6855   if (last_java_fp->is_valid()) {
6856     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
6857   }
6858 
6859   // last_java_pc is optional
6860 
6861   if (last_java_pc != NULL) {
6862     lea(Address(java_thread,
6863                  JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()),
6864         InternalAddress(last_java_pc));
6865 
6866   }
6867   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
6868 }
6869 
6870 void MacroAssembler::shlptr(Register dst, int imm8) {
6871   LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8));
6872 }
6873 
6874 void MacroAssembler::shrptr(Register dst, int imm8) {
6875   LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8));
6876 }
6877 
6878 void MacroAssembler::sign_extend_byte(Register reg) {
6879   if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) {
6880     movsbl(reg, reg); // movsxb
6881   } else {
6882     shll(reg, 24);
6883     sarl(reg, 24);
6884   }
6885 }
6886 
6887 void MacroAssembler::sign_extend_short(Register reg) {
6888   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6889     movswl(reg, reg); // movsxw
6890   } else {
6891     shll(reg, 16);
6892     sarl(reg, 16);
6893   }
6894 }
6895 
6896 void MacroAssembler::testl(Register dst, AddressLiteral src) {
6897   assert(reachable(src), "Address should be reachable");
6898   testl(dst, as_Address(src));
6899 }
6900 
6901 //////////////////////////////////////////////////////////////////////////////////
6902 #ifndef SERIALGC
6903 
6904 void MacroAssembler::g1_write_barrier_pre(Register obj,
6905 #ifndef _LP64
6906                                           Register thread,
6907 #endif
6908                                           Register tmp,
6909                                           Register tmp2,
6910                                           bool tosca_live) {
6911   LP64_ONLY(Register thread = r15_thread;)
6912   Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6913                                        PtrQueue::byte_offset_of_active()));
6914 
6915   Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6916                                        PtrQueue::byte_offset_of_index()));
6917   Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6918                                        PtrQueue::byte_offset_of_buf()));
6919 
6920 
6921   Label done;
6922   Label runtime;
6923 
6924   // if (!marking_in_progress) goto done;
6925   if (in_bytes(PtrQueue::byte_width_of_active()) == 4) {
6926     cmpl(in_progress, 0);
6927   } else {
6928     assert(in_bytes(PtrQueue::byte_width_of_active()) == 1, "Assumption");
6929     cmpb(in_progress, 0);
6930   }
6931   jcc(Assembler::equal, done);
6932 
6933   // if (x.f == NULL) goto done;
6934 #ifdef _LP64
6935   load_heap_oop(tmp2, Address(obj, 0));
6936 #else
6937   movptr(tmp2, Address(obj, 0));
6938 #endif
6939   cmpptr(tmp2, (int32_t) NULL_WORD);
6940   jcc(Assembler::equal, done);
6941 
6942   // Can we store original value in the thread's buffer?
6943 
6944 #ifdef _LP64
6945   movslq(tmp, index);
6946   cmpq(tmp, 0);
6947 #else
6948   cmpl(index, 0);
6949 #endif
6950   jcc(Assembler::equal, runtime);
6951 #ifdef _LP64
6952   subq(tmp, wordSize);
6953   movl(index, tmp);
6954   addq(tmp, buffer);
6955 #else
6956   subl(index, wordSize);
6957   movl(tmp, buffer);
6958   addl(tmp, index);
6959 #endif
6960   movptr(Address(tmp, 0), tmp2);
6961   jmp(done);
6962   bind(runtime);
6963   // save the live input values
6964   if(tosca_live) push(rax);
6965   push(obj);
6966 #ifdef _LP64
6967   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, r15_thread);
6968 #else
6969   push(thread);
6970   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, thread);
6971   pop(thread);
6972 #endif
6973   pop(obj);
6974   if(tosca_live) pop(rax);
6975   bind(done);
6976 
6977 }
6978 
6979 void MacroAssembler::g1_write_barrier_post(Register store_addr,
6980                                            Register new_val,
6981 #ifndef _LP64
6982                                            Register thread,
6983 #endif
6984                                            Register tmp,
6985                                            Register tmp2) {
6986 
6987   LP64_ONLY(Register thread = r15_thread;)
6988   Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
6989                                        PtrQueue::byte_offset_of_index()));
6990   Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
6991                                        PtrQueue::byte_offset_of_buf()));
6992   BarrierSet* bs = Universe::heap()->barrier_set();
6993   CardTableModRefBS* ct = (CardTableModRefBS*)bs;
6994   Label done;
6995   Label runtime;
6996 
6997   // Does store cross heap regions?
6998 
6999   movptr(tmp, store_addr);
7000   xorptr(tmp, new_val);
7001   shrptr(tmp, HeapRegion::LogOfHRGrainBytes);
7002   jcc(Assembler::equal, done);
7003 
7004   // crosses regions, storing NULL?
7005 
7006   cmpptr(new_val, (int32_t) NULL_WORD);
7007   jcc(Assembler::equal, done);
7008 
7009   // storing region crossing non-NULL, is card already dirty?
7010 
7011   ExternalAddress cardtable((address) ct->byte_map_base);
7012   assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
7013 #ifdef _LP64
7014   const Register card_addr = tmp;
7015 
7016   movq(card_addr, store_addr);
7017   shrq(card_addr, CardTableModRefBS::card_shift);
7018 
7019   lea(tmp2, cardtable);
7020 
7021   // get the address of the card
7022   addq(card_addr, tmp2);
7023 #else
7024   const Register card_index = tmp;
7025 
7026   movl(card_index, store_addr);
7027   shrl(card_index, CardTableModRefBS::card_shift);
7028 
7029   Address index(noreg, card_index, Address::times_1);
7030   const Register card_addr = tmp;
7031   lea(card_addr, as_Address(ArrayAddress(cardtable, index)));
7032 #endif
7033   cmpb(Address(card_addr, 0), 0);
7034   jcc(Assembler::equal, done);
7035 
7036   // storing a region crossing, non-NULL oop, card is clean.
7037   // dirty card and log.
7038 
7039   movb(Address(card_addr, 0), 0);
7040 
7041   cmpl(queue_index, 0);
7042   jcc(Assembler::equal, runtime);
7043   subl(queue_index, wordSize);
7044   movptr(tmp2, buffer);
7045 #ifdef _LP64
7046   movslq(rscratch1, queue_index);
7047   addq(tmp2, rscratch1);
7048   movq(Address(tmp2, 0), card_addr);
7049 #else
7050   addl(tmp2, queue_index);
7051   movl(Address(tmp2, 0), card_index);
7052 #endif
7053   jmp(done);
7054 
7055   bind(runtime);
7056   // save the live input values
7057   push(store_addr);
7058   push(new_val);
7059 #ifdef _LP64
7060   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, r15_thread);
7061 #else
7062   push(thread);
7063   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
7064   pop(thread);
7065 #endif
7066   pop(new_val);
7067   pop(store_addr);
7068 
7069   bind(done);
7070 
7071 }
7072 
7073 #endif // SERIALGC
7074 //////////////////////////////////////////////////////////////////////////////////
7075 
7076 
7077 void MacroAssembler::store_check(Register obj) {
7078   // Does a store check for the oop in register obj. The content of
7079   // register obj is destroyed afterwards.
7080   store_check_part_1(obj);
7081   store_check_part_2(obj);
7082 }
7083 
7084 void MacroAssembler::store_check(Register obj, Address dst) {
7085   store_check(obj);
7086 }
7087 
7088 
7089 // split the store check operation so that other instructions can be scheduled inbetween
7090 void MacroAssembler::store_check_part_1(Register obj) {
7091   BarrierSet* bs = Universe::heap()->barrier_set();
7092   assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
7093   shrptr(obj, CardTableModRefBS::card_shift);
7094 }
7095 
7096 void MacroAssembler::store_check_part_2(Register obj) {
7097   BarrierSet* bs = Universe::heap()->barrier_set();
7098   assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
7099   CardTableModRefBS* ct = (CardTableModRefBS*)bs;
7100   assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
7101 
7102   // The calculation for byte_map_base is as follows:
7103   // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift);
7104   // So this essentially converts an address to a displacement and
7105   // it will never need to be relocated. On 64bit however the value may be too
7106   // large for a 32bit displacement
7107 
7108   intptr_t disp = (intptr_t) ct->byte_map_base;
7109   if (is_simm32(disp)) {
7110     Address cardtable(noreg, obj, Address::times_1, disp);
7111     movb(cardtable, 0);
7112   } else {
7113     // By doing it as an ExternalAddress disp could be converted to a rip-relative
7114     // displacement and done in a single instruction given favorable mapping and
7115     // a smarter version of as_Address. Worst case it is two instructions which
7116     // is no worse off then loading disp into a register and doing as a simple
7117     // Address() as above.
7118     // We can't do as ExternalAddress as the only style since if disp == 0 we'll
7119     // assert since NULL isn't acceptable in a reloci (see 6644928). In any case
7120     // in some cases we'll get a single instruction version.
7121 
7122     ExternalAddress cardtable((address)disp);
7123     Address index(noreg, obj, Address::times_1);
7124     movb(as_Address(ArrayAddress(cardtable, index)), 0);
7125   }
7126 }
7127 
7128 void MacroAssembler::subptr(Register dst, int32_t imm32) {
7129   LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32));
7130 }
7131 
7132 void MacroAssembler::subptr(Register dst, Register src) {
7133   LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src));
7134 }
7135 
7136 // C++ bool manipulation
7137 void MacroAssembler::testbool(Register dst) {
7138   if(sizeof(bool) == 1)
7139     testb(dst, 0xff);
7140   else if(sizeof(bool) == 2) {
7141     // testw implementation needed for two byte bools
7142     ShouldNotReachHere();
7143   } else if(sizeof(bool) == 4)
7144     testl(dst, dst);
7145   else
7146     // unsupported
7147     ShouldNotReachHere();
7148 }
7149 
7150 void MacroAssembler::testptr(Register dst, Register src) {
7151   LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
7152 }
7153 
7154 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
7155 void MacroAssembler::tlab_allocate(Register obj,
7156                                    Register var_size_in_bytes,
7157                                    int con_size_in_bytes,
7158                                    Register t1,
7159                                    Register t2,
7160                                    Label& slow_case) {
7161   assert_different_registers(obj, t1, t2);
7162   assert_different_registers(obj, var_size_in_bytes, t1);
7163   Register end = t2;
7164   Register thread = NOT_LP64(t1) LP64_ONLY(r15_thread);
7165 
7166   verify_tlab();
7167 
7168   NOT_LP64(get_thread(thread));
7169 
7170   movptr(obj, Address(thread, JavaThread::tlab_top_offset()));
7171   if (var_size_in_bytes == noreg) {
7172     lea(end, Address(obj, con_size_in_bytes));
7173   } else {
7174     lea(end, Address(obj, var_size_in_bytes, Address::times_1));
7175   }
7176   cmpptr(end, Address(thread, JavaThread::tlab_end_offset()));
7177   jcc(Assembler::above, slow_case);
7178 
7179   // update the tlab top pointer
7180   movptr(Address(thread, JavaThread::tlab_top_offset()), end);
7181 
7182   // recover var_size_in_bytes if necessary
7183   if (var_size_in_bytes == end) {
7184     subptr(var_size_in_bytes, obj);
7185   }
7186   verify_tlab();
7187 }
7188 
7189 // Preserves rbx, and rdx.
7190 Register MacroAssembler::tlab_refill(Label& retry,
7191                                      Label& try_eden,
7192                                      Label& slow_case) {
7193   Register top = rax;
7194   Register t1  = rcx;
7195   Register t2  = rsi;
7196   Register thread_reg = NOT_LP64(rdi) LP64_ONLY(r15_thread);
7197   assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx);
7198   Label do_refill, discard_tlab;
7199 
7200   if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
7201     // No allocation in the shared eden.
7202     jmp(slow_case);
7203   }
7204 
7205   NOT_LP64(get_thread(thread_reg));
7206 
7207   movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7208   movptr(t1,  Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
7209 
7210   // calculate amount of free space
7211   subptr(t1, top);
7212   shrptr(t1, LogHeapWordSize);
7213 
7214   // Retain tlab and allocate object in shared space if
7215   // the amount free in the tlab is too large to discard.
7216   cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())));
7217   jcc(Assembler::lessEqual, discard_tlab);
7218 
7219   // Retain
7220   // %%% yuck as movptr...
7221   movptr(t2, (int32_t) ThreadLocalAllocBuffer::refill_waste_limit_increment());
7222   addptr(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2);
7223   if (TLABStats) {
7224     // increment number of slow_allocations
7225     addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1);
7226   }
7227   jmp(try_eden);
7228 
7229   bind(discard_tlab);
7230   if (TLABStats) {
7231     // increment number of refills
7232     addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1);
7233     // accumulate wastage -- t1 is amount free in tlab
7234     addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1);
7235   }
7236 
7237   // if tlab is currently allocated (top or end != null) then
7238   // fill [top, end + alignment_reserve) with array object
7239   testptr(top, top);
7240   jcc(Assembler::zero, do_refill);
7241 
7242   // set up the mark word
7243   movptr(Address(top, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::prototype()->copy_set_hash(0x2));
7244   // set the length to the remaining space
7245   subptr(t1, typeArrayOopDesc::header_size(T_INT));
7246   addptr(t1, (int32_t)ThreadLocalAllocBuffer::alignment_reserve());
7247   shlptr(t1, log2_intptr(HeapWordSize/sizeof(jint)));
7248   movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1);
7249   // set klass to intArrayKlass
7250   // dubious reloc why not an oop reloc?
7251   movptr(t1, ExternalAddress((address)Universe::intArrayKlassObj_addr()));
7252   // store klass last.  concurrent gcs assumes klass length is valid if
7253   // klass field is not null.
7254   store_klass(top, t1);
7255 
7256   movptr(t1, top);
7257   subptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
7258   incr_allocated_bytes(thread_reg, t1, 0);
7259 
7260   // refill the tlab with an eden allocation
7261   bind(do_refill);
7262   movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
7263   shlptr(t1, LogHeapWordSize);
7264   // allocate new tlab, address returned in top
7265   eden_allocate(top, t1, 0, t2, slow_case);
7266 
7267   // Check that t1 was preserved in eden_allocate.
7268 #ifdef ASSERT
7269   if (UseTLAB) {
7270     Label ok;
7271     Register tsize = rsi;
7272     assert_different_registers(tsize, thread_reg, t1);
7273     push(tsize);
7274     movptr(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
7275     shlptr(tsize, LogHeapWordSize);
7276     cmpptr(t1, tsize);
7277     jcc(Assembler::equal, ok);
7278     stop("assert(t1 != tlab size)");
7279     should_not_reach_here();
7280 
7281     bind(ok);
7282     pop(tsize);
7283   }
7284 #endif
7285   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top);
7286   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top);
7287   addptr(top, t1);
7288   subptr(top, (int32_t)ThreadLocalAllocBuffer::alignment_reserve_in_bytes());
7289   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top);
7290   verify_tlab();
7291   jmp(retry);
7292 
7293   return thread_reg; // for use by caller
7294 }
7295 
7296 void MacroAssembler::incr_allocated_bytes(Register thread,
7297                                           Register var_size_in_bytes,
7298                                           int con_size_in_bytes,
7299                                           Register t1) {
7300 #ifdef _LP64
7301   if (var_size_in_bytes->is_valid()) {
7302     addq(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), var_size_in_bytes);
7303   } else {
7304     addq(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), con_size_in_bytes);
7305   }
7306 #else
7307   if (!thread->is_valid()) {
7308     assert(t1->is_valid(), "need temp reg");
7309     thread = t1;
7310     get_thread(thread);
7311   }
7312 
7313   if (var_size_in_bytes->is_valid()) {
7314     addl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), var_size_in_bytes);
7315   } else {
7316     addl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), con_size_in_bytes);
7317   }
7318   adcl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())+4), 0);
7319 #endif
7320 }
7321 
7322 static const double     pi_4 =  0.7853981633974483;
7323 
7324 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) {
7325   // A hand-coded argument reduction for values in fabs(pi/4, pi/2)
7326   // was attempted in this code; unfortunately it appears that the
7327   // switch to 80-bit precision and back causes this to be
7328   // unprofitable compared with simply performing a runtime call if
7329   // the argument is out of the (-pi/4, pi/4) range.
7330 
7331   Register tmp = noreg;
7332   if (!VM_Version::supports_cmov()) {
7333     // fcmp needs a temporary so preserve rbx,
7334     tmp = rbx;
7335     push(tmp);
7336   }
7337 
7338   Label slow_case, done;
7339 
7340   ExternalAddress pi4_adr = (address)&pi_4;
7341   if (reachable(pi4_adr)) {
7342     // x ?<= pi/4
7343     fld_d(pi4_adr);
7344     fld_s(1);                // Stack:  X  PI/4  X
7345     fabs();                  // Stack: |X| PI/4  X
7346     fcmp(tmp);
7347     jcc(Assembler::above, slow_case);
7348 
7349     // fastest case: -pi/4 <= x <= pi/4
7350     switch(trig) {
7351     case 's':
7352       fsin();
7353       break;
7354     case 'c':
7355       fcos();
7356       break;
7357     case 't':
7358       ftan();
7359       break;
7360     default:
7361       assert(false, "bad intrinsic");
7362       break;
7363     }
7364     jmp(done);
7365   }
7366 
7367   // slow case: runtime call
7368   bind(slow_case);
7369   // Preserve registers across runtime call
7370   pusha();
7371   int incoming_argument_and_return_value_offset = -1;
7372   if (num_fpu_regs_in_use > 1) {
7373     // Must preserve all other FPU regs (could alternatively convert
7374     // SharedRuntime::dsin and dcos into assembly routines known not to trash
7375     // FPU state, but can not trust C compiler)
7376     NEEDS_CLEANUP;
7377     // NOTE that in this case we also push the incoming argument to
7378     // the stack and restore it later; we also use this stack slot to
7379     // hold the return value from dsin or dcos.
7380     for (int i = 0; i < num_fpu_regs_in_use; i++) {
7381       subptr(rsp, sizeof(jdouble));
7382       fstp_d(Address(rsp, 0));
7383     }
7384     incoming_argument_and_return_value_offset = sizeof(jdouble)*(num_fpu_regs_in_use-1);
7385     fld_d(Address(rsp, incoming_argument_and_return_value_offset));
7386   }
7387   subptr(rsp, sizeof(jdouble));
7388   fstp_d(Address(rsp, 0));
7389 #ifdef _LP64
7390   movdbl(xmm0, Address(rsp, 0));
7391 #endif // _LP64
7392 
7393   // NOTE: we must not use call_VM_leaf here because that requires a
7394   // complete interpreter frame in debug mode -- same bug as 4387334
7395   // MacroAssembler::call_VM_leaf_base is perfectly safe and will
7396   // do proper 64bit abi
7397 
7398   NEEDS_CLEANUP;
7399   // Need to add stack banging before this runtime call if it needs to
7400   // be taken; however, there is no generic stack banging routine at
7401   // the MacroAssembler level
7402   switch(trig) {
7403   case 's':
7404     {
7405       MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), 0);
7406     }
7407     break;
7408   case 'c':
7409     {
7410       MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), 0);
7411     }
7412     break;
7413   case 't':
7414     {
7415       MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 0);
7416     }
7417     break;
7418   default:
7419     assert(false, "bad intrinsic");
7420     break;
7421   }
7422 #ifdef _LP64
7423     movsd(Address(rsp, 0), xmm0);
7424     fld_d(Address(rsp, 0));
7425 #endif // _LP64
7426   addptr(rsp, sizeof(jdouble));
7427   if (num_fpu_regs_in_use > 1) {
7428     // Must save return value to stack and then restore entire FPU stack
7429     fstp_d(Address(rsp, incoming_argument_and_return_value_offset));
7430     for (int i = 0; i < num_fpu_regs_in_use; i++) {
7431       fld_d(Address(rsp, 0));
7432       addptr(rsp, sizeof(jdouble));
7433     }
7434   }
7435   popa();
7436 
7437   // Come here with result in F-TOS
7438   bind(done);
7439 
7440   if (tmp != noreg) {
7441     pop(tmp);
7442   }
7443 }
7444 
7445 
7446 // Look up the method for a megamorphic invokeinterface call.
7447 // The target method is determined by <intf_klass, itable_index>.
7448 // The receiver klass is in recv_klass.
7449 // On success, the result will be in method_result, and execution falls through.
7450 // On failure, execution transfers to the given label.
7451 void MacroAssembler::lookup_interface_method(Register recv_klass,
7452                                              Register intf_klass,
7453                                              RegisterOrConstant itable_index,
7454                                              Register method_result,
7455                                              Register scan_temp,
7456                                              Label& L_no_such_interface) {
7457   assert_different_registers(recv_klass, intf_klass, method_result, scan_temp);
7458   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
7459          "caller must use same register for non-constant itable index as for method");
7460 
7461   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
7462   int vtable_base = instanceKlass::vtable_start_offset() * wordSize;
7463   int itentry_off = itableMethodEntry::method_offset_in_bytes();
7464   int scan_step   = itableOffsetEntry::size() * wordSize;
7465   int vte_size    = vtableEntry::size() * wordSize;
7466   Address::ScaleFactor times_vte_scale = Address::times_ptr;
7467   assert(vte_size == wordSize, "else adjust times_vte_scale");
7468 
7469   movl(scan_temp, Address(recv_klass, instanceKlass::vtable_length_offset() * wordSize));
7470 
7471   // %%% Could store the aligned, prescaled offset in the klassoop.
7472   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
7473   if (HeapWordsPerLong > 1) {
7474     // Round up to align_object_offset boundary
7475     // see code for instanceKlass::start_of_itable!
7476     round_to(scan_temp, BytesPerLong);
7477   }
7478 
7479   // Adjust recv_klass by scaled itable_index, so we can free itable_index.
7480   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
7481   lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
7482 
7483   // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) {
7484   //   if (scan->interface() == intf) {
7485   //     result = (klass + scan->offset() + itable_index);
7486   //   }
7487   // }
7488   Label search, found_method;
7489 
7490   for (int peel = 1; peel >= 0; peel--) {
7491     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset_in_bytes()));
7492     cmpptr(intf_klass, method_result);
7493 
7494     if (peel) {
7495       jccb(Assembler::equal, found_method);
7496     } else {
7497       jccb(Assembler::notEqual, search);
7498       // (invert the test to fall through to found_method...)
7499     }
7500 
7501     if (!peel)  break;
7502 
7503     bind(search);
7504 
7505     // Check that the previous entry is non-null.  A null entry means that
7506     // the receiver class doesn't implement the interface, and wasn't the
7507     // same as when the caller was compiled.
7508     testptr(method_result, method_result);
7509     jcc(Assembler::zero, L_no_such_interface);
7510     addptr(scan_temp, scan_step);
7511   }
7512 
7513   bind(found_method);
7514 
7515   // Got a hit.
7516   movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset_in_bytes()));
7517   movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
7518 }
7519 
7520 
7521 void MacroAssembler::check_klass_subtype(Register sub_klass,
7522                            Register super_klass,
7523                            Register temp_reg,
7524                            Label& L_success) {
7525   Label L_failure;
7526   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, NULL);
7527   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL);
7528   bind(L_failure);
7529 }
7530 
7531 
7532 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
7533                                                    Register super_klass,
7534                                                    Register temp_reg,
7535                                                    Label* L_success,
7536                                                    Label* L_failure,
7537                                                    Label* L_slow_path,
7538                                         RegisterOrConstant super_check_offset) {
7539   assert_different_registers(sub_klass, super_klass, temp_reg);
7540   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
7541   if (super_check_offset.is_register()) {
7542     assert_different_registers(sub_klass, super_klass,
7543                                super_check_offset.as_register());
7544   } else if (must_load_sco) {
7545     assert(temp_reg != noreg, "supply either a temp or a register offset");
7546   }
7547 
7548   Label L_fallthrough;
7549   int label_nulls = 0;
7550   if (L_success == NULL)   { L_success   = &L_fallthrough; label_nulls++; }
7551   if (L_failure == NULL)   { L_failure   = &L_fallthrough; label_nulls++; }
7552   if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; }
7553   assert(label_nulls <= 1, "at most one NULL in the batch");
7554 
7555   int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
7556                    Klass::secondary_super_cache_offset_in_bytes());
7557   int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
7558                     Klass::super_check_offset_offset_in_bytes());
7559   Address super_check_offset_addr(super_klass, sco_offset);
7560 
7561   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
7562   // range of a jccb.  If this routine grows larger, reconsider at
7563   // least some of these.
7564 #define local_jcc(assembler_cond, label)                                \
7565   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
7566   else                             jcc( assembler_cond, label) /*omit semi*/
7567 
7568   // Hacked jmp, which may only be used just before L_fallthrough.
7569 #define final_jmp(label)                                                \
7570   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
7571   else                            jmp(label)                /*omit semi*/
7572 
7573   // If the pointers are equal, we are done (e.g., String[] elements).
7574   // This self-check enables sharing of secondary supertype arrays among
7575   // non-primary types such as array-of-interface.  Otherwise, each such
7576   // type would need its own customized SSA.
7577   // We move this check to the front of the fast path because many
7578   // type checks are in fact trivially successful in this manner,
7579   // so we get a nicely predicted branch right at the start of the check.
7580   cmpptr(sub_klass, super_klass);
7581   local_jcc(Assembler::equal, *L_success);
7582 
7583   // Check the supertype display:
7584   if (must_load_sco) {
7585     // Positive movl does right thing on LP64.
7586     movl(temp_reg, super_check_offset_addr);
7587     super_check_offset = RegisterOrConstant(temp_reg);
7588   }
7589   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
7590   cmpptr(super_klass, super_check_addr); // load displayed supertype
7591 
7592   // This check has worked decisively for primary supers.
7593   // Secondary supers are sought in the super_cache ('super_cache_addr').
7594   // (Secondary supers are interfaces and very deeply nested subtypes.)
7595   // This works in the same check above because of a tricky aliasing
7596   // between the super_cache and the primary super display elements.
7597   // (The 'super_check_addr' can address either, as the case requires.)
7598   // Note that the cache is updated below if it does not help us find
7599   // what we need immediately.
7600   // So if it was a primary super, we can just fail immediately.
7601   // Otherwise, it's the slow path for us (no success at this point).
7602 
7603   if (super_check_offset.is_register()) {
7604     local_jcc(Assembler::equal, *L_success);
7605     cmpl(super_check_offset.as_register(), sc_offset);
7606     if (L_failure == &L_fallthrough) {
7607       local_jcc(Assembler::equal, *L_slow_path);
7608     } else {
7609       local_jcc(Assembler::notEqual, *L_failure);
7610       final_jmp(*L_slow_path);
7611     }
7612   } else if (super_check_offset.as_constant() == sc_offset) {
7613     // Need a slow path; fast failure is impossible.
7614     if (L_slow_path == &L_fallthrough) {
7615       local_jcc(Assembler::equal, *L_success);
7616     } else {
7617       local_jcc(Assembler::notEqual, *L_slow_path);
7618       final_jmp(*L_success);
7619     }
7620   } else {
7621     // No slow path; it's a fast decision.
7622     if (L_failure == &L_fallthrough) {
7623       local_jcc(Assembler::equal, *L_success);
7624     } else {
7625       local_jcc(Assembler::notEqual, *L_failure);
7626       final_jmp(*L_success);
7627     }
7628   }
7629 
7630   bind(L_fallthrough);
7631 
7632 #undef local_jcc
7633 #undef final_jmp
7634 }
7635 
7636 
7637 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
7638                                                    Register super_klass,
7639                                                    Register temp_reg,
7640                                                    Register temp2_reg,
7641                                                    Label* L_success,
7642                                                    Label* L_failure,
7643                                                    bool set_cond_codes) {
7644   assert_different_registers(sub_klass, super_klass, temp_reg);
7645   if (temp2_reg != noreg)
7646     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
7647 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
7648 
7649   Label L_fallthrough;
7650   int label_nulls = 0;
7651   if (L_success == NULL)   { L_success   = &L_fallthrough; label_nulls++; }
7652   if (L_failure == NULL)   { L_failure   = &L_fallthrough; label_nulls++; }
7653   assert(label_nulls <= 1, "at most one NULL in the batch");
7654 
7655   // a couple of useful fields in sub_klass:
7656   int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
7657                    Klass::secondary_supers_offset_in_bytes());
7658   int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
7659                    Klass::secondary_super_cache_offset_in_bytes());
7660   Address secondary_supers_addr(sub_klass, ss_offset);
7661   Address super_cache_addr(     sub_klass, sc_offset);
7662 
7663   // Do a linear scan of the secondary super-klass chain.
7664   // This code is rarely used, so simplicity is a virtue here.
7665   // The repne_scan instruction uses fixed registers, which we must spill.
7666   // Don't worry too much about pre-existing connections with the input regs.
7667 
7668   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
7669   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
7670 
7671   // Get super_klass value into rax (even if it was in rdi or rcx).
7672   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
7673   if (super_klass != rax || UseCompressedOops) {
7674     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
7675     mov(rax, super_klass);
7676   }
7677   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
7678   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
7679 
7680 #ifndef PRODUCT
7681   int* pst_counter = &SharedRuntime::_partial_subtype_ctr;
7682   ExternalAddress pst_counter_addr((address) pst_counter);
7683   NOT_LP64(  incrementl(pst_counter_addr) );
7684   LP64_ONLY( lea(rcx, pst_counter_addr) );
7685   LP64_ONLY( incrementl(Address(rcx, 0)) );
7686 #endif //PRODUCT
7687 
7688   // We will consult the secondary-super array.
7689   movptr(rdi, secondary_supers_addr);
7690   // Load the array length.  (Positive movl does right thing on LP64.)
7691   movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
7692   // Skip to start of data.
7693   addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
7694 
7695   // Scan RCX words at [RDI] for an occurrence of RAX.
7696   // Set NZ/Z based on last compare.
7697   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
7698   // not change flags (only scas instruction which is repeated sets flags).
7699   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
7700 #ifdef _LP64
7701   // This part is tricky, as values in supers array could be 32 or 64 bit wide
7702   // and we store values in objArrays always encoded, thus we need to encode
7703   // the value of rax before repne.  Note that rax is dead after the repne.
7704   if (UseCompressedOops) {
7705     encode_heap_oop_not_null(rax); // Changes flags.
7706     // The superclass is never null; it would be a basic system error if a null
7707     // pointer were to sneak in here.  Note that we have already loaded the
7708     // Klass::super_check_offset from the super_klass in the fast path,
7709     // so if there is a null in that register, we are already in the afterlife.
7710     testl(rax,rax); // Set Z = 0
7711     repne_scanl();
7712   } else
7713 #endif // _LP64
7714   {
7715     testptr(rax,rax); // Set Z = 0
7716     repne_scan();
7717   }
7718   // Unspill the temp. registers:
7719   if (pushed_rdi)  pop(rdi);
7720   if (pushed_rcx)  pop(rcx);
7721   if (pushed_rax)  pop(rax);
7722 
7723   if (set_cond_codes) {
7724     // Special hack for the AD files:  rdi is guaranteed non-zero.
7725     assert(!pushed_rdi, "rdi must be left non-NULL");
7726     // Also, the condition codes are properly set Z/NZ on succeed/failure.
7727   }
7728 
7729   if (L_failure == &L_fallthrough)
7730         jccb(Assembler::notEqual, *L_failure);
7731   else  jcc(Assembler::notEqual, *L_failure);
7732 
7733   // Success.  Cache the super we found and proceed in triumph.
7734   movptr(super_cache_addr, super_klass);
7735 
7736   if (L_success != &L_fallthrough) {
7737     jmp(*L_success);
7738   }
7739 
7740 #undef IS_A_TEMP
7741 
7742   bind(L_fallthrough);
7743 }
7744 
7745 
7746 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) {
7747   ucomisd(dst, as_Address(src));
7748 }
7749 
7750 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) {
7751   ucomiss(dst, as_Address(src));
7752 }
7753 
7754 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) {
7755   if (reachable(src)) {
7756     xorpd(dst, as_Address(src));
7757   } else {
7758     lea(rscratch1, src);
7759     xorpd(dst, Address(rscratch1, 0));
7760   }
7761 }
7762 
7763 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) {
7764   if (reachable(src)) {
7765     xorps(dst, as_Address(src));
7766   } else {
7767     lea(rscratch1, src);
7768     xorps(dst, Address(rscratch1, 0));
7769   }
7770 }
7771 
7772 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
7773   if (VM_Version::supports_cmov()) {
7774     cmovl(cc, dst, src);
7775   } else {
7776     Label L;
7777     jccb(negate_condition(cc), L);
7778     movl(dst, src);
7779     bind(L);
7780   }
7781 }
7782 
7783 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
7784   if (VM_Version::supports_cmov()) {
7785     cmovl(cc, dst, src);
7786   } else {
7787     Label L;
7788     jccb(negate_condition(cc), L);
7789     movl(dst, src);
7790     bind(L);
7791   }
7792 }
7793 
7794 void MacroAssembler::verify_oop(Register reg, const char* s) {
7795   if (!VerifyOops) return;
7796 
7797   // Pass register number to verify_oop_subroutine
7798   char* b = new char[strlen(s) + 50];
7799   sprintf(b, "verify_oop: %s: %s", reg->name(), s);
7800 #ifdef _LP64
7801   push(rscratch1);                    // save r10, trashed by movptr()
7802 #endif
7803   push(rax);                          // save rax,
7804   push(reg);                          // pass register argument
7805   ExternalAddress buffer((address) b);
7806   // avoid using pushptr, as it modifies scratch registers
7807   // and our contract is not to modify anything
7808   movptr(rax, buffer.addr());
7809   push(rax);
7810   // call indirectly to solve generation ordering problem
7811   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
7812   call(rax);
7813   // Caller pops the arguments (oop, message) and restores rax, r10
7814 }
7815 
7816 
7817 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr,
7818                                                       Register tmp,
7819                                                       int offset) {
7820   intptr_t value = *delayed_value_addr;
7821   if (value != 0)
7822     return RegisterOrConstant(value + offset);
7823 
7824   // load indirectly to solve generation ordering problem
7825   movptr(tmp, ExternalAddress((address) delayed_value_addr));
7826 
7827 #ifdef ASSERT
7828   { Label L;
7829     testptr(tmp, tmp);
7830     if (WizardMode) {
7831       jcc(Assembler::notZero, L);
7832       char* buf = new char[40];
7833       sprintf(buf, "DelayedValue="INTPTR_FORMAT, delayed_value_addr[1]);
7834       stop(buf);
7835     } else {
7836       jccb(Assembler::notZero, L);
7837       hlt();
7838     }
7839     bind(L);
7840   }
7841 #endif
7842 
7843   if (offset != 0)
7844     addptr(tmp, offset);
7845 
7846   return RegisterOrConstant(tmp);
7847 }
7848 
7849 
7850 // registers on entry:
7851 //  - rax ('check' register): required MethodType
7852 //  - rcx: method handle
7853 //  - rdx, rsi, or ?: killable temp
7854 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg,
7855                                               Register temp_reg,
7856                                               Label& wrong_method_type) {
7857   Address type_addr(mh_reg, delayed_value(java_lang_invoke_MethodHandle::type_offset_in_bytes, temp_reg));
7858   // compare method type against that of the receiver
7859   if (UseCompressedOops) {
7860     load_heap_oop(temp_reg, type_addr);
7861     cmpptr(mtype_reg, temp_reg);
7862   } else {
7863     cmpptr(mtype_reg, type_addr);
7864   }
7865   jcc(Assembler::notEqual, wrong_method_type);
7866 }
7867 
7868 
7869 // A method handle has a "vmslots" field which gives the size of its
7870 // argument list in JVM stack slots.  This field is either located directly
7871 // in every method handle, or else is indirectly accessed through the
7872 // method handle's MethodType.  This macro hides the distinction.
7873 void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg,
7874                                                 Register temp_reg) {
7875   assert_different_registers(vmslots_reg, mh_reg, temp_reg);
7876   // load mh.type.form.vmslots
7877   if (java_lang_invoke_MethodHandle::vmslots_offset_in_bytes() != 0) {
7878     // hoist vmslots into every mh to avoid dependent load chain
7879     movl(vmslots_reg, Address(mh_reg, delayed_value(java_lang_invoke_MethodHandle::vmslots_offset_in_bytes, temp_reg)));
7880   } else {
7881     Register temp2_reg = vmslots_reg;
7882     load_heap_oop(temp2_reg, Address(mh_reg,    delayed_value(java_lang_invoke_MethodHandle::type_offset_in_bytes, temp_reg)));
7883     load_heap_oop(temp2_reg, Address(temp2_reg, delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes, temp_reg)));
7884     movl(vmslots_reg, Address(temp2_reg, delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)));
7885   }
7886 }
7887 
7888 
7889 // registers on entry:
7890 //  - rcx: method handle
7891 //  - rdx: killable temp (interpreted only)
7892 //  - rax: killable temp (compiled only)
7893 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) {
7894   assert(mh_reg == rcx, "caller must put MH object in rcx");
7895   assert_different_registers(mh_reg, temp_reg);
7896 
7897   // pick out the interpreted side of the handler
7898   // NOTE: vmentry is not an oop!
7899   movptr(temp_reg, Address(mh_reg, delayed_value(java_lang_invoke_MethodHandle::vmentry_offset_in_bytes, temp_reg)));
7900 
7901   // off we go...
7902   jmp(Address(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes()));
7903 
7904   // for the various stubs which take control at this point,
7905   // see MethodHandles::generate_method_handle_stub
7906 }
7907 
7908 
7909 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
7910                                          int extra_slot_offset) {
7911   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
7912   int stackElementSize = Interpreter::stackElementSize;
7913   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
7914 #ifdef ASSERT
7915   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
7916   assert(offset1 - offset == stackElementSize, "correct arithmetic");
7917 #endif
7918   Register             scale_reg    = noreg;
7919   Address::ScaleFactor scale_factor = Address::no_scale;
7920   if (arg_slot.is_constant()) {
7921     offset += arg_slot.as_constant() * stackElementSize;
7922   } else {
7923     scale_reg    = arg_slot.as_register();
7924     scale_factor = Address::times(stackElementSize);
7925   }
7926   offset += wordSize;           // return PC is on stack
7927   return Address(rsp, scale_reg, scale_factor, offset);
7928 }
7929 
7930 
7931 void MacroAssembler::verify_oop_addr(Address addr, const char* s) {
7932   if (!VerifyOops) return;
7933 
7934   // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord);
7935   // Pass register number to verify_oop_subroutine
7936   char* b = new char[strlen(s) + 50];
7937   sprintf(b, "verify_oop_addr: %s", s);
7938 
7939 #ifdef _LP64
7940   push(rscratch1);                    // save r10, trashed by movptr()
7941 #endif
7942   push(rax);                          // save rax,
7943   // addr may contain rsp so we will have to adjust it based on the push
7944   // we just did (and on 64 bit we do two pushes)
7945   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
7946   // stores rax into addr which is backwards of what was intended.
7947   if (addr.uses(rsp)) {
7948     lea(rax, addr);
7949     pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
7950   } else {
7951     pushptr(addr);
7952   }
7953 
7954   ExternalAddress buffer((address) b);
7955   // pass msg argument
7956   // avoid using pushptr, as it modifies scratch registers
7957   // and our contract is not to modify anything
7958   movptr(rax, buffer.addr());
7959   push(rax);
7960 
7961   // call indirectly to solve generation ordering problem
7962   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
7963   call(rax);
7964   // Caller pops the arguments (addr, message) and restores rax, r10.
7965 }
7966 
7967 void MacroAssembler::verify_tlab() {
7968 #ifdef ASSERT
7969   if (UseTLAB && VerifyOops) {
7970     Label next, ok;
7971     Register t1 = rsi;
7972     Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread);
7973 
7974     push(t1);
7975     NOT_LP64(push(thread_reg));
7976     NOT_LP64(get_thread(thread_reg));
7977 
7978     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7979     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
7980     jcc(Assembler::aboveEqual, next);
7981     stop("assert(top >= start)");
7982     should_not_reach_here();
7983 
7984     bind(next);
7985     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
7986     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7987     jcc(Assembler::aboveEqual, ok);
7988     stop("assert(top <= end)");
7989     should_not_reach_here();
7990 
7991     bind(ok);
7992     NOT_LP64(pop(thread_reg));
7993     pop(t1);
7994   }
7995 #endif
7996 }
7997 
7998 class ControlWord {
7999  public:
8000   int32_t _value;
8001 
8002   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
8003   int  precision_control() const       { return  (_value >>  8) & 3      ; }
8004   bool precision() const               { return ((_value >>  5) & 1) != 0; }
8005   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
8006   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
8007   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
8008   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
8009   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
8010 
8011   void print() const {
8012     // rounding control
8013     const char* rc;
8014     switch (rounding_control()) {
8015       case 0: rc = "round near"; break;
8016       case 1: rc = "round down"; break;
8017       case 2: rc = "round up  "; break;
8018       case 3: rc = "chop      "; break;
8019     };
8020     // precision control
8021     const char* pc;
8022     switch (precision_control()) {
8023       case 0: pc = "24 bits "; break;
8024       case 1: pc = "reserved"; break;
8025       case 2: pc = "53 bits "; break;
8026       case 3: pc = "64 bits "; break;
8027     };
8028     // flags
8029     char f[9];
8030     f[0] = ' ';
8031     f[1] = ' ';
8032     f[2] = (precision   ()) ? 'P' : 'p';
8033     f[3] = (underflow   ()) ? 'U' : 'u';
8034     f[4] = (overflow    ()) ? 'O' : 'o';
8035     f[5] = (zero_divide ()) ? 'Z' : 'z';
8036     f[6] = (denormalized()) ? 'D' : 'd';
8037     f[7] = (invalid     ()) ? 'I' : 'i';
8038     f[8] = '\x0';
8039     // output
8040     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
8041   }
8042 
8043 };
8044 
8045 class StatusWord {
8046  public:
8047   int32_t _value;
8048 
8049   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
8050   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
8051   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
8052   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
8053   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
8054   int  top() const                     { return  (_value >> 11) & 7      ; }
8055   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
8056   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
8057   bool precision() const               { return ((_value >>  5) & 1) != 0; }
8058   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
8059   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
8060   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
8061   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
8062   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
8063 
8064   void print() const {
8065     // condition codes
8066     char c[5];
8067     c[0] = (C3()) ? '3' : '-';
8068     c[1] = (C2()) ? '2' : '-';
8069     c[2] = (C1()) ? '1' : '-';
8070     c[3] = (C0()) ? '0' : '-';
8071     c[4] = '\x0';
8072     // flags
8073     char f[9];
8074     f[0] = (error_status()) ? 'E' : '-';
8075     f[1] = (stack_fault ()) ? 'S' : '-';
8076     f[2] = (precision   ()) ? 'P' : '-';
8077     f[3] = (underflow   ()) ? 'U' : '-';
8078     f[4] = (overflow    ()) ? 'O' : '-';
8079     f[5] = (zero_divide ()) ? 'Z' : '-';
8080     f[6] = (denormalized()) ? 'D' : '-';
8081     f[7] = (invalid     ()) ? 'I' : '-';
8082     f[8] = '\x0';
8083     // output
8084     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
8085   }
8086 
8087 };
8088 
8089 class TagWord {
8090  public:
8091   int32_t _value;
8092 
8093   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
8094 
8095   void print() const {
8096     printf("%04x", _value & 0xFFFF);
8097   }
8098 
8099 };
8100 
8101 class FPU_Register {
8102  public:
8103   int32_t _m0;
8104   int32_t _m1;
8105   int16_t _ex;
8106 
8107   bool is_indefinite() const           {
8108     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
8109   }
8110 
8111   void print() const {
8112     char  sign = (_ex < 0) ? '-' : '+';
8113     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
8114     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
8115   };
8116 
8117 };
8118 
8119 class FPU_State {
8120  public:
8121   enum {
8122     register_size       = 10,
8123     number_of_registers =  8,
8124     register_mask       =  7
8125   };
8126 
8127   ControlWord  _control_word;
8128   StatusWord   _status_word;
8129   TagWord      _tag_word;
8130   int32_t      _error_offset;
8131   int32_t      _error_selector;
8132   int32_t      _data_offset;
8133   int32_t      _data_selector;
8134   int8_t       _register[register_size * number_of_registers];
8135 
8136   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
8137   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
8138 
8139   const char* tag_as_string(int tag) const {
8140     switch (tag) {
8141       case 0: return "valid";
8142       case 1: return "zero";
8143       case 2: return "special";
8144       case 3: return "empty";
8145     }
8146     ShouldNotReachHere();
8147     return NULL;
8148   }
8149 
8150   void print() const {
8151     // print computation registers
8152     { int t = _status_word.top();
8153       for (int i = 0; i < number_of_registers; i++) {
8154         int j = (i - t) & register_mask;
8155         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
8156         st(j)->print();
8157         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
8158       }
8159     }
8160     printf("\n");
8161     // print control registers
8162     printf("ctrl = "); _control_word.print(); printf("\n");
8163     printf("stat = "); _status_word .print(); printf("\n");
8164     printf("tags = "); _tag_word    .print(); printf("\n");
8165   }
8166 
8167 };
8168 
8169 class Flag_Register {
8170  public:
8171   int32_t _value;
8172 
8173   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
8174   bool direction() const               { return ((_value >> 10) & 1) != 0; }
8175   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
8176   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
8177   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
8178   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
8179   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
8180 
8181   void print() const {
8182     // flags
8183     char f[8];
8184     f[0] = (overflow       ()) ? 'O' : '-';
8185     f[1] = (direction      ()) ? 'D' : '-';
8186     f[2] = (sign           ()) ? 'S' : '-';
8187     f[3] = (zero           ()) ? 'Z' : '-';
8188     f[4] = (auxiliary_carry()) ? 'A' : '-';
8189     f[5] = (parity         ()) ? 'P' : '-';
8190     f[6] = (carry          ()) ? 'C' : '-';
8191     f[7] = '\x0';
8192     // output
8193     printf("%08x  flags = %s", _value, f);
8194   }
8195 
8196 };
8197 
8198 class IU_Register {
8199  public:
8200   int32_t _value;
8201 
8202   void print() const {
8203     printf("%08x  %11d", _value, _value);
8204   }
8205 
8206 };
8207 
8208 class IU_State {
8209  public:
8210   Flag_Register _eflags;
8211   IU_Register   _rdi;
8212   IU_Register   _rsi;
8213   IU_Register   _rbp;
8214   IU_Register   _rsp;
8215   IU_Register   _rbx;
8216   IU_Register   _rdx;
8217   IU_Register   _rcx;
8218   IU_Register   _rax;
8219 
8220   void print() const {
8221     // computation registers
8222     printf("rax,  = "); _rax.print(); printf("\n");
8223     printf("rbx,  = "); _rbx.print(); printf("\n");
8224     printf("rcx  = "); _rcx.print(); printf("\n");
8225     printf("rdx  = "); _rdx.print(); printf("\n");
8226     printf("rdi  = "); _rdi.print(); printf("\n");
8227     printf("rsi  = "); _rsi.print(); printf("\n");
8228     printf("rbp,  = "); _rbp.print(); printf("\n");
8229     printf("rsp  = "); _rsp.print(); printf("\n");
8230     printf("\n");
8231     // control registers
8232     printf("flgs = "); _eflags.print(); printf("\n");
8233   }
8234 };
8235 
8236 
8237 class CPU_State {
8238  public:
8239   FPU_State _fpu_state;
8240   IU_State  _iu_state;
8241 
8242   void print() const {
8243     printf("--------------------------------------------------\n");
8244     _iu_state .print();
8245     printf("\n");
8246     _fpu_state.print();
8247     printf("--------------------------------------------------\n");
8248   }
8249 
8250 };
8251 
8252 
8253 static void _print_CPU_state(CPU_State* state) {
8254   state->print();
8255 };
8256 
8257 
8258 void MacroAssembler::print_CPU_state() {
8259   push_CPU_state();
8260   push(rsp);                // pass CPU state
8261   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
8262   addptr(rsp, wordSize);       // discard argument
8263   pop_CPU_state();
8264 }
8265 
8266 
8267 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) {
8268   static int counter = 0;
8269   FPU_State* fs = &state->_fpu_state;
8270   counter++;
8271   // For leaf calls, only verify that the top few elements remain empty.
8272   // We only need 1 empty at the top for C2 code.
8273   if( stack_depth < 0 ) {
8274     if( fs->tag_for_st(7) != 3 ) {
8275       printf("FPR7 not empty\n");
8276       state->print();
8277       assert(false, "error");
8278       return false;
8279     }
8280     return true;                // All other stack states do not matter
8281   }
8282 
8283   assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std,
8284          "bad FPU control word");
8285 
8286   // compute stack depth
8287   int i = 0;
8288   while (i < FPU_State::number_of_registers && fs->tag_for_st(i)  < 3) i++;
8289   int d = i;
8290   while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++;
8291   // verify findings
8292   if (i != FPU_State::number_of_registers) {
8293     // stack not contiguous
8294     printf("%s: stack not contiguous at ST%d\n", s, i);
8295     state->print();
8296     assert(false, "error");
8297     return false;
8298   }
8299   // check if computed stack depth corresponds to expected stack depth
8300   if (stack_depth < 0) {
8301     // expected stack depth is -stack_depth or less
8302     if (d > -stack_depth) {
8303       // too many elements on the stack
8304       printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d);
8305       state->print();
8306       assert(false, "error");
8307       return false;
8308     }
8309   } else {
8310     // expected stack depth is stack_depth
8311     if (d != stack_depth) {
8312       // wrong stack depth
8313       printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d);
8314       state->print();
8315       assert(false, "error");
8316       return false;
8317     }
8318   }
8319   // everything is cool
8320   return true;
8321 }
8322 
8323 
8324 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
8325   if (!VerifyFPU) return;
8326   push_CPU_state();
8327   push(rsp);                // pass CPU state
8328   ExternalAddress msg((address) s);
8329   // pass message string s
8330   pushptr(msg.addr());
8331   push(stack_depth);        // pass stack depth
8332   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU)));
8333   addptr(rsp, 3 * wordSize);   // discard arguments
8334   // check for error
8335   { Label L;
8336     testl(rax, rax);
8337     jcc(Assembler::notZero, L);
8338     int3();                  // break if error condition
8339     bind(L);
8340   }
8341   pop_CPU_state();
8342 }
8343 
8344 void MacroAssembler::load_klass(Register dst, Register src) {
8345 #ifdef _LP64
8346   if (UseCompressedOops) {
8347     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8348     decode_heap_oop_not_null(dst);
8349   } else
8350 #endif
8351     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8352 }
8353 
8354 void MacroAssembler::load_prototype_header(Register dst, Register src) {
8355 #ifdef _LP64
8356   if (UseCompressedOops) {
8357     assert (Universe::heap() != NULL, "java heap should be initialized");
8358     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8359     if (Universe::narrow_oop_shift() != 0) {
8360       assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8361       if (LogMinObjAlignmentInBytes == Address::times_8) {
8362         movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8363       } else {
8364         // OK to use shift since we don't need to preserve flags.
8365         shlq(dst, LogMinObjAlignmentInBytes);
8366         movq(dst, Address(r12_heapbase, dst, Address::times_1, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8367       }
8368     } else {
8369       movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8370     }
8371   } else
8372 #endif
8373   {
8374     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8375     movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8376   }
8377 }
8378 
8379 void MacroAssembler::store_klass(Register dst, Register src) {
8380 #ifdef _LP64
8381   if (UseCompressedOops) {
8382     encode_heap_oop_not_null(src);
8383     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
8384   } else
8385 #endif
8386     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
8387 }
8388 
8389 void MacroAssembler::load_heap_oop(Register dst, Address src) {
8390 #ifdef _LP64
8391   if (UseCompressedOops) {
8392     movl(dst, src);
8393     decode_heap_oop(dst);
8394   } else
8395 #endif
8396     movptr(dst, src);
8397 }
8398 
8399 // Doesn't do verfication, generates fixed size code
8400 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src) {
8401 #ifdef _LP64
8402   if (UseCompressedOops) {
8403     movl(dst, src);
8404     decode_heap_oop_not_null(dst);
8405   } else
8406 #endif
8407     movptr(dst, src);
8408 }
8409 
8410 void MacroAssembler::store_heap_oop(Address dst, Register src) {
8411 #ifdef _LP64
8412   if (UseCompressedOops) {
8413     assert(!dst.uses(src), "not enough registers");
8414     encode_heap_oop(src);
8415     movl(dst, src);
8416   } else
8417 #endif
8418     movptr(dst, src);
8419 }
8420 
8421 // Used for storing NULLs.
8422 void MacroAssembler::store_heap_oop_null(Address dst) {
8423 #ifdef _LP64
8424   if (UseCompressedOops) {
8425     movl(dst, (int32_t)NULL_WORD);
8426   } else {
8427     movslq(dst, (int32_t)NULL_WORD);
8428   }
8429 #else
8430   movl(dst, (int32_t)NULL_WORD);
8431 #endif
8432 }
8433 
8434 #ifdef _LP64
8435 void MacroAssembler::store_klass_gap(Register dst, Register src) {
8436   if (UseCompressedOops) {
8437     // Store to klass gap in destination
8438     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
8439   }
8440 }
8441 
8442 #ifdef ASSERT
8443 void MacroAssembler::verify_heapbase(const char* msg) {
8444   assert (UseCompressedOops, "should be compressed");
8445   assert (Universe::heap() != NULL, "java heap should be initialized");
8446   if (CheckCompressedOops) {
8447     Label ok;
8448     push(rscratch1); // cmpptr trashes rscratch1
8449     cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr()));
8450     jcc(Assembler::equal, ok);
8451     stop(msg);
8452     bind(ok);
8453     pop(rscratch1);
8454   }
8455 }
8456 #endif
8457 
8458 // Algorithm must match oop.inline.hpp encode_heap_oop.
8459 void MacroAssembler::encode_heap_oop(Register r) {
8460 #ifdef ASSERT
8461   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
8462 #endif
8463   verify_oop(r, "broken oop in encode_heap_oop");
8464   if (Universe::narrow_oop_base() == NULL) {
8465     if (Universe::narrow_oop_shift() != 0) {
8466       assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8467       shrq(r, LogMinObjAlignmentInBytes);
8468     }
8469     return;
8470   }
8471   testq(r, r);
8472   cmovq(Assembler::equal, r, r12_heapbase);
8473   subq(r, r12_heapbase);
8474   shrq(r, LogMinObjAlignmentInBytes);
8475 }
8476 
8477 void MacroAssembler::encode_heap_oop_not_null(Register r) {
8478 #ifdef ASSERT
8479   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
8480   if (CheckCompressedOops) {
8481     Label ok;
8482     testq(r, r);
8483     jcc(Assembler::notEqual, ok);
8484     stop("null oop passed to encode_heap_oop_not_null");
8485     bind(ok);
8486   }
8487 #endif
8488   verify_oop(r, "broken oop in encode_heap_oop_not_null");
8489   if (Universe::narrow_oop_base() != NULL) {
8490     subq(r, r12_heapbase);
8491   }
8492   if (Universe::narrow_oop_shift() != 0) {
8493     assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8494     shrq(r, LogMinObjAlignmentInBytes);
8495   }
8496 }
8497 
8498 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
8499 #ifdef ASSERT
8500   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
8501   if (CheckCompressedOops) {
8502     Label ok;
8503     testq(src, src);
8504     jcc(Assembler::notEqual, ok);
8505     stop("null oop passed to encode_heap_oop_not_null2");
8506     bind(ok);
8507   }
8508 #endif
8509   verify_oop(src, "broken oop in encode_heap_oop_not_null2");
8510   if (dst != src) {
8511     movq(dst, src);
8512   }
8513   if (Universe::narrow_oop_base() != NULL) {
8514     subq(dst, r12_heapbase);
8515   }
8516   if (Universe::narrow_oop_shift() != 0) {
8517     assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8518     shrq(dst, LogMinObjAlignmentInBytes);
8519   }
8520 }
8521 
8522 void  MacroAssembler::decode_heap_oop(Register r) {
8523 #ifdef ASSERT
8524   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
8525 #endif
8526   if (Universe::narrow_oop_base() == NULL) {
8527     if (Universe::narrow_oop_shift() != 0) {
8528       assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8529       shlq(r, LogMinObjAlignmentInBytes);
8530     }
8531   } else {
8532     Label done;
8533     shlq(r, LogMinObjAlignmentInBytes);
8534     jccb(Assembler::equal, done);
8535     addq(r, r12_heapbase);
8536     bind(done);
8537   }
8538   verify_oop(r, "broken oop in decode_heap_oop");
8539 }
8540 
8541 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
8542   // Note: it will change flags
8543   assert (UseCompressedOops, "should only be used for compressed headers");
8544   assert (Universe::heap() != NULL, "java heap should be initialized");
8545   // Cannot assert, unverified entry point counts instructions (see .ad file)
8546   // vtableStubs also counts instructions in pd_code_size_limit.
8547   // Also do not verify_oop as this is called by verify_oop.
8548   if (Universe::narrow_oop_shift() != 0) {
8549     assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8550     shlq(r, LogMinObjAlignmentInBytes);
8551     if (Universe::narrow_oop_base() != NULL) {
8552       addq(r, r12_heapbase);
8553     }
8554   } else {
8555     assert (Universe::narrow_oop_base() == NULL, "sanity");
8556   }
8557 }
8558 
8559 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
8560   // Note: it will change flags
8561   assert (UseCompressedOops, "should only be used for compressed headers");
8562   assert (Universe::heap() != NULL, "java heap should be initialized");
8563   // Cannot assert, unverified entry point counts instructions (see .ad file)
8564   // vtableStubs also counts instructions in pd_code_size_limit.
8565   // Also do not verify_oop as this is called by verify_oop.
8566   if (Universe::narrow_oop_shift() != 0) {
8567     assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8568     if (LogMinObjAlignmentInBytes == Address::times_8) {
8569       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
8570     } else {
8571       if (dst != src) {
8572         movq(dst, src);
8573       }
8574       shlq(dst, LogMinObjAlignmentInBytes);
8575       if (Universe::narrow_oop_base() != NULL) {
8576         addq(dst, r12_heapbase);
8577       }
8578     }
8579   } else {
8580     assert (Universe::narrow_oop_base() == NULL, "sanity");
8581     if (dst != src) {
8582       movq(dst, src);
8583     }
8584   }
8585 }
8586 
8587 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
8588   assert (UseCompressedOops, "should only be used for compressed headers");
8589   assert (Universe::heap() != NULL, "java heap should be initialized");
8590   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8591   int oop_index = oop_recorder()->find_index(obj);
8592   RelocationHolder rspec = oop_Relocation::spec(oop_index);
8593   mov_narrow_oop(dst, oop_index, rspec);
8594 }
8595 
8596 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
8597   assert (UseCompressedOops, "should only be used for compressed headers");
8598   assert (Universe::heap() != NULL, "java heap should be initialized");
8599   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8600   int oop_index = oop_recorder()->find_index(obj);
8601   RelocationHolder rspec = oop_Relocation::spec(oop_index);
8602   mov_narrow_oop(dst, oop_index, rspec);
8603 }
8604 
8605 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
8606   assert (UseCompressedOops, "should only be used for compressed headers");
8607   assert (Universe::heap() != NULL, "java heap should be initialized");
8608   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8609   int oop_index = oop_recorder()->find_index(obj);
8610   RelocationHolder rspec = oop_Relocation::spec(oop_index);
8611   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
8612 }
8613 
8614 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
8615   assert (UseCompressedOops, "should only be used for compressed headers");
8616   assert (Universe::heap() != NULL, "java heap should be initialized");
8617   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8618   int oop_index = oop_recorder()->find_index(obj);
8619   RelocationHolder rspec = oop_Relocation::spec(oop_index);
8620   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
8621 }
8622 
8623 void MacroAssembler::reinit_heapbase() {
8624   if (UseCompressedOops) {
8625     movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr()));
8626   }
8627 }
8628 #endif // _LP64
8629 
8630 // IndexOf for constant substrings with size >= 8 chars
8631 // which don't need to be loaded through stack.
8632 void MacroAssembler::string_indexofC8(Register str1, Register str2,
8633                                       Register cnt1, Register cnt2,
8634                                       int int_cnt2,  Register result,
8635                                       XMMRegister vec, Register tmp) {
8636   assert(UseSSE42Intrinsics, "SSE4.2 is required");
8637 
8638   // This method uses pcmpestri inxtruction with bound registers
8639   //   inputs:
8640   //     xmm - substring
8641   //     rax - substring length (elements count)
8642   //     mem - scanned string
8643   //     rdx - string length (elements count)
8644   //     0xd - mode: 1100 (substring search) + 01 (unsigned shorts)
8645   //   outputs:
8646   //     rcx - matched index in string
8647   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
8648 
8649   Label RELOAD_SUBSTR, SCAN_TO_SUBSTR, SCAN_SUBSTR,
8650         RET_FOUND, RET_NOT_FOUND, EXIT, FOUND_SUBSTR,
8651         MATCH_SUBSTR_HEAD, RELOAD_STR, FOUND_CANDIDATE;
8652 
8653   // Note, inline_string_indexOf() generates checks:
8654   // if (substr.count > string.count) return -1;
8655   // if (substr.count == 0) return 0;
8656   assert(int_cnt2 >= 8, "this code isused only for cnt2 >= 8 chars");
8657 
8658   // Load substring.
8659   movdqu(vec, Address(str2, 0));
8660   movl(cnt2, int_cnt2);
8661   movptr(result, str1); // string addr
8662 
8663   if (int_cnt2 > 8) {
8664     jmpb(SCAN_TO_SUBSTR);
8665 
8666     // Reload substr for rescan, this code
8667     // is executed only for large substrings (> 8 chars)
8668     bind(RELOAD_SUBSTR);
8669     movdqu(vec, Address(str2, 0));
8670     negptr(cnt2); // Jumped here with negative cnt2, convert to positive
8671 
8672     bind(RELOAD_STR);
8673     // We came here after the beginning of the substring was
8674     // matched but the rest of it was not so we need to search
8675     // again. Start from the next element after the previous match.
8676 
8677     // cnt2 is number of substring reminding elements and
8678     // cnt1 is number of string reminding elements when cmp failed.
8679     // Restored cnt1 = cnt1 - cnt2 + int_cnt2
8680     subl(cnt1, cnt2);
8681     addl(cnt1, int_cnt2);
8682     movl(cnt2, int_cnt2); // Now restore cnt2
8683 
8684     decrementl(cnt1);     // Shift to next element
8685     cmpl(cnt1, cnt2);
8686     jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
8687 
8688     addptr(result, 2);
8689 
8690   } // (int_cnt2 > 8)
8691 
8692   // Scan string for start of substr in 16-byte vectors
8693   bind(SCAN_TO_SUBSTR);
8694   pcmpestri(vec, Address(result, 0), 0x0d);
8695   jccb(Assembler::below, FOUND_CANDIDATE);   // CF == 1
8696   subl(cnt1, 8);
8697   jccb(Assembler::lessEqual, RET_NOT_FOUND); // Scanned full string
8698   cmpl(cnt1, cnt2);
8699   jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
8700   addptr(result, 16);
8701   jmpb(SCAN_TO_SUBSTR);
8702 
8703   // Found a potential substr
8704   bind(FOUND_CANDIDATE);
8705   // Matched whole vector if first element matched (tmp(rcx) == 0).
8706   if (int_cnt2 == 8) {
8707     jccb(Assembler::overflow, RET_FOUND);    // OF == 1
8708   } else { // int_cnt2 > 8
8709     jccb(Assembler::overflow, FOUND_SUBSTR);
8710   }
8711   // After pcmpestri tmp(rcx) contains matched element index
8712   // Compute start addr of substr
8713   lea(result, Address(result, tmp, Address::times_2));
8714 
8715   // Make sure string is still long enough
8716   subl(cnt1, tmp);
8717   cmpl(cnt1, cnt2);
8718   if (int_cnt2 == 8) {
8719     jccb(Assembler::greaterEqual, SCAN_TO_SUBSTR);
8720   } else { // int_cnt2 > 8
8721     jccb(Assembler::greaterEqual, MATCH_SUBSTR_HEAD);
8722   }
8723   // Left less then substring.
8724 
8725   bind(RET_NOT_FOUND);
8726   movl(result, -1);
8727   jmpb(EXIT);
8728 
8729   if (int_cnt2 > 8) {
8730     // This code is optimized for the case when whole substring
8731     // is matched if its head is matched.
8732     bind(MATCH_SUBSTR_HEAD);
8733     pcmpestri(vec, Address(result, 0), 0x0d);
8734     // Reload only string if does not match
8735     jccb(Assembler::noOverflow, RELOAD_STR); // OF == 0
8736 
8737     Label CONT_SCAN_SUBSTR;
8738     // Compare the rest of substring (> 8 chars).
8739     bind(FOUND_SUBSTR);
8740     // First 8 chars are already matched.
8741     negptr(cnt2);
8742     addptr(cnt2, 8);
8743 
8744     bind(SCAN_SUBSTR);
8745     subl(cnt1, 8);
8746     cmpl(cnt2, -8); // Do not read beyond substring
8747     jccb(Assembler::lessEqual, CONT_SCAN_SUBSTR);
8748     // Back-up strings to avoid reading beyond substring:
8749     // cnt1 = cnt1 - cnt2 + 8
8750     addl(cnt1, cnt2); // cnt2 is negative
8751     addl(cnt1, 8);
8752     movl(cnt2, 8); negptr(cnt2);
8753     bind(CONT_SCAN_SUBSTR);
8754     if (int_cnt2 < (int)G) {
8755       movdqu(vec, Address(str2, cnt2, Address::times_2, int_cnt2*2));
8756       pcmpestri(vec, Address(result, cnt2, Address::times_2, int_cnt2*2), 0x0d);
8757     } else {
8758       // calculate index in register to avoid integer overflow (int_cnt2*2)
8759       movl(tmp, int_cnt2);
8760       addptr(tmp, cnt2);
8761       movdqu(vec, Address(str2, tmp, Address::times_2, 0));
8762       pcmpestri(vec, Address(result, tmp, Address::times_2, 0), 0x0d);
8763     }
8764     // Need to reload strings pointers if not matched whole vector
8765     jccb(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0
8766     addptr(cnt2, 8);
8767     jccb(Assembler::negative, SCAN_SUBSTR);
8768     // Fall through if found full substring
8769 
8770   } // (int_cnt2 > 8)
8771 
8772   bind(RET_FOUND);
8773   // Found result if we matched full small substring.
8774   // Compute substr offset
8775   subptr(result, str1);
8776   shrl(result, 1); // index
8777   bind(EXIT);
8778 
8779 } // string_indexofC8
8780 
8781 // Small strings are loaded through stack if they cross page boundary.
8782 void MacroAssembler::string_indexof(Register str1, Register str2,
8783                                     Register cnt1, Register cnt2,
8784                                     int int_cnt2,  Register result,
8785                                     XMMRegister vec, Register tmp) {
8786   assert(UseSSE42Intrinsics, "SSE4.2 is required");
8787   //
8788   // int_cnt2 is length of small (< 8 chars) constant substring
8789   // or (-1) for non constant substring in which case its length
8790   // is in cnt2 register.
8791   //
8792   // Note, inline_string_indexOf() generates checks:
8793   // if (substr.count > string.count) return -1;
8794   // if (substr.count == 0) return 0;
8795   //
8796   assert(int_cnt2 == -1 || (0 < int_cnt2 && int_cnt2 < 8), "should be != 0");
8797 
8798   // This method uses pcmpestri inxtruction with bound registers
8799   //   inputs:
8800   //     xmm - substring
8801   //     rax - substring length (elements count)
8802   //     mem - scanned string
8803   //     rdx - string length (elements count)
8804   //     0xd - mode: 1100 (substring search) + 01 (unsigned shorts)
8805   //   outputs:
8806   //     rcx - matched index in string
8807   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
8808 
8809   Label RELOAD_SUBSTR, SCAN_TO_SUBSTR, SCAN_SUBSTR, ADJUST_STR,
8810         RET_FOUND, RET_NOT_FOUND, CLEANUP, FOUND_SUBSTR,
8811         FOUND_CANDIDATE;
8812 
8813   { //========================================================
8814     // We don't know where these strings are located
8815     // and we can't read beyond them. Load them through stack.
8816     Label BIG_STRINGS, CHECK_STR, COPY_SUBSTR, COPY_STR;
8817 
8818     movptr(tmp, rsp); // save old SP
8819 
8820     if (int_cnt2 > 0) {     // small (< 8 chars) constant substring
8821       if (int_cnt2 == 1) {  // One char
8822         load_unsigned_short(result, Address(str2, 0));
8823         movdl(vec, result); // move 32 bits
8824       } else if (int_cnt2 == 2) { // Two chars
8825         movdl(vec, Address(str2, 0)); // move 32 bits
8826       } else if (int_cnt2 == 4) { // Four chars
8827         movq(vec, Address(str2, 0));  // move 64 bits
8828       } else { // cnt2 = { 3, 5, 6, 7 }
8829         // Array header size is 12 bytes in 32-bit VM
8830         // + 6 bytes for 3 chars == 18 bytes,
8831         // enough space to load vec and shift.
8832         assert(HeapWordSize*typeArrayKlass::header_size() >= 12,"sanity");
8833         movdqu(vec, Address(str2, (int_cnt2*2)-16));
8834         psrldq(vec, 16-(int_cnt2*2));
8835       }
8836     } else { // not constant substring
8837       cmpl(cnt2, 8);
8838       jccb(Assembler::aboveEqual, BIG_STRINGS); // Both strings are big enough
8839 
8840       // We can read beyond string if srt+16 does not cross page boundary
8841       // since heaps are aligned and mapped by pages.
8842       assert(os::vm_page_size() < (int)G, "default page should be small");
8843       movl(result, str2); // We need only low 32 bits
8844       andl(result, (os::vm_page_size()-1));
8845       cmpl(result, (os::vm_page_size()-16));
8846       jccb(Assembler::belowEqual, CHECK_STR);
8847 
8848       // Move small strings to stack to allow load 16 bytes into vec.
8849       subptr(rsp, 16);
8850       int stk_offset = wordSize-2;
8851       push(cnt2);
8852 
8853       bind(COPY_SUBSTR);
8854       load_unsigned_short(result, Address(str2, cnt2, Address::times_2, -2));
8855       movw(Address(rsp, cnt2, Address::times_2, stk_offset), result);
8856       decrement(cnt2);
8857       jccb(Assembler::notZero, COPY_SUBSTR);
8858 
8859       pop(cnt2);
8860       movptr(str2, rsp);  // New substring address
8861     } // non constant
8862 
8863     bind(CHECK_STR);
8864     cmpl(cnt1, 8);
8865     jccb(Assembler::aboveEqual, BIG_STRINGS);
8866 
8867     // Check cross page boundary.
8868     movl(result, str1); // We need only low 32 bits
8869     andl(result, (os::vm_page_size()-1));
8870     cmpl(result, (os::vm_page_size()-16));
8871     jccb(Assembler::belowEqual, BIG_STRINGS);
8872 
8873     subptr(rsp, 16);
8874     int stk_offset = -2;
8875     if (int_cnt2 < 0) { // not constant
8876       push(cnt2);
8877       stk_offset += wordSize;
8878     }
8879     movl(cnt2, cnt1);
8880 
8881     bind(COPY_STR);
8882     load_unsigned_short(result, Address(str1, cnt2, Address::times_2, -2));
8883     movw(Address(rsp, cnt2, Address::times_2, stk_offset), result);
8884     decrement(cnt2);
8885     jccb(Assembler::notZero, COPY_STR);
8886 
8887     if (int_cnt2 < 0) { // not constant
8888       pop(cnt2);
8889     }
8890     movptr(str1, rsp);  // New string address
8891 
8892     bind(BIG_STRINGS);
8893     // Load substring.
8894     if (int_cnt2 < 0) { // -1
8895       movdqu(vec, Address(str2, 0));
8896       push(cnt2);       // substr count
8897       push(str2);       // substr addr
8898       push(str1);       // string addr
8899     } else {
8900       // Small (< 8 chars) constant substrings are loaded already.
8901       movl(cnt2, int_cnt2);
8902     }
8903     push(tmp);  // original SP
8904 
8905   } // Finished loading
8906 
8907   //========================================================
8908   // Start search
8909   //
8910 
8911   movptr(result, str1); // string addr
8912 
8913   if (int_cnt2  < 0) {  // Only for non constant substring
8914     jmpb(SCAN_TO_SUBSTR);
8915 
8916     // SP saved at sp+0
8917     // String saved at sp+1*wordSize
8918     // Substr saved at sp+2*wordSize
8919     // Substr count saved at sp+3*wordSize
8920 
8921     // Reload substr for rescan, this code
8922     // is executed only for large substrings (> 8 chars)
8923     bind(RELOAD_SUBSTR);
8924     movptr(str2, Address(rsp, 2*wordSize));
8925     movl(cnt2, Address(rsp, 3*wordSize));
8926     movdqu(vec, Address(str2, 0));
8927     // We came here after the beginning of the substring was
8928     // matched but the rest of it was not so we need to search
8929     // again. Start from the next element after the previous match.
8930     subptr(str1, result); // Restore counter
8931     shrl(str1, 1);
8932     addl(cnt1, str1);
8933     decrementl(cnt1);   // Shift to next element
8934     cmpl(cnt1, cnt2);
8935     jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
8936 
8937     addptr(result, 2);
8938   } // non constant
8939 
8940   // Scan string for start of substr in 16-byte vectors
8941   bind(SCAN_TO_SUBSTR);
8942   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
8943   pcmpestri(vec, Address(result, 0), 0x0d);
8944   jccb(Assembler::below, FOUND_CANDIDATE);   // CF == 1
8945   subl(cnt1, 8);
8946   jccb(Assembler::lessEqual, RET_NOT_FOUND); // Scanned full string
8947   cmpl(cnt1, cnt2);
8948   jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
8949   addptr(result, 16);
8950 
8951   bind(ADJUST_STR);
8952   cmpl(cnt1, 8); // Do not read beyond string
8953   jccb(Assembler::greaterEqual, SCAN_TO_SUBSTR);
8954   // Back-up string to avoid reading beyond string.
8955   lea(result, Address(result, cnt1, Address::times_2, -16));
8956   movl(cnt1, 8);
8957   jmpb(SCAN_TO_SUBSTR);
8958 
8959   // Found a potential substr
8960   bind(FOUND_CANDIDATE);
8961   // After pcmpestri tmp(rcx) contains matched element index
8962 
8963   // Make sure string is still long enough
8964   subl(cnt1, tmp);
8965   cmpl(cnt1, cnt2);
8966   jccb(Assembler::greaterEqual, FOUND_SUBSTR);
8967   // Left less then substring.
8968 
8969   bind(RET_NOT_FOUND);
8970   movl(result, -1);
8971   jmpb(CLEANUP);
8972 
8973   bind(FOUND_SUBSTR);
8974   // Compute start addr of substr
8975   lea(result, Address(result, tmp, Address::times_2));
8976 
8977   if (int_cnt2 > 0) { // Constant substring
8978     // Repeat search for small substring (< 8 chars)
8979     // from new point without reloading substring.
8980     // Have to check that we don't read beyond string.
8981     cmpl(tmp, 8-int_cnt2);
8982     jccb(Assembler::greater, ADJUST_STR);
8983     // Fall through if matched whole substring.
8984   } else { // non constant
8985     assert(int_cnt2 == -1, "should be != 0");
8986 
8987     addl(tmp, cnt2);
8988     // Found result if we matched whole substring.
8989     cmpl(tmp, 8);
8990     jccb(Assembler::lessEqual, RET_FOUND);
8991 
8992     // Repeat search for small substring (<= 8 chars)
8993     // from new point 'str1' without reloading substring.
8994     cmpl(cnt2, 8);
8995     // Have to check that we don't read beyond string.
8996     jccb(Assembler::lessEqual, ADJUST_STR);
8997 
8998     Label CHECK_NEXT, CONT_SCAN_SUBSTR, RET_FOUND_LONG;
8999     // Compare the rest of substring (> 8 chars).
9000     movptr(str1, result);
9001 
9002     cmpl(tmp, cnt2);
9003     // First 8 chars are already matched.
9004     jccb(Assembler::equal, CHECK_NEXT);
9005 
9006     bind(SCAN_SUBSTR);
9007     pcmpestri(vec, Address(str1, 0), 0x0d);
9008     // Need to reload strings pointers if not matched whole vector
9009     jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0
9010 
9011     bind(CHECK_NEXT);
9012     subl(cnt2, 8);
9013     jccb(Assembler::lessEqual, RET_FOUND_LONG); // Found full substring
9014     addptr(str1, 16);
9015     addptr(str2, 16);
9016     subl(cnt1, 8);
9017     cmpl(cnt2, 8); // Do not read beyond substring
9018     jccb(Assembler::greaterEqual, CONT_SCAN_SUBSTR);
9019     // Back-up strings to avoid reading beyond substring.
9020     lea(str2, Address(str2, cnt2, Address::times_2, -16));
9021     lea(str1, Address(str1, cnt2, Address::times_2, -16));
9022     subl(cnt1, cnt2);
9023     movl(cnt2, 8);
9024     addl(cnt1, 8);
9025     bind(CONT_SCAN_SUBSTR);
9026     movdqu(vec, Address(str2, 0));
9027     jmpb(SCAN_SUBSTR);
9028 
9029     bind(RET_FOUND_LONG);
9030     movptr(str1, Address(rsp, wordSize));
9031   } // non constant
9032 
9033   bind(RET_FOUND);
9034   // Compute substr offset
9035   subptr(result, str1);
9036   shrl(result, 1); // index
9037 
9038   bind(CLEANUP);
9039   pop(rsp); // restore SP
9040 
9041 } // string_indexof
9042 
9043 // Compare strings.
9044 void MacroAssembler::string_compare(Register str1, Register str2,
9045                                     Register cnt1, Register cnt2, Register result,
9046                                     XMMRegister vec1) {
9047   Label LENGTH_DIFF_LABEL, POP_LABEL, DONE_LABEL, WHILE_HEAD_LABEL;
9048 
9049   // Compute the minimum of the string lengths and the
9050   // difference of the string lengths (stack).
9051   // Do the conditional move stuff
9052   movl(result, cnt1);
9053   subl(cnt1, cnt2);
9054   push(cnt1);
9055   cmov32(Assembler::lessEqual, cnt2, result);
9056 
9057   // Is the minimum length zero?
9058   testl(cnt2, cnt2);
9059   jcc(Assembler::zero, LENGTH_DIFF_LABEL);
9060 
9061   // Load first characters
9062   load_unsigned_short(result, Address(str1, 0));
9063   load_unsigned_short(cnt1, Address(str2, 0));
9064 
9065   // Compare first characters
9066   subl(result, cnt1);
9067   jcc(Assembler::notZero,  POP_LABEL);
9068   decrementl(cnt2);
9069   jcc(Assembler::zero, LENGTH_DIFF_LABEL);
9070 
9071   {
9072     // Check after comparing first character to see if strings are equivalent
9073     Label LSkip2;
9074     // Check if the strings start at same location
9075     cmpptr(str1, str2);
9076     jccb(Assembler::notEqual, LSkip2);
9077 
9078     // Check if the length difference is zero (from stack)
9079     cmpl(Address(rsp, 0), 0x0);
9080     jcc(Assembler::equal,  LENGTH_DIFF_LABEL);
9081 
9082     // Strings might not be equivalent
9083     bind(LSkip2);
9084   }
9085 
9086   Address::ScaleFactor scale = Address::times_2;
9087   int stride = 8;
9088 
9089   // Advance to next element
9090   addptr(str1, 16/stride);
9091   addptr(str2, 16/stride);
9092 
9093   if (UseSSE42Intrinsics) {
9094     Label COMPARE_WIDE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
9095     int pcmpmask = 0x19;
9096     // Setup to compare 16-byte vectors
9097     movl(result, cnt2);
9098     andl(cnt2, ~(stride - 1));   // cnt2 holds the vector count
9099     jccb(Assembler::zero, COMPARE_TAIL);
9100 
9101     lea(str1, Address(str1, result, scale));
9102     lea(str2, Address(str2, result, scale));
9103     negptr(result);
9104 
9105     // pcmpestri
9106     //   inputs:
9107     //     vec1- substring
9108     //     rax - negative string length (elements count)
9109     //     mem - scaned string
9110     //     rdx - string length (elements count)
9111     //     pcmpmask - cmp mode: 11000 (string compare with negated result)
9112     //               + 00 (unsigned bytes) or  + 01 (unsigned shorts)
9113     //   outputs:
9114     //     rcx - first mismatched element index
9115     assert(result == rax && cnt2 == rdx && cnt1 == rcx, "pcmpestri");
9116 
9117     bind(COMPARE_WIDE_VECTORS);
9118     movdqu(vec1, Address(str1, result, scale));
9119     pcmpestri(vec1, Address(str2, result, scale), pcmpmask);
9120     // After pcmpestri cnt1(rcx) contains mismatched element index
9121 
9122     jccb(Assembler::below, VECTOR_NOT_EQUAL);  // CF==1
9123     addptr(result, stride);
9124     subptr(cnt2, stride);
9125     jccb(Assembler::notZero, COMPARE_WIDE_VECTORS);
9126 
9127     // compare wide vectors tail
9128     testl(result, result);
9129     jccb(Assembler::zero, LENGTH_DIFF_LABEL);
9130 
9131     movl(cnt2, stride);
9132     movl(result, stride);
9133     negptr(result);
9134     movdqu(vec1, Address(str1, result, scale));
9135     pcmpestri(vec1, Address(str2, result, scale), pcmpmask);
9136     jccb(Assembler::aboveEqual, LENGTH_DIFF_LABEL);
9137 
9138     // Mismatched characters in the vectors
9139     bind(VECTOR_NOT_EQUAL);
9140     addptr(result, cnt1);
9141     movptr(cnt2, result);
9142     load_unsigned_short(result, Address(str1, cnt2, scale));
9143     load_unsigned_short(cnt1, Address(str2, cnt2, scale));
9144     subl(result, cnt1);
9145     jmpb(POP_LABEL);
9146 
9147     bind(COMPARE_TAIL); // limit is zero
9148     movl(cnt2, result);
9149     // Fallthru to tail compare
9150   }
9151 
9152   // Shift str2 and str1 to the end of the arrays, negate min
9153   lea(str1, Address(str1, cnt2, scale, 0));
9154   lea(str2, Address(str2, cnt2, scale, 0));
9155   negptr(cnt2);
9156 
9157   // Compare the rest of the elements
9158   bind(WHILE_HEAD_LABEL);
9159   load_unsigned_short(result, Address(str1, cnt2, scale, 0));
9160   load_unsigned_short(cnt1, Address(str2, cnt2, scale, 0));
9161   subl(result, cnt1);
9162   jccb(Assembler::notZero, POP_LABEL);
9163   increment(cnt2);
9164   jccb(Assembler::notZero, WHILE_HEAD_LABEL);
9165 
9166   // Strings are equal up to min length.  Return the length difference.
9167   bind(LENGTH_DIFF_LABEL);
9168   pop(result);
9169   jmpb(DONE_LABEL);
9170 
9171   // Discard the stored length difference
9172   bind(POP_LABEL);
9173   pop(cnt1);
9174 
9175   // That's it
9176   bind(DONE_LABEL);
9177 }
9178 
9179 // Compare char[] arrays aligned to 4 bytes or substrings.
9180 void MacroAssembler::char_arrays_equals(bool is_array_equ, Register ary1, Register ary2,
9181                                         Register limit, Register result, Register chr,
9182                                         XMMRegister vec1, XMMRegister vec2) {
9183   Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
9184 
9185   int length_offset  = arrayOopDesc::length_offset_in_bytes();
9186   int base_offset    = arrayOopDesc::base_offset_in_bytes(T_CHAR);
9187 
9188   // Check the input args
9189   cmpptr(ary1, ary2);
9190   jcc(Assembler::equal, TRUE_LABEL);
9191 
9192   if (is_array_equ) {
9193     // Need additional checks for arrays_equals.
9194     testptr(ary1, ary1);
9195     jcc(Assembler::zero, FALSE_LABEL);
9196     testptr(ary2, ary2);
9197     jcc(Assembler::zero, FALSE_LABEL);
9198 
9199     // Check the lengths
9200     movl(limit, Address(ary1, length_offset));
9201     cmpl(limit, Address(ary2, length_offset));
9202     jcc(Assembler::notEqual, FALSE_LABEL);
9203   }
9204 
9205   // count == 0
9206   testl(limit, limit);
9207   jcc(Assembler::zero, TRUE_LABEL);
9208 
9209   if (is_array_equ) {
9210     // Load array address
9211     lea(ary1, Address(ary1, base_offset));
9212     lea(ary2, Address(ary2, base_offset));
9213   }
9214 
9215   shll(limit, 1);      // byte count != 0
9216   movl(result, limit); // copy
9217 
9218   if (UseSSE42Intrinsics) {
9219     // With SSE4.2, use double quad vector compare
9220     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
9221 
9222     // Compare 16-byte vectors
9223     andl(result, 0x0000000e);  //   tail count (in bytes)
9224     andl(limit, 0xfffffff0);   // vector count (in bytes)
9225     jccb(Assembler::zero, COMPARE_TAIL);
9226 
9227     lea(ary1, Address(ary1, limit, Address::times_1));
9228     lea(ary2, Address(ary2, limit, Address::times_1));
9229     negptr(limit);
9230 
9231     bind(COMPARE_WIDE_VECTORS);
9232     movdqu(vec1, Address(ary1, limit, Address::times_1));
9233     movdqu(vec2, Address(ary2, limit, Address::times_1));
9234     pxor(vec1, vec2);
9235 
9236     ptest(vec1, vec1);
9237     jccb(Assembler::notZero, FALSE_LABEL);
9238     addptr(limit, 16);
9239     jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
9240 
9241     testl(result, result);
9242     jccb(Assembler::zero, TRUE_LABEL);
9243 
9244     movdqu(vec1, Address(ary1, result, Address::times_1, -16));
9245     movdqu(vec2, Address(ary2, result, Address::times_1, -16));
9246     pxor(vec1, vec2);
9247 
9248     ptest(vec1, vec1);
9249     jccb(Assembler::notZero, FALSE_LABEL);
9250     jmpb(TRUE_LABEL);
9251 
9252     bind(COMPARE_TAIL); // limit is zero
9253     movl(limit, result);
9254     // Fallthru to tail compare
9255   }
9256 
9257   // Compare 4-byte vectors
9258   andl(limit, 0xfffffffc); // vector count (in bytes)
9259   jccb(Assembler::zero, COMPARE_CHAR);
9260 
9261   lea(ary1, Address(ary1, limit, Address::times_1));
9262   lea(ary2, Address(ary2, limit, Address::times_1));
9263   negptr(limit);
9264 
9265   bind(COMPARE_VECTORS);
9266   movl(chr, Address(ary1, limit, Address::times_1));
9267   cmpl(chr, Address(ary2, limit, Address::times_1));
9268   jccb(Assembler::notEqual, FALSE_LABEL);
9269   addptr(limit, 4);
9270   jcc(Assembler::notZero, COMPARE_VECTORS);
9271 
9272   // Compare trailing char (final 2 bytes), if any
9273   bind(COMPARE_CHAR);
9274   testl(result, 0x2);   // tail  char
9275   jccb(Assembler::zero, TRUE_LABEL);
9276   load_unsigned_short(chr, Address(ary1, 0));
9277   load_unsigned_short(limit, Address(ary2, 0));
9278   cmpl(chr, limit);
9279   jccb(Assembler::notEqual, FALSE_LABEL);
9280 
9281   bind(TRUE_LABEL);
9282   movl(result, 1);   // return true
9283   jmpb(DONE);
9284 
9285   bind(FALSE_LABEL);
9286   xorl(result, result); // return false
9287 
9288   // That's it
9289   bind(DONE);
9290 }
9291 
9292 #ifdef PRODUCT
9293 #define BLOCK_COMMENT(str) /* nothing */
9294 #else
9295 #define BLOCK_COMMENT(str) block_comment(str)
9296 #endif
9297 
9298 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
9299 void MacroAssembler::generate_fill(BasicType t, bool aligned,
9300                                    Register to, Register value, Register count,
9301                                    Register rtmp, XMMRegister xtmp) {
9302   assert_different_registers(to, value, count, rtmp);
9303   Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
9304   Label L_fill_2_bytes, L_fill_4_bytes;
9305 
9306   int shift = -1;
9307   switch (t) {
9308     case T_BYTE:
9309       shift = 2;
9310       break;
9311     case T_SHORT:
9312       shift = 1;
9313       break;
9314     case T_INT:
9315       shift = 0;
9316       break;
9317     default: ShouldNotReachHere();
9318   }
9319 
9320   if (t == T_BYTE) {
9321     andl(value, 0xff);
9322     movl(rtmp, value);
9323     shll(rtmp, 8);
9324     orl(value, rtmp);
9325   }
9326   if (t == T_SHORT) {
9327     andl(value, 0xffff);
9328   }
9329   if (t == T_BYTE || t == T_SHORT) {
9330     movl(rtmp, value);
9331     shll(rtmp, 16);
9332     orl(value, rtmp);
9333   }
9334 
9335   cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
9336   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
9337   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
9338     // align source address at 4 bytes address boundary
9339     if (t == T_BYTE) {
9340       // One byte misalignment happens only for byte arrays
9341       testptr(to, 1);
9342       jccb(Assembler::zero, L_skip_align1);
9343       movb(Address(to, 0), value);
9344       increment(to);
9345       decrement(count);
9346       BIND(L_skip_align1);
9347     }
9348     // Two bytes misalignment happens only for byte and short (char) arrays
9349     testptr(to, 2);
9350     jccb(Assembler::zero, L_skip_align2);
9351     movw(Address(to, 0), value);
9352     addptr(to, 2);
9353     subl(count, 1<<(shift-1));
9354     BIND(L_skip_align2);
9355   }
9356   if (UseSSE < 2) {
9357     Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
9358     // Fill 32-byte chunks
9359     subl(count, 8 << shift);
9360     jcc(Assembler::less, L_check_fill_8_bytes);
9361     align(16);
9362 
9363     BIND(L_fill_32_bytes_loop);
9364 
9365     for (int i = 0; i < 32; i += 4) {
9366       movl(Address(to, i), value);
9367     }
9368 
9369     addptr(to, 32);
9370     subl(count, 8 << shift);
9371     jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
9372     BIND(L_check_fill_8_bytes);
9373     addl(count, 8 << shift);
9374     jccb(Assembler::zero, L_exit);
9375     jmpb(L_fill_8_bytes);
9376 
9377     //
9378     // length is too short, just fill qwords
9379     //
9380     BIND(L_fill_8_bytes_loop);
9381     movl(Address(to, 0), value);
9382     movl(Address(to, 4), value);
9383     addptr(to, 8);
9384     BIND(L_fill_8_bytes);
9385     subl(count, 1 << (shift + 1));
9386     jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
9387     // fall through to fill 4 bytes
9388   } else {
9389     Label L_fill_32_bytes;
9390     if (!UseUnalignedLoadStores) {
9391       // align to 8 bytes, we know we are 4 byte aligned to start
9392       testptr(to, 4);
9393       jccb(Assembler::zero, L_fill_32_bytes);
9394       movl(Address(to, 0), value);
9395       addptr(to, 4);
9396       subl(count, 1<<shift);
9397     }
9398     BIND(L_fill_32_bytes);
9399     {
9400       assert( UseSSE >= 2, "supported cpu only" );
9401       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
9402       // Fill 32-byte chunks
9403       movdl(xtmp, value);
9404       pshufd(xtmp, xtmp, 0);
9405 
9406       subl(count, 8 << shift);
9407       jcc(Assembler::less, L_check_fill_8_bytes);
9408       align(16);
9409 
9410       BIND(L_fill_32_bytes_loop);
9411 
9412       if (UseUnalignedLoadStores) {
9413         movdqu(Address(to, 0), xtmp);
9414         movdqu(Address(to, 16), xtmp);
9415       } else {
9416         movq(Address(to, 0), xtmp);
9417         movq(Address(to, 8), xtmp);
9418         movq(Address(to, 16), xtmp);
9419         movq(Address(to, 24), xtmp);
9420       }
9421 
9422       addptr(to, 32);
9423       subl(count, 8 << shift);
9424       jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
9425       BIND(L_check_fill_8_bytes);
9426       addl(count, 8 << shift);
9427       jccb(Assembler::zero, L_exit);
9428       jmpb(L_fill_8_bytes);
9429 
9430       //
9431       // length is too short, just fill qwords
9432       //
9433       BIND(L_fill_8_bytes_loop);
9434       movq(Address(to, 0), xtmp);
9435       addptr(to, 8);
9436       BIND(L_fill_8_bytes);
9437       subl(count, 1 << (shift + 1));
9438       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
9439     }
9440   }
9441   // fill trailing 4 bytes
9442   BIND(L_fill_4_bytes);
9443   testl(count, 1<<shift);
9444   jccb(Assembler::zero, L_fill_2_bytes);
9445   movl(Address(to, 0), value);
9446   if (t == T_BYTE || t == T_SHORT) {
9447     addptr(to, 4);
9448     BIND(L_fill_2_bytes);
9449     // fill trailing 2 bytes
9450     testl(count, 1<<(shift-1));
9451     jccb(Assembler::zero, L_fill_byte);
9452     movw(Address(to, 0), value);
9453     if (t == T_BYTE) {
9454       addptr(to, 2);
9455       BIND(L_fill_byte);
9456       // fill trailing byte
9457       testl(count, 1);
9458       jccb(Assembler::zero, L_exit);
9459       movb(Address(to, 0), value);
9460     } else {
9461       BIND(L_fill_byte);
9462     }
9463   } else {
9464     BIND(L_fill_2_bytes);
9465   }
9466   BIND(L_exit);
9467 }
9468 #undef BIND
9469 #undef BLOCK_COMMENT
9470 
9471 
9472 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
9473   switch (cond) {
9474     // Note some conditions are synonyms for others
9475     case Assembler::zero:         return Assembler::notZero;
9476     case Assembler::notZero:      return Assembler::zero;
9477     case Assembler::less:         return Assembler::greaterEqual;
9478     case Assembler::lessEqual:    return Assembler::greater;
9479     case Assembler::greater:      return Assembler::lessEqual;
9480     case Assembler::greaterEqual: return Assembler::less;
9481     case Assembler::below:        return Assembler::aboveEqual;
9482     case Assembler::belowEqual:   return Assembler::above;
9483     case Assembler::above:        return Assembler::belowEqual;
9484     case Assembler::aboveEqual:   return Assembler::below;
9485     case Assembler::overflow:     return Assembler::noOverflow;
9486     case Assembler::noOverflow:   return Assembler::overflow;
9487     case Assembler::negative:     return Assembler::positive;
9488     case Assembler::positive:     return Assembler::negative;
9489     case Assembler::parity:       return Assembler::noParity;
9490     case Assembler::noParity:     return Assembler::parity;
9491   }
9492   ShouldNotReachHere(); return Assembler::overflow;
9493 }
9494 
9495 SkipIfEqual::SkipIfEqual(
9496     MacroAssembler* masm, const bool* flag_addr, bool value) {
9497   _masm = masm;
9498   _masm->cmp8(ExternalAddress((address)flag_addr), value);
9499   _masm->jcc(Assembler::equal, _label);
9500 }
9501 
9502 SkipIfEqual::~SkipIfEqual() {
9503   _masm->bind(_label);
9504 }