1 /*
   2  * Copyright 1997-2008 Sun Microsystems, Inc.  All Rights Reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 class BiasedLockingCounters;
  26 
  27 // Contains all the definitions needed for x86 assembly code generation.
  28 
  29 // Calling convention
  30 class Argument VALUE_OBJ_CLASS_SPEC {
  31  public:
  32   enum {
  33 #ifdef _LP64
  34 #ifdef _WIN64
  35     n_int_register_parameters_c   = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
  36     n_float_register_parameters_c = 4,  // xmm0 - xmm3 (c_farg0, c_farg1, ... )
  37 #else
  38     n_int_register_parameters_c   = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
  39     n_float_register_parameters_c = 8,  // xmm0 - xmm7 (c_farg0, c_farg1, ... )
  40 #endif // _WIN64
  41     n_int_register_parameters_j   = 6, // j_rarg0, j_rarg1, ...
  42     n_float_register_parameters_j = 8  // j_farg0, j_farg1, ...
  43 #else
  44     n_register_parameters = 0   // 0 registers used to pass arguments
  45 #endif // _LP64
  46   };
  47 };
  48 
  49 
  50 #ifdef _LP64
  51 // Symbolically name the register arguments used by the c calling convention.
  52 // Windows is different from linux/solaris. So much for standards...
  53 
  54 #ifdef _WIN64
  55 
  56 REGISTER_DECLARATION(Register, c_rarg0, rcx);
  57 REGISTER_DECLARATION(Register, c_rarg1, rdx);
  58 REGISTER_DECLARATION(Register, c_rarg2, r8);
  59 REGISTER_DECLARATION(Register, c_rarg3, r9);
  60 
  61 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0);
  62 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1);
  63 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2);
  64 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3);
  65 
  66 #else
  67 
  68 REGISTER_DECLARATION(Register, c_rarg0, rdi);
  69 REGISTER_DECLARATION(Register, c_rarg1, rsi);
  70 REGISTER_DECLARATION(Register, c_rarg2, rdx);
  71 REGISTER_DECLARATION(Register, c_rarg3, rcx);
  72 REGISTER_DECLARATION(Register, c_rarg4, r8);
  73 REGISTER_DECLARATION(Register, c_rarg5, r9);
  74 
  75 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0);
  76 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1);
  77 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2);
  78 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3);
  79 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4);
  80 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5);
  81 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6);
  82 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7);
  83 
  84 #endif // _WIN64
  85 
  86 // Symbolically name the register arguments used by the Java calling convention.
  87 // We have control over the convention for java so we can do what we please.
  88 // What pleases us is to offset the java calling convention so that when
  89 // we call a suitable jni method the arguments are lined up and we don't
  90 // have to do little shuffling. A suitable jni method is non-static and a
  91 // small number of arguments (two fewer args on windows)
  92 //
  93 //        |-------------------------------------------------------|
  94 //        | c_rarg0   c_rarg1  c_rarg2 c_rarg3 c_rarg4 c_rarg5    |
  95 //        |-------------------------------------------------------|
  96 //        | rcx       rdx      r8      r9      rdi*    rsi*       | windows (* not a c_rarg)
  97 //        | rdi       rsi      rdx     rcx     r8      r9         | solaris/linux
  98 //        |-------------------------------------------------------|
  99 //        | j_rarg5   j_rarg0  j_rarg1 j_rarg2 j_rarg3 j_rarg4    |
 100 //        |-------------------------------------------------------|
 101 
 102 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1);
 103 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2);
 104 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3);
 105 // Windows runs out of register args here
 106 #ifdef _WIN64
 107 REGISTER_DECLARATION(Register, j_rarg3, rdi);
 108 REGISTER_DECLARATION(Register, j_rarg4, rsi);
 109 #else
 110 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4);
 111 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5);
 112 #endif /* _WIN64 */
 113 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0);
 114 
 115 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0);
 116 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1);
 117 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2);
 118 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3);
 119 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4);
 120 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5);
 121 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6);
 122 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7);
 123 
 124 REGISTER_DECLARATION(Register, rscratch1, r10);  // volatile
 125 REGISTER_DECLARATION(Register, rscratch2, r11);  // volatile
 126 
 127 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved
 128 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved
 129 
 130 #else
 131 // rscratch1 will apear in 32bit code that is dead but of course must compile
 132 // Using noreg ensures if the dead code is incorrectly live and executed it
 133 // will cause an assertion failure
 134 #define rscratch1 noreg
 135 
 136 #endif // _LP64
 137 
 138 // Address is an abstraction used to represent a memory location
 139 // using any of the amd64 addressing modes with one object.
 140 //
 141 // Note: A register location is represented via a Register, not
 142 //       via an address for efficiency & simplicity reasons.
 143 
 144 class ArrayAddress;
 145 
 146 class Address VALUE_OBJ_CLASS_SPEC {
 147  public:
 148   enum ScaleFactor {
 149     no_scale = -1,
 150     times_1  =  0,
 151     times_2  =  1,
 152     times_4  =  2,
 153     times_8  =  3,
 154     times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4)
 155   };
 156 
 157  private:
 158   Register         _base;
 159   Register         _index;
 160   ScaleFactor      _scale;
 161   int              _disp;
 162   RelocationHolder _rspec;
 163 
 164   // Easily misused constructors make them private
 165   // %%% can we make these go away?
 166   NOT_LP64(Address(address loc, RelocationHolder spec);)
 167   Address(int disp, address loc, relocInfo::relocType rtype);
 168   Address(int disp, address loc, RelocationHolder spec);
 169 
 170  public:
 171 
 172  int disp() { return _disp; }
 173   // creation
 174   Address()
 175     : _base(noreg),
 176       _index(noreg),
 177       _scale(no_scale),
 178       _disp(0) {
 179   }
 180 
 181   // No default displacement otherwise Register can be implicitly
 182   // converted to 0(Register) which is quite a different animal.
 183 
 184   Address(Register base, int disp)
 185     : _base(base),
 186       _index(noreg),
 187       _scale(no_scale),
 188       _disp(disp) {
 189   }
 190 
 191   Address(Register base, Register index, ScaleFactor scale, int disp = 0)
 192     : _base (base),
 193       _index(index),
 194       _scale(scale),
 195       _disp (disp) {
 196     assert(!index->is_valid() == (scale == Address::no_scale),
 197            "inconsistent address");
 198   }
 199 
 200   // The following two overloads are used in connection with the
 201   // ByteSize type (see sizes.hpp).  They simplify the use of
 202   // ByteSize'd arguments in assembly code. Note that their equivalent
 203   // for the optimized build are the member functions with int disp
 204   // argument since ByteSize is mapped to an int type in that case.
 205   //
 206   // Note: DO NOT introduce similar overloaded functions for WordSize
 207   // arguments as in the optimized mode, both ByteSize and WordSize
 208   // are mapped to the same type and thus the compiler cannot make a
 209   // distinction anymore (=> compiler errors).
 210 
 211 #ifdef ASSERT
 212   Address(Register base, ByteSize disp)
 213     : _base(base),
 214       _index(noreg),
 215       _scale(no_scale),
 216       _disp(in_bytes(disp)) {
 217   }
 218 
 219   Address(Register base, Register index, ScaleFactor scale, ByteSize disp)
 220     : _base(base),
 221       _index(index),
 222       _scale(scale),
 223       _disp(in_bytes(disp)) {
 224     assert(!index->is_valid() == (scale == Address::no_scale),
 225            "inconsistent address");
 226   }
 227 #endif // ASSERT
 228 
 229   // accessors
 230   bool        uses(Register reg) const { return _base == reg || _index == reg; }
 231   Register    base()             const { return _base;  }
 232   Register    index()            const { return _index; }
 233   ScaleFactor scale()            const { return _scale; }
 234   int         disp()             const { return _disp;  }
 235 
 236   // Convert the raw encoding form into the form expected by the constructor for
 237   // Address.  An index of 4 (rsp) corresponds to having no index, so convert
 238   // that to noreg for the Address constructor.
 239   static Address make_raw(int base, int index, int scale, int disp);
 240 
 241   static Address make_array(ArrayAddress);
 242 
 243 
 244  private:
 245   bool base_needs_rex() const {
 246     return _base != noreg && _base->encoding() >= 8;
 247   }
 248 
 249   bool index_needs_rex() const {
 250     return _index != noreg &&_index->encoding() >= 8;
 251   }
 252 
 253   relocInfo::relocType reloc() const { return _rspec.type(); }
 254 
 255   friend class Assembler;
 256   friend class MacroAssembler;
 257   friend class LIR_Assembler; // base/index/scale/disp
 258 };
 259 
 260 //
 261 // AddressLiteral has been split out from Address because operands of this type
 262 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out
 263 // the few instructions that need to deal with address literals are unique and the
 264 // MacroAssembler does not have to implement every instruction in the Assembler
 265 // in order to search for address literals that may need special handling depending
 266 // on the instruction and the platform. As small step on the way to merging i486/amd64
 267 // directories.
 268 //
 269 class AddressLiteral VALUE_OBJ_CLASS_SPEC {
 270   friend class ArrayAddress;
 271   RelocationHolder _rspec;
 272   // Typically we use AddressLiterals we want to use their rval
 273   // However in some situations we want the lval (effect address) of the item.
 274   // We provide a special factory for making those lvals.
 275   bool _is_lval;
 276 
 277   // If the target is far we'll need to load the ea of this to
 278   // a register to reach it. Otherwise if near we can do rip
 279   // relative addressing.
 280 
 281   address          _target;
 282 
 283  protected:
 284   // creation
 285   AddressLiteral()
 286     : _is_lval(false),
 287       _target(NULL)
 288   {}
 289 
 290   public:
 291 
 292 
 293   AddressLiteral(address target, relocInfo::relocType rtype);
 294 
 295   AddressLiteral(address target, RelocationHolder const& rspec)
 296     : _rspec(rspec),
 297       _is_lval(false),
 298       _target(target)
 299   {}
 300 
 301   AddressLiteral addr() {
 302     AddressLiteral ret = *this;
 303     ret._is_lval = true;
 304     return ret;
 305   }
 306 
 307 
 308  private:
 309 
 310   address target() { return _target; }
 311   bool is_lval() { return _is_lval; }
 312 
 313   relocInfo::relocType reloc() const { return _rspec.type(); }
 314   const RelocationHolder& rspec() const { return _rspec; }
 315 
 316   friend class Assembler;
 317   friend class MacroAssembler;
 318   friend class Address;
 319   friend class LIR_Assembler;
 320 };
 321 
 322 // Convience classes
 323 class RuntimeAddress: public AddressLiteral {
 324 
 325   public:
 326 
 327   RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {}
 328 
 329 };
 330 
 331 class OopAddress: public AddressLiteral {
 332 
 333   public:
 334 
 335   OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){}
 336 
 337 };
 338 
 339 class ExternalAddress: public AddressLiteral {
 340 
 341   public:
 342 
 343   ExternalAddress(address target) : AddressLiteral(target, relocInfo::external_word_type){}
 344 
 345 };
 346 
 347 class InternalAddress: public AddressLiteral {
 348 
 349   public:
 350 
 351   InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {}
 352 
 353 };
 354 
 355 // x86 can do array addressing as a single operation since disp can be an absolute
 356 // address amd64 can't. We create a class that expresses the concept but does extra
 357 // magic on amd64 to get the final result
 358 
 359 class ArrayAddress VALUE_OBJ_CLASS_SPEC {
 360   private:
 361 
 362   AddressLiteral _base;
 363   Address        _index;
 364 
 365   public:
 366 
 367   ArrayAddress() {};
 368   ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {};
 369   AddressLiteral base() { return _base; }
 370   Address index() { return _index; }
 371 
 372 };
 373 
 374 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize);
 375 
 376 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction
 377 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write
 378 // is what you get. The Assembler is generating code into a CodeBuffer.
 379 
 380 class Assembler : public AbstractAssembler  {
 381   friend class AbstractAssembler; // for the non-virtual hack
 382   friend class LIR_Assembler; // as_Address()
 383   friend class StubGenerator;
 384 
 385  public:
 386   enum Condition {                     // The x86 condition codes used for conditional jumps/moves.
 387     zero          = 0x4,
 388     notZero       = 0x5,
 389     equal         = 0x4,
 390     notEqual      = 0x5,
 391     less          = 0xc,
 392     lessEqual     = 0xe,
 393     greater       = 0xf,
 394     greaterEqual  = 0xd,
 395     below         = 0x2,
 396     belowEqual    = 0x6,
 397     above         = 0x7,
 398     aboveEqual    = 0x3,
 399     overflow      = 0x0,
 400     noOverflow    = 0x1,
 401     carrySet      = 0x2,
 402     carryClear    = 0x3,
 403     negative      = 0x8,
 404     positive      = 0x9,
 405     parity        = 0xa,
 406     noParity      = 0xb
 407   };
 408 
 409   enum Prefix {
 410     // segment overrides
 411     CS_segment = 0x2e,
 412     SS_segment = 0x36,
 413     DS_segment = 0x3e,
 414     ES_segment = 0x26,
 415     FS_segment = 0x64,
 416     GS_segment = 0x65,
 417 
 418     REX        = 0x40,
 419 
 420     REX_B      = 0x41,
 421     REX_X      = 0x42,
 422     REX_XB     = 0x43,
 423     REX_R      = 0x44,
 424     REX_RB     = 0x45,
 425     REX_RX     = 0x46,
 426     REX_RXB    = 0x47,
 427 
 428     REX_W      = 0x48,
 429 
 430     REX_WB     = 0x49,
 431     REX_WX     = 0x4A,
 432     REX_WXB    = 0x4B,
 433     REX_WR     = 0x4C,
 434     REX_WRB    = 0x4D,
 435     REX_WRX    = 0x4E,
 436     REX_WRXB   = 0x4F
 437   };
 438 
 439   enum WhichOperand {
 440     // input to locate_operand, and format code for relocations
 441     imm_operand  = 0,            // embedded 32-bit|64-bit immediate operand
 442     disp32_operand = 1,          // embedded 32-bit displacement or address
 443     call32_operand = 2,          // embedded 32-bit self-relative displacement
 444 #ifndef _LP64
 445     _WhichOperand_limit = 3
 446 #else
 447      narrow_oop_operand = 3,     // embedded 32-bit immediate narrow oop
 448     _WhichOperand_limit = 4
 449 #endif
 450   };
 451 
 452 
 453 
 454   // NOTE: The general philopsophy of the declarations here is that 64bit versions
 455   // of instructions are freely declared without the need for wrapping them an ifdef.
 456   // (Some dangerous instructions are ifdef's out of inappropriate jvm's.)
 457   // In the .cpp file the implementations are wrapped so that they are dropped out
 458   // of the resulting jvm. This is done mostly to keep the footprint of KERNEL
 459   // to the size it was prior to merging up the 32bit and 64bit assemblers.
 460   //
 461   // This does mean you'll get a linker/runtime error if you use a 64bit only instruction
 462   // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down.
 463 
 464 private:
 465 
 466 
 467   // 64bit prefixes
 468   int prefix_and_encode(int reg_enc, bool byteinst = false);
 469   int prefixq_and_encode(int reg_enc);
 470 
 471   int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false);
 472   int prefixq_and_encode(int dst_enc, int src_enc);
 473 
 474   void prefix(Register reg);
 475   void prefix(Address adr);
 476   void prefixq(Address adr);
 477 
 478   void prefix(Address adr, Register reg,  bool byteinst = false);
 479   void prefixq(Address adr, Register reg);
 480 
 481   void prefix(Address adr, XMMRegister reg);
 482 
 483   void prefetch_prefix(Address src);
 484 
 485   // Helper functions for groups of instructions
 486   void emit_arith_b(int op1, int op2, Register dst, int imm8);
 487 
 488   void emit_arith(int op1, int op2, Register dst, int32_t imm32);
 489   // only 32bit??
 490   void emit_arith(int op1, int op2, Register dst, jobject obj);
 491   void emit_arith(int op1, int op2, Register dst, Register src);
 492 
 493   void emit_operand(Register reg,
 494                     Register base, Register index, Address::ScaleFactor scale,
 495                     int disp,
 496                     RelocationHolder const& rspec,
 497                     int rip_relative_correction = 0);
 498 
 499   void emit_operand(Register reg, Address adr, int rip_relative_correction = 0);
 500 
 501   // operands that only take the original 32bit registers
 502   void emit_operand32(Register reg, Address adr);
 503 
 504   void emit_operand(XMMRegister reg,
 505                     Register base, Register index, Address::ScaleFactor scale,
 506                     int disp,
 507                     RelocationHolder const& rspec);
 508 
 509   void emit_operand(XMMRegister reg, Address adr);
 510 
 511   void emit_operand(MMXRegister reg, Address adr);
 512 
 513   // workaround gcc (3.2.1-7) bug
 514   void emit_operand(Address adr, MMXRegister reg);
 515 
 516 
 517   // Immediate-to-memory forms
 518   void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32);
 519 
 520   void emit_farith(int b1, int b2, int i);
 521 
 522 
 523  protected:
 524   #ifdef ASSERT
 525   void check_relocation(RelocationHolder const& rspec, int format);
 526   #endif
 527 
 528   inline void emit_long64(jlong x);
 529 
 530   void emit_data(jint data, relocInfo::relocType    rtype, int format);
 531   void emit_data(jint data, RelocationHolder const& rspec, int format);
 532   void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0);
 533   void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0);
 534 
 535 
 536   bool reachable(AddressLiteral adr) NOT_LP64({ return true;});
 537 
 538   // These are all easily abused and hence protected
 539 
 540   void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec, int format = 0);
 541 
 542   // 32BIT ONLY SECTION
 543 #ifndef _LP64
 544   // Make these disappear in 64bit mode since they would never be correct
 545   void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec);   // 32BIT ONLY
 546   void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec);    // 32BIT ONLY
 547 
 548   void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec);     // 32BIT ONLY
 549 
 550   void push_literal32(int32_t imm32, RelocationHolder const& rspec);                 // 32BIT ONLY
 551 #else
 552   // 64BIT ONLY SECTION
 553   void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec);   // 64BIT ONLY
 554 #endif // _LP64
 555 
 556   // These are unique in that we are ensured by the caller that the 32bit
 557   // relative in these instructions will always be able to reach the potentially
 558   // 64bit address described by entry. Since they can take a 64bit address they
 559   // don't have the 32 suffix like the other instructions in this class.
 560 
 561   void call_literal(address entry, RelocationHolder const& rspec);
 562   void jmp_literal(address entry, RelocationHolder const& rspec);
 563 
 564   // Avoid using directly section
 565   // Instructions in this section are actually usable by anyone without danger
 566   // of failure but have performance issues that are addressed my enhanced
 567   // instructions which will do the proper thing base on the particular cpu.
 568   // We protect them because we don't trust you...
 569 
 570   // Don't use next inc() and dec() methods directly. INC & DEC instructions
 571   // could cause a partial flag stall since they don't set CF flag.
 572   // Use MacroAssembler::decrement() & MacroAssembler::increment() methods
 573   // which call inc() & dec() or add() & sub() in accordance with
 574   // the product flag UseIncDec value.
 575 
 576   void decl(Register dst);
 577   void decl(Address dst);
 578   void decq(Register dst);
 579   void decq(Address dst);
 580 
 581   void incl(Register dst);
 582   void incl(Address dst);
 583   void incq(Register dst);
 584   void incq(Address dst);
 585 
 586   // New cpus require use of movsd and movss to avoid partial register stall
 587   // when loading from memory. But for old Opteron use movlpd instead of movsd.
 588   // The selection is done in MacroAssembler::movdbl() and movflt().
 589 
 590   // Move Scalar Single-Precision Floating-Point Values
 591   void movss(XMMRegister dst, Address src);
 592   void movss(XMMRegister dst, XMMRegister src);
 593   void movss(Address dst, XMMRegister src);
 594 
 595   // Move Scalar Double-Precision Floating-Point Values
 596   void movsd(XMMRegister dst, Address src);
 597   void movsd(XMMRegister dst, XMMRegister src);
 598   void movsd(Address dst, XMMRegister src);
 599   void movlpd(XMMRegister dst, Address src);
 600 
 601   // New cpus require use of movaps and movapd to avoid partial register stall
 602   // when moving between registers.
 603   void movaps(XMMRegister dst, XMMRegister src);
 604   void movapd(XMMRegister dst, XMMRegister src);
 605 
 606   // End avoid using directly
 607 
 608 
 609   // Instruction prefixes
 610   void prefix(Prefix p);
 611 
 612   public:
 613 
 614   // Creation
 615   Assembler(CodeBuffer* code) : AbstractAssembler(code) {}
 616 
 617   // Decoding
 618   static address locate_operand(address inst, WhichOperand which);
 619   static address locate_next_instruction(address inst);
 620 
 621   // Utilities
 622 
 623 #ifdef _LP64
 624  static bool is_simm(int64_t x, int nbits) { return -( CONST64(1) << (nbits-1) )  <= x   &&   x  <  ( CONST64(1) << (nbits-1) ); }
 625  static bool is_simm32(int64_t x) { return x == (int64_t)(int32_t)x; }
 626 #else
 627  static bool is_simm(int32_t x, int nbits) { return -( 1 << (nbits-1) )  <= x   &&   x  <  ( 1 << (nbits-1) ); }
 628  static bool is_simm32(int32_t x) { return true; }
 629 #endif // LP64
 630 
 631   // Generic instructions
 632   // Does 32bit or 64bit as needed for the platform. In some sense these
 633   // belong in macro assembler but there is no need for both varieties to exist
 634 
 635   void lea(Register dst, Address src);
 636 
 637   void mov(Register dst, Register src);
 638 
 639   void pusha();
 640   void popa();
 641 
 642   void pushf();
 643   void popf();
 644 
 645   void push(int32_t imm32);
 646 
 647   void push(Register src);
 648 
 649   void pop(Register dst);
 650 
 651   // These are dummies to prevent surprise implicit conversions to Register
 652   void push(void* v);
 653   void pop(void* v);
 654 
 655 
 656   // These do register sized moves/scans
 657   void rep_mov();
 658   void rep_set();
 659   void repne_scan();
 660 #ifdef _LP64
 661   void repne_scanl();
 662 #endif
 663 
 664   // Vanilla instructions in lexical order
 665 
 666   void adcl(Register dst, int32_t imm32);
 667   void adcl(Register dst, Address src);
 668   void adcl(Register dst, Register src);
 669 
 670   void adcq(Register dst, int32_t imm32);
 671   void adcq(Register dst, Address src);
 672   void adcq(Register dst, Register src);
 673 
 674 
 675   void addl(Address dst, int32_t imm32);
 676   void addl(Address dst, Register src);
 677   void addl(Register dst, int32_t imm32);
 678   void addl(Register dst, Address src);
 679   void addl(Register dst, Register src);
 680 
 681   void addq(Address dst, int32_t imm32);
 682   void addq(Address dst, Register src);
 683   void addq(Register dst, int32_t imm32);
 684   void addq(Register dst, Address src);
 685   void addq(Register dst, Register src);
 686 
 687 
 688   void addr_nop_4();
 689   void addr_nop_5();
 690   void addr_nop_7();
 691   void addr_nop_8();
 692 
 693   // Add Scalar Double-Precision Floating-Point Values
 694   void addsd(XMMRegister dst, Address src);
 695   void addsd(XMMRegister dst, XMMRegister src);
 696 
 697   // Add Scalar Single-Precision Floating-Point Values
 698   void addss(XMMRegister dst, Address src);
 699   void addss(XMMRegister dst, XMMRegister src);
 700 
 701   void andl(Register dst, int32_t imm32);
 702   void andl(Register dst, Address src);
 703   void andl(Register dst, Register src);
 704 
 705   void andq(Register dst, int32_t imm32);
 706   void andq(Register dst, Address src);
 707   void andq(Register dst, Register src);
 708 
 709 
 710   // Bitwise Logical AND of Packed Double-Precision Floating-Point Values
 711   void andpd(XMMRegister dst, Address src);
 712   void andpd(XMMRegister dst, XMMRegister src);
 713 
 714   void bswapl(Register reg);
 715 
 716   void bswapq(Register reg);
 717 
 718   void call(Label& L, relocInfo::relocType rtype);
 719   void call(Register reg);  // push pc; pc <- reg
 720   void call(Address adr);   // push pc; pc <- adr
 721 
 722   void cdql();
 723 
 724   void cdqq();
 725 
 726   void cld() { emit_byte(0xfc); }
 727 
 728   void clflush(Address adr);
 729 
 730   void cmovl(Condition cc, Register dst, Register src);
 731   void cmovl(Condition cc, Register dst, Address src);
 732 
 733   void cmovq(Condition cc, Register dst, Register src);
 734   void cmovq(Condition cc, Register dst, Address src);
 735 
 736 
 737   void cmpb(Address dst, int imm8);
 738 
 739   void cmpl(Address dst, int32_t imm32);
 740 
 741   void cmpl(Register dst, int32_t imm32);
 742   void cmpl(Register dst, Register src);
 743   void cmpl(Register dst, Address src);
 744 
 745   void cmpq(Address dst, int32_t imm32);
 746   void cmpq(Address dst, Register src);
 747 
 748   void cmpq(Register dst, int32_t imm32);
 749   void cmpq(Register dst, Register src);
 750   void cmpq(Register dst, Address src);
 751 
 752   // these are dummies used to catch attempting to convert NULL to Register
 753   void cmpl(Register dst, void* junk); // dummy
 754   void cmpq(Register dst, void* junk); // dummy
 755 
 756   void cmpw(Address dst, int imm16);
 757 
 758   void cmpxchg8 (Address adr);
 759 
 760   void cmpxchgl(Register reg, Address adr);
 761 
 762   void cmpxchgq(Register reg, Address adr);
 763 
 764   // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
 765   void comisd(XMMRegister dst, Address src);
 766 
 767   // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
 768   void comiss(XMMRegister dst, Address src);
 769 
 770   // Identify processor type and features
 771   void cpuid() {
 772     emit_byte(0x0F);
 773     emit_byte(0xA2);
 774   }
 775 
 776   // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value
 777   void cvtsd2ss(XMMRegister dst, XMMRegister src);
 778 
 779   // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value
 780   void cvtsi2sdl(XMMRegister dst, Register src);
 781   void cvtsi2sdq(XMMRegister dst, Register src);
 782 
 783   // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value
 784   void cvtsi2ssl(XMMRegister dst, Register src);
 785   void cvtsi2ssq(XMMRegister dst, Register src);
 786 
 787   // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value
 788   void cvtdq2pd(XMMRegister dst, XMMRegister src);
 789 
 790   // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value
 791   void cvtdq2ps(XMMRegister dst, XMMRegister src);
 792 
 793   // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value
 794   void cvtss2sd(XMMRegister dst, XMMRegister src);
 795 
 796   // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer
 797   void cvttsd2sil(Register dst, Address src);
 798   void cvttsd2sil(Register dst, XMMRegister src);
 799   void cvttsd2siq(Register dst, XMMRegister src);
 800 
 801   // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer
 802   void cvttss2sil(Register dst, XMMRegister src);
 803   void cvttss2siq(Register dst, XMMRegister src);
 804 
 805   // Divide Scalar Double-Precision Floating-Point Values
 806   void divsd(XMMRegister dst, Address src);
 807   void divsd(XMMRegister dst, XMMRegister src);
 808 
 809   // Divide Scalar Single-Precision Floating-Point Values
 810   void divss(XMMRegister dst, Address src);
 811   void divss(XMMRegister dst, XMMRegister src);
 812 
 813   void emms();
 814 
 815   void fabs();
 816 
 817   void fadd(int i);
 818 
 819   void fadd_d(Address src);
 820   void fadd_s(Address src);
 821 
 822   // "Alternate" versions of x87 instructions place result down in FPU
 823   // stack instead of on TOS
 824 
 825   void fadda(int i); // "alternate" fadd
 826   void faddp(int i = 1);
 827 
 828   void fchs();
 829 
 830   void fcom(int i);
 831 
 832   void fcomp(int i = 1);
 833   void fcomp_d(Address src);
 834   void fcomp_s(Address src);
 835 
 836   void fcompp();
 837 
 838   void fcos();
 839 
 840   void fdecstp();
 841 
 842   void fdiv(int i);
 843   void fdiv_d(Address src);
 844   void fdivr_s(Address src);
 845   void fdiva(int i);  // "alternate" fdiv
 846   void fdivp(int i = 1);
 847 
 848   void fdivr(int i);
 849   void fdivr_d(Address src);
 850   void fdiv_s(Address src);
 851 
 852   void fdivra(int i); // "alternate" reversed fdiv
 853 
 854   void fdivrp(int i = 1);
 855 
 856   void ffree(int i = 0);
 857 
 858   void fild_d(Address adr);
 859   void fild_s(Address adr);
 860 
 861   void fincstp();
 862 
 863   void finit();
 864 
 865   void fist_s (Address adr);
 866   void fistp_d(Address adr);
 867   void fistp_s(Address adr);
 868 
 869   void fld1();
 870 
 871   void fld_d(Address adr);
 872   void fld_s(Address adr);
 873   void fld_s(int index);
 874   void fld_x(Address adr);  // extended-precision (80-bit) format
 875 
 876   void fldcw(Address src);
 877 
 878   void fldenv(Address src);
 879 
 880   void fldlg2();
 881 
 882   void fldln2();
 883 
 884   void fldz();
 885 
 886   void flog();
 887   void flog10();
 888 
 889   void fmul(int i);
 890 
 891   void fmul_d(Address src);
 892   void fmul_s(Address src);
 893 
 894   void fmula(int i);  // "alternate" fmul
 895 
 896   void fmulp(int i = 1);
 897 
 898   void fnsave(Address dst);
 899 
 900   void fnstcw(Address src);
 901 
 902   void fnstsw_ax();
 903 
 904   void fprem();
 905   void fprem1();
 906 
 907   void frstor(Address src);
 908 
 909   void fsin();
 910 
 911   void fsqrt();
 912 
 913   void fst_d(Address adr);
 914   void fst_s(Address adr);
 915 
 916   void fstp_d(Address adr);
 917   void fstp_d(int index);
 918   void fstp_s(Address adr);
 919   void fstp_x(Address adr); // extended-precision (80-bit) format
 920 
 921   void fsub(int i);
 922   void fsub_d(Address src);
 923   void fsub_s(Address src);
 924 
 925   void fsuba(int i);  // "alternate" fsub
 926 
 927   void fsubp(int i = 1);
 928 
 929   void fsubr(int i);
 930   void fsubr_d(Address src);
 931   void fsubr_s(Address src);
 932 
 933   void fsubra(int i); // "alternate" reversed fsub
 934 
 935   void fsubrp(int i = 1);
 936 
 937   void ftan();
 938 
 939   void ftst();
 940 
 941   void fucomi(int i = 1);
 942   void fucomip(int i = 1);
 943 
 944   void fwait();
 945 
 946   void fxch(int i = 1);
 947 
 948   void fxrstor(Address src);
 949 
 950   void fxsave(Address dst);
 951 
 952   void fyl2x();
 953 
 954   void hlt();
 955 
 956   void idivl(Register src);
 957 
 958   void idivq(Register src);
 959 
 960   void imull(Register dst, Register src);
 961   void imull(Register dst, Register src, int value);
 962 
 963   void imulq(Register dst, Register src);
 964   void imulq(Register dst, Register src, int value);
 965 
 966 
 967   // jcc is the generic conditional branch generator to run-
 968   // time routines, jcc is used for branches to labels. jcc
 969   // takes a branch opcode (cc) and a label (L) and generates
 970   // either a backward branch or a forward branch and links it
 971   // to the label fixup chain. Usage:
 972   //
 973   // Label L;      // unbound label
 974   // jcc(cc, L);   // forward branch to unbound label
 975   // bind(L);      // bind label to the current pc
 976   // jcc(cc, L);   // backward branch to bound label
 977   // bind(L);      // illegal: a label may be bound only once
 978   //
 979   // Note: The same Label can be used for forward and backward branches
 980   // but it may be bound only once.
 981 
 982   void jcc(Condition cc, Label& L,
 983            relocInfo::relocType rtype = relocInfo::none);
 984 
 985   // Conditional jump to a 8-bit offset to L.
 986   // WARNING: be very careful using this for forward jumps.  If the label is
 987   // not bound within an 8-bit offset of this instruction, a run-time error
 988   // will occur.
 989   void jccb(Condition cc, Label& L);
 990 
 991   void jmp(Address entry);    // pc <- entry
 992 
 993   // Label operations & relative jumps (PPUM Appendix D)
 994   void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none);   // unconditional jump to L
 995 
 996   void jmp(Register entry); // pc <- entry
 997 
 998   // Unconditional 8-bit offset jump to L.
 999   // WARNING: be very careful using this for forward jumps.  If the label is
1000   // not bound within an 8-bit offset of this instruction, a run-time error
1001   // will occur.
1002   void jmpb(Label& L);
1003 
1004   void ldmxcsr( Address src );
1005 
1006   void leal(Register dst, Address src);
1007 
1008   void leaq(Register dst, Address src);
1009 
1010   void lfence() {
1011     emit_byte(0x0F);
1012     emit_byte(0xAE);
1013     emit_byte(0xE8);
1014   }
1015 
1016   void lock();
1017 
1018   enum Membar_mask_bits {
1019     StoreStore = 1 << 3,
1020     LoadStore  = 1 << 2,
1021     StoreLoad  = 1 << 1,
1022     LoadLoad   = 1 << 0
1023   };
1024 
1025   // Serializes memory.
1026   void membar(Membar_mask_bits order_constraint) {
1027     // We only have to handle StoreLoad and LoadLoad
1028     if (order_constraint & StoreLoad) {
1029       // MFENCE subsumes LFENCE
1030       mfence();
1031     } /* [jk] not needed currently: else if (order_constraint & LoadLoad) {
1032          lfence();
1033     } */
1034   }
1035 
1036   void mfence();
1037 
1038   // Moves
1039 
1040   void mov64(Register dst, int64_t imm64);
1041 
1042   void movb(Address dst, Register src);
1043   void movb(Address dst, int imm8);
1044   void movb(Register dst, Address src);
1045 
1046   void movdl(XMMRegister dst, Register src);
1047   void movdl(Register dst, XMMRegister src);
1048 
1049   // Move Double Quadword
1050   void movdq(XMMRegister dst, Register src);
1051   void movdq(Register dst, XMMRegister src);
1052 
1053   // Move Aligned Double Quadword
1054   void movdqa(Address     dst, XMMRegister src);
1055   void movdqa(XMMRegister dst, Address src);
1056   void movdqa(XMMRegister dst, XMMRegister src);
1057 
1058   void movl(Register dst, int32_t imm32);
1059   void movl(Address dst, int32_t imm32);
1060   void movl(Register dst, Register src);
1061   void movl(Register dst, Address src);
1062   void movl(Address dst, Register src);
1063 
1064   // These dummies prevent using movl from converting a zero (like NULL) into Register
1065   // by giving the compiler two choices it can't resolve
1066 
1067   void movl(Address  dst, void* junk);
1068   void movl(Register dst, void* junk);
1069 
1070 #ifdef _LP64
1071   void movq(Register dst, Register src);
1072   void movq(Register dst, Address src);
1073   void movq(Address dst, Register src);
1074 #endif
1075 
1076   void movq(Address     dst, MMXRegister src );
1077   void movq(MMXRegister dst, Address src );
1078 
1079 #ifdef _LP64
1080   // These dummies prevent using movq from converting a zero (like NULL) into Register
1081   // by giving the compiler two choices it can't resolve
1082 
1083   void movq(Address  dst, void* dummy);
1084   void movq(Register dst, void* dummy);
1085 #endif
1086 
1087   // Move Quadword
1088   void movq(Address     dst, XMMRegister src);
1089   void movq(XMMRegister dst, Address src);
1090 
1091   void movsbl(Register dst, Address src);
1092   void movsbl(Register dst, Register src);
1093 
1094 #ifdef _LP64
1095   // Move signed 32bit immediate to 64bit extending sign
1096   void movslq(Address dst, int32_t imm64);
1097   void movslq(Register dst, int32_t imm64);
1098 
1099   void movslq(Register dst, Address src);
1100   void movslq(Register dst, Register src);
1101   void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous
1102 #endif
1103 
1104   void movswl(Register dst, Address src);
1105   void movswl(Register dst, Register src);
1106 
1107   void movw(Address dst, int imm16);
1108   void movw(Register dst, Address src);
1109   void movw(Address dst, Register src);
1110 
1111   void movzbl(Register dst, Address src);
1112   void movzbl(Register dst, Register src);
1113 
1114   void movzwl(Register dst, Address src);
1115   void movzwl(Register dst, Register src);
1116 
1117   void mull(Address src);
1118   void mull(Register src);
1119 
1120   // Multiply Scalar Double-Precision Floating-Point Values
1121   void mulsd(XMMRegister dst, Address src);
1122   void mulsd(XMMRegister dst, XMMRegister src);
1123 
1124   // Multiply Scalar Single-Precision Floating-Point Values
1125   void mulss(XMMRegister dst, Address src);
1126   void mulss(XMMRegister dst, XMMRegister src);
1127 
1128   void negl(Register dst);
1129 
1130 #ifdef _LP64
1131   void negq(Register dst);
1132 #endif
1133 
1134   void nop(int i = 1);
1135 
1136   void notl(Register dst);
1137 
1138 #ifdef _LP64
1139   void notq(Register dst);
1140 #endif
1141 
1142   void orl(Address dst, int32_t imm32);
1143   void orl(Register dst, int32_t imm32);
1144   void orl(Register dst, Address src);
1145   void orl(Register dst, Register src);
1146 
1147   void orq(Address dst, int32_t imm32);
1148   void orq(Register dst, int32_t imm32);
1149   void orq(Register dst, Address src);
1150   void orq(Register dst, Register src);
1151 
1152   void popl(Address dst);
1153 
1154 #ifdef _LP64
1155   void popq(Address dst);
1156 #endif
1157 
1158   // Prefetches (SSE, SSE2, 3DNOW only)
1159 
1160   void prefetchnta(Address src);
1161   void prefetchr(Address src);
1162   void prefetcht0(Address src);
1163   void prefetcht1(Address src);
1164   void prefetcht2(Address src);
1165   void prefetchw(Address src);
1166 
1167   // Shuffle Packed Doublewords
1168   void pshufd(XMMRegister dst, XMMRegister src, int mode);
1169   void pshufd(XMMRegister dst, Address src,     int mode);
1170 
1171   // Shuffle Packed Low Words
1172   void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1173   void pshuflw(XMMRegister dst, Address src,     int mode);
1174 
1175   // Shift Right Logical Quadword Immediate
1176   void psrlq(XMMRegister dst, int shift);
1177 
1178   // Interleave Low Bytes
1179   void punpcklbw(XMMRegister dst, XMMRegister src);
1180 
1181   void pushl(Address src);
1182 
1183   void pushq(Address src);
1184 
1185   // Xor Packed Byte Integer Values
1186   void pxor(XMMRegister dst, Address src);
1187   void pxor(XMMRegister dst, XMMRegister src);
1188 
1189   void rcll(Register dst, int imm8);
1190 
1191   void rclq(Register dst, int imm8);
1192 
1193   void ret(int imm16);
1194 
1195   void sahf();
1196 
1197   void sarl(Register dst, int imm8);
1198   void sarl(Register dst);
1199 
1200   void sarq(Register dst, int imm8);
1201   void sarq(Register dst);
1202 
1203   void sbbl(Address dst, int32_t imm32);
1204   void sbbl(Register dst, int32_t imm32);
1205   void sbbl(Register dst, Address src);
1206   void sbbl(Register dst, Register src);
1207 
1208   void sbbq(Address dst, int32_t imm32);
1209   void sbbq(Register dst, int32_t imm32);
1210   void sbbq(Register dst, Address src);
1211   void sbbq(Register dst, Register src);
1212 
1213   void setb(Condition cc, Register dst);
1214 
1215   void shldl(Register dst, Register src);
1216 
1217   void shll(Register dst, int imm8);
1218   void shll(Register dst);
1219 
1220   void shlq(Register dst, int imm8);
1221   void shlq(Register dst);
1222 
1223   void shrdl(Register dst, Register src);
1224 
1225   void shrl(Register dst, int imm8);
1226   void shrl(Register dst);
1227 
1228   void shrq(Register dst, int imm8);
1229   void shrq(Register dst);
1230 
1231   void smovl(); // QQQ generic?
1232 
1233   // Compute Square Root of Scalar Double-Precision Floating-Point Value
1234   void sqrtsd(XMMRegister dst, Address src);
1235   void sqrtsd(XMMRegister dst, XMMRegister src);
1236 
1237   void std() { emit_byte(0xfd); }
1238 
1239   void stmxcsr( Address dst );
1240 
1241   void subl(Address dst, int32_t imm32);
1242   void subl(Address dst, Register src);
1243   void subl(Register dst, int32_t imm32);
1244   void subl(Register dst, Address src);
1245   void subl(Register dst, Register src);
1246 
1247   void subq(Address dst, int32_t imm32);
1248   void subq(Address dst, Register src);
1249   void subq(Register dst, int32_t imm32);
1250   void subq(Register dst, Address src);
1251   void subq(Register dst, Register src);
1252 
1253 
1254   // Subtract Scalar Double-Precision Floating-Point Values
1255   void subsd(XMMRegister dst, Address src);
1256   void subsd(XMMRegister dst, XMMRegister src);
1257 
1258   // Subtract Scalar Single-Precision Floating-Point Values
1259   void subss(XMMRegister dst, Address src);
1260   void subss(XMMRegister dst, XMMRegister src);
1261 
1262   void testb(Register dst, int imm8);
1263 
1264   void testl(Register dst, int32_t imm32);
1265   void testl(Register dst, Register src);
1266   void testl(Register dst, Address src);
1267 
1268   void testq(Register dst, int32_t imm32);
1269   void testq(Register dst, Register src);
1270 
1271 
1272   // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
1273   void ucomisd(XMMRegister dst, Address src);
1274   void ucomisd(XMMRegister dst, XMMRegister src);
1275 
1276   // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
1277   void ucomiss(XMMRegister dst, Address src);
1278   void ucomiss(XMMRegister dst, XMMRegister src);
1279 
1280   void xaddl(Address dst, Register src);
1281 
1282   void xaddq(Address dst, Register src);
1283 
1284   void xchgl(Register reg, Address adr);
1285   void xchgl(Register dst, Register src);
1286 
1287   void xchgq(Register reg, Address adr);
1288   void xchgq(Register dst, Register src);
1289 
1290   void xorl(Register dst, int32_t imm32);
1291   void xorl(Register dst, Address src);
1292   void xorl(Register dst, Register src);
1293 
1294   void xorq(Register dst, Address src);
1295   void xorq(Register dst, Register src);
1296 
1297   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1298   void xorpd(XMMRegister dst, Address src);
1299   void xorpd(XMMRegister dst, XMMRegister src);
1300 
1301   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1302   void xorps(XMMRegister dst, Address src);
1303   void xorps(XMMRegister dst, XMMRegister src);
1304 
1305   void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0
1306 };
1307 
1308 
1309 // MacroAssembler extends Assembler by frequently used macros.
1310 //
1311 // Instructions for which a 'better' code sequence exists depending
1312 // on arguments should also go in here.
1313 
1314 class MacroAssembler: public Assembler {
1315   friend class LIR_Assembler;
1316   friend class Runtime1;      // as_Address()
1317  protected:
1318 
1319   Address as_Address(AddressLiteral adr);
1320   Address as_Address(ArrayAddress adr);
1321 
1322   // Support for VM calls
1323   //
1324   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
1325   // may customize this version by overriding it for its purposes (e.g., to save/restore
1326   // additional registers when doing a VM call).
1327 #ifdef CC_INTERP
1328   // c++ interpreter never wants to use interp_masm version of call_VM
1329   #define VIRTUAL
1330 #else
1331   #define VIRTUAL virtual
1332 #endif
1333 
1334   VIRTUAL void call_VM_leaf_base(
1335     address entry_point,               // the entry point
1336     int     number_of_arguments        // the number of arguments to pop after the call
1337   );
1338 
1339   // This is the base routine called by the different versions of call_VM. The interpreter
1340   // may customize this version by overriding it for its purposes (e.g., to save/restore
1341   // additional registers when doing a VM call).
1342   //
1343   // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base
1344   // returns the register which contains the thread upon return. If a thread register has been
1345   // specified, the return value will correspond to that register. If no last_java_sp is specified
1346   // (noreg) than rsp will be used instead.
1347   VIRTUAL void call_VM_base(           // returns the register containing the thread upon return
1348     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
1349     Register java_thread,              // the thread if computed before     ; use noreg otherwise
1350     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
1351     address  entry_point,              // the entry point
1352     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
1353     bool     check_exceptions          // whether to check for pending exceptions after return
1354   );
1355 
1356   // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
1357   // The implementation is only non-empty for the InterpreterMacroAssembler,
1358   // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
1359   virtual void check_and_handle_popframe(Register java_thread);
1360   virtual void check_and_handle_earlyret(Register java_thread);
1361 
1362   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
1363 
1364   // helpers for FPU flag access
1365   // tmp is a temporary register, if none is available use noreg
1366   void save_rax   (Register tmp);
1367   void restore_rax(Register tmp);
1368 
1369  public:
1370   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
1371 
1372   // Support for NULL-checks
1373   //
1374   // Generates code that causes a NULL OS exception if the content of reg is NULL.
1375   // If the accessed location is M[reg + offset] and the offset is known, provide the
1376   // offset. No explicit code generation is needed if the offset is within a certain
1377   // range (0 <= offset <= page_size).
1378 
1379   void null_check(Register reg, int offset = -1);
1380   static bool needs_explicit_null_check(intptr_t offset);
1381 
1382   // Required platform-specific helpers for Label::patch_instructions.
1383   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
1384   void pd_patch_instruction(address branch, address target);
1385 #ifndef PRODUCT
1386   static void pd_print_patched_instruction(address branch);
1387 #endif
1388 
1389   // The following 4 methods return the offset of the appropriate move instruction
1390 
1391   // Support for fast byte/word loading with zero extension (depending on particular CPU)
1392   int load_unsigned_byte(Register dst, Address src);
1393   int load_unsigned_word(Register dst, Address src);
1394 
1395   // Support for fast byte/word loading with sign extension (depending on particular CPU)
1396   int load_signed_byte(Register dst, Address src);
1397   int load_signed_word(Register dst, Address src);
1398 
1399   // Support for sign-extension (hi:lo = extend_sign(lo))
1400   void extend_sign(Register hi, Register lo);
1401 
1402   // Support for inc/dec with optimal instruction selection depending on value
1403 
1404   void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
1405   void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
1406 
1407   void decrementl(Address dst, int value = 1);
1408   void decrementl(Register reg, int value = 1);
1409 
1410   void decrementq(Register reg, int value = 1);
1411   void decrementq(Address dst, int value = 1);
1412 
1413   void incrementl(Address dst, int value = 1);
1414   void incrementl(Register reg, int value = 1);
1415 
1416   void incrementq(Register reg, int value = 1);
1417   void incrementq(Address dst, int value = 1);
1418 
1419 
1420   // Support optimal SSE move instructions.
1421   void movflt(XMMRegister dst, XMMRegister src) {
1422     if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
1423     else                       { movss (dst, src); return; }
1424   }
1425   void movflt(XMMRegister dst, Address src) { movss(dst, src); }
1426   void movflt(XMMRegister dst, AddressLiteral src);
1427   void movflt(Address dst, XMMRegister src) { movss(dst, src); }
1428 
1429   void movdbl(XMMRegister dst, XMMRegister src) {
1430     if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
1431     else                       { movsd (dst, src); return; }
1432   }
1433 
1434   void movdbl(XMMRegister dst, AddressLiteral src);
1435 
1436   void movdbl(XMMRegister dst, Address src) {
1437     if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
1438     else                         { movlpd(dst, src); return; }
1439   }
1440   void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
1441 
1442   void incrementl(AddressLiteral dst);
1443   void incrementl(ArrayAddress dst);
1444 
1445   // Alignment
1446   void align(int modulus);
1447 
1448   // Misc
1449   void fat_nop(); // 5 byte nop
1450 
1451   // Stack frame creation/removal
1452   void enter();
1453   void leave();
1454 
1455   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
1456   // The pointer will be loaded into the thread register.
1457   void get_thread(Register thread);
1458 
1459 
1460   // Support for VM calls
1461   //
1462   // It is imperative that all calls into the VM are handled via the call_VM macros.
1463   // They make sure that the stack linkage is setup correctly. call_VM's correspond
1464   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
1465 
1466 
1467   void call_VM(Register oop_result,
1468                address entry_point,
1469                bool check_exceptions = true);
1470   void call_VM(Register oop_result,
1471                address entry_point,
1472                Register arg_1,
1473                bool check_exceptions = true);
1474   void call_VM(Register oop_result,
1475                address entry_point,
1476                Register arg_1, Register arg_2,
1477                bool check_exceptions = true);
1478   void call_VM(Register oop_result,
1479                address entry_point,
1480                Register arg_1, Register arg_2, Register arg_3,
1481                bool check_exceptions = true);
1482 
1483   // Overloadings with last_Java_sp
1484   void call_VM(Register oop_result,
1485                Register last_java_sp,
1486                address entry_point,
1487                int number_of_arguments = 0,
1488                bool check_exceptions = true);
1489   void call_VM(Register oop_result,
1490                Register last_java_sp,
1491                address entry_point,
1492                Register arg_1, bool
1493                check_exceptions = true);
1494   void call_VM(Register oop_result,
1495                Register last_java_sp,
1496                address entry_point,
1497                Register arg_1, Register arg_2,
1498                bool check_exceptions = true);
1499   void call_VM(Register oop_result,
1500                Register last_java_sp,
1501                address entry_point,
1502                Register arg_1, Register arg_2, Register arg_3,
1503                bool check_exceptions = true);
1504 
1505   void call_VM_leaf(address entry_point,
1506                     int number_of_arguments = 0);
1507   void call_VM_leaf(address entry_point,
1508                     Register arg_1);
1509   void call_VM_leaf(address entry_point,
1510                     Register arg_1, Register arg_2);
1511   void call_VM_leaf(address entry_point,
1512                     Register arg_1, Register arg_2, Register arg_3);
1513 
1514   // last Java Frame (fills frame anchor)
1515   void set_last_Java_frame(Register thread,
1516                            Register last_java_sp,
1517                            Register last_java_fp,
1518                            address last_java_pc);
1519 
1520   // thread in the default location (r15_thread on 64bit)
1521   void set_last_Java_frame(Register last_java_sp,
1522                            Register last_java_fp,
1523                            address last_java_pc);
1524 
1525   void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc);
1526 
1527   // thread in the default location (r15_thread on 64bit)
1528   void reset_last_Java_frame(bool clear_fp, bool clear_pc);
1529 
1530   // Stores
1531   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
1532   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
1533 
1534   void g1_write_barrier_pre(Register obj,
1535 #ifndef _LP64
1536                             Register thread,
1537 #endif
1538                             Register tmp,
1539                             Register tmp2,
1540                             bool     tosca_live);
1541   void g1_write_barrier_post(Register store_addr,
1542                              Register new_val,
1543 #ifndef _LP64
1544                              Register thread,
1545 #endif
1546                              Register tmp,
1547                              Register tmp2);
1548 
1549 
1550   // split store_check(Register obj) to enhance instruction interleaving
1551   void store_check_part_1(Register obj);
1552   void store_check_part_2(Register obj);
1553 
1554   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
1555   void c2bool(Register x);
1556 
1557   // C++ bool manipulation
1558 
1559   void movbool(Register dst, Address src);
1560   void movbool(Address dst, bool boolconst);
1561   void movbool(Address dst, Register src);
1562   void testbool(Register dst);
1563 
1564   // oop manipulations
1565   void load_klass(Register dst, Register src);
1566   void store_klass(Register dst, Register src);
1567 
1568   void load_prototype_header(Register dst, Register src);
1569 
1570 #ifdef _LP64
1571   void store_klass_gap(Register dst, Register src);
1572 
1573   void load_heap_oop(Register dst, Address src);
1574   void store_heap_oop(Address dst, Register src);
1575   void encode_heap_oop(Register r);
1576   void decode_heap_oop(Register r);
1577   void encode_heap_oop_not_null(Register r);
1578   void decode_heap_oop_not_null(Register r);
1579   void encode_heap_oop_not_null(Register dst, Register src);
1580   void decode_heap_oop_not_null(Register dst, Register src);
1581 
1582   void set_narrow_oop(Register dst, jobject obj);
1583 
1584   // if heap base register is used - reinit it with the correct value
1585   void reinit_heapbase();
1586 #endif // _LP64
1587 
1588   // Int division/remainder for Java
1589   // (as idivl, but checks for special case as described in JVM spec.)
1590   // returns idivl instruction offset for implicit exception handling
1591   int corrected_idivl(Register reg);
1592 
1593   // Long division/remainder for Java
1594   // (as idivq, but checks for special case as described in JVM spec.)
1595   // returns idivq instruction offset for implicit exception handling
1596   int corrected_idivq(Register reg);
1597 
1598   void int3();
1599 
1600   // Long operation macros for a 32bit cpu
1601   // Long negation for Java
1602   void lneg(Register hi, Register lo);
1603 
1604   // Long multiplication for Java
1605   // (destroys contents of eax, ebx, ecx and edx)
1606   void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
1607 
1608   // Long shifts for Java
1609   // (semantics as described in JVM spec.)
1610   void lshl(Register hi, Register lo);                               // hi:lo << (rcx & 0x3f)
1611   void lshr(Register hi, Register lo, bool sign_extension = false);  // hi:lo >> (rcx & 0x3f)
1612 
1613   // Long compare for Java
1614   // (semantics as described in JVM spec.)
1615   void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
1616 
1617 
1618   // misc
1619 
1620   // Sign extension
1621   void sign_extend_short(Register reg);
1622   void sign_extend_byte(Register reg);
1623 
1624   // Division by power of 2, rounding towards 0
1625   void division_with_shift(Register reg, int shift_value);
1626 
1627   // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
1628   //
1629   // CF (corresponds to C0) if x < y
1630   // PF (corresponds to C2) if unordered
1631   // ZF (corresponds to C3) if x = y
1632   //
1633   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
1634   // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
1635   void fcmp(Register tmp);
1636   // Variant of the above which allows y to be further down the stack
1637   // and which only pops x and y if specified. If pop_right is
1638   // specified then pop_left must also be specified.
1639   void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
1640 
1641   // Floating-point comparison for Java
1642   // Compares the top-most stack entries on the FPU stack and stores the result in dst.
1643   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
1644   // (semantics as described in JVM spec.)
1645   void fcmp2int(Register dst, bool unordered_is_less);
1646   // Variant of the above which allows y to be further down the stack
1647   // and which only pops x and y if specified. If pop_right is
1648   // specified then pop_left must also be specified.
1649   void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
1650 
1651   // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
1652   // tmp is a temporary register, if none is available use noreg
1653   void fremr(Register tmp);
1654 
1655 
1656   // same as fcmp2int, but using SSE2
1657   void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
1658   void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
1659 
1660   // Inlined sin/cos generator for Java; must not use CPU instruction
1661   // directly on Intel as it does not have high enough precision
1662   // outside of the range [-pi/4, pi/4]. Extra argument indicate the
1663   // number of FPU stack slots in use; all but the topmost will
1664   // require saving if a slow case is necessary. Assumes argument is
1665   // on FP TOS; result is on FP TOS.  No cpu registers are changed by
1666   // this code.
1667   void trigfunc(char trig, int num_fpu_regs_in_use = 1);
1668 
1669   // branch to L if FPU flag C2 is set/not set
1670   // tmp is a temporary register, if none is available use noreg
1671   void jC2 (Register tmp, Label& L);
1672   void jnC2(Register tmp, Label& L);
1673 
1674   // Pop ST (ffree & fincstp combined)
1675   void fpop();
1676 
1677   // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1678   void push_fTOS();
1679 
1680   // pops double TOS element from CPU stack and pushes on FPU stack
1681   void pop_fTOS();
1682 
1683   void empty_FPU_stack();
1684 
1685   void push_IU_state();
1686   void pop_IU_state();
1687 
1688   void push_FPU_state();
1689   void pop_FPU_state();
1690 
1691   void push_CPU_state();
1692   void pop_CPU_state();
1693 
1694   // Round up to a power of two
1695   void round_to(Register reg, int modulus);
1696 
1697   // Callee saved registers handling
1698   void push_callee_saved_registers();
1699   void pop_callee_saved_registers();
1700 
1701   // allocation
1702   void eden_allocate(
1703     Register obj,                      // result: pointer to object after successful allocation
1704     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1705     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1706     Register t1,                       // temp register
1707     Label&   slow_case                 // continuation point if fast allocation fails
1708   );
1709   void tlab_allocate(
1710     Register obj,                      // result: pointer to object after successful allocation
1711     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1712     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1713     Register t1,                       // temp register
1714     Register t2,                       // temp register
1715     Label&   slow_case                 // continuation point if fast allocation fails
1716   );
1717   void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
1718 
1719   //----
1720   void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
1721 
1722   // Debugging
1723 
1724   // only if +VerifyOops
1725   void verify_oop(Register reg, const char* s = "broken oop");
1726   void verify_oop_addr(Address addr, const char * s = "broken oop addr");
1727 
1728   // only if +VerifyFPU
1729   void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
1730 
1731   // prints msg, dumps registers and stops execution
1732   void stop(const char* msg);
1733 
1734   // prints msg and continues
1735   void warn(const char* msg);
1736 
1737   static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
1738   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1739 
1740   void os_breakpoint();
1741 
1742   void untested()                                { stop("untested"); }
1743 
1744   void unimplemented(const char* what = "")      { char* b = new char[1024];  jio_snprintf(b, sizeof(b), "unimplemented: %s", what);  stop(b); }
1745 
1746   void should_not_reach_here()                   { stop("should not reach here"); }
1747 
1748   void print_CPU_state();
1749 
1750   // Stack overflow checking
1751   void bang_stack_with_offset(int offset) {
1752     // stack grows down, caller passes positive offset
1753     assert(offset > 0, "must bang with negative offset");
1754     movl(Address(rsp, (-offset)), rax);
1755   }
1756 
1757   // Writes to stack successive pages until offset reached to check for
1758   // stack overflow + shadow pages.  Also, clobbers tmp
1759   void bang_stack_size(Register size, Register tmp);
1760 
1761   // Support for serializing memory accesses between threads
1762   void serialize_memory(Register thread, Register tmp);
1763 
1764   void verify_tlab();
1765 
1766   // Biased locking support
1767   // lock_reg and obj_reg must be loaded up with the appropriate values.
1768   // swap_reg must be rax, and is killed.
1769   // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
1770   // be killed; if not supplied, push/pop will be used internally to
1771   // allocate a temporary (inefficient, avoid if possible).
1772   // Optional slow case is for implementations (interpreter and C1) which branch to
1773   // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
1774   // Returns offset of first potentially-faulting instruction for null
1775   // check info (currently consumed only by C1). If
1776   // swap_reg_contains_mark is true then returns -1 as it is assumed
1777   // the calling code has already passed any potential faults.
1778   int biased_locking_enter(Register lock_reg, Register obj_reg, Register swap_reg, Register tmp_reg,
1779                            bool swap_reg_contains_mark,
1780                            Label& done, Label* slow_case = NULL,
1781                            BiasedLockingCounters* counters = NULL);
1782   void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
1783 
1784 
1785   Condition negate_condition(Condition cond);
1786 
1787   // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
1788   // operands. In general the names are modified to avoid hiding the instruction in Assembler
1789   // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
1790   // here in MacroAssembler. The major exception to this rule is call
1791 
1792   // Arithmetics
1793 
1794 
1795   void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
1796   void addptr(Address dst, Register src);
1797 
1798   void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
1799   void addptr(Register dst, int32_t src);
1800   void addptr(Register dst, Register src);
1801 
1802   void andptr(Register dst, int32_t src);
1803   void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
1804 
1805   void cmp8(AddressLiteral src1, int imm);
1806 
1807   // renamed to drag out the casting of address to int32_t/intptr_t
1808   void cmp32(Register src1, int32_t imm);
1809 
1810   void cmp32(AddressLiteral src1, int32_t imm);
1811   // compare reg - mem, or reg - &mem
1812   void cmp32(Register src1, AddressLiteral src2);
1813 
1814   void cmp32(Register src1, Address src2);
1815 
1816 #ifndef _LP64
1817   void cmpoop(Address dst, jobject obj);
1818   void cmpoop(Register dst, jobject obj);
1819 #endif // _LP64
1820 
1821   // NOTE src2 must be the lval. This is NOT an mem-mem compare
1822   void cmpptr(Address src1, AddressLiteral src2);
1823 
1824   void cmpptr(Register src1, AddressLiteral src2);
1825 
1826   void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1827   void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1828   // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1829 
1830   void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1831   void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1832 
1833   // cmp64 to avoild hiding cmpq
1834   void cmp64(Register src1, AddressLiteral src);
1835 
1836   void cmpxchgptr(Register reg, Address adr);
1837 
1838   void locked_cmpxchgptr(Register reg, AddressLiteral adr);
1839 
1840 
1841   void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
1842 
1843 
1844   void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
1845 
1846   void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
1847 
1848   void shlptr(Register dst, int32_t shift);
1849   void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
1850 
1851   void shrptr(Register dst, int32_t shift);
1852   void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
1853 
1854   void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
1855   void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
1856 
1857   void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
1858 
1859   void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
1860   void subptr(Register dst, int32_t src);
1861   void subptr(Register dst, Register src);
1862 
1863 
1864   void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
1865   void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
1866 
1867   void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
1868   void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
1869 
1870   void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
1871 
1872 
1873 
1874   // Helper functions for statistics gathering.
1875   // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
1876   void cond_inc32(Condition cond, AddressLiteral counter_addr);
1877   // Unconditional atomic increment.
1878   void atomic_incl(AddressLiteral counter_addr);
1879 
1880   void lea(Register dst, AddressLiteral adr);
1881   void lea(Address dst, AddressLiteral adr);
1882   void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
1883 
1884   void leal32(Register dst, Address src) { leal(dst, src); }
1885 
1886   void test32(Register src1, AddressLiteral src2);
1887 
1888   void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1889   void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1890   void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1891 
1892   void testptr(Register src, int32_t imm32) {  LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
1893   void testptr(Register src1, Register src2);
1894 
1895   void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
1896   void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
1897 
1898   // Calls
1899 
1900   void call(Label& L, relocInfo::relocType rtype);
1901   void call(Register entry);
1902 
1903   // NOTE: this call tranfers to the effective address of entry NOT
1904   // the address contained by entry. This is because this is more natural
1905   // for jumps/calls.
1906   void call(AddressLiteral entry);
1907 
1908   // Jumps
1909 
1910   // NOTE: these jumps tranfer to the effective address of dst NOT
1911   // the address contained by dst. This is because this is more natural
1912   // for jumps/calls.
1913   void jump(AddressLiteral dst);
1914   void jump_cc(Condition cc, AddressLiteral dst);
1915 
1916   // 32bit can do a case table jump in one instruction but we no longer allow the base
1917   // to be installed in the Address class. This jump will tranfers to the address
1918   // contained in the location described by entry (not the address of entry)
1919   void jump(ArrayAddress entry);
1920 
1921   // Floating
1922 
1923   void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
1924   void andpd(XMMRegister dst, AddressLiteral src);
1925 
1926   void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
1927   void comiss(XMMRegister dst, AddressLiteral src);
1928 
1929   void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
1930   void comisd(XMMRegister dst, AddressLiteral src);
1931 
1932   void fldcw(Address src) { Assembler::fldcw(src); }
1933   void fldcw(AddressLiteral src);
1934 
1935   void fld_s(int index)   { Assembler::fld_s(index); }
1936   void fld_s(Address src) { Assembler::fld_s(src); }
1937   void fld_s(AddressLiteral src);
1938 
1939   void fld_d(Address src) { Assembler::fld_d(src); }
1940   void fld_d(AddressLiteral src);
1941 
1942   void fld_x(Address src) { Assembler::fld_x(src); }
1943   void fld_x(AddressLiteral src);
1944 
1945   void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
1946   void ldmxcsr(AddressLiteral src);
1947 
1948 private:
1949   // these are private because users should be doing movflt/movdbl
1950 
1951   void movss(Address dst, XMMRegister src)     { Assembler::movss(dst, src); }
1952   void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
1953   void movss(XMMRegister dst, Address src)     { Assembler::movss(dst, src); }
1954   void movss(XMMRegister dst, AddressLiteral src);
1955 
1956   void movlpd(XMMRegister dst, Address src)      {Assembler::movlpd(dst, src); }
1957   void movlpd(XMMRegister dst, AddressLiteral src);
1958 
1959 public:
1960 
1961   void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
1962   void movsd(Address dst, XMMRegister src)     { Assembler::movsd(dst, src); }
1963   void movsd(XMMRegister dst, Address src)     { Assembler::movsd(dst, src); }
1964   void movsd(XMMRegister dst, AddressLiteral src);
1965 
1966   void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
1967   void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
1968   void ucomiss(XMMRegister dst, AddressLiteral src);
1969 
1970   void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1971   void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
1972   void ucomisd(XMMRegister dst, AddressLiteral src);
1973 
1974   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1975   void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); }
1976   void xorpd(XMMRegister dst, Address src)     { Assembler::xorpd(dst, src); }
1977   void xorpd(XMMRegister dst, AddressLiteral src);
1978 
1979   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1980   void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); }
1981   void xorps(XMMRegister dst, Address src)     { Assembler::xorps(dst, src); }
1982   void xorps(XMMRegister dst, AddressLiteral src);
1983 
1984   // Data
1985 
1986   void cmov(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
1987 
1988   void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
1989   void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
1990 
1991   void movoop(Register dst, jobject obj);
1992   void movoop(Address dst, jobject obj);
1993 
1994   void movptr(ArrayAddress dst, Register src);
1995   // can this do an lea?
1996   void movptr(Register dst, ArrayAddress src);
1997 
1998   void movptr(Register dst, Address src);
1999 
2000   void movptr(Register dst, AddressLiteral src);
2001 
2002   void movptr(Register dst, intptr_t src);
2003   void movptr(Register dst, Register src);
2004   void movptr(Address dst, intptr_t src);
2005 
2006   void movptr(Address dst, Register src);
2007 
2008 #ifdef _LP64
2009   // Generally the next two are only used for moving NULL
2010   // Although there are situations in initializing the mark word where
2011   // they could be used. They are dangerous.
2012 
2013   // They only exist on LP64 so that int32_t and intptr_t are not the same
2014   // and we have ambiguous declarations.
2015 
2016   void movptr(Address dst, int32_t imm32);
2017   void movptr(Register dst, int32_t imm32);
2018 #endif // _LP64
2019 
2020   // to avoid hiding movl
2021   void mov32(AddressLiteral dst, Register src);
2022   void mov32(Register dst, AddressLiteral src);
2023 
2024   // to avoid hiding movb
2025   void movbyte(ArrayAddress dst, int src);
2026 
2027   // Can push value or effective address
2028   void pushptr(AddressLiteral src);
2029 
2030   void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
2031   void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
2032 
2033   void pushoop(jobject obj);
2034 
2035   // sign extend as need a l to ptr sized element
2036   void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
2037   void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
2038 
2039 
2040 #undef VIRTUAL
2041 
2042 };
2043 
2044 /**
2045  * class SkipIfEqual:
2046  *
2047  * Instantiating this class will result in assembly code being output that will
2048  * jump around any code emitted between the creation of the instance and it's
2049  * automatic destruction at the end of a scope block, depending on the value of
2050  * the flag passed to the constructor, which will be checked at run-time.
2051  */
2052 class SkipIfEqual {
2053  private:
2054   MacroAssembler* _masm;
2055   Label _label;
2056 
2057  public:
2058    SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
2059    ~SkipIfEqual();
2060 };
2061 
2062 #ifdef ASSERT
2063 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; }
2064 #endif