rev 522 : [mq]: meth.patch

   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 // A union type for code which has to assemble both constant and non-constant operands.
 139 class RegisterConstant VALUE_OBJ_CLASS_SPEC {
 140  private:
 141   Register _r;
 142   intptr_t _c;
 143 
 144  public:
 145   RegisterConstant(): _r(noreg), _c(0) {}
 146   RegisterConstant(Register r): _r(r), _c(0) {}
 147   RegisterConstant(intptr_t c): _r(noreg), _c(c) {}
 148 
 149   Register as_register() const { assert(is_register(),""); return _r; }
 150   intptr_t as_constant() const { assert(is_constant(),""); return _c; }
 151 
 152   Register register_or_noreg() const { return _r; }
 153   intptr_t constant_or_zero() const  { return _c; }
 154 
 155   bool is_register() const { return _r != noreg; }
 156   bool is_constant() const { return _r == noreg; }
 157 };
 158 
 159 // Address is an abstraction used to represent a memory location
 160 // using any of the amd64 addressing modes with one object.
 161 //
 162 // Note: A register location is represented via a Register, not
 163 //       via an address for efficiency & simplicity reasons.
 164 
 165 class ArrayAddress;
 166 
 167 class Address VALUE_OBJ_CLASS_SPEC {
 168  public:
 169   enum ScaleFactor {
 170     no_scale = -1,
 171     times_1  =  0,
 172     times_2  =  1,
 173     times_4  =  2,
 174     times_8  =  3,
 175     times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4)
 176   };
 177 
 178   static ScaleFactor times(int size) {
 179     assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size");
 180     if (size == 8)  return times_8;
 181     if (size == 4)  return times_4;
 182     if (size == 2)  return times_2;
 183     return times_1;
 184   }
 185 
 186  private:
 187   Register         _base;
 188   Register         _index;
 189   ScaleFactor      _scale;
 190   int              _disp;
 191   RelocationHolder _rspec;
 192 
 193   // Easily misused constructors make them private
 194   // %%% can we make these go away?
 195   NOT_LP64(Address(address loc, RelocationHolder spec);)
 196   Address(int disp, address loc, relocInfo::relocType rtype);
 197   Address(int disp, address loc, RelocationHolder spec);
 198 
 199  public:
 200 
 201  int disp() { return _disp; }
 202   // creation
 203   Address()
 204     : _base(noreg),
 205       _index(noreg),
 206       _scale(no_scale),
 207       _disp(0) {
 208   }
 209 
 210   // No default displacement otherwise Register can be implicitly
 211   // converted to 0(Register) which is quite a different animal.
 212 
 213   Address(Register base, int disp)
 214     : _base(base),
 215       _index(noreg),
 216       _scale(no_scale),
 217       _disp(disp) {
 218   }
 219 
 220   Address(Register base, Register index, ScaleFactor scale, int disp = 0)
 221     : _base (base),
 222       _index(index),
 223       _scale(scale),
 224       _disp (disp) {
 225     assert(!index->is_valid() == (scale == Address::no_scale),
 226            "inconsistent address");
 227   }
 228 
 229   Address(Register base, RegisterConstant index, ScaleFactor scale = times_1, int disp = 0)
 230     : _base (base),
 231       _index(index.register_or_noreg()),
 232       _scale(scale),
 233       _disp (disp + (index.constant_or_zero() * scale_size(scale))) {
 234     if (!index.is_register())  scale = Address::no_scale;
 235     assert(!_index->is_valid() == (scale == Address::no_scale),
 236            "inconsistent address");
 237   }
 238 
 239   Address plus_disp(int disp) {
 240     Address result = (*this);
 241     result._disp += disp;
 242     return result;
 243   }
 244 
 245   // The following two overloads are used in connection with the
 246   // ByteSize type (see sizes.hpp).  They simplify the use of
 247   // ByteSize'd arguments in assembly code. Note that their equivalent
 248   // for the optimized build are the member functions with int disp
 249   // argument since ByteSize is mapped to an int type in that case.
 250   //
 251   // Note: DO NOT introduce similar overloaded functions for WordSize
 252   // arguments as in the optimized mode, both ByteSize and WordSize
 253   // are mapped to the same type and thus the compiler cannot make a
 254   // distinction anymore (=> compiler errors).
 255 
 256 #ifdef ASSERT
 257   Address(Register base, ByteSize disp)
 258     : _base(base),
 259       _index(noreg),
 260       _scale(no_scale),
 261       _disp(in_bytes(disp)) {
 262   }
 263 
 264   Address(Register base, Register index, ScaleFactor scale, ByteSize disp)
 265     : _base(base),
 266       _index(index),
 267       _scale(scale),
 268       _disp(in_bytes(disp)) {
 269     assert(!index->is_valid() == (scale == Address::no_scale),
 270            "inconsistent address");
 271   }
 272 
 273   Address(Register base, RegisterConstant index, ScaleFactor scale, ByteSize disp)
 274     : _base (base),
 275       _index(index.register_or_noreg()),
 276       _scale(scale),
 277       _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) {
 278     if (!index.is_register())  scale = Address::no_scale;
 279     assert(!_index->is_valid() == (scale == Address::no_scale),
 280            "inconsistent address");
 281   }
 282 
 283 #endif // ASSERT
 284 
 285   // accessors
 286   bool        uses(Register reg) const { return _base == reg || _index == reg; }
 287   Register    base()             const { return _base;  }
 288   Register    index()            const { return _index; }
 289   ScaleFactor scale()            const { return _scale; }
 290   int         disp()             const { return _disp;  }
 291 
 292   // Convert the raw encoding form into the form expected by the constructor for
 293   // Address.  An index of 4 (rsp) corresponds to having no index, so convert
 294   // that to noreg for the Address constructor.
 295   static Address make_raw(int base, int index, int scale, int disp);
 296 
 297   static Address make_array(ArrayAddress);
 298 
 299   static int scale_size(ScaleFactor scale) {
 300     assert(scale != no_scale, "");
 301     assert(((1 << (int)times_1) == 1 &&
 302             (1 << (int)times_2) == 2 &&
 303             (1 << (int)times_4) == 4 &&
 304             (1 << (int)times_8) == 8), "");
 305     return (1 << (int)scale);
 306   }
 307 
 308  private:
 309   bool base_needs_rex() const {
 310     return _base != noreg && _base->encoding() >= 8;
 311   }
 312 
 313   bool index_needs_rex() const {
 314     return _index != noreg &&_index->encoding() >= 8;
 315   }
 316 
 317   relocInfo::relocType reloc() const { return _rspec.type(); }
 318 
 319   friend class Assembler;
 320   friend class MacroAssembler;
 321   friend class LIR_Assembler; // base/index/scale/disp
 322 };
 323 
 324 //
 325 // AddressLiteral has been split out from Address because operands of this type
 326 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out
 327 // the few instructions that need to deal with address literals are unique and the
 328 // MacroAssembler does not have to implement every instruction in the Assembler
 329 // in order to search for address literals that may need special handling depending
 330 // on the instruction and the platform. As small step on the way to merging i486/amd64
 331 // directories.
 332 //
 333 class AddressLiteral VALUE_OBJ_CLASS_SPEC {
 334   friend class ArrayAddress;
 335   RelocationHolder _rspec;
 336   // Typically we use AddressLiterals we want to use their rval
 337   // However in some situations we want the lval (effect address) of the item.
 338   // We provide a special factory for making those lvals.
 339   bool _is_lval;
 340 
 341   // If the target is far we'll need to load the ea of this to
 342   // a register to reach it. Otherwise if near we can do rip
 343   // relative addressing.
 344 
 345   address          _target;
 346 
 347  protected:
 348   // creation
 349   AddressLiteral()
 350     : _is_lval(false),
 351       _target(NULL)
 352   {}
 353 
 354   public:
 355 
 356 
 357   AddressLiteral(address target, relocInfo::relocType rtype);
 358 
 359   AddressLiteral(address target, RelocationHolder const& rspec)
 360     : _rspec(rspec),
 361       _is_lval(false),
 362       _target(target)
 363   {}
 364 
 365   AddressLiteral addr() {
 366     AddressLiteral ret = *this;
 367     ret._is_lval = true;
 368     return ret;
 369   }
 370 
 371 
 372  private:
 373 
 374   address target() { return _target; }
 375   bool is_lval() { return _is_lval; }
 376 
 377   relocInfo::relocType reloc() const { return _rspec.type(); }
 378   const RelocationHolder& rspec() const { return _rspec; }
 379 
 380   friend class Assembler;
 381   friend class MacroAssembler;
 382   friend class Address;
 383   friend class LIR_Assembler;
 384 };
 385 
 386 // Convience classes
 387 class RuntimeAddress: public AddressLiteral {
 388 
 389   public:
 390 
 391   RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {}
 392 
 393 };
 394 
 395 class OopAddress: public AddressLiteral {
 396 
 397   public:
 398 
 399   OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){}
 400 
 401 };
 402 
 403 class ExternalAddress: public AddressLiteral {
 404 
 405   public:
 406 
 407   ExternalAddress(address target) : AddressLiteral(target, relocInfo::external_word_type){}
 408 
 409 };
 410 
 411 class InternalAddress: public AddressLiteral {
 412 
 413   public:
 414 
 415   InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {}
 416 
 417 };
 418 
 419 // x86 can do array addressing as a single operation since disp can be an absolute
 420 // address amd64 can't. We create a class that expresses the concept but does extra
 421 // magic on amd64 to get the final result
 422 
 423 class ArrayAddress VALUE_OBJ_CLASS_SPEC {
 424   private:
 425 
 426   AddressLiteral _base;
 427   Address        _index;
 428 
 429   public:
 430 
 431   ArrayAddress() {};
 432   ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {};
 433   AddressLiteral base() { return _base; }
 434   Address index() { return _index; }
 435 
 436 };
 437 
 438 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize);
 439 
 440 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction
 441 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write
 442 // is what you get. The Assembler is generating code into a CodeBuffer.
 443 
 444 class Assembler : public AbstractAssembler  {
 445   friend class AbstractAssembler; // for the non-virtual hack
 446   friend class LIR_Assembler; // as_Address()
 447   friend class StubGenerator;
 448 
 449  public:
 450   enum Condition {                     // The x86 condition codes used for conditional jumps/moves.
 451     zero          = 0x4,
 452     notZero       = 0x5,
 453     equal         = 0x4,
 454     notEqual      = 0x5,
 455     less          = 0xc,
 456     lessEqual     = 0xe,
 457     greater       = 0xf,
 458     greaterEqual  = 0xd,
 459     below         = 0x2,
 460     belowEqual    = 0x6,
 461     above         = 0x7,
 462     aboveEqual    = 0x3,
 463     overflow      = 0x0,
 464     noOverflow    = 0x1,
 465     carrySet      = 0x2,
 466     carryClear    = 0x3,
 467     negative      = 0x8,
 468     positive      = 0x9,
 469     parity        = 0xa,
 470     noParity      = 0xb
 471   };
 472 
 473   enum Prefix {
 474     // segment overrides
 475     CS_segment = 0x2e,
 476     SS_segment = 0x36,
 477     DS_segment = 0x3e,
 478     ES_segment = 0x26,
 479     FS_segment = 0x64,
 480     GS_segment = 0x65,
 481 
 482     REX        = 0x40,
 483 
 484     REX_B      = 0x41,
 485     REX_X      = 0x42,
 486     REX_XB     = 0x43,
 487     REX_R      = 0x44,
 488     REX_RB     = 0x45,
 489     REX_RX     = 0x46,
 490     REX_RXB    = 0x47,
 491 
 492     REX_W      = 0x48,
 493 
 494     REX_WB     = 0x49,
 495     REX_WX     = 0x4A,
 496     REX_WXB    = 0x4B,
 497     REX_WR     = 0x4C,
 498     REX_WRB    = 0x4D,
 499     REX_WRX    = 0x4E,
 500     REX_WRXB   = 0x4F
 501   };
 502 
 503   enum WhichOperand {
 504     // input to locate_operand, and format code for relocations
 505     imm_operand  = 0,            // embedded 32-bit|64-bit immediate operand
 506     disp32_operand = 1,          // embedded 32-bit displacement or address
 507     call32_operand = 2,          // embedded 32-bit self-relative displacement
 508 #ifndef _LP64
 509     _WhichOperand_limit = 3
 510 #else
 511      narrow_oop_operand = 3,     // embedded 32-bit immediate narrow oop
 512     _WhichOperand_limit = 4
 513 #endif
 514   };
 515 
 516 
 517 
 518   // NOTE: The general philopsophy of the declarations here is that 64bit versions
 519   // of instructions are freely declared without the need for wrapping them an ifdef.
 520   // (Some dangerous instructions are ifdef's out of inappropriate jvm's.)
 521   // In the .cpp file the implementations are wrapped so that they are dropped out
 522   // of the resulting jvm. This is done mostly to keep the footprint of KERNEL
 523   // to the size it was prior to merging up the 32bit and 64bit assemblers.
 524   //
 525   // This does mean you'll get a linker/runtime error if you use a 64bit only instruction
 526   // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down.
 527 
 528 private:
 529 
 530 
 531   // 64bit prefixes
 532   int prefix_and_encode(int reg_enc, bool byteinst = false);
 533   int prefixq_and_encode(int reg_enc);
 534 
 535   int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false);
 536   int prefixq_and_encode(int dst_enc, int src_enc);
 537 
 538   void prefix(Register reg);
 539   void prefix(Address adr);
 540   void prefixq(Address adr);
 541 
 542   void prefix(Address adr, Register reg,  bool byteinst = false);
 543   void prefixq(Address adr, Register reg);
 544 
 545   void prefix(Address adr, XMMRegister reg);
 546 
 547   void prefetch_prefix(Address src);
 548 
 549   // Helper functions for groups of instructions
 550   void emit_arith_b(int op1, int op2, Register dst, int imm8);
 551 
 552   void emit_arith(int op1, int op2, Register dst, int32_t imm32);
 553   // only 32bit??
 554   void emit_arith(int op1, int op2, Register dst, jobject obj);
 555   void emit_arith(int op1, int op2, Register dst, Register src);
 556 
 557   void emit_operand(Register reg,
 558                     Register base, Register index, Address::ScaleFactor scale,
 559                     int disp,
 560                     RelocationHolder const& rspec,
 561                     int rip_relative_correction = 0);
 562 
 563   void emit_operand(Register reg, Address adr, int rip_relative_correction = 0);
 564 
 565   // operands that only take the original 32bit registers
 566   void emit_operand32(Register reg, Address adr);
 567 
 568   void emit_operand(XMMRegister reg,
 569                     Register base, Register index, Address::ScaleFactor scale,
 570                     int disp,
 571                     RelocationHolder const& rspec);
 572 
 573   void emit_operand(XMMRegister reg, Address adr);
 574 
 575   void emit_operand(MMXRegister reg, Address adr);
 576 
 577   // workaround gcc (3.2.1-7) bug
 578   void emit_operand(Address adr, MMXRegister reg);
 579 
 580 
 581   // Immediate-to-memory forms
 582   void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32);
 583 
 584   void emit_farith(int b1, int b2, int i);
 585 
 586 
 587  protected:
 588   #ifdef ASSERT
 589   void check_relocation(RelocationHolder const& rspec, int format);
 590   #endif
 591 
 592   inline void emit_long64(jlong x);
 593 
 594   void emit_data(jint data, relocInfo::relocType    rtype, int format);
 595   void emit_data(jint data, RelocationHolder const& rspec, int format);
 596   void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0);
 597   void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0);
 598 
 599 
 600   bool reachable(AddressLiteral adr) NOT_LP64({ return true;});
 601 
 602   // These are all easily abused and hence protected
 603 
 604   void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec, int format = 0);
 605 
 606   // 32BIT ONLY SECTION
 607 #ifndef _LP64
 608   // Make these disappear in 64bit mode since they would never be correct
 609   void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec);   // 32BIT ONLY
 610   void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec);    // 32BIT ONLY
 611 
 612   void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec);     // 32BIT ONLY
 613 
 614   void push_literal32(int32_t imm32, RelocationHolder const& rspec);                 // 32BIT ONLY
 615 #else
 616   // 64BIT ONLY SECTION
 617   void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec);   // 64BIT ONLY
 618 #endif // _LP64
 619 
 620   // These are unique in that we are ensured by the caller that the 32bit
 621   // relative in these instructions will always be able to reach the potentially
 622   // 64bit address described by entry. Since they can take a 64bit address they
 623   // don't have the 32 suffix like the other instructions in this class.
 624 
 625   void call_literal(address entry, RelocationHolder const& rspec);
 626   void jmp_literal(address entry, RelocationHolder const& rspec);
 627 
 628   // Avoid using directly section
 629   // Instructions in this section are actually usable by anyone without danger
 630   // of failure but have performance issues that are addressed my enhanced
 631   // instructions which will do the proper thing base on the particular cpu.
 632   // We protect them because we don't trust you...
 633 
 634   // Don't use next inc() and dec() methods directly. INC & DEC instructions
 635   // could cause a partial flag stall since they don't set CF flag.
 636   // Use MacroAssembler::decrement() & MacroAssembler::increment() methods
 637   // which call inc() & dec() or add() & sub() in accordance with
 638   // the product flag UseIncDec value.
 639 
 640   void decl(Register dst);
 641   void decl(Address dst);
 642   void decq(Register dst);
 643   void decq(Address dst);
 644 
 645   void incl(Register dst);
 646   void incl(Address dst);
 647   void incq(Register dst);
 648   void incq(Address dst);
 649 
 650   // New cpus require use of movsd and movss to avoid partial register stall
 651   // when loading from memory. But for old Opteron use movlpd instead of movsd.
 652   // The selection is done in MacroAssembler::movdbl() and movflt().
 653 
 654   // Move Scalar Single-Precision Floating-Point Values
 655   void movss(XMMRegister dst, Address src);
 656   void movss(XMMRegister dst, XMMRegister src);
 657   void movss(Address dst, XMMRegister src);
 658 
 659   // Move Scalar Double-Precision Floating-Point Values
 660   void movsd(XMMRegister dst, Address src);
 661   void movsd(XMMRegister dst, XMMRegister src);
 662   void movsd(Address dst, XMMRegister src);
 663   void movlpd(XMMRegister dst, Address src);
 664 
 665   // New cpus require use of movaps and movapd to avoid partial register stall
 666   // when moving between registers.
 667   void movaps(XMMRegister dst, XMMRegister src);
 668   void movapd(XMMRegister dst, XMMRegister src);
 669 
 670   // End avoid using directly
 671 
 672 
 673   // Instruction prefixes
 674   void prefix(Prefix p);
 675 
 676   public:
 677 
 678   // Creation
 679   Assembler(CodeBuffer* code) : AbstractAssembler(code) {}
 680 
 681   // Decoding
 682   static address locate_operand(address inst, WhichOperand which);
 683   static address locate_next_instruction(address inst);
 684 
 685   // Utilities
 686 
 687 #ifdef _LP64
 688  static bool is_simm(int64_t x, int nbits) { return -( CONST64(1) << (nbits-1) )  <= x   &&   x  <  ( CONST64(1) << (nbits-1) ); }
 689  static bool is_simm32(int64_t x) { return x == (int64_t)(int32_t)x; }
 690 #else
 691  static bool is_simm(int32_t x, int nbits) { return -( 1 << (nbits-1) )  <= x   &&   x  <  ( 1 << (nbits-1) ); }
 692  static bool is_simm32(int32_t x) { return true; }
 693 #endif // LP64
 694 
 695   // Generic instructions
 696   // Does 32bit or 64bit as needed for the platform. In some sense these
 697   // belong in macro assembler but there is no need for both varieties to exist
 698 
 699   void lea(Register dst, Address src);
 700 
 701   void mov(Register dst, Register src);
 702 
 703   void pusha();
 704   void popa();
 705 
 706   void pushf();
 707   void popf();
 708 
 709   void push(int32_t imm32);
 710 
 711   void push(Register src);
 712 
 713   void pop(Register dst);
 714 
 715   // These are dummies to prevent surprise implicit conversions to Register
 716   void push(void* v);
 717   void pop(void* v);
 718 
 719 
 720   // These do register sized moves/scans
 721   void rep_mov();
 722   void rep_set();
 723   void repne_scan();
 724 #ifdef _LP64
 725   void repne_scanl();
 726 #endif
 727 
 728   // Vanilla instructions in lexical order
 729 
 730   void adcl(Register dst, int32_t imm32);
 731   void adcl(Register dst, Address src);
 732   void adcl(Register dst, Register src);
 733 
 734   void adcq(Register dst, int32_t imm32);
 735   void adcq(Register dst, Address src);
 736   void adcq(Register dst, Register src);
 737 
 738 
 739   void addl(Address dst, int32_t imm32);
 740   void addl(Address dst, Register src);
 741   void addl(Register dst, int32_t imm32);
 742   void addl(Register dst, Address src);
 743   void addl(Register dst, Register src);
 744 
 745   void addq(Address dst, int32_t imm32);
 746   void addq(Address dst, Register src);
 747   void addq(Register dst, int32_t imm32);
 748   void addq(Register dst, Address src);
 749   void addq(Register dst, Register src);
 750 
 751 
 752   void addr_nop_4();
 753   void addr_nop_5();
 754   void addr_nop_7();
 755   void addr_nop_8();
 756 
 757   // Add Scalar Double-Precision Floating-Point Values
 758   void addsd(XMMRegister dst, Address src);
 759   void addsd(XMMRegister dst, XMMRegister src);
 760 
 761   // Add Scalar Single-Precision Floating-Point Values
 762   void addss(XMMRegister dst, Address src);
 763   void addss(XMMRegister dst, XMMRegister src);
 764 
 765   void andl(Register dst, int32_t imm32);
 766   void andl(Register dst, Address src);
 767   void andl(Register dst, Register src);
 768 
 769   void andq(Register dst, int32_t imm32);
 770   void andq(Register dst, Address src);
 771   void andq(Register dst, Register src);
 772 
 773 
 774   // Bitwise Logical AND of Packed Double-Precision Floating-Point Values
 775   void andpd(XMMRegister dst, Address src);
 776   void andpd(XMMRegister dst, XMMRegister src);
 777 
 778   void bswapl(Register reg);
 779 
 780   void bswapq(Register reg);
 781 
 782   void call(Label& L, relocInfo::relocType rtype);
 783   void call(Register reg);  // push pc; pc <- reg
 784   void call(Address adr);   // push pc; pc <- adr
 785 
 786   void cdql();
 787 
 788   void cdqq();
 789 
 790   void cld() { emit_byte(0xfc); }
 791 
 792   void clflush(Address adr);
 793 
 794   void cmovl(Condition cc, Register dst, Register src);
 795   void cmovl(Condition cc, Register dst, Address src);
 796 
 797   void cmovq(Condition cc, Register dst, Register src);
 798   void cmovq(Condition cc, Register dst, Address src);
 799 
 800 
 801   void cmpb(Address dst, int imm8);
 802 
 803   void cmpl(Address dst, int32_t imm32);
 804 
 805   void cmpl(Register dst, int32_t imm32);
 806   void cmpl(Register dst, Register src);
 807   void cmpl(Register dst, Address src);
 808 
 809   void cmpq(Address dst, int32_t imm32);
 810   void cmpq(Address dst, Register src);
 811 
 812   void cmpq(Register dst, int32_t imm32);
 813   void cmpq(Register dst, Register src);
 814   void cmpq(Register dst, Address src);
 815 
 816   // these are dummies used to catch attempting to convert NULL to Register
 817   void cmpl(Register dst, void* junk); // dummy
 818   void cmpq(Register dst, void* junk); // dummy
 819 
 820   void cmpw(Address dst, int imm16);
 821 
 822   void cmpxchg8 (Address adr);
 823 
 824   void cmpxchgl(Register reg, Address adr);
 825 
 826   void cmpxchgq(Register reg, Address adr);
 827 
 828   // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
 829   void comisd(XMMRegister dst, Address src);
 830 
 831   // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
 832   void comiss(XMMRegister dst, Address src);
 833 
 834   // Identify processor type and features
 835   void cpuid() {
 836     emit_byte(0x0F);
 837     emit_byte(0xA2);
 838   }
 839 
 840   // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value
 841   void cvtsd2ss(XMMRegister dst, XMMRegister src);
 842 
 843   // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value
 844   void cvtsi2sdl(XMMRegister dst, Register src);
 845   void cvtsi2sdq(XMMRegister dst, Register src);
 846 
 847   // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value
 848   void cvtsi2ssl(XMMRegister dst, Register src);
 849   void cvtsi2ssq(XMMRegister dst, Register src);
 850 
 851   // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value
 852   void cvtdq2pd(XMMRegister dst, XMMRegister src);
 853 
 854   // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value
 855   void cvtdq2ps(XMMRegister dst, XMMRegister src);
 856 
 857   // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value
 858   void cvtss2sd(XMMRegister dst, XMMRegister src);
 859 
 860   // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer
 861   void cvttsd2sil(Register dst, Address src);
 862   void cvttsd2sil(Register dst, XMMRegister src);
 863   void cvttsd2siq(Register dst, XMMRegister src);
 864 
 865   // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer
 866   void cvttss2sil(Register dst, XMMRegister src);
 867   void cvttss2siq(Register dst, XMMRegister src);
 868 
 869   // Divide Scalar Double-Precision Floating-Point Values
 870   void divsd(XMMRegister dst, Address src);
 871   void divsd(XMMRegister dst, XMMRegister src);
 872 
 873   // Divide Scalar Single-Precision Floating-Point Values
 874   void divss(XMMRegister dst, Address src);
 875   void divss(XMMRegister dst, XMMRegister src);
 876 
 877   void emms();
 878 
 879   void fabs();
 880 
 881   void fadd(int i);
 882 
 883   void fadd_d(Address src);
 884   void fadd_s(Address src);
 885 
 886   // "Alternate" versions of x87 instructions place result down in FPU
 887   // stack instead of on TOS
 888 
 889   void fadda(int i); // "alternate" fadd
 890   void faddp(int i = 1);
 891 
 892   void fchs();
 893 
 894   void fcom(int i);
 895 
 896   void fcomp(int i = 1);
 897   void fcomp_d(Address src);
 898   void fcomp_s(Address src);
 899 
 900   void fcompp();
 901 
 902   void fcos();
 903 
 904   void fdecstp();
 905 
 906   void fdiv(int i);
 907   void fdiv_d(Address src);
 908   void fdivr_s(Address src);
 909   void fdiva(int i);  // "alternate" fdiv
 910   void fdivp(int i = 1);
 911 
 912   void fdivr(int i);
 913   void fdivr_d(Address src);
 914   void fdiv_s(Address src);
 915 
 916   void fdivra(int i); // "alternate" reversed fdiv
 917 
 918   void fdivrp(int i = 1);
 919 
 920   void ffree(int i = 0);
 921 
 922   void fild_d(Address adr);
 923   void fild_s(Address adr);
 924 
 925   void fincstp();
 926 
 927   void finit();
 928 
 929   void fist_s (Address adr);
 930   void fistp_d(Address adr);
 931   void fistp_s(Address adr);
 932 
 933   void fld1();
 934 
 935   void fld_d(Address adr);
 936   void fld_s(Address adr);
 937   void fld_s(int index);
 938   void fld_x(Address adr);  // extended-precision (80-bit) format
 939 
 940   void fldcw(Address src);
 941 
 942   void fldenv(Address src);
 943 
 944   void fldlg2();
 945 
 946   void fldln2();
 947 
 948   void fldz();
 949 
 950   void flog();
 951   void flog10();
 952 
 953   void fmul(int i);
 954 
 955   void fmul_d(Address src);
 956   void fmul_s(Address src);
 957 
 958   void fmula(int i);  // "alternate" fmul
 959 
 960   void fmulp(int i = 1);
 961 
 962   void fnsave(Address dst);
 963 
 964   void fnstcw(Address src);
 965 
 966   void fnstsw_ax();
 967 
 968   void fprem();
 969   void fprem1();
 970 
 971   void frstor(Address src);
 972 
 973   void fsin();
 974 
 975   void fsqrt();
 976 
 977   void fst_d(Address adr);
 978   void fst_s(Address adr);
 979 
 980   void fstp_d(Address adr);
 981   void fstp_d(int index);
 982   void fstp_s(Address adr);
 983   void fstp_x(Address adr); // extended-precision (80-bit) format
 984 
 985   void fsub(int i);
 986   void fsub_d(Address src);
 987   void fsub_s(Address src);
 988 
 989   void fsuba(int i);  // "alternate" fsub
 990 
 991   void fsubp(int i = 1);
 992 
 993   void fsubr(int i);
 994   void fsubr_d(Address src);
 995   void fsubr_s(Address src);
 996 
 997   void fsubra(int i); // "alternate" reversed fsub
 998 
 999   void fsubrp(int i = 1);
1000 
1001   void ftan();
1002 
1003   void ftst();
1004 
1005   void fucomi(int i = 1);
1006   void fucomip(int i = 1);
1007 
1008   void fwait();
1009 
1010   void fxch(int i = 1);
1011 
1012   void fxrstor(Address src);
1013 
1014   void fxsave(Address dst);
1015 
1016   void fyl2x();
1017 
1018   void hlt();
1019 
1020   void idivl(Register src);
1021 
1022   void idivq(Register src);
1023 
1024   void imull(Register dst, Register src);
1025   void imull(Register dst, Register src, int value);
1026 
1027   void imulq(Register dst, Register src);
1028   void imulq(Register dst, Register src, int value);
1029 
1030 
1031   // jcc is the generic conditional branch generator to run-
1032   // time routines, jcc is used for branches to labels. jcc
1033   // takes a branch opcode (cc) and a label (L) and generates
1034   // either a backward branch or a forward branch and links it
1035   // to the label fixup chain. Usage:
1036   //
1037   // Label L;      // unbound label
1038   // jcc(cc, L);   // forward branch to unbound label
1039   // bind(L);      // bind label to the current pc
1040   // jcc(cc, L);   // backward branch to bound label
1041   // bind(L);      // illegal: a label may be bound only once
1042   //
1043   // Note: The same Label can be used for forward and backward branches
1044   // but it may be bound only once.
1045 
1046   void jcc(Condition cc, Label& L,
1047            relocInfo::relocType rtype = relocInfo::none);
1048 
1049   // Conditional jump to a 8-bit offset to L.
1050   // WARNING: be very careful using this for forward jumps.  If the label is
1051   // not bound within an 8-bit offset of this instruction, a run-time error
1052   // will occur.
1053   void jccb(Condition cc, Label& L);
1054 
1055   void jmp(Address entry);    // pc <- entry
1056 
1057   // Label operations & relative jumps (PPUM Appendix D)
1058   void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none);   // unconditional jump to L
1059 
1060   void jmp(Register entry); // pc <- entry
1061 
1062   // Unconditional 8-bit offset jump to L.
1063   // WARNING: be very careful using this for forward jumps.  If the label is
1064   // not bound within an 8-bit offset of this instruction, a run-time error
1065   // will occur.
1066   void jmpb(Label& L);
1067 
1068   void ldmxcsr( Address src );
1069 
1070   void leal(Register dst, Address src);
1071 
1072   void leaq(Register dst, Address src);
1073 
1074   void lfence() {
1075     emit_byte(0x0F);
1076     emit_byte(0xAE);
1077     emit_byte(0xE8);
1078   }
1079 
1080   void lock();
1081 
1082   enum Membar_mask_bits {
1083     StoreStore = 1 << 3,
1084     LoadStore  = 1 << 2,
1085     StoreLoad  = 1 << 1,
1086     LoadLoad   = 1 << 0
1087   };
1088 
1089   // Serializes memory.
1090   void membar(Membar_mask_bits order_constraint) {
1091     // We only have to handle StoreLoad and LoadLoad
1092     if (order_constraint & StoreLoad) {
1093       // MFENCE subsumes LFENCE
1094       mfence();
1095     } /* [jk] not needed currently: else if (order_constraint & LoadLoad) {
1096          lfence();
1097     } */
1098   }
1099 
1100   void mfence();
1101 
1102   // Moves
1103 
1104   void mov64(Register dst, int64_t imm64);
1105 
1106   void movb(Address dst, Register src);
1107   void movb(Address dst, int imm8);
1108   void movb(Register dst, Address src);
1109 
1110   void movdl(XMMRegister dst, Register src);
1111   void movdl(Register dst, XMMRegister src);
1112 
1113   // Move Double Quadword
1114   void movdq(XMMRegister dst, Register src);
1115   void movdq(Register dst, XMMRegister src);
1116 
1117   // Move Aligned Double Quadword
1118   void movdqa(Address     dst, XMMRegister src);
1119   void movdqa(XMMRegister dst, Address src);
1120   void movdqa(XMMRegister dst, XMMRegister src);
1121 
1122   // Move Unaligned Double Quadword
1123   void movdqu(Address     dst, XMMRegister src);
1124   void movdqu(XMMRegister dst, Address src);
1125   void movdqu(XMMRegister dst, XMMRegister src);
1126 
1127   void movl(Register dst, int32_t imm32);
1128   void movl(Address dst, int32_t imm32);
1129   void movl(Register dst, Register src);
1130   void movl(Register dst, Address src);
1131   void movl(Address dst, Register src);
1132 
1133   // These dummies prevent using movl from converting a zero (like NULL) into Register
1134   // by giving the compiler two choices it can't resolve
1135 
1136   void movl(Address  dst, void* junk);
1137   void movl(Register dst, void* junk);
1138 
1139 #ifdef _LP64
1140   void movq(Register dst, Register src);
1141   void movq(Register dst, Address src);
1142   void movq(Address dst, Register src);
1143 #endif
1144 
1145   void movq(Address     dst, MMXRegister src );
1146   void movq(MMXRegister dst, Address src );
1147 
1148 #ifdef _LP64
1149   // These dummies prevent using movq from converting a zero (like NULL) into Register
1150   // by giving the compiler two choices it can't resolve
1151 
1152   void movq(Address  dst, void* dummy);
1153   void movq(Register dst, void* dummy);
1154 #endif
1155 
1156   // Move Quadword
1157   void movq(Address     dst, XMMRegister src);
1158   void movq(XMMRegister dst, Address src);
1159 
1160   void movsbl(Register dst, Address src);
1161   void movsbl(Register dst, Register src);
1162 
1163 #ifdef _LP64
1164   // Move signed 32bit immediate to 64bit extending sign
1165   void movslq(Address dst, int32_t imm64);
1166   void movslq(Register dst, int32_t imm64);
1167 
1168   void movslq(Register dst, Address src);
1169   void movslq(Register dst, Register src);
1170   void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous
1171 #endif
1172 
1173   void movswl(Register dst, Address src);
1174   void movswl(Register dst, Register src);
1175 
1176   void movw(Address dst, int imm16);
1177   void movw(Register dst, Address src);
1178   void movw(Address dst, Register src);
1179 
1180   void movzbl(Register dst, Address src);
1181   void movzbl(Register dst, Register src);
1182 
1183   void movzwl(Register dst, Address src);
1184   void movzwl(Register dst, Register src);
1185 
1186   void mull(Address src);
1187   void mull(Register src);
1188 
1189   // Multiply Scalar Double-Precision Floating-Point Values
1190   void mulsd(XMMRegister dst, Address src);
1191   void mulsd(XMMRegister dst, XMMRegister src);
1192 
1193   // Multiply Scalar Single-Precision Floating-Point Values
1194   void mulss(XMMRegister dst, Address src);
1195   void mulss(XMMRegister dst, XMMRegister src);
1196 
1197   void negl(Register dst);
1198 
1199 #ifdef _LP64
1200   void negq(Register dst);
1201 #endif
1202 
1203   void nop(int i = 1);
1204 
1205   void notl(Register dst);
1206 
1207 #ifdef _LP64
1208   void notq(Register dst);
1209 #endif
1210 
1211   void orl(Address dst, int32_t imm32);
1212   void orl(Register dst, int32_t imm32);
1213   void orl(Register dst, Address src);
1214   void orl(Register dst, Register src);
1215 
1216   void orq(Address dst, int32_t imm32);
1217   void orq(Register dst, int32_t imm32);
1218   void orq(Register dst, Address src);
1219   void orq(Register dst, Register src);
1220 
1221   void popl(Address dst);
1222 
1223 #ifdef _LP64
1224   void popq(Address dst);
1225 #endif
1226 
1227   // Prefetches (SSE, SSE2, 3DNOW only)
1228 
1229   void prefetchnta(Address src);
1230   void prefetchr(Address src);
1231   void prefetcht0(Address src);
1232   void prefetcht1(Address src);
1233   void prefetcht2(Address src);
1234   void prefetchw(Address src);
1235 
1236   // Shuffle Packed Doublewords
1237   void pshufd(XMMRegister dst, XMMRegister src, int mode);
1238   void pshufd(XMMRegister dst, Address src,     int mode);
1239 
1240   // Shuffle Packed Low Words
1241   void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1242   void pshuflw(XMMRegister dst, Address src,     int mode);
1243 
1244   // Shift Right Logical Quadword Immediate
1245   void psrlq(XMMRegister dst, int shift);
1246 
1247   // Interleave Low Bytes
1248   void punpcklbw(XMMRegister dst, XMMRegister src);
1249 
1250   void pushl(Address src);
1251 
1252   void pushq(Address src);
1253 
1254   // Xor Packed Byte Integer Values
1255   void pxor(XMMRegister dst, Address src);
1256   void pxor(XMMRegister dst, XMMRegister src);
1257 
1258   void rcll(Register dst, int imm8);
1259 
1260   void rclq(Register dst, int imm8);
1261 
1262   void ret(int imm16);
1263 
1264   void sahf();
1265 
1266   void sarl(Register dst, int imm8);
1267   void sarl(Register dst);
1268 
1269   void sarq(Register dst, int imm8);
1270   void sarq(Register dst);
1271 
1272   void sbbl(Address dst, int32_t imm32);
1273   void sbbl(Register dst, int32_t imm32);
1274   void sbbl(Register dst, Address src);
1275   void sbbl(Register dst, Register src);
1276 
1277   void sbbq(Address dst, int32_t imm32);
1278   void sbbq(Register dst, int32_t imm32);
1279   void sbbq(Register dst, Address src);
1280   void sbbq(Register dst, Register src);
1281 
1282   void setb(Condition cc, Register dst);
1283 
1284   void shldl(Register dst, Register src);
1285 
1286   void shll(Register dst, int imm8);
1287   void shll(Register dst);
1288 
1289   void shlq(Register dst, int imm8);
1290   void shlq(Register dst);
1291 
1292   void shrdl(Register dst, Register src);
1293 
1294   void shrl(Register dst, int imm8);
1295   void shrl(Register dst);
1296 
1297   void shrq(Register dst, int imm8);
1298   void shrq(Register dst);
1299 
1300   void smovl(); // QQQ generic?
1301 
1302   // Compute Square Root of Scalar Double-Precision Floating-Point Value
1303   void sqrtsd(XMMRegister dst, Address src);
1304   void sqrtsd(XMMRegister dst, XMMRegister src);
1305 
1306   void std() { emit_byte(0xfd); }
1307 
1308   void stmxcsr( Address dst );
1309 
1310   void subl(Address dst, int32_t imm32);
1311   void subl(Address dst, Register src);
1312   void subl(Register dst, int32_t imm32);
1313   void subl(Register dst, Address src);
1314   void subl(Register dst, Register src);
1315 
1316   void subq(Address dst, int32_t imm32);
1317   void subq(Address dst, Register src);
1318   void subq(Register dst, int32_t imm32);
1319   void subq(Register dst, Address src);
1320   void subq(Register dst, Register src);
1321 
1322 
1323   // Subtract Scalar Double-Precision Floating-Point Values
1324   void subsd(XMMRegister dst, Address src);
1325   void subsd(XMMRegister dst, XMMRegister src);
1326 
1327   // Subtract Scalar Single-Precision Floating-Point Values
1328   void subss(XMMRegister dst, Address src);
1329   void subss(XMMRegister dst, XMMRegister src);
1330 
1331   void testb(Register dst, int imm8);
1332 
1333   void testl(Register dst, int32_t imm32);
1334   void testl(Register dst, Register src);
1335   void testl(Register dst, Address src);
1336 
1337   void testq(Register dst, int32_t imm32);
1338   void testq(Register dst, Register src);
1339 
1340 
1341   // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
1342   void ucomisd(XMMRegister dst, Address src);
1343   void ucomisd(XMMRegister dst, XMMRegister src);
1344 
1345   // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
1346   void ucomiss(XMMRegister dst, Address src);
1347   void ucomiss(XMMRegister dst, XMMRegister src);
1348 
1349   void xaddl(Address dst, Register src);
1350 
1351   void xaddq(Address dst, Register src);
1352 
1353   void xchgl(Register reg, Address adr);
1354   void xchgl(Register dst, Register src);
1355 
1356   void xchgq(Register reg, Address adr);
1357   void xchgq(Register dst, Register src);
1358 
1359   void xorl(Register dst, int32_t imm32);
1360   void xorl(Register dst, Address src);
1361   void xorl(Register dst, Register src);
1362 
1363   void xorq(Register dst, Address src);
1364   void xorq(Register dst, Register src);
1365 
1366   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1367   void xorpd(XMMRegister dst, Address src);
1368   void xorpd(XMMRegister dst, XMMRegister src);
1369 
1370   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1371   void xorps(XMMRegister dst, Address src);
1372   void xorps(XMMRegister dst, XMMRegister src);
1373 
1374   void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0
1375 };
1376 
1377 
1378 // MacroAssembler extends Assembler by frequently used macros.
1379 //
1380 // Instructions for which a 'better' code sequence exists depending
1381 // on arguments should also go in here.
1382 
1383 class MacroAssembler: public Assembler {
1384   friend class LIR_Assembler;
1385   friend class Runtime1;      // as_Address()
1386  protected:
1387 
1388   Address as_Address(AddressLiteral adr);
1389   Address as_Address(ArrayAddress adr);
1390 
1391   // Support for VM calls
1392   //
1393   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
1394   // may customize this version by overriding it for its purposes (e.g., to save/restore
1395   // additional registers when doing a VM call).
1396 #ifdef CC_INTERP
1397   // c++ interpreter never wants to use interp_masm version of call_VM
1398   #define VIRTUAL
1399 #else
1400   #define VIRTUAL virtual
1401 #endif
1402 
1403   VIRTUAL void call_VM_leaf_base(
1404     address entry_point,               // the entry point
1405     int     number_of_arguments        // the number of arguments to pop after the call
1406   );
1407 
1408   // This is the base routine called by the different versions of call_VM. The interpreter
1409   // may customize this version by overriding it for its purposes (e.g., to save/restore
1410   // additional registers when doing a VM call).
1411   //
1412   // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base
1413   // returns the register which contains the thread upon return. If a thread register has been
1414   // specified, the return value will correspond to that register. If no last_java_sp is specified
1415   // (noreg) than rsp will be used instead.
1416   VIRTUAL void call_VM_base(           // returns the register containing the thread upon return
1417     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
1418     Register java_thread,              // the thread if computed before     ; use noreg otherwise
1419     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
1420     address  entry_point,              // the entry point
1421     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
1422     bool     check_exceptions          // whether to check for pending exceptions after return
1423   );
1424 
1425   // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
1426   // The implementation is only non-empty for the InterpreterMacroAssembler,
1427   // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
1428   virtual void check_and_handle_popframe(Register java_thread);
1429   virtual void check_and_handle_earlyret(Register java_thread);
1430 
1431   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
1432 
1433   // helpers for FPU flag access
1434   // tmp is a temporary register, if none is available use noreg
1435   void save_rax   (Register tmp);
1436   void restore_rax(Register tmp);
1437 
1438  public:
1439   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
1440 
1441   // Support for NULL-checks
1442   //
1443   // Generates code that causes a NULL OS exception if the content of reg is NULL.
1444   // If the accessed location is M[reg + offset] and the offset is known, provide the
1445   // offset. No explicit code generation is needed if the offset is within a certain
1446   // range (0 <= offset <= page_size).
1447 
1448   void null_check(Register reg, int offset = -1);
1449   static bool needs_explicit_null_check(intptr_t offset);
1450 
1451   // Required platform-specific helpers for Label::patch_instructions.
1452   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
1453   void pd_patch_instruction(address branch, address target);
1454 #ifndef PRODUCT
1455   static void pd_print_patched_instruction(address branch);
1456 #endif
1457 
1458   // The following 4 methods return the offset of the appropriate move instruction
1459 
1460   // Support for fast byte/word loading with zero extension (depending on particular CPU)
1461   int load_unsigned_byte(Register dst, Address src);
1462   int load_unsigned_word(Register dst, Address src);
1463 
1464   // Support for fast byte/word loading with sign extension (depending on particular CPU)
1465   int load_signed_byte(Register dst, Address src);
1466   int load_signed_word(Register dst, Address src);
1467 
1468   // Support for sign-extension (hi:lo = extend_sign(lo))
1469   void extend_sign(Register hi, Register lo);
1470 
1471   // Loading values by size and signed-ness
1472   void load_sized_value(Register dst, Address src, int size_in_bytes, bool is_signed);
1473 
1474   // Support for inc/dec with optimal instruction selection depending on value
1475 
1476   void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
1477   void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
1478 
1479   void decrementl(Address dst, int value = 1);
1480   void decrementl(Register reg, int value = 1);
1481 
1482   void decrementq(Register reg, int value = 1);
1483   void decrementq(Address dst, int value = 1);
1484 
1485   void incrementl(Address dst, int value = 1);
1486   void incrementl(Register reg, int value = 1);
1487 
1488   void incrementq(Register reg, int value = 1);
1489   void incrementq(Address dst, int value = 1);
1490 
1491 
1492   // Support optimal SSE move instructions.
1493   void movflt(XMMRegister dst, XMMRegister src) {
1494     if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
1495     else                       { movss (dst, src); return; }
1496   }
1497   void movflt(XMMRegister dst, Address src) { movss(dst, src); }
1498   void movflt(XMMRegister dst, AddressLiteral src);
1499   void movflt(Address dst, XMMRegister src) { movss(dst, src); }
1500 
1501   void movdbl(XMMRegister dst, XMMRegister src) {
1502     if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
1503     else                       { movsd (dst, src); return; }
1504   }
1505 
1506   void movdbl(XMMRegister dst, AddressLiteral src);
1507 
1508   void movdbl(XMMRegister dst, Address src) {
1509     if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
1510     else                         { movlpd(dst, src); return; }
1511   }
1512   void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
1513 
1514   void incrementl(AddressLiteral dst);
1515   void incrementl(ArrayAddress dst);
1516 
1517   // Alignment
1518   void align(int modulus);
1519 
1520   // Misc
1521   void fat_nop(); // 5 byte nop
1522 
1523   // Stack frame creation/removal
1524   void enter();
1525   void leave();
1526 
1527   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
1528   // The pointer will be loaded into the thread register.
1529   void get_thread(Register thread);
1530 
1531 
1532   // Support for VM calls
1533   //
1534   // It is imperative that all calls into the VM are handled via the call_VM macros.
1535   // They make sure that the stack linkage is setup correctly. call_VM's correspond
1536   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
1537 
1538 
1539   void call_VM(Register oop_result,
1540                address entry_point,
1541                bool check_exceptions = true);
1542   void call_VM(Register oop_result,
1543                address entry_point,
1544                Register arg_1,
1545                bool check_exceptions = true);
1546   void call_VM(Register oop_result,
1547                address entry_point,
1548                Register arg_1, Register arg_2,
1549                bool check_exceptions = true);
1550   void call_VM(Register oop_result,
1551                address entry_point,
1552                Register arg_1, Register arg_2, Register arg_3,
1553                bool check_exceptions = true);
1554 
1555   // Overloadings with last_Java_sp
1556   void call_VM(Register oop_result,
1557                Register last_java_sp,
1558                address entry_point,
1559                int number_of_arguments = 0,
1560                bool check_exceptions = true);
1561   void call_VM(Register oop_result,
1562                Register last_java_sp,
1563                address entry_point,
1564                Register arg_1, bool
1565                check_exceptions = true);
1566   void call_VM(Register oop_result,
1567                Register last_java_sp,
1568                address entry_point,
1569                Register arg_1, Register arg_2,
1570                bool check_exceptions = true);
1571   void call_VM(Register oop_result,
1572                Register last_java_sp,
1573                address entry_point,
1574                Register arg_1, Register arg_2, Register arg_3,
1575                bool check_exceptions = true);
1576 
1577   void call_VM_leaf(address entry_point,
1578                     int number_of_arguments = 0);
1579   void call_VM_leaf(address entry_point,
1580                     Register arg_1);
1581   void call_VM_leaf(address entry_point,
1582                     Register arg_1, Register arg_2);
1583   void call_VM_leaf(address entry_point,
1584                     Register arg_1, Register arg_2, Register arg_3);
1585 
1586   // last Java Frame (fills frame anchor)
1587   void set_last_Java_frame(Register thread,
1588                            Register last_java_sp,
1589                            Register last_java_fp,
1590                            address last_java_pc);
1591 
1592   // thread in the default location (r15_thread on 64bit)
1593   void set_last_Java_frame(Register last_java_sp,
1594                            Register last_java_fp,
1595                            address last_java_pc);
1596 
1597   void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc);
1598 
1599   // thread in the default location (r15_thread on 64bit)
1600   void reset_last_Java_frame(bool clear_fp, bool clear_pc);
1601 
1602   // Stores
1603   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
1604   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
1605 
1606   void g1_write_barrier_pre(Register obj,
1607 #ifndef _LP64
1608                             Register thread,
1609 #endif
1610                             Register tmp,
1611                             Register tmp2,
1612                             bool     tosca_live);
1613   void g1_write_barrier_post(Register store_addr,
1614                              Register new_val,
1615 #ifndef _LP64
1616                              Register thread,
1617 #endif
1618                              Register tmp,
1619                              Register tmp2);
1620 
1621 
1622   // split store_check(Register obj) to enhance instruction interleaving
1623   void store_check_part_1(Register obj);
1624   void store_check_part_2(Register obj);
1625 
1626   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
1627   void c2bool(Register x);
1628 
1629   // C++ bool manipulation
1630 
1631   void movbool(Register dst, Address src);
1632   void movbool(Address dst, bool boolconst);
1633   void movbool(Address dst, Register src);
1634   void testbool(Register dst);
1635 
1636   // oop manipulations
1637   void load_klass(Register dst, Register src);
1638   void store_klass(Register dst, Register src);
1639 
1640   void load_prototype_header(Register dst, Register src);
1641 
1642 #ifdef _LP64
1643   void store_klass_gap(Register dst, Register src);
1644 
1645   void load_heap_oop(Register dst, Address src);
1646   void store_heap_oop(Address dst, Register src);
1647   void encode_heap_oop(Register r);
1648   void decode_heap_oop(Register r);
1649   void encode_heap_oop_not_null(Register r);
1650   void decode_heap_oop_not_null(Register r);
1651   void encode_heap_oop_not_null(Register dst, Register src);
1652   void decode_heap_oop_not_null(Register dst, Register src);
1653 
1654   void set_narrow_oop(Register dst, jobject obj);
1655 
1656   // if heap base register is used - reinit it with the correct value
1657   void reinit_heapbase();
1658 #endif // _LP64
1659 
1660   // Int division/remainder for Java
1661   // (as idivl, but checks for special case as described in JVM spec.)
1662   // returns idivl instruction offset for implicit exception handling
1663   int corrected_idivl(Register reg);
1664 
1665   // Long division/remainder for Java
1666   // (as idivq, but checks for special case as described in JVM spec.)
1667   // returns idivq instruction offset for implicit exception handling
1668   int corrected_idivq(Register reg);
1669 
1670   void int3();
1671 
1672   // Long operation macros for a 32bit cpu
1673   // Long negation for Java
1674   void lneg(Register hi, Register lo);
1675 
1676   // Long multiplication for Java
1677   // (destroys contents of eax, ebx, ecx and edx)
1678   void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
1679 
1680   // Long shifts for Java
1681   // (semantics as described in JVM spec.)
1682   void lshl(Register hi, Register lo);                               // hi:lo << (rcx & 0x3f)
1683   void lshr(Register hi, Register lo, bool sign_extension = false);  // hi:lo >> (rcx & 0x3f)
1684 
1685   // Long compare for Java
1686   // (semantics as described in JVM spec.)
1687   void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
1688 
1689 
1690   // misc
1691 
1692   // Sign extension
1693   void sign_extend_short(Register reg);
1694   void sign_extend_byte(Register reg);
1695 
1696   // Division by power of 2, rounding towards 0
1697   void division_with_shift(Register reg, int shift_value);
1698 
1699   // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
1700   //
1701   // CF (corresponds to C0) if x < y
1702   // PF (corresponds to C2) if unordered
1703   // ZF (corresponds to C3) if x = y
1704   //
1705   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
1706   // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
1707   void fcmp(Register tmp);
1708   // Variant of the above which allows y to be further down the stack
1709   // and which only pops x and y if specified. If pop_right is
1710   // specified then pop_left must also be specified.
1711   void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
1712 
1713   // Floating-point comparison for Java
1714   // Compares the top-most stack entries on the FPU stack and stores the result in dst.
1715   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
1716   // (semantics as described in JVM spec.)
1717   void fcmp2int(Register dst, bool unordered_is_less);
1718   // Variant of the above which allows y to be further down the stack
1719   // and which only pops x and y if specified. If pop_right is
1720   // specified then pop_left must also be specified.
1721   void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
1722 
1723   // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
1724   // tmp is a temporary register, if none is available use noreg
1725   void fremr(Register tmp);
1726 
1727 
1728   // same as fcmp2int, but using SSE2
1729   void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
1730   void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
1731 
1732   // Inlined sin/cos generator for Java; must not use CPU instruction
1733   // directly on Intel as it does not have high enough precision
1734   // outside of the range [-pi/4, pi/4]. Extra argument indicate the
1735   // number of FPU stack slots in use; all but the topmost will
1736   // require saving if a slow case is necessary. Assumes argument is
1737   // on FP TOS; result is on FP TOS.  No cpu registers are changed by
1738   // this code.
1739   void trigfunc(char trig, int num_fpu_regs_in_use = 1);
1740 
1741   // branch to L if FPU flag C2 is set/not set
1742   // tmp is a temporary register, if none is available use noreg
1743   void jC2 (Register tmp, Label& L);
1744   void jnC2(Register tmp, Label& L);
1745 
1746   // Pop ST (ffree & fincstp combined)
1747   void fpop();
1748 
1749   // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1750   void push_fTOS();
1751 
1752   // pops double TOS element from CPU stack and pushes on FPU stack
1753   void pop_fTOS();
1754 
1755   void empty_FPU_stack();
1756 
1757   void push_IU_state();
1758   void pop_IU_state();
1759 
1760   void push_FPU_state();
1761   void pop_FPU_state();
1762 
1763   void push_CPU_state();
1764   void pop_CPU_state();
1765 
1766   // Round up to a power of two
1767   void round_to(Register reg, int modulus);
1768 
1769   // Callee saved registers handling
1770   void push_callee_saved_registers();
1771   void pop_callee_saved_registers();
1772 
1773   // allocation
1774   void eden_allocate(
1775     Register obj,                      // result: pointer to object after successful allocation
1776     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1777     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1778     Register t1,                       // temp register
1779     Label&   slow_case                 // continuation point if fast allocation fails
1780   );
1781   void tlab_allocate(
1782     Register obj,                      // result: pointer to object after successful allocation
1783     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1784     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1785     Register t1,                       // temp register
1786     Register t2,                       // temp register
1787     Label&   slow_case                 // continuation point if fast allocation fails
1788   );
1789   void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
1790 
1791   // small bootstrap problems
1792   RegisterConstant delayed_value(intptr_t* delayed_value_addr, Register tmp);
1793   RegisterConstant delayed_value(int(*value_fn)(), Register tmp) {
1794     return delayed_value(delayed_value_addr(value_fn), tmp);
1795   }
1796   RegisterConstant delayed_value(address(*value_fn)(), Register tmp) {
1797     return delayed_value((intptr_t*) delayed_value_addr(value_fn), tmp);
1798   }
1799 
1800   // interface method calling
1801   void lookup_interface_method(Register recv_klass,
1802                                Register intf_klass,
1803                                RegisterConstant itable_index,
1804                                Register method_result,
1805                                Register scan_temp,
1806                                Label& no_such_interface);
1807 
1808   // method handles (JSR 292)
1809   void check_method_handle_type(Register mtype_reg, Register mh_reg,
1810                                 Register temp_reg,
1811                                 Label& wrong_method_type);
1812   void load_method_handle_vmslots(Register vmslots_reg, Register mh_reg,
1813                                   Register temp_reg);
1814   void jump_to_method_handle_entry(Register mh_reg, Register temp_reg);
1815   Address argument_address(RegisterConstant arg_slot, int extra_slot_offset = 0);
1816 
1817 
1818   // klass type checking (falls through on failure)
1819   void check_klass_subtype(Register sub_klass,
1820                            Register super_klass,
1821                            Register temp_reg,
1822                            Label& L_success);
1823 
1824   //----
1825   void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
1826 
1827   // Debugging
1828 
1829   // only if +VerifyOops
1830   void verify_oop(Register reg, const char* s = "broken oop");
1831   void verify_oop_addr(Address addr, const char * s = "broken oop addr");
1832 
1833   // only if +VerifyFPU
1834   void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
1835 
1836   // prints msg, dumps registers and stops execution
1837   void stop(const char* msg);
1838 
1839   // prints msg and continues
1840   void warn(const char* msg);
1841 
1842   static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
1843   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1844 
1845   void os_breakpoint();
1846 
1847   void untested()                                { stop("untested"); }
1848 
1849   void unimplemented(const char* what = "")      { char* b = new char[1024];  jio_snprintf(b, sizeof(b), "unimplemented: %s", what);  stop(b); }
1850 
1851   void should_not_reach_here()                   { stop("should not reach here"); }
1852 
1853   void print_CPU_state();
1854 
1855   // Stack overflow checking
1856   void bang_stack_with_offset(int offset) {
1857     // stack grows down, caller passes positive offset
1858     assert(offset > 0, "must bang with negative offset");
1859     movl(Address(rsp, (-offset)), rax);
1860   }
1861 
1862   // Writes to stack successive pages until offset reached to check for
1863   // stack overflow + shadow pages.  Also, clobbers tmp
1864   void bang_stack_size(Register size, Register tmp);
1865 
1866   // Support for serializing memory accesses between threads
1867   void serialize_memory(Register thread, Register tmp);
1868 
1869   void verify_tlab();
1870 
1871   // Biased locking support
1872   // lock_reg and obj_reg must be loaded up with the appropriate values.
1873   // swap_reg must be rax, and is killed.
1874   // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
1875   // be killed; if not supplied, push/pop will be used internally to
1876   // allocate a temporary (inefficient, avoid if possible).
1877   // Optional slow case is for implementations (interpreter and C1) which branch to
1878   // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
1879   // Returns offset of first potentially-faulting instruction for null
1880   // check info (currently consumed only by C1). If
1881   // swap_reg_contains_mark is true then returns -1 as it is assumed
1882   // the calling code has already passed any potential faults.
1883   int biased_locking_enter(Register lock_reg, Register obj_reg,
1884                            Register swap_reg, Register tmp_reg,
1885                            bool swap_reg_contains_mark,
1886                            Label& done, Label* slow_case = NULL,
1887                            BiasedLockingCounters* counters = NULL);
1888   void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
1889 
1890 
1891   Condition negate_condition(Condition cond);
1892 
1893   // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
1894   // operands. In general the names are modified to avoid hiding the instruction in Assembler
1895   // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
1896   // here in MacroAssembler. The major exception to this rule is call
1897 
1898   // Arithmetics
1899 
1900 
1901   void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
1902   void addptr(Address dst, Register src);
1903 
1904   void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
1905   void addptr(Register dst, int32_t src);
1906   void addptr(Register dst, Register src);
1907 
1908   void andptr(Register dst, int32_t src);
1909   void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
1910 
1911   void cmp8(AddressLiteral src1, int imm);
1912 
1913   // renamed to drag out the casting of address to int32_t/intptr_t
1914   void cmp32(Register src1, int32_t imm);
1915 
1916   void cmp32(AddressLiteral src1, int32_t imm);
1917   // compare reg - mem, or reg - &mem
1918   void cmp32(Register src1, AddressLiteral src2);
1919 
1920   void cmp32(Register src1, Address src2);
1921 
1922 #ifndef _LP64
1923   void cmpoop(Address dst, jobject obj);
1924   void cmpoop(Register dst, jobject obj);
1925 #endif // _LP64
1926 
1927   // NOTE src2 must be the lval. This is NOT an mem-mem compare
1928   void cmpptr(Address src1, AddressLiteral src2);
1929 
1930   void cmpptr(Register src1, AddressLiteral src2);
1931 
1932   void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1933   void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1934   // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1935 
1936   void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1937   void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
1938 
1939   // cmp64 to avoild hiding cmpq
1940   void cmp64(Register src1, AddressLiteral src);
1941 
1942   void cmpxchgptr(Register reg, Address adr);
1943 
1944   void locked_cmpxchgptr(Register reg, AddressLiteral adr);
1945 
1946 
1947   void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
1948 
1949 
1950   void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
1951 
1952   void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
1953 
1954   void shlptr(Register dst, int32_t shift);
1955   void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
1956 
1957   void shrptr(Register dst, int32_t shift);
1958   void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
1959 
1960   void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
1961   void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
1962 
1963   void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
1964 
1965   void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
1966   void subptr(Register dst, int32_t src);
1967   void subptr(Register dst, Register src);
1968 
1969 
1970   void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
1971   void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
1972 
1973   void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
1974   void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
1975 
1976   void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
1977 
1978 
1979 
1980   // Helper functions for statistics gathering.
1981   // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
1982   void cond_inc32(Condition cond, AddressLiteral counter_addr);
1983   // Unconditional atomic increment.
1984   void atomic_incl(AddressLiteral counter_addr);
1985 
1986   void lea(Register dst, AddressLiteral adr);
1987   void lea(Address dst, AddressLiteral adr);
1988   void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
1989 
1990   void leal32(Register dst, Address src) { leal(dst, src); }
1991 
1992   void test32(Register src1, AddressLiteral src2);
1993 
1994   void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1995   void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1996   void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
1997 
1998   void testptr(Register src, int32_t imm32) {  LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
1999   void testptr(Register src1, Register src2);
2000 
2001   void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
2002   void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
2003 
2004   // Calls
2005 
2006   void call(Label& L, relocInfo::relocType rtype);
2007   void call(Register entry);
2008 
2009   // NOTE: this call tranfers to the effective address of entry NOT
2010   // the address contained by entry. This is because this is more natural
2011   // for jumps/calls.
2012   void call(AddressLiteral entry);
2013 
2014   // Jumps
2015 
2016   // NOTE: these jumps tranfer to the effective address of dst NOT
2017   // the address contained by dst. This is because this is more natural
2018   // for jumps/calls.
2019   void jump(AddressLiteral dst);
2020   void jump_cc(Condition cc, AddressLiteral dst);
2021 
2022   // 32bit can do a case table jump in one instruction but we no longer allow the base
2023   // to be installed in the Address class. This jump will tranfers to the address
2024   // contained in the location described by entry (not the address of entry)
2025   void jump(ArrayAddress entry);
2026 
2027   // Floating
2028 
2029   void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
2030   void andpd(XMMRegister dst, AddressLiteral src);
2031 
2032   void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
2033   void comiss(XMMRegister dst, AddressLiteral src);
2034 
2035   void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
2036   void comisd(XMMRegister dst, AddressLiteral src);
2037 
2038   void fldcw(Address src) { Assembler::fldcw(src); }
2039   void fldcw(AddressLiteral src);
2040 
2041   void fld_s(int index)   { Assembler::fld_s(index); }
2042   void fld_s(Address src) { Assembler::fld_s(src); }
2043   void fld_s(AddressLiteral src);
2044 
2045   void fld_d(Address src) { Assembler::fld_d(src); }
2046   void fld_d(AddressLiteral src);
2047 
2048   void fld_x(Address src) { Assembler::fld_x(src); }
2049   void fld_x(AddressLiteral src);
2050 
2051   void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
2052   void ldmxcsr(AddressLiteral src);
2053 
2054 private:
2055   // these are private because users should be doing movflt/movdbl
2056 
2057   void movss(Address dst, XMMRegister src)     { Assembler::movss(dst, src); }
2058   void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
2059   void movss(XMMRegister dst, Address src)     { Assembler::movss(dst, src); }
2060   void movss(XMMRegister dst, AddressLiteral src);
2061 
2062   void movlpd(XMMRegister dst, Address src)      {Assembler::movlpd(dst, src); }
2063   void movlpd(XMMRegister dst, AddressLiteral src);
2064 
2065 public:
2066 
2067   void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
2068   void movsd(Address dst, XMMRegister src)     { Assembler::movsd(dst, src); }
2069   void movsd(XMMRegister dst, Address src)     { Assembler::movsd(dst, src); }
2070   void movsd(XMMRegister dst, AddressLiteral src);
2071 
2072   void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
2073   void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
2074   void ucomiss(XMMRegister dst, AddressLiteral src);
2075 
2076   void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
2077   void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
2078   void ucomisd(XMMRegister dst, AddressLiteral src);
2079 
2080   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
2081   void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); }
2082   void xorpd(XMMRegister dst, Address src)     { Assembler::xorpd(dst, src); }
2083   void xorpd(XMMRegister dst, AddressLiteral src);
2084 
2085   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
2086   void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); }
2087   void xorps(XMMRegister dst, Address src)     { Assembler::xorps(dst, src); }
2088   void xorps(XMMRegister dst, AddressLiteral src);
2089 
2090   // Data
2091 
2092   void cmov(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
2093 
2094   void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
2095   void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); }
2096 
2097   void movoop(Register dst, jobject obj);
2098   void movoop(Address dst, jobject obj);
2099 
2100   void movptr(ArrayAddress dst, Register src);
2101   // can this do an lea?
2102   void movptr(Register dst, ArrayAddress src);
2103 
2104   void movptr(Register dst, Address src);
2105 
2106   void movptr(Register dst, AddressLiteral src);
2107 
2108   void movptr(Register dst, intptr_t src);
2109   void movptr(Register dst, Register src);
2110   void movptr(Address dst, intptr_t src);
2111 
2112   void movptr(Address dst, Register src);
2113 
2114 #ifdef _LP64
2115   // Generally the next two are only used for moving NULL
2116   // Although there are situations in initializing the mark word where
2117   // they could be used. They are dangerous.
2118 
2119   // They only exist on LP64 so that int32_t and intptr_t are not the same
2120   // and we have ambiguous declarations.
2121 
2122   void movptr(Address dst, int32_t imm32);
2123   void movptr(Register dst, int32_t imm32);
2124 #endif // _LP64
2125 
2126   // to avoid hiding movl
2127   void mov32(AddressLiteral dst, Register src);
2128   void mov32(Register dst, AddressLiteral src);
2129 
2130   // to avoid hiding movb
2131   void movbyte(ArrayAddress dst, int src);
2132 
2133   // Can push value or effective address
2134   void pushptr(AddressLiteral src);
2135 
2136   void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
2137   void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
2138 
2139   void pushoop(jobject obj);
2140 
2141   // sign extend as need a l to ptr sized element
2142   void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
2143   void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
2144 
2145 
2146 #undef VIRTUAL
2147 
2148 };
2149 
2150 /**
2151  * class SkipIfEqual:
2152  *
2153  * Instantiating this class will result in assembly code being output that will
2154  * jump around any code emitted between the creation of the instance and it's
2155  * automatic destruction at the end of a scope block, depending on the value of
2156  * the flag passed to the constructor, which will be checked at run-time.
2157  */
2158 class SkipIfEqual {
2159  private:
2160   MacroAssembler* _masm;
2161   Label _label;
2162 
2163  public:
2164    SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
2165    ~SkipIfEqual();
2166 };
2167 
2168 #ifdef ASSERT
2169 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; }
2170 #endif
--- EOF ---