1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP
  26 #define CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP
  27 
  28 #include "asm/assembler.hpp"
  29 #include "utilities/macros.hpp"
  30 
  31 // <sys/trap.h> promises that the system will not use traps 16-31
  32 #define ST_RESERVED_FOR_USER_0 0x10
  33 
  34 class BiasedLockingCounters;
  35 
  36 
  37 // Register aliases for parts of the system:
  38 
  39 // 64 bit values can be kept in g1-g5, o1-o5 and o7 and all 64 bits are safe
  40 // across context switches in V8+ ABI.  Of course, there are no 64 bit regs
  41 // in V8 ABI. All 64 bits are preserved in V9 ABI for all registers.
  42 
  43 // g2-g4 are scratch registers called "application globals".  Their
  44 // meaning is reserved to the "compilation system"--which means us!
  45 // They are are not supposed to be touched by ordinary C code, although
  46 // highly-optimized C code might steal them for temps.  They are safe
  47 // across thread switches, and the ABI requires that they be safe
  48 // across function calls.
  49 //
  50 // g1 and g3 are touched by more modules.  V8 allows g1 to be clobbered
  51 // across func calls, and V8+ also allows g5 to be clobbered across
  52 // func calls.  Also, g1 and g5 can get touched while doing shared
  53 // library loading.
  54 //
  55 // We must not touch g7 (it is the thread-self register) and g6 is
  56 // reserved for certain tools.  g0, of course, is always zero.
  57 //
  58 // (Sources:  SunSoft Compilers Group, thread library engineers.)
  59 
  60 // %%%% The interpreter should be revisited to reduce global scratch regs.
  61 
  62 // This global always holds the current JavaThread pointer:
  63 
  64 REGISTER_DECLARATION(Register, G2_thread , G2);
  65 REGISTER_DECLARATION(Register, G6_heapbase , G6);
  66 
  67 // The following globals are part of the Java calling convention:
  68 
  69 REGISTER_DECLARATION(Register, G5_method             , G5);
  70 REGISTER_DECLARATION(Register, G5_megamorphic_method , G5_method);
  71 REGISTER_DECLARATION(Register, G5_inline_cache_reg   , G5_method);
  72 
  73 // The following globals are used for the new C1 & interpreter calling convention:
  74 REGISTER_DECLARATION(Register, Gargs        , G4); // pointing to the last argument
  75 
  76 // This local is used to preserve G2_thread in the interpreter and in stubs:
  77 REGISTER_DECLARATION(Register, L7_thread_cache , L7);
  78 
  79 // These globals are used as scratch registers in the interpreter:
  80 
  81 REGISTER_DECLARATION(Register, Gframe_size   , G1); // SAME REG as G1_scratch
  82 REGISTER_DECLARATION(Register, G1_scratch    , G1); // also SAME
  83 REGISTER_DECLARATION(Register, G3_scratch    , G3);
  84 REGISTER_DECLARATION(Register, G4_scratch    , G4);
  85 
  86 // These globals are used as short-lived scratch registers in the compiler:
  87 
  88 REGISTER_DECLARATION(Register, Gtemp  , G5);
  89 
  90 // JSR 292 fixed register usages:
  91 REGISTER_DECLARATION(Register, G5_method_type        , G5);
  92 REGISTER_DECLARATION(Register, G3_method_handle      , G3);
  93 REGISTER_DECLARATION(Register, L7_mh_SP_save         , L7);
  94 
  95 // The compiler requires that G5_megamorphic_method is G5_inline_cache_klass,
  96 // because a single patchable "set" instruction (NativeMovConstReg,
  97 // or NativeMovConstPatching for compiler1) instruction
  98 // serves to set up either quantity, depending on whether the compiled
  99 // call site is an inline cache or is megamorphic.  See the function
 100 // CompiledIC::set_to_megamorphic.
 101 //
 102 // If a inline cache targets an interpreted method, then the
 103 // G5 register will be used twice during the call.  First,
 104 // the call site will be patched to load a compiledICHolder
 105 // into G5. (This is an ordered pair of ic_klass, method.)
 106 // The c2i adapter will first check the ic_klass, then load
 107 // G5_method with the method part of the pair just before
 108 // jumping into the interpreter.
 109 //
 110 // Note that G5_method is only the method-self for the interpreter,
 111 // and is logically unrelated to G5_megamorphic_method.
 112 //
 113 // Invariants on G2_thread (the JavaThread pointer):
 114 //  - it should not be used for any other purpose anywhere
 115 //  - it must be re-initialized by StubRoutines::call_stub()
 116 //  - it must be preserved around every use of call_VM
 117 
 118 // We can consider using g2/g3/g4 to cache more values than the
 119 // JavaThread, such as the card-marking base or perhaps pointers into
 120 // Eden.  It's something of a waste to use them as scratch temporaries,
 121 // since they are not supposed to be volatile.  (Of course, if we find
 122 // that Java doesn't benefit from application globals, then we can just
 123 // use them as ordinary temporaries.)
 124 //
 125 // Since g1 and g5 (and/or g6) are the volatile (caller-save) registers,
 126 // it makes sense to use them routinely for procedure linkage,
 127 // whenever the On registers are not applicable.  Examples:  G5_method,
 128 // G5_inline_cache_klass, and a double handful of miscellaneous compiler
 129 // stubs.  This means that compiler stubs, etc., should be kept to a
 130 // maximum of two or three G-register arguments.
 131 
 132 
 133 // stub frames
 134 
 135 REGISTER_DECLARATION(Register, Lentry_args      , L0); // pointer to args passed to callee (interpreter) not stub itself
 136 
 137 // Interpreter frames
 138 
 139 REGISTER_DECLARATION(Register, Lesp             , L0); // expression stack pointer
 140 REGISTER_DECLARATION(Register, Lbcp             , L1); // pointer to next bytecode
 141 REGISTER_DECLARATION(Register, Lmethod          , L2);
 142 REGISTER_DECLARATION(Register, Llocals          , L3);
 143 REGISTER_DECLARATION(Register, Largs            , L3); // pointer to locals for signature handler
 144                                                        // must match Llocals in asm interpreter
 145 REGISTER_DECLARATION(Register, Lmonitors        , L4);
 146 REGISTER_DECLARATION(Register, Lbyte_code       , L5);
 147 // When calling out from the interpreter we record SP so that we can remove any extra stack
 148 // space allocated during adapter transitions. This register is only live from the point
 149 // of the call until we return.
 150 REGISTER_DECLARATION(Register, Llast_SP         , L5);
 151 REGISTER_DECLARATION(Register, Lscratch         , L5);
 152 REGISTER_DECLARATION(Register, Lscratch2        , L6);
 153 REGISTER_DECLARATION(Register, LcpoolCache      , L6); // constant pool cache
 154 
 155 REGISTER_DECLARATION(Register, O5_savedSP       , O5);
 156 REGISTER_DECLARATION(Register, I5_savedSP       , I5); // Saved SP before bumping for locals.  This is simply
 157                                                        // a copy SP, so in 64-bit it's a biased value.  The bias
 158                                                        // is added and removed as needed in the frame code.
 159 REGISTER_DECLARATION(Register, IdispatchTables  , I4); // Base address of the bytecode dispatch tables
 160 REGISTER_DECLARATION(Register, IdispatchAddress , I3); // Register which saves the dispatch address for each bytecode
 161 REGISTER_DECLARATION(Register, ImethodDataPtr   , I2); // Pointer to the current method data
 162 
 163 // NOTE: Lscratch2 and LcpoolCache point to the same registers in
 164 //       the interpreter code. If Lscratch2 needs to be used for some
 165 //       purpose than LcpoolCache should be restore after that for
 166 //       the interpreter to work right
 167 // (These assignments must be compatible with L7_thread_cache; see above.)
 168 
 169 // Lbcp points into the middle of the method object.
 170 
 171 // Exception processing
 172 // These registers are passed into exception handlers.
 173 // All exception handlers require the exception object being thrown.
 174 // In addition, an nmethod's exception handler must be passed
 175 // the address of the call site within the nmethod, to allow
 176 // proper selection of the applicable catch block.
 177 // (Interpreter frames use their own bcp() for this purpose.)
 178 //
 179 // The Oissuing_pc value is not always needed.  When jumping to a
 180 // handler that is known to be interpreted, the Oissuing_pc value can be
 181 // omitted.  An actual catch block in compiled code receives (from its
 182 // nmethod's exception handler) the thrown exception in the Oexception,
 183 // but it doesn't need the Oissuing_pc.
 184 //
 185 // If an exception handler (either interpreted or compiled)
 186 // discovers there is no applicable catch block, it updates
 187 // the Oissuing_pc to the continuation PC of its own caller,
 188 // pops back to that caller's stack frame, and executes that
 189 // caller's exception handler.  Obviously, this process will
 190 // iterate until the control stack is popped back to a method
 191 // containing an applicable catch block.  A key invariant is
 192 // that the Oissuing_pc value is always a value local to
 193 // the method whose exception handler is currently executing.
 194 //
 195 // Note:  The issuing PC value is __not__ a raw return address (I7 value).
 196 // It is a "return pc", the address __following__ the call.
 197 // Raw return addresses are converted to issuing PCs by frame::pc(),
 198 // or by stubs.  Issuing PCs can be used directly with PC range tables.
 199 //
 200 REGISTER_DECLARATION(Register, Oexception  , O0); // exception being thrown
 201 REGISTER_DECLARATION(Register, Oissuing_pc , O1); // where the exception is coming from
 202 
 203 
 204 // These must occur after the declarations above
 205 #ifndef DONT_USE_REGISTER_DEFINES
 206 
 207 #define Gthread             AS_REGISTER(Register, Gthread)
 208 #define Gmethod             AS_REGISTER(Register, Gmethod)
 209 #define Gmegamorphic_method AS_REGISTER(Register, Gmegamorphic_method)
 210 #define Ginline_cache_reg   AS_REGISTER(Register, Ginline_cache_reg)
 211 #define Gargs               AS_REGISTER(Register, Gargs)
 212 #define Lthread_cache       AS_REGISTER(Register, Lthread_cache)
 213 #define Gframe_size         AS_REGISTER(Register, Gframe_size)
 214 #define Gtemp               AS_REGISTER(Register, Gtemp)
 215 
 216 #define Lesp                AS_REGISTER(Register, Lesp)
 217 #define Lbcp                AS_REGISTER(Register, Lbcp)
 218 #define Lmethod             AS_REGISTER(Register, Lmethod)
 219 #define Llocals             AS_REGISTER(Register, Llocals)
 220 #define Lmonitors           AS_REGISTER(Register, Lmonitors)
 221 #define Lbyte_code          AS_REGISTER(Register, Lbyte_code)
 222 #define Lscratch            AS_REGISTER(Register, Lscratch)
 223 #define Lscratch2           AS_REGISTER(Register, Lscratch2)
 224 #define LcpoolCache         AS_REGISTER(Register, LcpoolCache)
 225 
 226 #define Lentry_args         AS_REGISTER(Register, Lentry_args)
 227 #define I5_savedSP          AS_REGISTER(Register, I5_savedSP)
 228 #define O5_savedSP          AS_REGISTER(Register, O5_savedSP)
 229 #define IdispatchAddress    AS_REGISTER(Register, IdispatchAddress)
 230 #define ImethodDataPtr      AS_REGISTER(Register, ImethodDataPtr)
 231 #define IdispatchTables     AS_REGISTER(Register, IdispatchTables)
 232 
 233 #define Oexception          AS_REGISTER(Register, Oexception)
 234 #define Oissuing_pc         AS_REGISTER(Register, Oissuing_pc)
 235 
 236 #endif
 237 
 238 
 239 // Address is an abstraction used to represent a memory location.
 240 //
 241 // Note: A register location is represented via a Register, not
 242 //       via an address for efficiency & simplicity reasons.
 243 
 244 class Address VALUE_OBJ_CLASS_SPEC {
 245  private:
 246   Register           _base;           // Base register.
 247   RegisterOrConstant _index_or_disp;  // Index register or constant displacement.
 248   RelocationHolder   _rspec;
 249 
 250  public:
 251   Address() : _base(noreg), _index_or_disp(noreg) {}
 252 
 253   Address(Register base, RegisterOrConstant index_or_disp)
 254     : _base(base),
 255       _index_or_disp(index_or_disp) {
 256   }
 257 
 258   Address(Register base, Register index)
 259     : _base(base),
 260       _index_or_disp(index) {
 261   }
 262 
 263   Address(Register base, int disp)
 264     : _base(base),
 265       _index_or_disp(disp) {
 266   }
 267 
 268 #ifdef ASSERT
 269   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 270   Address(Register base, ByteSize disp)
 271     : _base(base),
 272       _index_or_disp(in_bytes(disp)) {
 273   }
 274 #endif
 275 
 276   // accessors
 277   Register base()             const { return _base; }
 278   Register index()            const { return _index_or_disp.as_register(); }
 279   int      disp()             const { return _index_or_disp.as_constant(); }
 280 
 281   bool     has_index()        const { return _index_or_disp.is_register(); }
 282   bool     has_disp()         const { return _index_or_disp.is_constant(); }
 283 
 284   bool     uses(Register reg) const { return base() == reg || (has_index() && index() == reg); }
 285 
 286   const relocInfo::relocType rtype() { return _rspec.type(); }
 287   const RelocationHolder&    rspec() { return _rspec; }
 288 
 289   RelocationHolder rspec(int offset) const {
 290     return offset == 0 ? _rspec : _rspec.plus(offset);
 291   }
 292 
 293   inline bool is_simm13(int offset = 0);  // check disp+offset for overflow
 294 
 295   Address plus_disp(int plusdisp) const {     // bump disp by a small amount
 296     assert(_index_or_disp.is_constant(), "must have a displacement");
 297     Address a(base(), disp() + plusdisp);
 298     return a;
 299   }
 300   bool is_same_address(Address a) const {
 301     // disregard _rspec
 302     return base() == a.base() && (has_index() ? index() == a.index() : disp() == a.disp());
 303   }
 304 
 305   Address after_save() const {
 306     Address a = (*this);
 307     a._base = a._base->after_save();
 308     return a;
 309   }
 310 
 311   Address after_restore() const {
 312     Address a = (*this);
 313     a._base = a._base->after_restore();
 314     return a;
 315   }
 316 
 317   // Convert the raw encoding form into the form expected by the
 318   // constructor for Address.
 319   static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc);
 320 
 321   friend class Assembler;
 322 };
 323 
 324 
 325 class AddressLiteral VALUE_OBJ_CLASS_SPEC {
 326  private:
 327   address          _address;
 328   RelocationHolder _rspec;
 329 
 330   RelocationHolder rspec_from_rtype(relocInfo::relocType rtype, address addr) {
 331     switch (rtype) {
 332     case relocInfo::external_word_type:
 333       return external_word_Relocation::spec(addr);
 334     case relocInfo::internal_word_type:
 335       return internal_word_Relocation::spec(addr);
 336     case relocInfo::opt_virtual_call_type:
 337       return opt_virtual_call_Relocation::spec();
 338     case relocInfo::static_call_type:
 339       return static_call_Relocation::spec();
 340     case relocInfo::runtime_call_type:
 341       return runtime_call_Relocation::spec();
 342     case relocInfo::none:
 343       return RelocationHolder();
 344     default:
 345       ShouldNotReachHere();
 346       return RelocationHolder();
 347     }
 348   }
 349 
 350  protected:
 351   // creation
 352   AddressLiteral() : _address(NULL), _rspec(NULL) {}
 353 
 354  public:
 355   AddressLiteral(address addr, RelocationHolder const& rspec)
 356     : _address(addr),
 357       _rspec(rspec) {}
 358 
 359   // Some constructors to avoid casting at the call site.
 360   AddressLiteral(jobject obj, RelocationHolder const& rspec)
 361     : _address((address) obj),
 362       _rspec(rspec) {}
 363 
 364   AddressLiteral(intptr_t value, RelocationHolder const& rspec)
 365     : _address((address) value),
 366       _rspec(rspec) {}
 367 
 368   AddressLiteral(address addr, relocInfo::relocType rtype = relocInfo::none)
 369     : _address((address) addr),
 370     _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 371 
 372   // Some constructors to avoid casting at the call site.
 373   AddressLiteral(address* addr, relocInfo::relocType rtype = relocInfo::none)
 374     : _address((address) addr),
 375     _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 376 
 377   AddressLiteral(bool* addr, relocInfo::relocType rtype = relocInfo::none)
 378     : _address((address) addr),
 379       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 380 
 381   AddressLiteral(const bool* addr, relocInfo::relocType rtype = relocInfo::none)
 382     : _address((address) addr),
 383       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 384 
 385   AddressLiteral(signed char* addr, relocInfo::relocType rtype = relocInfo::none)
 386     : _address((address) addr),
 387       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 388 
 389   AddressLiteral(int* addr, relocInfo::relocType rtype = relocInfo::none)
 390     : _address((address) addr),
 391       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 392 
 393   AddressLiteral(intptr_t addr, relocInfo::relocType rtype = relocInfo::none)
 394     : _address((address) addr),
 395       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 396 
 397   // 32-bit complains about a multiple declaration for int*.
 398   AddressLiteral(intptr_t* addr, relocInfo::relocType rtype = relocInfo::none)
 399     : _address((address) addr),
 400       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 401 
 402   AddressLiteral(Metadata* addr, relocInfo::relocType rtype = relocInfo::none)
 403     : _address((address) addr),
 404       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 405 
 406   AddressLiteral(Metadata** addr, relocInfo::relocType rtype = relocInfo::none)
 407     : _address((address) addr),
 408       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 409 
 410   AddressLiteral(float* addr, relocInfo::relocType rtype = relocInfo::none)
 411     : _address((address) addr),
 412       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 413 
 414   AddressLiteral(double* addr, relocInfo::relocType rtype = relocInfo::none)
 415     : _address((address) addr),
 416       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 417 
 418   intptr_t value() const { return (intptr_t) _address; }
 419   int      low10() const;
 420 
 421   const relocInfo::relocType rtype() const { return _rspec.type(); }
 422   const RelocationHolder&    rspec() const { return _rspec; }
 423 
 424   RelocationHolder rspec(int offset) const {
 425     return offset == 0 ? _rspec : _rspec.plus(offset);
 426   }
 427 };
 428 
 429 // Convenience classes
 430 class ExternalAddress: public AddressLiteral {
 431  private:
 432   static relocInfo::relocType reloc_for_target(address target) {
 433     // Sometimes ExternalAddress is used for values which aren't
 434     // exactly addresses, like the card table base.
 435     // external_word_type can't be used for values in the first page
 436     // so just skip the reloc in that case.
 437     return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
 438   }
 439 
 440  public:
 441   ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(          target)) {}
 442   ExternalAddress(Metadata** target) : AddressLiteral(target, reloc_for_target((address) target)) {}
 443 };
 444 
 445 inline Address RegisterImpl::address_in_saved_window() const {
 446    return (Address(SP, (sp_offset_in_saved_window() * wordSize) + STACK_BIAS));
 447 }
 448 
 449 
 450 
 451 // Argument is an abstraction used to represent an outgoing
 452 // actual argument or an incoming formal parameter, whether
 453 // it resides in memory or in a register, in a manner consistent
 454 // with the SPARC Application Binary Interface, or ABI.  This is
 455 // often referred to as the native or C calling convention.
 456 
 457 class Argument VALUE_OBJ_CLASS_SPEC {
 458  private:
 459   int _number;
 460   bool _is_in;
 461 
 462  public:
 463   enum {
 464     n_register_parameters = 6,          // only 6 registers may contain integer parameters
 465     n_float_register_parameters = 16    // Can have up to 16 floating registers
 466   };
 467 
 468   // creation
 469   Argument(int number, bool is_in) : _number(number), _is_in(is_in) {}
 470 
 471   int  number() const  { return _number;  }
 472   bool is_in()  const  { return _is_in;   }
 473   bool is_out() const  { return !is_in(); }
 474 
 475   Argument successor() const  { return Argument(number() + 1, is_in()); }
 476   Argument as_in()     const  { return Argument(number(), true ); }
 477   Argument as_out()    const  { return Argument(number(), false); }
 478 
 479   // locating register-based arguments:
 480   bool is_register() const { return _number < n_register_parameters; }
 481 
 482   // locating Floating Point register-based arguments:
 483   bool is_float_register() const { return _number < n_float_register_parameters; }
 484 
 485   FloatRegister as_float_register() const {
 486     assert(is_float_register(), "must be a register argument");
 487     return as_FloatRegister(( number() *2 ) + 1);
 488   }
 489   FloatRegister as_double_register() const {
 490     assert(is_float_register(), "must be a register argument");
 491     return as_FloatRegister(( number() *2 ));
 492   }
 493 
 494   Register as_register() const {
 495     assert(is_register(), "must be a register argument");
 496     return is_in() ? as_iRegister(number()) : as_oRegister(number());
 497   }
 498 
 499   // locating memory-based arguments
 500   Address as_address() const {
 501     assert(!is_register(), "must be a memory argument");
 502     return address_in_frame();
 503   }
 504 
 505   // When applied to a register-based argument, give the corresponding address
 506   // into the 6-word area "into which callee may store register arguments"
 507   // (This is a different place than the corresponding register-save area location.)
 508   Address address_in_frame() const;
 509 
 510   // debugging
 511   const char* name() const;
 512 
 513   friend class Assembler;
 514 };
 515 
 516 
 517 class RegistersForDebugging : public StackObj {
 518  public:
 519   intptr_t i[8], l[8], o[8], g[8];
 520   float    f[32];
 521   double   d[32];
 522 
 523   void print(outputStream* s);
 524 
 525   static int i_offset(int j) { return offset_of(RegistersForDebugging, i[j]); }
 526   static int l_offset(int j) { return offset_of(RegistersForDebugging, l[j]); }
 527   static int o_offset(int j) { return offset_of(RegistersForDebugging, o[j]); }
 528   static int g_offset(int j) { return offset_of(RegistersForDebugging, g[j]); }
 529   static int f_offset(int j) { return offset_of(RegistersForDebugging, f[j]); }
 530   static int d_offset(int j) { return offset_of(RegistersForDebugging, d[j / 2]); }
 531 
 532   // gen asm code to save regs
 533   static void save_registers(MacroAssembler* a);
 534 
 535   // restore global registers in case C code disturbed them
 536   static void restore_registers(MacroAssembler* a, Register r);
 537 };
 538 
 539 
 540 // MacroAssembler extends Assembler by a few frequently used macros.
 541 //
 542 // Most of the standard SPARC synthetic ops are defined here.
 543 // Instructions for which a 'better' code sequence exists depending
 544 // on arguments should also go in here.
 545 
 546 #define JMP2(r1, r2) jmp(r1, r2, __FILE__, __LINE__)
 547 #define JMP(r1, off) jmp(r1, off, __FILE__, __LINE__)
 548 #define JUMP(a, temp, off)     jump(a, temp, off, __FILE__, __LINE__)
 549 #define JUMPL(a, temp, d, off) jumpl(a, temp, d, off, __FILE__, __LINE__)
 550 
 551 
 552 class MacroAssembler : public Assembler {
 553   // code patchers need various routines like inv_wdisp()
 554   friend class NativeInstruction;
 555   friend class NativeGeneralJump;
 556   friend class Relocation;
 557   friend class Label;
 558 
 559  protected:
 560   static int  patched_branch(int dest_pos, int inst, int inst_pos);
 561   static int  branch_destination(int inst, int pos);
 562 
 563   // Support for VM calls
 564   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
 565   // may customize this version by overriding it for its purposes (e.g., to save/restore
 566   // additional registers when doing a VM call).
 567   virtual void call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments);
 568 
 569   //
 570   // It is imperative that all calls into the VM are handled via the call_VM macros.
 571   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 572   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 573   //
 574   // This is the base routine called by the different versions of call_VM. The interpreter
 575   // may customize this version by overriding it for its purposes (e.g., to save/restore
 576   // additional registers when doing a VM call).
 577   //
 578   // A non-volatile java_thread_cache register should be specified so
 579   // that the G2_thread value can be preserved across the call.
 580   // (If java_thread_cache is noreg, then a slow get_thread call
 581   // will re-initialize the G2_thread.) call_VM_base returns the register that contains the
 582   // thread.
 583   //
 584   // If no last_java_sp is specified (noreg) than SP will be used instead.
 585 
 586   virtual void call_VM_base(
 587     Register        oop_result,             // where an oop-result ends up if any; use noreg otherwise
 588     Register        java_thread_cache,      // the thread if computed before     ; use noreg otherwise
 589     Register        last_java_sp,           // to set up last_Java_frame in stubs; use noreg otherwise
 590     address         entry_point,            // the entry point
 591     int             number_of_arguments,    // the number of arguments (w/o thread) to pop after call
 592     bool            check_exception=true    // flag which indicates if exception should be checked
 593   );
 594 
 595  public:
 596   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 597 
 598   // This routine should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 599   // The implementation is only non-empty for the InterpreterMacroAssembler,
 600   // as only the interpreter handles and ForceEarlyReturn PopFrame requests.
 601   virtual void check_and_handle_popframe(Register scratch_reg);
 602   virtual void check_and_handle_earlyret(Register scratch_reg);
 603 
 604   // Support for NULL-checks
 605   //
 606   // Generates code that causes a NULL OS exception if the content of reg is NULL.
 607   // If the accessed location is M[reg + offset] and the offset is known, provide the
 608   // offset.  No explicit code generation is needed if the offset is within a certain
 609   // range (0 <= offset <= page_size).
 610   //
 611   // %%%%%% Currently not done for SPARC
 612 
 613   void null_check(Register reg, int offset = -1);
 614   static bool needs_explicit_null_check(intptr_t offset);
 615 
 616   // support for delayed instructions
 617   MacroAssembler* delayed() { Assembler::delayed();  return this; }
 618 
 619   // branches that use right instruction for v8 vs. v9
 620   inline void br( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 621   inline void br( Condition c, bool a, Predict p, Label& L );
 622 
 623   inline void fb( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 624   inline void fb( Condition c, bool a, Predict p, Label& L );
 625 
 626   // compares register with zero (32 bit) and branches (V9 and V8 instructions)
 627   void cmp_zero_and_br( Condition c, Register s1, Label& L, bool a = false, Predict p = pn );
 628   // Compares a pointer register with zero and branches on (not)null.
 629   // Does a test & branch on 32-bit systems and a register-branch on 64-bit.
 630   void br_null   ( Register s1, bool a, Predict p, Label& L );
 631   void br_notnull( Register s1, bool a, Predict p, Label& L );
 632 
 633   //
 634   // Compare registers and branch with nop in delay slot or cbcond without delay slot.
 635   //
 636   // ATTENTION: use these instructions with caution because cbcond instruction
 637   //            has very short distance: 512 instructions (2Kbyte).
 638 
 639   // Compare integer (32 bit) values (icc only).
 640   void cmp_and_br_short(Register s1, Register s2, Condition c, Predict p, Label& L);
 641   void cmp_and_br_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
 642   // Platform depending version for pointer compare (icc on !LP64 and xcc on LP64).
 643   void cmp_and_brx_short(Register s1, Register s2, Condition c, Predict p, Label& L);
 644   void cmp_and_brx_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
 645 
 646   // Short branch version for compares a pointer pwith zero.
 647   void br_null_short   ( Register s1, Predict p, Label& L );
 648   void br_notnull_short( Register s1, Predict p, Label& L );
 649 
 650   // unconditional short branch
 651   void ba_short(Label& L);
 652 
 653   inline void bp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 654   inline void bp( Condition c, bool a, CC cc, Predict p, Label& L );
 655 
 656   // Branch that tests xcc in LP64 and icc in !LP64
 657   inline void brx( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 658   inline void brx( Condition c, bool a, Predict p, Label& L );
 659 
 660   // unconditional branch
 661   inline void ba( Label& L );
 662 
 663   // Branch that tests fp condition codes
 664   inline void fbp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 665   inline void fbp( Condition c, bool a, CC cc, Predict p, Label& L );
 666 
 667   // get PC the best way
 668   inline int get_pc( Register d );
 669 
 670   // Sparc shorthands(pp 85, V8 manual, pp 289 V9 manual)
 671   inline void cmp(  Register s1, Register s2 );
 672   inline void cmp(  Register s1, int simm13a );
 673 
 674   inline void jmp( Register s1, Register s2 );
 675   inline void jmp( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
 676 
 677   // Check if the call target is out of wdisp30 range (relative to the code cache)
 678   static inline bool is_far_target(address d);
 679   inline void call( address d,  relocInfo::relocType rt = relocInfo::runtime_call_type );
 680   inline void call( address d,  RelocationHolder const& rspec);
 681 
 682   inline void call( Label& L,   relocInfo::relocType rt = relocInfo::runtime_call_type );
 683   inline void call( Label& L,  RelocationHolder const& rspec);
 684 
 685   inline void callr( Register s1, Register s2 );
 686   inline void callr( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
 687 
 688   // Emits nothing on V8
 689   inline void iprefetch( address d, relocInfo::relocType rt = relocInfo::none );
 690   inline void iprefetch( Label& L);
 691 
 692   inline void tst( Register s );
 693 
 694   inline void ret(  bool trace = false );
 695   inline void retl( bool trace = false );
 696 
 697   // Required platform-specific helpers for Label::patch_instructions.
 698   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 699   void pd_patch_instruction(address branch, address target);
 700 
 701   // sethi Macro handles optimizations and relocations
 702 private:
 703   void internal_sethi(const AddressLiteral& addrlit, Register d, bool ForceRelocatable);
 704 public:
 705   void sethi(const AddressLiteral& addrlit, Register d);
 706   void patchable_sethi(const AddressLiteral& addrlit, Register d);
 707 
 708   // compute the number of instructions for a sethi/set
 709   static int  insts_for_sethi( address a, bool worst_case = false );
 710   static int  worst_case_insts_for_set();
 711 
 712   // set may be either setsw or setuw (high 32 bits may be zero or sign)
 713 private:
 714   void internal_set(const AddressLiteral& al, Register d, bool ForceRelocatable);
 715   static int insts_for_internal_set(intptr_t value);
 716 public:
 717   void set(const AddressLiteral& addrlit, Register d);
 718   void set(intptr_t value, Register d);
 719   void set(address addr, Register d, RelocationHolder const& rspec);
 720   static int insts_for_set(intptr_t value) { return insts_for_internal_set(value); }
 721 
 722   void patchable_set(const AddressLiteral& addrlit, Register d);
 723   void patchable_set(intptr_t value, Register d);
 724   void set64(jlong value, Register d, Register tmp);
 725   static int insts_for_set64(jlong value);
 726 
 727   // sign-extend 32 to 64
 728   inline void signx( Register s, Register d );
 729   inline void signx( Register d );
 730 
 731   inline void not1( Register s, Register d );
 732   inline void not1( Register d );
 733 
 734   inline void neg( Register s, Register d );
 735   inline void neg( Register d );
 736 
 737   inline void cas(  Register s1, Register s2, Register d);
 738   inline void casx( Register s1, Register s2, Register d);
 739   // Functions for isolating 64 bit atomic swaps for LP64
 740   // cas_ptr will perform cas for 32 bit VM's and casx for 64 bit VM's
 741   inline void cas_ptr(  Register s1, Register s2, Register d);
 742 
 743   // Resolve a jobject or jweak
 744   void resolve_jobject(Register value, Register tmp);
 745 
 746   // Functions for isolating 64 bit shifts for LP64
 747   inline void sll_ptr( Register s1, Register s2, Register d );
 748   inline void sll_ptr( Register s1, int imm6a,   Register d );
 749   inline void sll_ptr( Register s1, RegisterOrConstant s2, Register d );
 750   inline void srl_ptr( Register s1, Register s2, Register d );
 751   inline void srl_ptr( Register s1, int imm6a,   Register d );
 752 
 753   // little-endian
 754   inline void casl(  Register s1, Register s2, Register d);
 755   inline void casxl( Register s1, Register s2, Register d);
 756 
 757   inline void inc(   Register d,  int const13 = 1 );
 758   inline void inccc( Register d,  int const13 = 1 );
 759 
 760   inline void dec(   Register d,  int const13 = 1 );
 761   inline void deccc( Register d,  int const13 = 1 );
 762 
 763   using Assembler::add;
 764   inline void add(Register s1, int simm13a, Register d, relocInfo::relocType rtype);
 765   inline void add(Register s1, int simm13a, Register d, RelocationHolder const& rspec);
 766   inline void add(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
 767   inline void add(const Address& a, Register d, int offset = 0);
 768 
 769   using Assembler::andn;
 770   inline void andn(  Register s1, RegisterOrConstant s2, Register d);
 771 
 772   inline void btst( Register s1,  Register s2 );
 773   inline void btst( int simm13a,  Register s );
 774 
 775   inline void bset( Register s1,  Register s2 );
 776   inline void bset( int simm13a,  Register s );
 777 
 778   inline void bclr( Register s1,  Register s2 );
 779   inline void bclr( int simm13a,  Register s );
 780 
 781   inline void btog( Register s1,  Register s2 );
 782   inline void btog( int simm13a,  Register s );
 783 
 784   inline void clr( Register d );
 785 
 786   inline void clrb( Register s1, Register s2);
 787   inline void clrh( Register s1, Register s2);
 788   inline void clr(  Register s1, Register s2);
 789   inline void clrx( Register s1, Register s2);
 790 
 791   inline void clrb( Register s1, int simm13a);
 792   inline void clrh( Register s1, int simm13a);
 793   inline void clr(  Register s1, int simm13a);
 794   inline void clrx( Register s1, int simm13a);
 795 
 796   // copy & clear upper word
 797   inline void clruw( Register s, Register d );
 798   // clear upper word
 799   inline void clruwu( Register d );
 800 
 801   using Assembler::ldsb;
 802   using Assembler::ldsh;
 803   using Assembler::ldsw;
 804   using Assembler::ldub;
 805   using Assembler::lduh;
 806   using Assembler::lduw;
 807   using Assembler::ldx;
 808   using Assembler::ldd;
 809 
 810 #ifdef ASSERT
 811   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 812   inline void ld(Register s1, ByteSize simm13a, Register d);
 813 #endif
 814 
 815   inline void ld(Register s1, Register s2, Register d);
 816   inline void ld(Register s1, int simm13a, Register d);
 817 
 818   inline void ldsb(const Address& a, Register d, int offset = 0);
 819   inline void ldsh(const Address& a, Register d, int offset = 0);
 820   inline void ldsw(const Address& a, Register d, int offset = 0);
 821   inline void ldub(const Address& a, Register d, int offset = 0);
 822   inline void lduh(const Address& a, Register d, int offset = 0);
 823   inline void lduw(const Address& a, Register d, int offset = 0);
 824   inline void ldx( const Address& a, Register d, int offset = 0);
 825   inline void ld(  const Address& a, Register d, int offset = 0);
 826   inline void ldd( const Address& a, Register d, int offset = 0);
 827 
 828   inline void ldub(Register s1, RegisterOrConstant s2, Register d );
 829   inline void ldsb(Register s1, RegisterOrConstant s2, Register d );
 830   inline void lduh(Register s1, RegisterOrConstant s2, Register d );
 831   inline void ldsh(Register s1, RegisterOrConstant s2, Register d );
 832   inline void lduw(Register s1, RegisterOrConstant s2, Register d );
 833   inline void ldsw(Register s1, RegisterOrConstant s2, Register d );
 834   inline void ldx( Register s1, RegisterOrConstant s2, Register d );
 835   inline void ld(  Register s1, RegisterOrConstant s2, Register d );
 836   inline void ldd( Register s1, RegisterOrConstant s2, Register d );
 837 
 838   using Assembler::ldf;
 839   inline void ldf(FloatRegisterImpl::Width w, Register s1, RegisterOrConstant s2, FloatRegister d);
 840   inline void ldf(FloatRegisterImpl::Width w, const Address& a, FloatRegister d, int offset = 0);
 841 
 842   // little-endian
 843   inline void lduwl(Register s1, Register s2, Register d);
 844   inline void ldswl(Register s1, Register s2, Register d);
 845   inline void ldxl( Register s1, Register s2, Register d);
 846   inline void ldfl(FloatRegisterImpl::Width w, Register s1, Register s2, FloatRegister d);
 847 
 848   // membar psuedo instruction.  takes into account target memory model.
 849   inline void membar( Assembler::Membar_mask_bits const7a );
 850 
 851   // returns if membar generates anything.
 852   inline bool membar_has_effect( Assembler::Membar_mask_bits const7a );
 853 
 854   // mov pseudo instructions
 855   inline void mov( Register s,  Register d);
 856 
 857   inline void mov_or_nop( Register s,  Register d);
 858 
 859   inline void mov( int simm13a, Register d);
 860 
 861   using Assembler::prefetch;
 862   inline void prefetch(const Address& a, PrefetchFcn F, int offset = 0);
 863 
 864   using Assembler::stb;
 865   using Assembler::sth;
 866   using Assembler::stw;
 867   using Assembler::stx;
 868   using Assembler::std;
 869 
 870 #ifdef ASSERT
 871   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 872   inline void st(Register d, Register s1, ByteSize simm13a);
 873 #endif
 874 
 875   inline void st(Register d, Register s1, Register s2);
 876   inline void st(Register d, Register s1, int simm13a);
 877 
 878   inline void stb(Register d, const Address& a, int offset = 0 );
 879   inline void sth(Register d, const Address& a, int offset = 0 );
 880   inline void stw(Register d, const Address& a, int offset = 0 );
 881   inline void stx(Register d, const Address& a, int offset = 0 );
 882   inline void st( Register d, const Address& a, int offset = 0 );
 883   inline void std(Register d, const Address& a, int offset = 0 );
 884 
 885   inline void stb(Register d, Register s1, RegisterOrConstant s2 );
 886   inline void sth(Register d, Register s1, RegisterOrConstant s2 );
 887   inline void stw(Register d, Register s1, RegisterOrConstant s2 );
 888   inline void stx(Register d, Register s1, RegisterOrConstant s2 );
 889   inline void std(Register d, Register s1, RegisterOrConstant s2 );
 890   inline void st( Register d, Register s1, RegisterOrConstant s2 );
 891 
 892   using Assembler::stf;
 893   inline void stf(FloatRegisterImpl::Width w, FloatRegister d, Register s1, RegisterOrConstant s2);
 894   inline void stf(FloatRegisterImpl::Width w, FloatRegister d, const Address& a, int offset = 0);
 895 
 896   // Note: offset is added to s2.
 897   using Assembler::sub;
 898   inline void sub(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
 899 
 900   using Assembler::swap;
 901   inline void swap(const Address& a, Register d, int offset = 0);
 902 
 903   // address pseudos: make these names unlike instruction names to avoid confusion
 904   inline intptr_t load_pc_address( Register reg, int bytes_to_skip );
 905   inline void load_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 906   inline void load_bool_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 907   inline void load_ptr_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 908   inline void store_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
 909   inline void store_ptr_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
 910   inline void jumpl_to(const AddressLiteral& addrlit, Register temp, Register d, int offset = 0);
 911   inline void jump_to(const AddressLiteral& addrlit, Register temp, int offset = 0);
 912   inline void jump_indirect_to(Address& a, Register temp, int ld_offset = 0, int jmp_offset = 0);
 913 
 914   // ring buffer traceable jumps
 915 
 916   void jmp2( Register r1, Register r2, const char* file, int line );
 917   void jmp ( Register r1, int offset,  const char* file, int line );
 918 
 919   void jumpl(const AddressLiteral& addrlit, Register temp, Register d, int offset, const char* file, int line);
 920   void jump (const AddressLiteral& addrlit, Register temp,             int offset, const char* file, int line);
 921 
 922 
 923   // argument pseudos:
 924 
 925   inline void load_argument( Argument& a, Register  d );
 926   inline void store_argument( Register s, Argument& a );
 927   inline void store_ptr_argument( Register s, Argument& a );
 928   inline void store_float_argument( FloatRegister s, Argument& a );
 929   inline void store_double_argument( FloatRegister s, Argument& a );
 930   inline void store_long_argument( Register s, Argument& a );
 931 
 932   // handy macros:
 933 
 934   inline void round_to( Register r, int modulus );
 935 
 936   // --------------------------------------------------
 937 
 938   // Functions for isolating 64 bit loads for LP64
 939   // ld_ptr will perform ld for 32 bit VM's and ldx for 64 bit VM's
 940   // st_ptr will perform st for 32 bit VM's and stx for 64 bit VM's
 941   inline void ld_ptr(Register s1, Register s2, Register d);
 942   inline void ld_ptr(Register s1, int simm13a, Register d);
 943   inline void ld_ptr(Register s1, RegisterOrConstant s2, Register d);
 944   inline void ld_ptr(const Address& a, Register d, int offset = 0);
 945   inline void st_ptr(Register d, Register s1, Register s2);
 946   inline void st_ptr(Register d, Register s1, int simm13a);
 947   inline void st_ptr(Register d, Register s1, RegisterOrConstant s2);
 948   inline void st_ptr(Register d, const Address& a, int offset = 0);
 949 
 950 #ifdef ASSERT
 951   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 952   inline void ld_ptr(Register s1, ByteSize simm13a, Register d);
 953   inline void st_ptr(Register d, Register s1, ByteSize simm13a);
 954 #endif
 955 
 956   // ld_long will perform ldd for 32 bit VM's and ldx for 64 bit VM's
 957   // st_long will perform std for 32 bit VM's and stx for 64 bit VM's
 958   inline void ld_long(Register s1, Register s2, Register d);
 959   inline void ld_long(Register s1, int simm13a, Register d);
 960   inline void ld_long(Register s1, RegisterOrConstant s2, Register d);
 961   inline void ld_long(const Address& a, Register d, int offset = 0);
 962   inline void st_long(Register d, Register s1, Register s2);
 963   inline void st_long(Register d, Register s1, int simm13a);
 964   inline void st_long(Register d, Register s1, RegisterOrConstant s2);
 965   inline void st_long(Register d, const Address& a, int offset = 0);
 966 
 967   // Helpers for address formation.
 968   // - They emit only a move if s2 is a constant zero.
 969   // - If dest is a constant and either s1 or s2 is a register, the temp argument is required and becomes the result.
 970   // - If dest is a register and either s1 or s2 is a non-simm13 constant, the temp argument is required and used to materialize the constant.
 971   RegisterOrConstant regcon_andn_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
 972   RegisterOrConstant regcon_inc_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
 973   RegisterOrConstant regcon_sll_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
 974 
 975   RegisterOrConstant ensure_simm13_or_reg(RegisterOrConstant src, Register temp) {
 976     if (is_simm13(src.constant_or_zero()))
 977       return src;               // register or short constant
 978     guarantee(temp != noreg, "constant offset overflow");
 979     set(src.as_constant(), temp);
 980     return temp;
 981   }
 982 
 983   // --------------------------------------------------
 984 
 985  public:
 986   // traps as per trap.h (SPARC ABI?)
 987 
 988   void breakpoint_trap();
 989   void breakpoint_trap(Condition c, CC cc);
 990 
 991   // Support for serializing memory accesses between threads
 992   void serialize_memory(Register thread, Register tmp1, Register tmp2);
 993 
 994   // Stack frame creation/removal
 995   void enter();
 996   void leave();
 997 
 998   // Manipulation of C++ bools
 999   // These are idioms to flag the need for care with accessing bools but on
1000   // this platform we assume byte size
1001 
1002   inline void stbool(Register d, const Address& a);
1003   inline void ldbool(const Address& a, Register d);
1004   inline void movbool( bool boolconst, Register d);
1005 
1006   void load_mirror(Register mirror, Register method);
1007 
1008   // klass oop manipulations if compressed
1009   void load_klass(Register src_oop, Register klass);
1010   void store_klass(Register klass, Register dst_oop);
1011   void store_klass_gap(Register s, Register dst_oop);
1012 
1013    // oop manipulations
1014   void load_heap_oop(const Address& s, Register d);
1015   void load_heap_oop(Register s1, Register s2, Register d);
1016   void load_heap_oop(Register s1, int simm13a, Register d);
1017   void load_heap_oop(Register s1, RegisterOrConstant s2, Register d);
1018   void store_heap_oop(Register d, Register s1, Register s2);
1019   void store_heap_oop(Register d, Register s1, int simm13a);
1020   void store_heap_oop(Register d, const Address& a, int offset = 0);
1021 
1022   void encode_heap_oop(Register src, Register dst);
1023   void encode_heap_oop(Register r) {
1024     encode_heap_oop(r, r);
1025   }
1026   void decode_heap_oop(Register src, Register dst);
1027   void decode_heap_oop(Register r) {
1028     decode_heap_oop(r, r);
1029   }
1030   void encode_heap_oop_not_null(Register r);
1031   void decode_heap_oop_not_null(Register r);
1032   void encode_heap_oop_not_null(Register src, Register dst);
1033   void decode_heap_oop_not_null(Register src, Register dst);
1034 
1035   void encode_klass_not_null(Register r);
1036   void decode_klass_not_null(Register r);
1037   void encode_klass_not_null(Register src, Register dst);
1038   void decode_klass_not_null(Register src, Register dst);
1039 
1040   // Support for managing the JavaThread pointer (i.e.; the reference to
1041   // thread-local information).
1042   void get_thread();                                // load G2_thread
1043   void verify_thread();                             // verify G2_thread contents
1044   void save_thread   (const Register threache); // save to cache
1045   void restore_thread(const Register thread_cache); // restore from cache
1046 
1047   // Support for last Java frame (but use call_VM instead where possible)
1048   void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
1049   void reset_last_Java_frame(void);
1050 
1051   // Call into the VM.
1052   // Passes the thread pointer (in O0) as a prepended argument.
1053   // Makes sure oop return values are visible to the GC.
1054   void call_VM(Register oop_result, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1055   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
1056   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1057   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
1058 
1059   // these overloadings are not presently used on SPARC:
1060   void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1061   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
1062   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1063   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
1064 
1065   void call_VM_leaf(Register thread_cache, address entry_point, int number_of_arguments = 0);
1066   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1);
1067   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2);
1068   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3);
1069 
1070   void get_vm_result  (Register oop_result);
1071   void get_vm_result_2(Register metadata_result);
1072 
1073   // vm result is currently getting hijacked to for oop preservation
1074   void set_vm_result(Register oop_result);
1075 
1076   // Emit the CompiledIC call idiom
1077   void ic_call(address entry, bool emit_delay = true, jint method_index = 0);
1078 
1079   // if call_VM_base was called with check_exceptions=false, then call
1080   // check_and_forward_exception to handle exceptions when it is safe
1081   void check_and_forward_exception(Register scratch_reg);
1082 
1083   // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1084   void push_fTOS();
1085 
1086   // pops double TOS element from CPU stack and pushes on FPU stack
1087   void pop_fTOS();
1088 
1089   void empty_FPU_stack();
1090 
1091   void push_IU_state();
1092   void pop_IU_state();
1093 
1094   void push_FPU_state();
1095   void pop_FPU_state();
1096 
1097   void push_CPU_state();
1098   void pop_CPU_state();
1099 
1100   // Returns the byte size of the instructions generated by decode_klass_not_null().
1101   static int instr_size_for_decode_klass_not_null();
1102 
1103   // if heap base register is used - reinit it with the correct value
1104   void reinit_heapbase();
1105 
1106   // Debugging
1107   void _verify_oop(Register reg, const char * msg, const char * file, int line);
1108   void _verify_oop_addr(Address addr, const char * msg, const char * file, int line);
1109 
1110   // TODO: verify_method and klass metadata (compare against vptr?)
1111   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1112   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1113 
1114 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
1115 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop addr ", __FILE__, __LINE__)
1116 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1117 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1118 
1119         // only if +VerifyOops
1120   void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
1121         // only if +VerifyFPU
1122   void stop(const char* msg);                          // prints msg, dumps registers and stops execution
1123   void warn(const char* msg);                          // prints msg, but don't stop
1124   void untested(const char* what = "");
1125   void unimplemented(const char* what = "")      { char* b = new char[1024];  jio_snprintf(b, 1024, "unimplemented: %s", what);  stop(b); }
1126   void should_not_reach_here()                   { stop("should not reach here"); }
1127   void print_CPU_state();
1128 
1129   // oops in code
1130   AddressLiteral allocate_oop_address(jobject obj);                          // allocate_index
1131   AddressLiteral constant_oop_address(jobject obj);                          // find_index
1132   inline void    set_oop             (jobject obj, Register d);              // uses allocate_oop_address
1133   inline void    set_oop_constant    (jobject obj, Register d);              // uses constant_oop_address
1134   inline void    set_oop             (const AddressLiteral& obj_addr, Register d); // same as load_address
1135 
1136   // metadata in code that we have to keep track of
1137   AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
1138   AddressLiteral constant_metadata_address(Metadata* obj); // find_index
1139   inline void    set_metadata             (Metadata* obj, Register d);              // uses allocate_metadata_address
1140   inline void    set_metadata_constant    (Metadata* obj, Register d);              // uses constant_metadata_address
1141   inline void    set_metadata             (const AddressLiteral& obj_addr, Register d); // same as load_address
1142 
1143   void set_narrow_oop( jobject obj, Register d );
1144   void set_narrow_klass( Klass* k, Register d );
1145 
1146   // nop padding
1147   void align(int modulus);
1148 
1149   // declare a safepoint
1150   void safepoint();
1151 
1152   // factor out part of stop into subroutine to save space
1153   void stop_subroutine();
1154   // factor out part of verify_oop into subroutine to save space
1155   void verify_oop_subroutine();
1156 
1157   // side-door communication with signalHandler in os_solaris.cpp
1158   static address _verify_oop_implicit_branch[3];
1159 
1160   int total_frame_size_in_bytes(int extraWords);
1161 
1162   // used when extraWords known statically
1163   void save_frame(int extraWords = 0);
1164   void save_frame_c1(int size_in_bytes);
1165   // make a frame, and simultaneously pass up one or two register value
1166   // into the new register window
1167   void save_frame_and_mov(int extraWords, Register s1, Register d1, Register s2 = Register(), Register d2 = Register());
1168 
1169   // give no. (outgoing) params, calc # of words will need on frame
1170   void calc_mem_param_words(Register Rparam_words, Register Rresult);
1171 
1172   // used to calculate frame size dynamically
1173   // result is in bytes and must be negated for save inst
1174   void calc_frame_size(Register extraWords, Register resultReg);
1175 
1176   // calc and also save
1177   void calc_frame_size_and_save(Register extraWords, Register resultReg);
1178 
1179   static void debug(char* msg, RegistersForDebugging* outWindow);
1180 
1181   // implementations of bytecodes used by both interpreter and compiler
1182 
1183   void lcmp( Register Ra_hi, Register Ra_low,
1184              Register Rb_hi, Register Rb_low,
1185              Register Rresult);
1186 
1187   void lneg( Register Rhi, Register Rlow );
1188 
1189   void lshl(  Register Rin_high,  Register Rin_low,  Register Rcount,
1190               Register Rout_high, Register Rout_low, Register Rtemp );
1191 
1192   void lshr(  Register Rin_high,  Register Rin_low,  Register Rcount,
1193               Register Rout_high, Register Rout_low, Register Rtemp );
1194 
1195   void lushr( Register Rin_high,  Register Rin_low,  Register Rcount,
1196               Register Rout_high, Register Rout_low, Register Rtemp );
1197 
1198   void lcmp( Register Ra, Register Rb, Register Rresult);
1199 
1200   // Load and store values by size and signed-ness
1201   void load_sized_value( Address src, Register dst, size_t size_in_bytes, bool is_signed);
1202   void store_sized_value(Register src, Address dst, size_t size_in_bytes);
1203 
1204   void float_cmp( bool is_float, int unordered_result,
1205                   FloatRegister Fa, FloatRegister Fb,
1206                   Register Rresult);
1207 
1208   void save_all_globals_into_locals();
1209   void restore_globals_from_locals();
1210 
1211   // These set the icc condition code to equal if the lock succeeded
1212   // and notEqual if it failed and requires a slow case
1213   void compiler_lock_object(Register Roop, Register Rmark, Register Rbox,
1214                             Register Rscratch,
1215                             BiasedLockingCounters* counters = NULL,
1216                             bool try_bias = UseBiasedLocking);
1217   void compiler_unlock_object(Register Roop, Register Rmark, Register Rbox,
1218                               Register Rscratch,
1219                               bool try_bias = UseBiasedLocking);
1220 
1221   // Biased locking support
1222   // Upon entry, lock_reg must point to the lock record on the stack,
1223   // obj_reg must contain the target object, and mark_reg must contain
1224   // the target object's header.
1225   // Destroys mark_reg if an attempt is made to bias an anonymously
1226   // biased lock. In this case a failure will go either to the slow
1227   // case or fall through with the notEqual condition code set with
1228   // the expectation that the slow case in the runtime will be called.
1229   // In the fall-through case where the CAS-based lock is done,
1230   // mark_reg is not destroyed.
1231   void biased_locking_enter(Register obj_reg, Register mark_reg, Register temp_reg,
1232                             Label& done, Label* slow_case = NULL,
1233                             BiasedLockingCounters* counters = NULL);
1234   // Upon entry, the base register of mark_addr must contain the oop.
1235   // Destroys temp_reg.
1236 
1237   // If allow_delay_slot_filling is set to true, the next instruction
1238   // emitted after this one will go in an annulled delay slot if the
1239   // biased locking exit case failed.
1240   void biased_locking_exit(Address mark_addr, Register temp_reg, Label& done, bool allow_delay_slot_filling = false);
1241 
1242   // allocation
1243   void eden_allocate(
1244     Register obj,                      // result: pointer to object after successful allocation
1245     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1246     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1247     Register t1,                       // temp register
1248     Register t2,                       // temp register
1249     Label&   slow_case                 // continuation point if fast allocation fails
1250   );
1251   void tlab_allocate(
1252     Register obj,                      // result: pointer to object after successful allocation
1253     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1254     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1255     Register t1,                       // temp register
1256     Label&   slow_case                 // continuation point if fast allocation fails
1257   );
1258   void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
1259   void zero_memory(Register base, Register index);
1260   void incr_allocated_bytes(RegisterOrConstant size_in_bytes,
1261                             Register t1, Register t2);
1262 
1263   // interface method calling
1264   void lookup_interface_method(Register recv_klass,
1265                                Register intf_klass,
1266                                RegisterOrConstant itable_index,
1267                                Register method_result,
1268                                Register temp_reg, Register temp2_reg,
1269                                Label& no_such_interface);
1270 
1271   // virtual method calling
1272   void lookup_virtual_method(Register recv_klass,
1273                              RegisterOrConstant vtable_index,
1274                              Register method_result);
1275 
1276   // Test sub_klass against super_klass, with fast and slow paths.
1277 
1278   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1279   // One of the three labels can be NULL, meaning take the fall-through.
1280   // If super_check_offset is -1, the value is loaded up from super_klass.
1281   // No registers are killed, except temp_reg and temp2_reg.
1282   // If super_check_offset is not -1, temp2_reg is not used and can be noreg.
1283   void check_klass_subtype_fast_path(Register sub_klass,
1284                                      Register super_klass,
1285                                      Register temp_reg,
1286                                      Register temp2_reg,
1287                                      Label* L_success,
1288                                      Label* L_failure,
1289                                      Label* L_slow_path,
1290                 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
1291 
1292   // The rest of the type check; must be wired to a corresponding fast path.
1293   // It does not repeat the fast path logic, so don't use it standalone.
1294   // The temp_reg can be noreg, if no temps are available.
1295   // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
1296   // Updates the sub's secondary super cache as necessary.
1297   void check_klass_subtype_slow_path(Register sub_klass,
1298                                      Register super_klass,
1299                                      Register temp_reg,
1300                                      Register temp2_reg,
1301                                      Register temp3_reg,
1302                                      Register temp4_reg,
1303                                      Label* L_success,
1304                                      Label* L_failure);
1305 
1306   // Simplified, combined version, good for typical uses.
1307   // Falls through on failure.
1308   void check_klass_subtype(Register sub_klass,
1309                            Register super_klass,
1310                            Register temp_reg,
1311                            Register temp2_reg,
1312                            Label& L_success);
1313 
1314   // method handles (JSR 292)
1315   // offset relative to Gargs of argument at tos[arg_slot].
1316   // (arg_slot == 0 means the last argument, not the first).
1317   RegisterOrConstant argument_offset(RegisterOrConstant arg_slot,
1318                                      Register temp_reg,
1319                                      int extra_slot_offset = 0);
1320   // Address of Gargs and argument_offset.
1321   Address            argument_address(RegisterOrConstant arg_slot,
1322                                       Register temp_reg = noreg,
1323                                       int extra_slot_offset = 0);
1324 
1325   // Stack overflow checking
1326 
1327   // Note: this clobbers G3_scratch
1328   inline void bang_stack_with_offset(int offset);
1329 
1330   // Writes to stack successive pages until offset reached to check for
1331   // stack overflow + shadow pages.  Clobbers tsp and scratch registers.
1332   void bang_stack_size(Register Rsize, Register Rtsp, Register Rscratch);
1333 
1334   // Check for reserved stack access in method being exited (for JIT)
1335   void reserved_stack_check();
1336 
1337   virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset);
1338 
1339   void verify_tlab();
1340 
1341   Condition negate_condition(Condition cond);
1342 
1343   // Helper functions for statistics gathering.
1344   // Conditionally (non-atomically) increments passed counter address, preserving condition codes.
1345   void cond_inc(Condition cond, address counter_addr, Register Rtemp1, Register Rtemp2);
1346   // Unconditional increment.
1347   void inc_counter(address counter_addr, Register Rtmp1, Register Rtmp2);
1348   void inc_counter(int*    counter_addr, Register Rtmp1, Register Rtmp2);
1349 
1350 #ifdef COMPILER2
1351   // Compress char[] to byte[] by compressing 16 bytes at once. Return 0 on failure.
1352   void string_compress_16(Register src, Register dst, Register cnt, Register result,
1353                           Register tmp1, Register tmp2, Register tmp3, Register tmp4,
1354                           FloatRegister ftmp1, FloatRegister ftmp2, FloatRegister ftmp3, Label& Ldone);
1355 
1356   // Compress char[] to byte[]. Return 0 on failure.
1357   void string_compress(Register src, Register dst, Register cnt, Register tmp, Register result, Label& Ldone);
1358 
1359   // Inflate byte[] to char[] by inflating 16 bytes at once.
1360   void string_inflate_16(Register src, Register dst, Register cnt, Register tmp,
1361                          FloatRegister ftmp1, FloatRegister ftmp2, FloatRegister ftmp3, FloatRegister ftmp4, Label& Ldone);
1362 
1363   // Inflate byte[] to char[].
1364   void string_inflate(Register src, Register dst, Register cnt, Register tmp, Label& Ldone);
1365 
1366   void string_compare(Register str1, Register str2,
1367                       Register cnt1, Register cnt2,
1368                       Register tmp1, Register tmp2,
1369                       Register result, int ae);
1370 
1371   void array_equals(bool is_array_equ, Register ary1, Register ary2,
1372                     Register limit, Register tmp, Register result, bool is_byte);
1373   // test for negative bytes in input string of a given size, result 0 if none
1374   void has_negatives(Register inp, Register size, Register result,
1375                      Register t2, Register t3, Register t4,
1376                      Register t5);
1377 
1378 #endif
1379 
1380   // Use BIS for zeroing
1381   void bis_zeroing(Register to, Register count, Register temp, Label& Ldone);
1382 
1383   // Update CRC-32[C] with a byte value according to constants in table
1384   void update_byte_crc32(Register crc, Register val, Register table);
1385 
1386   // Reverse byte order of lower 32 bits, assuming upper 32 bits all zeros
1387   void reverse_bytes_32(Register src, Register dst, Register tmp);
1388   void movitof_revbytes(Register src, FloatRegister dst, Register tmp1, Register tmp2);
1389   void movftoi_revbytes(FloatRegister src, Register dst, Register tmp1, Register tmp2);
1390 
1391   // CRC32 code for java.util.zip.CRC32::updateBytes0() instrinsic.
1392   void kernel_crc32(Register crc, Register buf, Register len, Register table);
1393   // Fold 128-bit data chunk
1394   void fold_128bit_crc32(Register xcrc_hi, Register xcrc_lo, Register xK_hi, Register xK_lo, Register xtmp_hi, Register xtmp_lo, Register buf, int offset);
1395   void fold_128bit_crc32(Register xcrc_hi, Register xcrc_lo, Register xK_hi, Register xK_lo, Register xtmp_hi, Register xtmp_lo, Register xbuf_hi, Register xbuf_lo);
1396   // Fold 8-bit data
1397   void fold_8bit_crc32(Register xcrc, Register table, Register xtmp, Register tmp);
1398   void fold_8bit_crc32(Register crc, Register table, Register tmp);
1399   // CRC32C code for java.util.zip.CRC32C::updateBytes/updateDirectByteBuffer instrinsic.
1400   void kernel_crc32c(Register crc, Register buf, Register len, Register table);
1401 
1402 };
1403 
1404 /**
1405  * class SkipIfEqual:
1406  *
1407  * Instantiating this class will result in assembly code being output that will
1408  * jump around any code emitted between the creation of the instance and it's
1409  * automatic destruction at the end of a scope block, depending on the value of
1410  * the flag passed to the constructor, which will be checked at run-time.
1411  */
1412 class SkipIfEqual : public StackObj {
1413  private:
1414   MacroAssembler* _masm;
1415   Label _label;
1416 
1417  public:
1418    // 'temp' is a temp register that this object can use (and trash)
1419    SkipIfEqual(MacroAssembler*, Register temp,
1420                const bool* flag_addr, Assembler::Condition condition);
1421    ~SkipIfEqual();
1422 };
1423 
1424 #endif // CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP