1 /*
   2  * Copyright (c) 1997, 2015, 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 #ifdef CC_INTERP
 140 REGISTER_DECLARATION(Register, Lstate           , L0); // interpreter state object pointer
 141 REGISTER_DECLARATION(Register, L1_scratch       , L1); // scratch
 142 REGISTER_DECLARATION(Register, Lmirror          , L1); // mirror (for native methods only)
 143 REGISTER_DECLARATION(Register, L2_scratch       , L2);
 144 REGISTER_DECLARATION(Register, L3_scratch       , L3);
 145 REGISTER_DECLARATION(Register, L4_scratch       , L4);
 146 REGISTER_DECLARATION(Register, Lscratch         , L5); // C1 uses
 147 REGISTER_DECLARATION(Register, Lscratch2        , L6); // C1 uses
 148 REGISTER_DECLARATION(Register, L7_scratch       , L7); // constant pool cache
 149 REGISTER_DECLARATION(Register, O5_savedSP       , O5);
 150 REGISTER_DECLARATION(Register, I5_savedSP       , I5); // Saved SP before bumping for locals.  This is simply
 151                                                        // a copy SP, so in 64-bit it's a biased value.  The bias
 152                                                        // is added and removed as needed in the frame code.
 153 // Interface to signature handler
 154 REGISTER_DECLARATION(Register, Llocals          , L7); // pointer to locals for signature handler
 155 REGISTER_DECLARATION(Register, Lmethod          , L6); // Method* when calling signature handler
 156 
 157 #else
 158 REGISTER_DECLARATION(Register, Lesp             , L0); // expression stack pointer
 159 REGISTER_DECLARATION(Register, Lbcp             , L1); // pointer to next bytecode
 160 REGISTER_DECLARATION(Register, Lmethod          , L2);
 161 REGISTER_DECLARATION(Register, Llocals          , L3);
 162 REGISTER_DECLARATION(Register, Largs            , L3); // pointer to locals for signature handler
 163                                                        // must match Llocals in asm interpreter
 164 REGISTER_DECLARATION(Register, Lmonitors        , L4);
 165 REGISTER_DECLARATION(Register, Lbyte_code       , L5);
 166 // When calling out from the interpreter we record SP so that we can remove any extra stack
 167 // space allocated during adapter transitions. This register is only live from the point
 168 // of the call until we return.
 169 REGISTER_DECLARATION(Register, Llast_SP         , L5);
 170 REGISTER_DECLARATION(Register, Lscratch         , L5);
 171 REGISTER_DECLARATION(Register, Lscratch2        , L6);
 172 REGISTER_DECLARATION(Register, LcpoolCache      , L6); // constant pool cache
 173 
 174 REGISTER_DECLARATION(Register, O5_savedSP       , O5);
 175 REGISTER_DECLARATION(Register, I5_savedSP       , I5); // Saved SP before bumping for locals.  This is simply
 176                                                        // a copy SP, so in 64-bit it's a biased value.  The bias
 177                                                        // is added and removed as needed in the frame code.
 178 REGISTER_DECLARATION(Register, IdispatchTables  , I4); // Base address of the bytecode dispatch tables
 179 REGISTER_DECLARATION(Register, IdispatchAddress , I3); // Register which saves the dispatch address for each bytecode
 180 REGISTER_DECLARATION(Register, ImethodDataPtr   , I2); // Pointer to the current method data
 181 #endif /* CC_INTERP */
 182 
 183 // NOTE: Lscratch2 and LcpoolCache point to the same registers in
 184 //       the interpreter code. If Lscratch2 needs to be used for some
 185 //       purpose than LcpoolCache should be restore after that for
 186 //       the interpreter to work right
 187 // (These assignments must be compatible with L7_thread_cache; see above.)
 188 
 189 // Lbcp points into the middle of the method object.
 190 
 191 // Exception processing
 192 // These registers are passed into exception handlers.
 193 // All exception handlers require the exception object being thrown.
 194 // In addition, an nmethod's exception handler must be passed
 195 // the address of the call site within the nmethod, to allow
 196 // proper selection of the applicable catch block.
 197 // (Interpreter frames use their own bcp() for this purpose.)
 198 //
 199 // The Oissuing_pc value is not always needed.  When jumping to a
 200 // handler that is known to be interpreted, the Oissuing_pc value can be
 201 // omitted.  An actual catch block in compiled code receives (from its
 202 // nmethod's exception handler) the thrown exception in the Oexception,
 203 // but it doesn't need the Oissuing_pc.
 204 //
 205 // If an exception handler (either interpreted or compiled)
 206 // discovers there is no applicable catch block, it updates
 207 // the Oissuing_pc to the continuation PC of its own caller,
 208 // pops back to that caller's stack frame, and executes that
 209 // caller's exception handler.  Obviously, this process will
 210 // iterate until the control stack is popped back to a method
 211 // containing an applicable catch block.  A key invariant is
 212 // that the Oissuing_pc value is always a value local to
 213 // the method whose exception handler is currently executing.
 214 //
 215 // Note:  The issuing PC value is __not__ a raw return address (I7 value).
 216 // It is a "return pc", the address __following__ the call.
 217 // Raw return addresses are converted to issuing PCs by frame::pc(),
 218 // or by stubs.  Issuing PCs can be used directly with PC range tables.
 219 //
 220 REGISTER_DECLARATION(Register, Oexception  , O0); // exception being thrown
 221 REGISTER_DECLARATION(Register, Oissuing_pc , O1); // where the exception is coming from
 222 
 223 
 224 // These must occur after the declarations above
 225 #ifndef DONT_USE_REGISTER_DEFINES
 226 
 227 #define Gthread             AS_REGISTER(Register, Gthread)
 228 #define Gmethod             AS_REGISTER(Register, Gmethod)
 229 #define Gmegamorphic_method AS_REGISTER(Register, Gmegamorphic_method)
 230 #define Ginline_cache_reg   AS_REGISTER(Register, Ginline_cache_reg)
 231 #define Gargs               AS_REGISTER(Register, Gargs)
 232 #define Lthread_cache       AS_REGISTER(Register, Lthread_cache)
 233 #define Gframe_size         AS_REGISTER(Register, Gframe_size)
 234 #define Gtemp               AS_REGISTER(Register, Gtemp)
 235 
 236 #ifdef CC_INTERP
 237 #define Lstate              AS_REGISTER(Register, Lstate)
 238 #define Lesp                AS_REGISTER(Register, Lesp)
 239 #define L1_scratch          AS_REGISTER(Register, L1_scratch)
 240 #define Lmirror             AS_REGISTER(Register, Lmirror)
 241 #define L2_scratch          AS_REGISTER(Register, L2_scratch)
 242 #define L3_scratch          AS_REGISTER(Register, L3_scratch)
 243 #define L4_scratch          AS_REGISTER(Register, L4_scratch)
 244 #define Lscratch            AS_REGISTER(Register, Lscratch)
 245 #define Lscratch2           AS_REGISTER(Register, Lscratch2)
 246 #define L7_scratch          AS_REGISTER(Register, L7_scratch)
 247 #define Ostate              AS_REGISTER(Register, Ostate)
 248 #else
 249 #define Lesp                AS_REGISTER(Register, Lesp)
 250 #define Lbcp                AS_REGISTER(Register, Lbcp)
 251 #define Lmethod             AS_REGISTER(Register, Lmethod)
 252 #define Llocals             AS_REGISTER(Register, Llocals)
 253 #define Lmonitors           AS_REGISTER(Register, Lmonitors)
 254 #define Lbyte_code          AS_REGISTER(Register, Lbyte_code)
 255 #define Lscratch            AS_REGISTER(Register, Lscratch)
 256 #define Lscratch2           AS_REGISTER(Register, Lscratch2)
 257 #define LcpoolCache         AS_REGISTER(Register, LcpoolCache)
 258 #endif /* ! CC_INTERP */
 259 
 260 #define Lentry_args         AS_REGISTER(Register, Lentry_args)
 261 #define I5_savedSP          AS_REGISTER(Register, I5_savedSP)
 262 #define O5_savedSP          AS_REGISTER(Register, O5_savedSP)
 263 #define IdispatchAddress    AS_REGISTER(Register, IdispatchAddress)
 264 #define ImethodDataPtr      AS_REGISTER(Register, ImethodDataPtr)
 265 #define IdispatchTables     AS_REGISTER(Register, IdispatchTables)
 266 
 267 #define Oexception          AS_REGISTER(Register, Oexception)
 268 #define Oissuing_pc         AS_REGISTER(Register, Oissuing_pc)
 269 
 270 #endif
 271 
 272 
 273 // Address is an abstraction used to represent a memory location.
 274 //
 275 // Note: A register location is represented via a Register, not
 276 //       via an address for efficiency & simplicity reasons.
 277 
 278 class Address VALUE_OBJ_CLASS_SPEC {
 279  private:
 280   Register           _base;           // Base register.
 281   RegisterOrConstant _index_or_disp;  // Index register or constant displacement.
 282   RelocationHolder   _rspec;
 283 
 284  public:
 285   Address() : _base(noreg), _index_or_disp(noreg) {}
 286 
 287   Address(Register base, RegisterOrConstant index_or_disp)
 288     : _base(base),
 289       _index_or_disp(index_or_disp) {
 290   }
 291 
 292   Address(Register base, Register index)
 293     : _base(base),
 294       _index_or_disp(index) {
 295   }
 296 
 297   Address(Register base, int disp)
 298     : _base(base),
 299       _index_or_disp(disp) {
 300   }
 301 
 302 #ifdef ASSERT
 303   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 304   Address(Register base, ByteSize disp)
 305     : _base(base),
 306       _index_or_disp(in_bytes(disp)) {
 307   }
 308 #endif
 309 
 310   // accessors
 311   Register base()             const { return _base; }
 312   Register index()            const { return _index_or_disp.as_register(); }
 313   int      disp()             const { return _index_or_disp.as_constant(); }
 314 
 315   bool     has_index()        const { return _index_or_disp.is_register(); }
 316   bool     has_disp()         const { return _index_or_disp.is_constant(); }
 317 
 318   bool     uses(Register reg) const { return base() == reg || (has_index() && index() == reg); }
 319 
 320   const relocInfo::relocType rtype() { return _rspec.type(); }
 321   const RelocationHolder&    rspec() { return _rspec; }
 322 
 323   RelocationHolder rspec(int offset) const {
 324     return offset == 0 ? _rspec : _rspec.plus(offset);
 325   }
 326 
 327   inline bool is_simm13(int offset = 0);  // check disp+offset for overflow
 328 
 329   Address plus_disp(int plusdisp) const {     // bump disp by a small amount
 330     assert(_index_or_disp.is_constant(), "must have a displacement");
 331     Address a(base(), disp() + plusdisp);
 332     return a;
 333   }
 334   bool is_same_address(Address a) const {
 335     // disregard _rspec
 336     return base() == a.base() && (has_index() ? index() == a.index() : disp() == a.disp());
 337   }
 338 
 339   Address after_save() const {
 340     Address a = (*this);
 341     a._base = a._base->after_save();
 342     return a;
 343   }
 344 
 345   Address after_restore() const {
 346     Address a = (*this);
 347     a._base = a._base->after_restore();
 348     return a;
 349   }
 350 
 351   // Convert the raw encoding form into the form expected by the
 352   // constructor for Address.
 353   static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc);
 354 
 355   friend class Assembler;
 356 };
 357 
 358 
 359 class AddressLiteral VALUE_OBJ_CLASS_SPEC {
 360  private:
 361   address          _address;
 362   RelocationHolder _rspec;
 363 
 364   RelocationHolder rspec_from_rtype(relocInfo::relocType rtype, address addr) {
 365     switch (rtype) {
 366     case relocInfo::external_word_type:
 367       return external_word_Relocation::spec(addr);
 368     case relocInfo::internal_word_type:
 369       return internal_word_Relocation::spec(addr);
 370 #ifdef _LP64
 371     case relocInfo::opt_virtual_call_type:
 372       return opt_virtual_call_Relocation::spec();
 373     case relocInfo::static_call_type:
 374       return static_call_Relocation::spec();
 375     case relocInfo::runtime_call_type:
 376       return runtime_call_Relocation::spec();
 377 #endif
 378     case relocInfo::none:
 379       return RelocationHolder();
 380     default:
 381       ShouldNotReachHere();
 382       return RelocationHolder();
 383     }
 384   }
 385 
 386  protected:
 387   // creation
 388   AddressLiteral() : _address(NULL), _rspec(NULL) {}
 389 
 390  public:
 391   AddressLiteral(address addr, RelocationHolder const& rspec)
 392     : _address(addr),
 393       _rspec(rspec) {}
 394 
 395   // Some constructors to avoid casting at the call site.
 396   AddressLiteral(jobject obj, RelocationHolder const& rspec)
 397     : _address((address) obj),
 398       _rspec(rspec) {}
 399 
 400   AddressLiteral(intptr_t value, RelocationHolder const& rspec)
 401     : _address((address) value),
 402       _rspec(rspec) {}
 403 
 404   AddressLiteral(address addr, relocInfo::relocType rtype = relocInfo::none)
 405     : _address((address) addr),
 406     _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 407 
 408   // Some constructors to avoid casting at the call site.
 409   AddressLiteral(address* addr, relocInfo::relocType rtype = relocInfo::none)
 410     : _address((address) addr),
 411     _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 412 
 413   AddressLiteral(bool* addr, relocInfo::relocType rtype = relocInfo::none)
 414     : _address((address) addr),
 415       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 416 
 417   AddressLiteral(const bool* addr, relocInfo::relocType rtype = relocInfo::none)
 418     : _address((address) addr),
 419       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 420 
 421   AddressLiteral(signed char* addr, relocInfo::relocType rtype = relocInfo::none)
 422     : _address((address) addr),
 423       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 424 
 425   AddressLiteral(int* addr, relocInfo::relocType rtype = relocInfo::none)
 426     : _address((address) addr),
 427       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 428 
 429   AddressLiteral(intptr_t addr, relocInfo::relocType rtype = relocInfo::none)
 430     : _address((address) addr),
 431       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 432 
 433 #ifdef _LP64
 434   // 32-bit complains about a multiple declaration for int*.
 435   AddressLiteral(intptr_t* addr, relocInfo::relocType rtype = relocInfo::none)
 436     : _address((address) addr),
 437       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 438 #endif
 439 
 440   AddressLiteral(Metadata* addr, relocInfo::relocType rtype = relocInfo::none)
 441     : _address((address) addr),
 442       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 443 
 444   AddressLiteral(Metadata** addr, relocInfo::relocType rtype = relocInfo::none)
 445     : _address((address) addr),
 446       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 447 
 448   AddressLiteral(float* addr, relocInfo::relocType rtype = relocInfo::none)
 449     : _address((address) addr),
 450       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 451 
 452   AddressLiteral(double* addr, relocInfo::relocType rtype = relocInfo::none)
 453     : _address((address) addr),
 454       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 455 
 456   intptr_t value() const { return (intptr_t) _address; }
 457   int      low10() const;
 458 
 459   const relocInfo::relocType rtype() const { return _rspec.type(); }
 460   const RelocationHolder&    rspec() const { return _rspec; }
 461 
 462   RelocationHolder rspec(int offset) const {
 463     return offset == 0 ? _rspec : _rspec.plus(offset);
 464   }
 465 };
 466 
 467 // Convenience classes
 468 class ExternalAddress: public AddressLiteral {
 469  private:
 470   static relocInfo::relocType reloc_for_target(address target) {
 471     // Sometimes ExternalAddress is used for values which aren't
 472     // exactly addresses, like the card table base.
 473     // external_word_type can't be used for values in the first page
 474     // so just skip the reloc in that case.
 475     return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
 476   }
 477 
 478  public:
 479   ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(          target)) {}
 480   ExternalAddress(Metadata** target) : AddressLiteral(target, reloc_for_target((address) target)) {}
 481 };
 482 
 483 inline Address RegisterImpl::address_in_saved_window() const {
 484    return (Address(SP, (sp_offset_in_saved_window() * wordSize) + STACK_BIAS));
 485 }
 486 
 487 
 488 
 489 // Argument is an abstraction used to represent an outgoing
 490 // actual argument or an incoming formal parameter, whether
 491 // it resides in memory or in a register, in a manner consistent
 492 // with the SPARC Application Binary Interface, or ABI.  This is
 493 // often referred to as the native or C calling convention.
 494 
 495 class Argument VALUE_OBJ_CLASS_SPEC {
 496  private:
 497   int _number;
 498   bool _is_in;
 499 
 500  public:
 501 #ifdef _LP64
 502   enum {
 503     n_register_parameters = 6,          // only 6 registers may contain integer parameters
 504     n_float_register_parameters = 16    // Can have up to 16 floating registers
 505   };
 506 #else
 507   enum {
 508     n_register_parameters = 6           // only 6 registers may contain integer parameters
 509   };
 510 #endif
 511 
 512   // creation
 513   Argument(int number, bool is_in) : _number(number), _is_in(is_in) {}
 514 
 515   int  number() const  { return _number;  }
 516   bool is_in()  const  { return _is_in;   }
 517   bool is_out() const  { return !is_in(); }
 518 
 519   Argument successor() const  { return Argument(number() + 1, is_in()); }
 520   Argument as_in()     const  { return Argument(number(), true ); }
 521   Argument as_out()    const  { return Argument(number(), false); }
 522 
 523   // locating register-based arguments:
 524   bool is_register() const { return _number < n_register_parameters; }
 525 
 526 #ifdef _LP64
 527   // locating Floating Point register-based arguments:
 528   bool is_float_register() const { return _number < n_float_register_parameters; }
 529 
 530   FloatRegister as_float_register() const {
 531     assert(is_float_register(), "must be a register argument");
 532     return as_FloatRegister(( number() *2 ) + 1);
 533   }
 534   FloatRegister as_double_register() const {
 535     assert(is_float_register(), "must be a register argument");
 536     return as_FloatRegister(( number() *2 ));
 537   }
 538 #endif
 539 
 540   Register as_register() const {
 541     assert(is_register(), "must be a register argument");
 542     return is_in() ? as_iRegister(number()) : as_oRegister(number());
 543   }
 544 
 545   // locating memory-based arguments
 546   Address as_address() const {
 547     assert(!is_register(), "must be a memory argument");
 548     return address_in_frame();
 549   }
 550 
 551   // When applied to a register-based argument, give the corresponding address
 552   // into the 6-word area "into which callee may store register arguments"
 553   // (This is a different place than the corresponding register-save area location.)
 554   Address address_in_frame() const;
 555 
 556   // debugging
 557   const char* name() const;
 558 
 559   friend class Assembler;
 560 };
 561 
 562 
 563 class RegistersForDebugging : public StackObj {
 564  public:
 565   intptr_t i[8], l[8], o[8], g[8];
 566   float    f[32];
 567   double   d[32];
 568 
 569   void print(outputStream* s);
 570 
 571   static int i_offset(int j) { return offset_of(RegistersForDebugging, i[j]); }
 572   static int l_offset(int j) { return offset_of(RegistersForDebugging, l[j]); }
 573   static int o_offset(int j) { return offset_of(RegistersForDebugging, o[j]); }
 574   static int g_offset(int j) { return offset_of(RegistersForDebugging, g[j]); }
 575   static int f_offset(int j) { return offset_of(RegistersForDebugging, f[j]); }
 576   static int d_offset(int j) { return offset_of(RegistersForDebugging, d[j / 2]); }
 577 
 578   // gen asm code to save regs
 579   static void save_registers(MacroAssembler* a);
 580 
 581   // restore global registers in case C code disturbed them
 582   static void restore_registers(MacroAssembler* a, Register r);
 583 };
 584 
 585 
 586 // MacroAssembler extends Assembler by a few frequently used macros.
 587 //
 588 // Most of the standard SPARC synthetic ops are defined here.
 589 // Instructions for which a 'better' code sequence exists depending
 590 // on arguments should also go in here.
 591 
 592 #define JMP2(r1, r2) jmp(r1, r2, __FILE__, __LINE__)
 593 #define JMP(r1, off) jmp(r1, off, __FILE__, __LINE__)
 594 #define JUMP(a, temp, off)     jump(a, temp, off, __FILE__, __LINE__)
 595 #define JUMPL(a, temp, d, off) jumpl(a, temp, d, off, __FILE__, __LINE__)
 596 
 597 
 598 class MacroAssembler : public Assembler {
 599   // code patchers need various routines like inv_wdisp()
 600   friend class NativeInstruction;
 601   friend class NativeGeneralJump;
 602   friend class Relocation;
 603   friend class Label;
 604 
 605  protected:
 606   static int  patched_branch(int dest_pos, int inst, int inst_pos);
 607   static int  branch_destination(int inst, int pos);
 608 
 609   // Support for VM calls
 610   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
 611   // may customize this version by overriding it for its purposes (e.g., to save/restore
 612   // additional registers when doing a VM call).
 613 #ifdef CC_INTERP
 614   #define VIRTUAL
 615 #else
 616   #define VIRTUAL virtual
 617 #endif
 618 
 619   VIRTUAL void call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments);
 620 
 621   //
 622   // It is imperative that all calls into the VM are handled via the call_VM macros.
 623   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 624   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 625   //
 626   // This is the base routine called by the different versions of call_VM. The interpreter
 627   // may customize this version by overriding it for its purposes (e.g., to save/restore
 628   // additional registers when doing a VM call).
 629   //
 630   // A non-volatile java_thread_cache register should be specified so
 631   // that the G2_thread value can be preserved across the call.
 632   // (If java_thread_cache is noreg, then a slow get_thread call
 633   // will re-initialize the G2_thread.) call_VM_base returns the register that contains the
 634   // thread.
 635   //
 636   // If no last_java_sp is specified (noreg) than SP will be used instead.
 637 
 638   virtual void call_VM_base(
 639     Register        oop_result,             // where an oop-result ends up if any; use noreg otherwise
 640     Register        java_thread_cache,      // the thread if computed before     ; use noreg otherwise
 641     Register        last_java_sp,           // to set up last_Java_frame in stubs; use noreg otherwise
 642     address         entry_point,            // the entry point
 643     int             number_of_arguments,    // the number of arguments (w/o thread) to pop after call
 644     bool            check_exception=true    // flag which indicates if exception should be checked
 645   );
 646 
 647   // This routine should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 648   // The implementation is only non-empty for the InterpreterMacroAssembler,
 649   // as only the interpreter handles and ForceEarlyReturn PopFrame requests.
 650   virtual void check_and_handle_popframe(Register scratch_reg);
 651   virtual void check_and_handle_earlyret(Register scratch_reg);
 652 
 653  public:
 654   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 655 
 656   // Support for NULL-checks
 657   //
 658   // Generates code that causes a NULL OS exception if the content of reg is NULL.
 659   // If the accessed location is M[reg + offset] and the offset is known, provide the
 660   // offset.  No explicit code generation is needed if the offset is within a certain
 661   // range (0 <= offset <= page_size).
 662   //
 663   // %%%%%% Currently not done for SPARC
 664 
 665   void null_check(Register reg, int offset = -1);
 666   static bool needs_explicit_null_check(intptr_t offset);
 667 
 668   // support for delayed instructions
 669   MacroAssembler* delayed() { Assembler::delayed();  return this; }
 670 
 671   // branches that use right instruction for v8 vs. v9
 672   inline void br( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 673   inline void br( Condition c, bool a, Predict p, Label& L );
 674 
 675   inline void fb( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 676   inline void fb( Condition c, bool a, Predict p, Label& L );
 677 
 678   // compares register with zero (32 bit) and branches (V9 and V8 instructions)
 679   void cmp_zero_and_br( Condition c, Register s1, Label& L, bool a = false, Predict p = pn );
 680   // Compares a pointer register with zero and branches on (not)null.
 681   // Does a test & branch on 32-bit systems and a register-branch on 64-bit.
 682   void br_null   ( Register s1, bool a, Predict p, Label& L );
 683   void br_notnull( Register s1, bool a, Predict p, Label& L );
 684 
 685   //
 686   // Compare registers and branch with nop in delay slot or cbcond without delay slot.
 687   //
 688   // ATTENTION: use these instructions with caution because cbcond instruction
 689   //            has very short distance: 512 instructions (2Kbyte).
 690 
 691   // Compare integer (32 bit) values (icc only).
 692   void cmp_and_br_short(Register s1, Register s2, Condition c, Predict p, Label& L);
 693   void cmp_and_br_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
 694   // Platform depending version for pointer compare (icc on !LP64 and xcc on LP64).
 695   void cmp_and_brx_short(Register s1, Register s2, Condition c, Predict p, Label& L);
 696   void cmp_and_brx_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
 697 
 698   // Short branch version for compares a pointer pwith zero.
 699   void br_null_short   ( Register s1, Predict p, Label& L );
 700   void br_notnull_short( Register s1, Predict p, Label& L );
 701 
 702   // unconditional short branch
 703   void ba_short(Label& L);
 704 
 705   inline void bp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 706   inline void bp( Condition c, bool a, CC cc, Predict p, Label& L );
 707 
 708   // Branch that tests xcc in LP64 and icc in !LP64
 709   inline void brx( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 710   inline void brx( Condition c, bool a, Predict p, Label& L );
 711 
 712   // unconditional branch
 713   inline void ba( Label& L );
 714 
 715   // Branch that tests fp condition codes
 716   inline void fbp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
 717   inline void fbp( Condition c, bool a, CC cc, Predict p, Label& L );
 718 
 719   // get PC the best way
 720   inline int get_pc( Register d );
 721 
 722   // Sparc shorthands(pp 85, V8 manual, pp 289 V9 manual)
 723   inline void cmp(  Register s1, Register s2 ) { subcc( s1, s2, G0 ); }
 724   inline void cmp(  Register s1, int simm13a ) { subcc( s1, simm13a, G0 ); }
 725 
 726   inline void jmp( Register s1, Register s2 );
 727   inline void jmp( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
 728 
 729   // Check if the call target is out of wdisp30 range (relative to the code cache)
 730   static inline bool is_far_target(address d);
 731   inline void call( address d,  relocInfo::relocType rt = relocInfo::runtime_call_type );
 732   inline void call( Label& L,   relocInfo::relocType rt = relocInfo::runtime_call_type );
 733   inline void callr( Register s1, Register s2 );
 734   inline void callr( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
 735 
 736   // Emits nothing on V8
 737   inline void iprefetch( address d, relocInfo::relocType rt = relocInfo::none );
 738   inline void iprefetch( Label& L);
 739 
 740   inline void tst( Register s ) { orcc( G0, s, G0 ); }
 741 
 742 #ifdef PRODUCT
 743   inline void ret(  bool trace = TraceJumps )   { if (trace) {
 744                                                     mov(I7, O7); // traceable register
 745                                                     JMP(O7, 2 * BytesPerInstWord);
 746                                                   } else {
 747                                                     jmpl( I7, 2 * BytesPerInstWord, G0 );
 748                                                   }
 749                                                 }
 750 
 751   inline void retl( bool trace = TraceJumps )  { if (trace) JMP(O7, 2 * BytesPerInstWord);
 752                                                  else jmpl( O7, 2 * BytesPerInstWord, G0 ); }
 753 #else
 754   void ret(  bool trace = TraceJumps );
 755   void retl( bool trace = TraceJumps );
 756 #endif /* PRODUCT */
 757 
 758   // Required platform-specific helpers for Label::patch_instructions.
 759   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 760   void pd_patch_instruction(address branch, address target);
 761 
 762   // sethi Macro handles optimizations and relocations
 763 private:
 764   void internal_sethi(const AddressLiteral& addrlit, Register d, bool ForceRelocatable);
 765 public:
 766   void sethi(const AddressLiteral& addrlit, Register d);
 767   void patchable_sethi(const AddressLiteral& addrlit, Register d);
 768 
 769   // compute the number of instructions for a sethi/set
 770   static int  insts_for_sethi( address a, bool worst_case = false );
 771   static int  worst_case_insts_for_set();
 772 
 773   // set may be either setsw or setuw (high 32 bits may be zero or sign)
 774 private:
 775   void internal_set(const AddressLiteral& al, Register d, bool ForceRelocatable);
 776   static int insts_for_internal_set(intptr_t value);
 777 public:
 778   void set(const AddressLiteral& addrlit, Register d);
 779   void set(intptr_t value, Register d);
 780   void set(address addr, Register d, RelocationHolder const& rspec);
 781   static int insts_for_set(intptr_t value) { return insts_for_internal_set(value); }
 782 
 783   void patchable_set(const AddressLiteral& addrlit, Register d);
 784   void patchable_set(intptr_t value, Register d);
 785   void set64(jlong value, Register d, Register tmp);
 786   static int insts_for_set64(jlong value);
 787 
 788   // sign-extend 32 to 64
 789   inline void signx( Register s, Register d ) { sra( s, G0, d); }
 790   inline void signx( Register d )             { sra( d, G0, d); }
 791 
 792   inline void not1( Register s, Register d ) { xnor( s, G0, d ); }
 793   inline void not1( Register d )             { xnor( d, G0, d ); }
 794 
 795   inline void neg( Register s, Register d ) { sub( G0, s, d ); }
 796   inline void neg( Register d )             { sub( G0, d, d ); }
 797 
 798   inline void cas(  Register s1, Register s2, Register d) { casa( s1, s2, d, ASI_PRIMARY); }
 799   inline void casx( Register s1, Register s2, Register d) { casxa(s1, s2, d, ASI_PRIMARY); }
 800   // Functions for isolating 64 bit atomic swaps for LP64
 801   // cas_ptr will perform cas for 32 bit VM's and casx for 64 bit VM's
 802   inline void cas_ptr(  Register s1, Register s2, Register d) {
 803 #ifdef _LP64
 804     casx( s1, s2, d );
 805 #else
 806     cas( s1, s2, d );
 807 #endif
 808   }
 809 
 810   // Functions for isolating 64 bit shifts for LP64
 811   inline void sll_ptr( Register s1, Register s2, Register d );
 812   inline void sll_ptr( Register s1, int imm6a,   Register d );
 813   inline void sll_ptr( Register s1, RegisterOrConstant s2, Register d );
 814   inline void srl_ptr( Register s1, Register s2, Register d );
 815   inline void srl_ptr( Register s1, int imm6a,   Register d );
 816 
 817   // little-endian
 818   inline void casl(  Register s1, Register s2, Register d) { casa( s1, s2, d, ASI_PRIMARY_LITTLE); }
 819   inline void casxl( Register s1, Register s2, Register d) { casxa(s1, s2, d, ASI_PRIMARY_LITTLE); }
 820 
 821   inline void inc(   Register d,  int const13 = 1 ) { add(   d, const13, d); }
 822   inline void inccc( Register d,  int const13 = 1 ) { addcc( d, const13, d); }
 823 
 824   inline void dec(   Register d,  int const13 = 1 ) { sub(   d, const13, d); }
 825   inline void deccc( Register d,  int const13 = 1 ) { subcc( d, const13, d); }
 826 
 827   using Assembler::add;
 828   inline void add(Register s1, int simm13a, Register d, relocInfo::relocType rtype);
 829   inline void add(Register s1, int simm13a, Register d, RelocationHolder const& rspec);
 830   inline void add(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
 831   inline void add(const Address& a, Register d, int offset = 0);
 832 
 833   using Assembler::andn;
 834   inline void andn(  Register s1, RegisterOrConstant s2, Register d);
 835 
 836   inline void btst( Register s1,  Register s2 ) { andcc( s1, s2, G0 ); }
 837   inline void btst( int simm13a,  Register s )  { andcc( s,  simm13a, G0 ); }
 838 
 839   inline void bset( Register s1,  Register s2 ) { or3( s1, s2, s2 ); }
 840   inline void bset( int simm13a,  Register s )  { or3( s,  simm13a, s ); }
 841 
 842   inline void bclr( Register s1,  Register s2 ) { andn( s1, s2, s2 ); }
 843   inline void bclr( int simm13a,  Register s )  { andn( s,  simm13a, s ); }
 844 
 845   inline void btog( Register s1,  Register s2 ) { xor3( s1, s2, s2 ); }
 846   inline void btog( int simm13a,  Register s )  { xor3( s,  simm13a, s ); }
 847 
 848   inline void clr( Register d ) { or3( G0, G0, d ); }
 849 
 850   inline void clrb( Register s1, Register s2);
 851   inline void clrh( Register s1, Register s2);
 852   inline void clr(  Register s1, Register s2);
 853   inline void clrx( Register s1, Register s2);
 854 
 855   inline void clrb( Register s1, int simm13a);
 856   inline void clrh( Register s1, int simm13a);
 857   inline void clr(  Register s1, int simm13a);
 858   inline void clrx( Register s1, int simm13a);
 859 
 860   // copy & clear upper word
 861   inline void clruw( Register s, Register d ) { srl( s, G0, d); }
 862   // clear upper word
 863   inline void clruwu( Register d ) { srl( d, G0, d); }
 864 
 865   using Assembler::ldsb;
 866   using Assembler::ldsh;
 867   using Assembler::ldsw;
 868   using Assembler::ldub;
 869   using Assembler::lduh;
 870   using Assembler::lduw;
 871   using Assembler::ldx;
 872   using Assembler::ldd;
 873 
 874 #ifdef ASSERT
 875   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 876   inline void ld(Register s1, ByteSize simm13a, Register d);
 877 #endif
 878 
 879   inline void ld(Register s1, Register s2, Register d);
 880   inline void ld(Register s1, int simm13a, Register d);
 881 
 882   inline void ldsb(const Address& a, Register d, int offset = 0);
 883   inline void ldsh(const Address& a, Register d, int offset = 0);
 884   inline void ldsw(const Address& a, Register d, int offset = 0);
 885   inline void ldub(const Address& a, Register d, int offset = 0);
 886   inline void lduh(const Address& a, Register d, int offset = 0);
 887   inline void lduw(const Address& a, Register d, int offset = 0);
 888   inline void ldx( const Address& a, Register d, int offset = 0);
 889   inline void ld(  const Address& a, Register d, int offset = 0);
 890   inline void ldd( const Address& a, Register d, int offset = 0);
 891 
 892   inline void ldub(Register s1, RegisterOrConstant s2, Register d );
 893   inline void ldsb(Register s1, RegisterOrConstant s2, Register d );
 894   inline void lduh(Register s1, RegisterOrConstant s2, Register d );
 895   inline void ldsh(Register s1, RegisterOrConstant s2, Register d );
 896   inline void lduw(Register s1, RegisterOrConstant s2, Register d );
 897   inline void ldsw(Register s1, RegisterOrConstant s2, Register d );
 898   inline void ldx( Register s1, RegisterOrConstant s2, Register d );
 899   inline void ld(  Register s1, RegisterOrConstant s2, Register d );
 900   inline void ldd( Register s1, RegisterOrConstant s2, Register d );
 901 
 902   using Assembler::ldf;
 903   inline void ldf(FloatRegisterImpl::Width w, Register s1, RegisterOrConstant s2, FloatRegister d);
 904   inline void ldf(FloatRegisterImpl::Width w, const Address& a, FloatRegister d, int offset = 0);
 905 
 906   // little-endian
 907   inline void ldxl(Register s1, Register s2, Register d) { ldxa(s1, s2, ASI_PRIMARY_LITTLE, d); }
 908   inline void ldfl(FloatRegisterImpl::Width w, Register s1, Register s2, FloatRegister d) { ldfa(w, s1, s2, ASI_PRIMARY_LITTLE, d); }
 909 
 910   // membar psuedo instruction.  takes into account target memory model.
 911   inline void membar( Assembler::Membar_mask_bits const7a );
 912 
 913   // returns if membar generates anything.
 914   inline bool membar_has_effect( Assembler::Membar_mask_bits const7a );
 915 
 916   // mov pseudo instructions
 917   inline void mov( Register s,  Register d) {
 918     if ( s != d )    or3( G0, s, d);
 919     else             assert_not_delayed();  // Put something useful in the delay slot!
 920   }
 921 
 922   inline void mov_or_nop( Register s,  Register d) {
 923     if ( s != d )    or3( G0, s, d);
 924     else             nop();
 925   }
 926 
 927   inline void mov( int simm13a, Register d) { or3( G0, simm13a, d); }
 928 
 929   using Assembler::prefetch;
 930   inline void prefetch(const Address& a, PrefetchFcn F, int offset = 0);
 931 
 932   using Assembler::stb;
 933   using Assembler::sth;
 934   using Assembler::stw;
 935   using Assembler::stx;
 936   using Assembler::std;
 937 
 938 #ifdef ASSERT
 939   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
 940   inline void st(Register d, Register s1, ByteSize simm13a);
 941 #endif
 942 
 943   inline void st(Register d, Register s1, Register s2);
 944   inline void st(Register d, Register s1, int simm13a);
 945 
 946   inline void stb(Register d, const Address& a, int offset = 0 );
 947   inline void sth(Register d, const Address& a, int offset = 0 );
 948   inline void stw(Register d, const Address& a, int offset = 0 );
 949   inline void stx(Register d, const Address& a, int offset = 0 );
 950   inline void st( Register d, const Address& a, int offset = 0 );
 951   inline void std(Register d, const Address& a, int offset = 0 );
 952 
 953   inline void stb(Register d, Register s1, RegisterOrConstant s2 );
 954   inline void sth(Register d, Register s1, RegisterOrConstant s2 );
 955   inline void stw(Register d, Register s1, RegisterOrConstant s2 );
 956   inline void stx(Register d, Register s1, RegisterOrConstant s2 );
 957   inline void std(Register d, Register s1, RegisterOrConstant s2 );
 958   inline void st( Register d, Register s1, RegisterOrConstant s2 );
 959 
 960   using Assembler::stf;
 961   inline void stf(FloatRegisterImpl::Width w, FloatRegister d, Register s1, RegisterOrConstant s2);
 962   inline void stf(FloatRegisterImpl::Width w, FloatRegister d, const Address& a, int offset = 0);
 963 
 964   // Note: offset is added to s2.
 965   using Assembler::sub;
 966   inline void sub(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
 967 
 968   using Assembler::swap;
 969   inline void swap(const Address& a, Register d, int offset = 0);
 970 
 971   // address pseudos: make these names unlike instruction names to avoid confusion
 972   inline intptr_t load_pc_address( Register reg, int bytes_to_skip );
 973   inline void load_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 974   inline void load_bool_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 975   inline void load_ptr_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
 976   inline void store_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
 977   inline void store_ptr_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
 978   inline void jumpl_to(const AddressLiteral& addrlit, Register temp, Register d, int offset = 0);
 979   inline void jump_to(const AddressLiteral& addrlit, Register temp, int offset = 0);
 980   inline void jump_indirect_to(Address& a, Register temp, int ld_offset = 0, int jmp_offset = 0);
 981 
 982   // ring buffer traceable jumps
 983 
 984   void jmp2( Register r1, Register r2, const char* file, int line );
 985   void jmp ( Register r1, int offset,  const char* file, int line );
 986 
 987   void jumpl(const AddressLiteral& addrlit, Register temp, Register d, int offset, const char* file, int line);
 988   void jump (const AddressLiteral& addrlit, Register temp,             int offset, const char* file, int line);
 989 
 990 
 991   // argument pseudos:
 992 
 993   inline void load_argument( Argument& a, Register  d );
 994   inline void store_argument( Register s, Argument& a );
 995   inline void store_ptr_argument( Register s, Argument& a );
 996   inline void store_float_argument( FloatRegister s, Argument& a );
 997   inline void store_double_argument( FloatRegister s, Argument& a );
 998   inline void store_long_argument( Register s, Argument& a );
 999 
1000   // handy macros:
1001 
1002   inline void round_to( Register r, int modulus ) {
1003     assert_not_delayed();
1004     inc( r, modulus - 1 );
1005     and3( r, -modulus, r );
1006   }
1007 
1008   // --------------------------------------------------
1009 
1010   // Functions for isolating 64 bit loads for LP64
1011   // ld_ptr will perform ld for 32 bit VM's and ldx for 64 bit VM's
1012   // st_ptr will perform st for 32 bit VM's and stx for 64 bit VM's
1013   inline void ld_ptr(Register s1, Register s2, Register d);
1014   inline void ld_ptr(Register s1, int simm13a, Register d);
1015   inline void ld_ptr(Register s1, RegisterOrConstant s2, Register d);
1016   inline void ld_ptr(const Address& a, Register d, int offset = 0);
1017   inline void st_ptr(Register d, Register s1, Register s2);
1018   inline void st_ptr(Register d, Register s1, int simm13a);
1019   inline void st_ptr(Register d, Register s1, RegisterOrConstant s2);
1020   inline void st_ptr(Register d, const Address& a, int offset = 0);
1021 
1022 #ifdef ASSERT
1023   // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
1024   inline void ld_ptr(Register s1, ByteSize simm13a, Register d);
1025   inline void st_ptr(Register d, Register s1, ByteSize simm13a);
1026 #endif
1027 
1028   // ld_long will perform ldd for 32 bit VM's and ldx for 64 bit VM's
1029   // st_long will perform std for 32 bit VM's and stx for 64 bit VM's
1030   inline void ld_long(Register s1, Register s2, Register d);
1031   inline void ld_long(Register s1, int simm13a, Register d);
1032   inline void ld_long(Register s1, RegisterOrConstant s2, Register d);
1033   inline void ld_long(const Address& a, Register d, int offset = 0);
1034   inline void st_long(Register d, Register s1, Register s2);
1035   inline void st_long(Register d, Register s1, int simm13a);
1036   inline void st_long(Register d, Register s1, RegisterOrConstant s2);
1037   inline void st_long(Register d, const Address& a, int offset = 0);
1038 
1039   // Helpers for address formation.
1040   // - They emit only a move if s2 is a constant zero.
1041   // - If dest is a constant and either s1 or s2 is a register, the temp argument is required and becomes the result.
1042   // - 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.
1043   RegisterOrConstant regcon_andn_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1044   RegisterOrConstant regcon_inc_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1045   RegisterOrConstant regcon_sll_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1046 
1047   RegisterOrConstant ensure_simm13_or_reg(RegisterOrConstant src, Register temp) {
1048     if (is_simm13(src.constant_or_zero()))
1049       return src;               // register or short constant
1050     guarantee(temp != noreg, "constant offset overflow");
1051     set(src.as_constant(), temp);
1052     return temp;
1053   }
1054 
1055   // --------------------------------------------------
1056 
1057  public:
1058   // traps as per trap.h (SPARC ABI?)
1059 
1060   void breakpoint_trap();
1061   void breakpoint_trap(Condition c, CC cc);
1062 
1063   // Support for serializing memory accesses between threads
1064   void serialize_memory(Register thread, Register tmp1, Register tmp2);
1065 
1066   // Stack frame creation/removal
1067   void enter();
1068   void leave();
1069 
1070   // Manipulation of C++ bools
1071   // These are idioms to flag the need for care with accessing bools but on
1072   // this platform we assume byte size
1073 
1074   inline void stbool(Register d, const Address& a) { stb(d, a); }
1075   inline void ldbool(const Address& a, Register d) { ldub(a, d); }
1076   inline void movbool( bool boolconst, Register d) { mov( (int) boolconst, d); }
1077 
1078   // klass oop manipulations if compressed
1079   void load_klass(Register src_oop, Register klass);
1080   void store_klass(Register klass, Register dst_oop);
1081   void store_klass_gap(Register s, Register dst_oop);
1082 
1083    // oop manipulations
1084   void load_heap_oop(const Address& s, Register d);
1085   void load_heap_oop(Register s1, Register s2, Register d);
1086   void load_heap_oop(Register s1, int simm13a, Register d);
1087   void load_heap_oop(Register s1, RegisterOrConstant s2, Register d);
1088   void store_heap_oop(Register d, Register s1, Register s2);
1089   void store_heap_oop(Register d, Register s1, int simm13a);
1090   void store_heap_oop(Register d, const Address& a, int offset = 0);
1091 
1092   void encode_heap_oop(Register src, Register dst);
1093   void encode_heap_oop(Register r) {
1094     encode_heap_oop(r, r);
1095   }
1096   void decode_heap_oop(Register src, Register dst);
1097   void decode_heap_oop(Register r) {
1098     decode_heap_oop(r, r);
1099   }
1100   void encode_heap_oop_not_null(Register r);
1101   void decode_heap_oop_not_null(Register r);
1102   void encode_heap_oop_not_null(Register src, Register dst);
1103   void decode_heap_oop_not_null(Register src, Register dst);
1104 
1105   void encode_klass_not_null(Register r);
1106   void decode_klass_not_null(Register r);
1107   void encode_klass_not_null(Register src, Register dst);
1108   void decode_klass_not_null(Register src, Register dst);
1109 
1110   // Support for managing the JavaThread pointer (i.e.; the reference to
1111   // thread-local information).
1112   void get_thread();                                // load G2_thread
1113   void verify_thread();                             // verify G2_thread contents
1114   void save_thread   (const Register threache); // save to cache
1115   void restore_thread(const Register thread_cache); // restore from cache
1116 
1117   // Support for last Java frame (but use call_VM instead where possible)
1118   void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
1119   void reset_last_Java_frame(void);
1120 
1121   // Call into the VM.
1122   // Passes the thread pointer (in O0) as a prepended argument.
1123   // Makes sure oop return values are visible to the GC.
1124   void call_VM(Register oop_result, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1125   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
1126   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1127   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
1128 
1129   // these overloadings are not presently used on SPARC:
1130   void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1131   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
1132   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1133   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);
1134 
1135   void call_VM_leaf(Register thread_cache, address entry_point, int number_of_arguments = 0);
1136   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1);
1137   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2);
1138   void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3);
1139 
1140   void get_vm_result  (Register oop_result);
1141   void get_vm_result_2(Register metadata_result);
1142 
1143   // vm result is currently getting hijacked to for oop preservation
1144   void set_vm_result(Register oop_result);
1145 
1146   // Emit the CompiledIC call idiom
1147   void ic_call(address entry, bool emit_delay = true);
1148 
1149   // if call_VM_base was called with check_exceptions=false, then call
1150   // check_and_forward_exception to handle exceptions when it is safe
1151   void check_and_forward_exception(Register scratch_reg);
1152 
1153   // Write to card table for - register is destroyed afterwards.
1154   void card_table_write(jbyte* byte_map_base, Register tmp, Register obj);
1155 
1156   void card_write_barrier_post(Register store_addr, Register new_val, Register tmp);
1157 
1158 #if INCLUDE_ALL_GCS
1159   // General G1 pre-barrier generator.
1160   void g1_write_barrier_pre(Register obj, Register index, int offset, Register pre_val, Register tmp, bool preserve_o_regs);
1161 
1162   // General G1 post-barrier generator
1163   void g1_write_barrier_post(Register store_addr, Register new_val, Register tmp);
1164 #endif // INCLUDE_ALL_GCS
1165 
1166   // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1167   void push_fTOS();
1168 
1169   // pops double TOS element from CPU stack and pushes on FPU stack
1170   void pop_fTOS();
1171 
1172   void empty_FPU_stack();
1173 
1174   void push_IU_state();
1175   void pop_IU_state();
1176 
1177   void push_FPU_state();
1178   void pop_FPU_state();
1179 
1180   void push_CPU_state();
1181   void pop_CPU_state();
1182 
1183   // Returns the byte size of the instructions generated by decode_klass_not_null().
1184   static int instr_size_for_decode_klass_not_null();
1185 
1186   // if heap base register is used - reinit it with the correct value
1187   void reinit_heapbase();
1188 
1189   // Debugging
1190   void _verify_oop(Register reg, const char * msg, const char * file, int line);
1191   void _verify_oop_addr(Address addr, const char * msg, const char * file, int line);
1192 
1193   // TODO: verify_method and klass metadata (compare against vptr?)
1194   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1195   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1196 
1197 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
1198 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop addr ", __FILE__, __LINE__)
1199 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1200 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1201 
1202         // only if +VerifyOops
1203   void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
1204         // only if +VerifyFPU
1205   void stop(const char* msg);                          // prints msg, dumps registers and stops execution
1206   void warn(const char* msg);                          // prints msg, but don't stop
1207   void untested(const char* what = "");
1208   void unimplemented(const char* what = "")      { char* b = new char[1024];  jio_snprintf(b, 1024, "unimplemented: %s", what);  stop(b); }
1209   void should_not_reach_here()                   { stop("should not reach here"); }
1210   void print_CPU_state();
1211 
1212   // oops in code
1213   AddressLiteral allocate_oop_address(jobject obj);                          // allocate_index
1214   AddressLiteral constant_oop_address(jobject obj);                          // find_index
1215   inline void    set_oop             (jobject obj, Register d);              // uses allocate_oop_address
1216   inline void    set_oop_constant    (jobject obj, Register d);              // uses constant_oop_address
1217   inline void    set_oop             (const AddressLiteral& obj_addr, Register d); // same as load_address
1218 
1219   // metadata in code that we have to keep track of
1220   AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
1221   AddressLiteral constant_metadata_address(Metadata* obj); // find_index
1222   inline void    set_metadata             (Metadata* obj, Register d);              // uses allocate_metadata_address
1223   inline void    set_metadata_constant    (Metadata* obj, Register d);              // uses constant_metadata_address
1224   inline void    set_metadata             (const AddressLiteral& obj_addr, Register d); // same as load_address
1225 
1226   void set_narrow_oop( jobject obj, Register d );
1227   void set_narrow_klass( Klass* k, Register d );
1228 
1229   // nop padding
1230   void align(int modulus);
1231 
1232   // declare a safepoint
1233   void safepoint();
1234 
1235   // factor out part of stop into subroutine to save space
1236   void stop_subroutine();
1237   // factor out part of verify_oop into subroutine to save space
1238   void verify_oop_subroutine();
1239 
1240   // side-door communication with signalHandler in os_solaris.cpp
1241   static address _verify_oop_implicit_branch[3];
1242 
1243   int total_frame_size_in_bytes(int extraWords);
1244 
1245   // used when extraWords known statically
1246   void save_frame(int extraWords = 0);
1247   void save_frame_c1(int size_in_bytes);
1248   // make a frame, and simultaneously pass up one or two register value
1249   // into the new register window
1250   void save_frame_and_mov(int extraWords, Register s1, Register d1, Register s2 = Register(), Register d2 = Register());
1251 
1252   // give no. (outgoing) params, calc # of words will need on frame
1253   void calc_mem_param_words(Register Rparam_words, Register Rresult);
1254 
1255   // used to calculate frame size dynamically
1256   // result is in bytes and must be negated for save inst
1257   void calc_frame_size(Register extraWords, Register resultReg);
1258 
1259   // calc and also save
1260   void calc_frame_size_and_save(Register extraWords, Register resultReg);
1261 
1262   static void debug(char* msg, RegistersForDebugging* outWindow);
1263 
1264   // implementations of bytecodes used by both interpreter and compiler
1265 
1266   void lcmp( Register Ra_hi, Register Ra_low,
1267              Register Rb_hi, Register Rb_low,
1268              Register Rresult);
1269 
1270   void lneg( Register Rhi, Register Rlow );
1271 
1272   void lshl(  Register Rin_high,  Register Rin_low,  Register Rcount,
1273               Register Rout_high, Register Rout_low, Register Rtemp );
1274 
1275   void lshr(  Register Rin_high,  Register Rin_low,  Register Rcount,
1276               Register Rout_high, Register Rout_low, Register Rtemp );
1277 
1278   void lushr( Register Rin_high,  Register Rin_low,  Register Rcount,
1279               Register Rout_high, Register Rout_low, Register Rtemp );
1280 
1281 #ifdef _LP64
1282   void lcmp( Register Ra, Register Rb, Register Rresult);
1283 #endif
1284 
1285   // Load and store values by size and signed-ness
1286   void load_sized_value( Address src, Register dst, size_t size_in_bytes, bool is_signed);
1287   void store_sized_value(Register src, Address dst, size_t size_in_bytes);
1288 
1289   void float_cmp( bool is_float, int unordered_result,
1290                   FloatRegister Fa, FloatRegister Fb,
1291                   Register Rresult);
1292 
1293   void save_all_globals_into_locals();
1294   void restore_globals_from_locals();
1295 
1296   // These set the icc condition code to equal if the lock succeeded
1297   // and notEqual if it failed and requires a slow case
1298   void compiler_lock_object(Register Roop, Register Rmark, Register Rbox,
1299                             Register Rscratch,
1300                             BiasedLockingCounters* counters = NULL,
1301                             bool try_bias = UseBiasedLocking);
1302   void compiler_unlock_object(Register Roop, Register Rmark, Register Rbox,
1303                               Register Rscratch,
1304                               bool try_bias = UseBiasedLocking);
1305 
1306   // Biased locking support
1307   // Upon entry, lock_reg must point to the lock record on the stack,
1308   // obj_reg must contain the target object, and mark_reg must contain
1309   // the target object's header.
1310   // Destroys mark_reg if an attempt is made to bias an anonymously
1311   // biased lock. In this case a failure will go either to the slow
1312   // case or fall through with the notEqual condition code set with
1313   // the expectation that the slow case in the runtime will be called.
1314   // In the fall-through case where the CAS-based lock is done,
1315   // mark_reg is not destroyed.
1316   void biased_locking_enter(Register obj_reg, Register mark_reg, Register temp_reg,
1317                             Label& done, Label* slow_case = NULL,
1318                             BiasedLockingCounters* counters = NULL);
1319   // Upon entry, the base register of mark_addr must contain the oop.
1320   // Destroys temp_reg.
1321 
1322   // If allow_delay_slot_filling is set to true, the next instruction
1323   // emitted after this one will go in an annulled delay slot if the
1324   // biased locking exit case failed.
1325   void biased_locking_exit(Address mark_addr, Register temp_reg, Label& done, bool allow_delay_slot_filling = false);
1326 
1327   // allocation
1328   void eden_allocate(
1329     Register obj,                      // result: pointer to object after successful allocation
1330     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1331     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1332     Register t1,                       // temp register
1333     Register t2,                       // temp register
1334     Label&   slow_case                 // continuation point if fast allocation fails
1335   );
1336   void tlab_allocate(
1337     Register obj,                      // result: pointer to object after successful allocation
1338     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1339     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1340     Register t1,                       // temp register
1341     Label&   slow_case                 // continuation point if fast allocation fails
1342   );
1343   void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
1344   void incr_allocated_bytes(RegisterOrConstant size_in_bytes,
1345                             Register t1, Register t2);
1346 
1347   // interface method calling
1348   void lookup_interface_method(Register recv_klass,
1349                                Register intf_klass,
1350                                RegisterOrConstant itable_index,
1351                                Register method_result,
1352                                Register temp_reg, Register temp2_reg,
1353                                Label& no_such_interface);
1354 
1355   // virtual method calling
1356   void lookup_virtual_method(Register recv_klass,
1357                              RegisterOrConstant vtable_index,
1358                              Register method_result);
1359 
1360   // Test sub_klass against super_klass, with fast and slow paths.
1361 
1362   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1363   // One of the three labels can be NULL, meaning take the fall-through.
1364   // If super_check_offset is -1, the value is loaded up from super_klass.
1365   // No registers are killed, except temp_reg and temp2_reg.
1366   // If super_check_offset is not -1, temp2_reg is not used and can be noreg.
1367   void check_klass_subtype_fast_path(Register sub_klass,
1368                                      Register super_klass,
1369                                      Register temp_reg,
1370                                      Register temp2_reg,
1371                                      Label* L_success,
1372                                      Label* L_failure,
1373                                      Label* L_slow_path,
1374                 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
1375 
1376   // The rest of the type check; must be wired to a corresponding fast path.
1377   // It does not repeat the fast path logic, so don't use it standalone.
1378   // The temp_reg can be noreg, if no temps are available.
1379   // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
1380   // Updates the sub's secondary super cache as necessary.
1381   void check_klass_subtype_slow_path(Register sub_klass,
1382                                      Register super_klass,
1383                                      Register temp_reg,
1384                                      Register temp2_reg,
1385                                      Register temp3_reg,
1386                                      Register temp4_reg,
1387                                      Label* L_success,
1388                                      Label* L_failure);
1389 
1390   // Simplified, combined version, good for typical uses.
1391   // Falls through on failure.
1392   void check_klass_subtype(Register sub_klass,
1393                            Register super_klass,
1394                            Register temp_reg,
1395                            Register temp2_reg,
1396                            Label& L_success);
1397 
1398   // method handles (JSR 292)
1399   // offset relative to Gargs of argument at tos[arg_slot].
1400   // (arg_slot == 0 means the last argument, not the first).
1401   RegisterOrConstant argument_offset(RegisterOrConstant arg_slot,
1402                                      Register temp_reg,
1403                                      int extra_slot_offset = 0);
1404   // Address of Gargs and argument_offset.
1405   Address            argument_address(RegisterOrConstant arg_slot,
1406                                       Register temp_reg = noreg,
1407                                       int extra_slot_offset = 0);
1408 
1409   // Stack overflow checking
1410 
1411   // Note: this clobbers G3_scratch
1412   void bang_stack_with_offset(int offset) {
1413     // stack grows down, caller passes positive offset
1414     assert(offset > 0, "must bang with negative offset");
1415     set((-offset)+STACK_BIAS, G3_scratch);
1416     st(G0, SP, G3_scratch);
1417   }
1418 
1419   // Writes to stack successive pages until offset reached to check for
1420   // stack overflow + shadow pages.  Clobbers tsp and scratch registers.
1421   void bang_stack_size(Register Rsize, Register Rtsp, Register Rscratch);
1422 
1423   virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset);
1424 
1425   void verify_tlab();
1426 
1427   Condition negate_condition(Condition cond);
1428 
1429   // Helper functions for statistics gathering.
1430   // Conditionally (non-atomically) increments passed counter address, preserving condition codes.
1431   void cond_inc(Condition cond, address counter_addr, Register Rtemp1, Register Rtemp2);
1432   // Unconditional increment.
1433   void inc_counter(address counter_addr, Register Rtmp1, Register Rtmp2);
1434   void inc_counter(int*    counter_addr, Register Rtmp1, Register Rtmp2);
1435 
1436   // Compare char[] arrays aligned to 4 bytes.
1437   void char_arrays_equals(Register ary1, Register ary2,
1438                           Register limit, Register result,
1439                           Register chr1, Register chr2, Label& Ldone);
1440   // Use BIS for zeroing
1441   void bis_zeroing(Register to, Register count, Register temp, Label& Ldone);
1442 
1443   // Update CRC-32[C] with a byte value according to constants in table
1444   void update_byte_crc32(Register crc, Register val, Register table);
1445 
1446   // Reverse byte order of lower 32 bits, assuming upper 32 bits all zeros
1447   void reverse_bytes_32(Register src, Register dst, Register tmp);
1448   void movitof_revbytes(Register src, FloatRegister dst, Register tmp1, Register tmp2);
1449   void movftoi_revbytes(FloatRegister src, Register dst, Register tmp1, Register tmp2);
1450 
1451 #undef VIRTUAL
1452 };
1453 
1454 /**
1455  * class SkipIfEqual:
1456  *
1457  * Instantiating this class will result in assembly code being output that will
1458  * jump around any code emitted between the creation of the instance and it's
1459  * automatic destruction at the end of a scope block, depending on the value of
1460  * the flag passed to the constructor, which will be checked at run-time.
1461  */
1462 class SkipIfEqual : public StackObj {
1463  private:
1464   MacroAssembler* _masm;
1465   Label _label;
1466 
1467  public:
1468    // 'temp' is a temp register that this object can use (and trash)
1469    SkipIfEqual(MacroAssembler*, Register temp,
1470                const bool* flag_addr, Assembler::Condition condition);
1471    ~SkipIfEqual();
1472 };
1473 
1474 #endif // CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP