1 /*
   2  * Copyright 1997-2008 Sun Microsystems, Inc.  All Rights Reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 #include "incls/_precompiled.incl"
  26 #include "incls/_assembler_sparc.cpp.incl"
  27 
  28 // Implementation of Address
  29 
  30 Address::Address( addr_type t, int which ) {
  31   switch (t) {
  32    case extra_in_argument:
  33    case extra_out_argument:
  34      _base = t == extra_in_argument ? FP : SP;
  35      _hi   = 0;
  36 // Warning:  In LP64 mode, _disp will occupy more than 10 bits.
  37 //           This is inconsistent with the other constructors but op
  38 //           codes such as ld or ldx, only access disp() to get their
  39 //           simm13 argument.
  40      _disp = ((which - Argument::n_register_parameters + frame::memory_parameter_word_sp_offset) * BytesPerWord) + STACK_BIAS;
  41     break;
  42    default:
  43     ShouldNotReachHere();
  44     break;
  45   }
  46 }
  47 
  48 static const char* argumentNames[][2] = {
  49   {"A0","P0"}, {"A1","P1"}, {"A2","P2"}, {"A3","P3"}, {"A4","P4"},
  50   {"A5","P5"}, {"A6","P6"}, {"A7","P7"}, {"A8","P8"}, {"A9","P9"},
  51   {"A(n>9)","P(n>9)"}
  52 };
  53 
  54 const char* Argument::name() const {
  55   int nofArgs = sizeof argumentNames / sizeof argumentNames[0];
  56   int num = number();
  57   if (num >= nofArgs)  num = nofArgs - 1;
  58   return argumentNames[num][is_in() ? 1 : 0];
  59 }
  60 
  61 void Assembler::print_instruction(int inst) {
  62   const char* s;
  63   switch (inv_op(inst)) {
  64   default:         s = "????"; break;
  65   case call_op:    s = "call"; break;
  66   case branch_op:
  67     switch (inv_op2(inst)) {
  68       case bpr_op2:    s = "bpr";  break;
  69       case fb_op2:     s = "fb";   break;
  70       case fbp_op2:    s = "fbp";  break;
  71       case br_op2:     s = "br";   break;
  72       case bp_op2:     s = "bp";   break;
  73       case cb_op2:     s = "cb";   break;
  74       default:         s = "????"; break;
  75     }
  76   }
  77   ::tty->print("%s", s);
  78 }
  79 
  80 
  81 // Patch instruction inst at offset inst_pos to refer to dest_pos
  82 // and return the resulting instruction.
  83 // We should have pcs, not offsets, but since all is relative, it will work out
  84 // OK.
  85 int Assembler::patched_branch(int dest_pos, int inst, int inst_pos) {
  86 
  87   int m; // mask for displacement field
  88   int v; // new value for displacement field
  89   const int word_aligned_ones = -4;
  90   switch (inv_op(inst)) {
  91   default: ShouldNotReachHere();
  92   case call_op:    m = wdisp(word_aligned_ones, 0, 30);  v = wdisp(dest_pos, inst_pos, 30); break;
  93   case branch_op:
  94     switch (inv_op2(inst)) {
  95       case bpr_op2:    m = wdisp16(word_aligned_ones, 0);      v = wdisp16(dest_pos, inst_pos);     break;
  96       case fbp_op2:    m = wdisp(  word_aligned_ones, 0, 19);  v = wdisp(  dest_pos, inst_pos, 19); break;
  97       case bp_op2:     m = wdisp(  word_aligned_ones, 0, 19);  v = wdisp(  dest_pos, inst_pos, 19); break;
  98       case fb_op2:     m = wdisp(  word_aligned_ones, 0, 22);  v = wdisp(  dest_pos, inst_pos, 22); break;
  99       case br_op2:     m = wdisp(  word_aligned_ones, 0, 22);  v = wdisp(  dest_pos, inst_pos, 22); break;
 100       case cb_op2:     m = wdisp(  word_aligned_ones, 0, 22);  v = wdisp(  dest_pos, inst_pos, 22); break;
 101       default: ShouldNotReachHere();
 102     }
 103   }
 104   return  inst & ~m  |  v;
 105 }
 106 
 107 // Return the offset of the branch destionation of instruction inst
 108 // at offset pos.
 109 // Should have pcs, but since all is relative, it works out.
 110 int Assembler::branch_destination(int inst, int pos) {
 111   int r;
 112   switch (inv_op(inst)) {
 113   default: ShouldNotReachHere();
 114   case call_op:        r = inv_wdisp(inst, pos, 30);  break;
 115   case branch_op:
 116     switch (inv_op2(inst)) {
 117       case bpr_op2:    r = inv_wdisp16(inst, pos);    break;
 118       case fbp_op2:    r = inv_wdisp(  inst, pos, 19);  break;
 119       case bp_op2:     r = inv_wdisp(  inst, pos, 19);  break;
 120       case fb_op2:     r = inv_wdisp(  inst, pos, 22);  break;
 121       case br_op2:     r = inv_wdisp(  inst, pos, 22);  break;
 122       case cb_op2:     r = inv_wdisp(  inst, pos, 22);  break;
 123       default: ShouldNotReachHere();
 124     }
 125   }
 126   return r;
 127 }
 128 
 129 int AbstractAssembler::code_fill_byte() {
 130   return 0x00;                  // illegal instruction 0x00000000
 131 }
 132 
 133 Assembler::Condition Assembler::reg_cond_to_cc_cond(Assembler::RCondition in) {
 134   switch (in) {
 135   case rc_z:   return equal;
 136   case rc_lez: return lessEqual;
 137   case rc_lz:  return less;
 138   case rc_nz:  return notEqual;
 139   case rc_gz:  return greater;
 140   case rc_gez: return greaterEqual;
 141   default:
 142     ShouldNotReachHere();
 143   }
 144   return equal;
 145 }
 146 
 147 // Generate a bunch 'o stuff (including v9's
 148 #ifndef PRODUCT
 149 void Assembler::test_v9() {
 150   add(    G0, G1, G2 );
 151   add(    G3,  0, G4 );
 152 
 153   addcc(  G5, G6, G7 );
 154   addcc(  I0,  1, I1 );
 155   addc(   I2, I3, I4 );
 156   addc(   I5, -1, I6 );
 157   addccc( I7, L0, L1 );
 158   addccc( L2, (1 << 12) - 2, L3 );
 159 
 160   Label lbl1, lbl2, lbl3;
 161 
 162   bind(lbl1);
 163 
 164   bpr( rc_z,    true, pn, L4, pc(),  relocInfo::oop_type );
 165   delayed()->nop();
 166   bpr( rc_lez, false, pt, L5, lbl1);
 167   delayed()->nop();
 168 
 169   fb( f_never,     true, pc() + 4,  relocInfo::none);
 170   delayed()->nop();
 171   fb( f_notEqual, false, lbl2 );
 172   delayed()->nop();
 173 
 174   fbp( f_notZero,        true, fcc0, pn, pc() - 4,  relocInfo::none);
 175   delayed()->nop();
 176   fbp( f_lessOrGreater, false, fcc1, pt, lbl3 );
 177   delayed()->nop();
 178 
 179   br( equal,  true, pc() + 1024, relocInfo::none);
 180   delayed()->nop();
 181   br( lessEqual, false, lbl1 );
 182   delayed()->nop();
 183   br( never, false, lbl1 );
 184   delayed()->nop();
 185 
 186   bp( less,               true, icc, pn, pc(), relocInfo::none);
 187   delayed()->nop();
 188   bp( lessEqualUnsigned, false, xcc, pt, lbl2 );
 189   delayed()->nop();
 190 
 191   call( pc(), relocInfo::none);
 192   delayed()->nop();
 193   call( lbl3 );
 194   delayed()->nop();
 195 
 196 
 197   casa(  L6, L7, O0 );
 198   casxa( O1, O2, O3, 0 );
 199 
 200   udiv(   O4, O5, O7 );
 201   udiv(   G0, (1 << 12) - 1, G1 );
 202   sdiv(   G1, G2, G3 );
 203   sdiv(   G4, -((1 << 12) - 1), G5 );
 204   udivcc( G6, G7, I0 );
 205   udivcc( I1, -((1 << 12) - 2), I2 );
 206   sdivcc( I3, I4, I5 );
 207   sdivcc( I6, -((1 << 12) - 0), I7 );
 208 
 209   done();
 210   retry();
 211 
 212   fadd( FloatRegisterImpl::S, F0,  F1, F2 );
 213   fsub( FloatRegisterImpl::D, F34, F0, F62 );
 214 
 215   fcmp(  FloatRegisterImpl::Q, fcc0, F0, F60);
 216   fcmpe( FloatRegisterImpl::S, fcc1, F31, F30);
 217 
 218   ftox( FloatRegisterImpl::D, F2, F4 );
 219   ftoi( FloatRegisterImpl::Q, F4, F8 );
 220 
 221   ftof( FloatRegisterImpl::S, FloatRegisterImpl::Q, F3, F12 );
 222 
 223   fxtof( FloatRegisterImpl::S, F4, F5 );
 224   fitof( FloatRegisterImpl::D, F6, F8 );
 225 
 226   fmov( FloatRegisterImpl::Q, F16, F20 );
 227   fneg( FloatRegisterImpl::S, F6, F7 );
 228   fabs( FloatRegisterImpl::D, F10, F12 );
 229 
 230   fmul( FloatRegisterImpl::Q,  F24, F28, F32 );
 231   fmul( FloatRegisterImpl::S,  FloatRegisterImpl::D,  F8, F9, F14 );
 232   fdiv( FloatRegisterImpl::S,  F10, F11, F12 );
 233 
 234   fsqrt( FloatRegisterImpl::S, F13, F14 );
 235 
 236   flush( L0, L1 );
 237   flush( L2, -1 );
 238 
 239   flushw();
 240 
 241   illtrap( (1 << 22) - 2);
 242 
 243   impdep1( 17, (1 << 19) - 1 );
 244   impdep2( 3,  0 );
 245 
 246   jmpl( L3, L4, L5 );
 247   delayed()->nop();
 248   jmpl( L6, -1, L7, Relocation::spec_simple(relocInfo::none));
 249   delayed()->nop();
 250 
 251 
 252   ldf(    FloatRegisterImpl::S, O0, O1, F15 );
 253   ldf(    FloatRegisterImpl::D, O2, -1, F14 );
 254 
 255 
 256   ldfsr(  O3, O4 );
 257   ldfsr(  O5, -1 );
 258   ldxfsr( O6, O7 );
 259   ldxfsr( I0, -1 );
 260 
 261   ldfa(  FloatRegisterImpl::D, I1, I2, 1, F16 );
 262   ldfa(  FloatRegisterImpl::Q, I3, -1,    F36 );
 263 
 264   ldsb(  I4, I5, I6 );
 265   ldsb(  I7, -1, G0 );
 266   ldsh(  G1, G3, G4 );
 267   ldsh(  G5, -1, G6 );
 268   ldsw(  G7, L0, L1 );
 269   ldsw(  L2, -1, L3 );
 270   ldub(  L4, L5, L6 );
 271   ldub(  L7, -1, O0 );
 272   lduh(  O1, O2, O3 );
 273   lduh(  O4, -1, O5 );
 274   lduw(  O6, O7, G0 );
 275   lduw(  G1, -1, G2 );
 276   ldx(   G3, G4, G5 );
 277   ldx(   G6, -1, G7 );
 278   ldd(   I0, I1, I2 );
 279   ldd(   I3, -1, I4 );
 280 
 281   ldsba(  I5, I6, 2, I7 );
 282   ldsba(  L0, -1, L1 );
 283   ldsha(  L2, L3, 3, L4 );
 284   ldsha(  L5, -1, L6 );
 285   ldswa(  L7, O0, (1 << 8) - 1, O1 );
 286   ldswa(  O2, -1, O3 );
 287   lduba(  O4, O5, 0, O6 );
 288   lduba(  O7, -1, I0 );
 289   lduha(  I1, I2, 1, I3 );
 290   lduha(  I4, -1, I5 );
 291   lduwa(  I6, I7, 2, L0 );
 292   lduwa(  L1, -1, L2 );
 293   ldxa(   L3, L4, 3, L5 );
 294   ldxa(   L6, -1, L7 );
 295   ldda(   G0, G1, 4, G2 );
 296   ldda(   G3, -1, G4 );
 297 
 298   ldstub(  G5, G6, G7 );
 299   ldstub(  O0, -1, O1 );
 300 
 301   ldstuba( O2, O3, 5, O4 );
 302   ldstuba( O5, -1, O6 );
 303 
 304   and3(    I0, L0, O0 );
 305   and3(    G7, -1, O7 );
 306   andcc(   L2, I2, G2 );
 307   andcc(   L4, -1, G4 );
 308   andn(    I5, I6, I7 );
 309   andn(    I6, -1, I7 );
 310   andncc(  I5, I6, I7 );
 311   andncc(  I7, -1, I6 );
 312   or3(     I5, I6, I7 );
 313   or3(     I7, -1, I6 );
 314   orcc(    I5, I6, I7 );
 315   orcc(    I7, -1, I6 );
 316   orn(     I5, I6, I7 );
 317   orn(     I7, -1, I6 );
 318   orncc(   I5, I6, I7 );
 319   orncc(   I7, -1, I6 );
 320   xor3(    I5, I6, I7 );
 321   xor3(    I7, -1, I6 );
 322   xorcc(   I5, I6, I7 );
 323   xorcc(   I7, -1, I6 );
 324   xnor(    I5, I6, I7 );
 325   xnor(    I7, -1, I6 );
 326   xnorcc(  I5, I6, I7 );
 327   xnorcc(  I7, -1, I6 );
 328 
 329   membar( Membar_mask_bits(StoreStore | LoadStore | StoreLoad | LoadLoad | Sync | MemIssue | Lookaside ) );
 330   membar( StoreStore );
 331   membar( LoadStore );
 332   membar( StoreLoad );
 333   membar( LoadLoad );
 334   membar( Sync );
 335   membar( MemIssue );
 336   membar( Lookaside );
 337 
 338   fmov( FloatRegisterImpl::S, f_ordered,  true, fcc2, F16, F17 );
 339   fmov( FloatRegisterImpl::D, rc_lz, L5, F18, F20 );
 340 
 341   movcc( overflowClear,  false, icc, I6, L4 );
 342   movcc( f_unorderedOrEqual, true, fcc2, (1 << 10) - 1, O0 );
 343 
 344   movr( rc_nz, I5, I6, I7 );
 345   movr( rc_gz, L1, -1,  L2 );
 346 
 347   mulx(  I5, I6, I7 );
 348   mulx(  I7, -1, I6 );
 349   sdivx( I5, I6, I7 );
 350   sdivx( I7, -1, I6 );
 351   udivx( I5, I6, I7 );
 352   udivx( I7, -1, I6 );
 353 
 354   umul(   I5, I6, I7 );
 355   umul(   I7, -1, I6 );
 356   smul(   I5, I6, I7 );
 357   smul(   I7, -1, I6 );
 358   umulcc( I5, I6, I7 );
 359   umulcc( I7, -1, I6 );
 360   smulcc( I5, I6, I7 );
 361   smulcc( I7, -1, I6 );
 362 
 363   mulscc(   I5, I6, I7 );
 364   mulscc(   I7, -1, I6 );
 365 
 366   nop();
 367 
 368 
 369   popc( G0,  G1);
 370   popc( -1, G2);
 371 
 372   prefetch(   L1, L2,    severalReads );
 373   prefetch(   L3, -1,    oneRead );
 374   prefetcha(  O3, O2, 6, severalWritesAndPossiblyReads );
 375   prefetcha(  G2, -1,    oneWrite );
 376 
 377   rett( I7, I7);
 378   delayed()->nop();
 379   rett( G0, -1, relocInfo::none);
 380   delayed()->nop();
 381 
 382   save(    I5, I6, I7 );
 383   save(    I7, -1, I6 );
 384   restore( I5, I6, I7 );
 385   restore( I7, -1, I6 );
 386 
 387   saved();
 388   restored();
 389 
 390   sethi( 0xaaaaaaaa, I3, Relocation::spec_simple(relocInfo::none));
 391 
 392   sll(  I5, I6, I7 );
 393   sll(  I7, 31, I6 );
 394   srl(  I5, I6, I7 );
 395   srl(  I7,  0, I6 );
 396   sra(  I5, I6, I7 );
 397   sra(  I7, 30, I6 );
 398   sllx( I5, I6, I7 );
 399   sllx( I7, 63, I6 );
 400   srlx( I5, I6, I7 );
 401   srlx( I7,  0, I6 );
 402   srax( I5, I6, I7 );
 403   srax( I7, 62, I6 );
 404 
 405   sir( -1 );
 406 
 407   stbar();
 408 
 409   stf(    FloatRegisterImpl::Q, F40, G0, I7 );
 410   stf(    FloatRegisterImpl::S, F18, I3, -1 );
 411 
 412   stfsr(  L1, L2 );
 413   stfsr(  I7, -1 );
 414   stxfsr( I6, I5 );
 415   stxfsr( L4, -1 );
 416 
 417   stfa(  FloatRegisterImpl::D, F22, I6, I7, 7 );
 418   stfa(  FloatRegisterImpl::Q, F44, G0, -1 );
 419 
 420   stb(  L5, O2, I7 );
 421   stb(  I7, I6, -1 );
 422   sth(  L5, O2, I7 );
 423   sth(  I7, I6, -1 );
 424   stw(  L5, O2, I7 );
 425   stw(  I7, I6, -1 );
 426   stx(  L5, O2, I7 );
 427   stx(  I7, I6, -1 );
 428   std(  L5, O2, I7 );
 429   std(  I7, I6, -1 );
 430 
 431   stba(  L5, O2, I7, 8 );
 432   stba(  I7, I6, -1    );
 433   stha(  L5, O2, I7, 9 );
 434   stha(  I7, I6, -1    );
 435   stwa(  L5, O2, I7, 0 );
 436   stwa(  I7, I6, -1    );
 437   stxa(  L5, O2, I7, 11 );
 438   stxa(  I7, I6, -1     );
 439   stda(  L5, O2, I7, 12 );
 440   stda(  I7, I6, -1     );
 441 
 442   sub(    I5, I6, I7 );
 443   sub(    I7, -1, I6 );
 444   subcc(  I5, I6, I7 );
 445   subcc(  I7, -1, I6 );
 446   subc(   I5, I6, I7 );
 447   subc(   I7, -1, I6 );
 448   subccc( I5, I6, I7 );
 449   subccc( I7, -1, I6 );
 450 
 451   swap( I5, I6, I7 );
 452   swap( I7, -1, I6 );
 453 
 454   swapa(   G0, G1, 13, G2 );
 455   swapa(   I7, -1,     I6 );
 456 
 457   taddcc(    I5, I6, I7 );
 458   taddcc(    I7, -1, I6 );
 459   taddcctv(  I5, I6, I7 );
 460   taddcctv(  I7, -1, I6 );
 461 
 462   tsubcc(    I5, I6, I7 );
 463   tsubcc(    I7, -1, I6 );
 464   tsubcctv(  I5, I6, I7 );
 465   tsubcctv(  I7, -1, I6 );
 466 
 467   trap( overflowClear, xcc, G0, G1 );
 468   trap( lessEqual,     icc, I7, 17 );
 469 
 470   bind(lbl2);
 471   bind(lbl3);
 472 
 473   code()->decode();
 474 }
 475 
 476 // Generate a bunch 'o stuff unique to V8
 477 void Assembler::test_v8_onlys() {
 478   Label lbl1;
 479 
 480   cb( cp_0or1or2, false, pc() - 4, relocInfo::none);
 481   delayed()->nop();
 482   cb( cp_never,    true, lbl1);
 483   delayed()->nop();
 484 
 485   cpop1(1, 2, 3, 4);
 486   cpop2(5, 6, 7, 8);
 487 
 488   ldc( I0, I1, 31);
 489   ldc( I2, -1,  0);
 490 
 491   lddc( I4, I4, 30);
 492   lddc( I6,  0, 1 );
 493 
 494   ldcsr( L0, L1, 0);
 495   ldcsr( L1, (1 << 12) - 1, 17 );
 496 
 497   stc( 31, L4, L5);
 498   stc( 30, L6, -(1 << 12) );
 499 
 500   stdc( 0, L7, G0);
 501   stdc( 1, G1, 0 );
 502 
 503   stcsr( 16, G2, G3);
 504   stcsr( 17, G4, 1 );
 505 
 506   stdcq( 4, G5, G6);
 507   stdcq( 5, G7, -1 );
 508 
 509   bind(lbl1);
 510 
 511   code()->decode();
 512 }
 513 #endif
 514 
 515 // Implementation of MacroAssembler
 516 
 517 void MacroAssembler::null_check(Register reg, int offset) {
 518   if (needs_explicit_null_check((intptr_t)offset)) {
 519     // provoke OS NULL exception if reg = NULL by
 520     // accessing M[reg] w/o changing any registers
 521     ld_ptr(reg, 0, G0);
 522   }
 523   else {
 524     // nothing to do, (later) access of M[reg + offset]
 525     // will provoke OS NULL exception if reg = NULL
 526   }
 527 }
 528 
 529 // Ring buffer jumps
 530 
 531 #ifndef PRODUCT
 532 void MacroAssembler::ret(  bool trace )   { if (trace) {
 533                                                     mov(I7, O7); // traceable register
 534                                                     JMP(O7, 2 * BytesPerInstWord);
 535                                                   } else {
 536                                                     jmpl( I7, 2 * BytesPerInstWord, G0 );
 537                                                   }
 538                                                 }
 539 
 540 void MacroAssembler::retl( bool trace )  { if (trace) JMP(O7, 2 * BytesPerInstWord);
 541                                                  else jmpl( O7, 2 * BytesPerInstWord, G0 ); }
 542 #endif /* PRODUCT */
 543 
 544 
 545 void MacroAssembler::jmp2(Register r1, Register r2, const char* file, int line ) {
 546   assert_not_delayed();
 547   // This can only be traceable if r1 & r2 are visible after a window save
 548   if (TraceJumps) {
 549 #ifndef PRODUCT
 550     save_frame(0);
 551     verify_thread();
 552     ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0);
 553     add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1);
 554     sll(O0, exact_log2(4*sizeof(intptr_t)), O2);
 555     add(O2, O1, O1);
 556 
 557     add(r1->after_save(), r2->after_save(), O2);
 558     set((intptr_t)file, O3);
 559     set(line, O4);
 560     Label L;
 561     // get nearby pc, store jmp target
 562     call(L, relocInfo::none);  // No relocation for call to pc+0x8
 563     delayed()->st(O2, O1, 0);
 564     bind(L);
 565 
 566     // store nearby pc
 567     st(O7, O1, sizeof(intptr_t));
 568     // store file
 569     st(O3, O1, 2*sizeof(intptr_t));
 570     // store line
 571     st(O4, O1, 3*sizeof(intptr_t));
 572     add(O0, 1, O0);
 573     and3(O0, JavaThread::jump_ring_buffer_size  - 1, O0);
 574     st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()));
 575     restore();
 576 #endif /* PRODUCT */
 577   }
 578   jmpl(r1, r2, G0);
 579 }
 580 void MacroAssembler::jmp(Register r1, int offset, const char* file, int line ) {
 581   assert_not_delayed();
 582   // This can only be traceable if r1 is visible after a window save
 583   if (TraceJumps) {
 584 #ifndef PRODUCT
 585     save_frame(0);
 586     verify_thread();
 587     ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0);
 588     add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1);
 589     sll(O0, exact_log2(4*sizeof(intptr_t)), O2);
 590     add(O2, O1, O1);
 591 
 592     add(r1->after_save(), offset, O2);
 593     set((intptr_t)file, O3);
 594     set(line, O4);
 595     Label L;
 596     // get nearby pc, store jmp target
 597     call(L, relocInfo::none);  // No relocation for call to pc+0x8
 598     delayed()->st(O2, O1, 0);
 599     bind(L);
 600 
 601     // store nearby pc
 602     st(O7, O1, sizeof(intptr_t));
 603     // store file
 604     st(O3, O1, 2*sizeof(intptr_t));
 605     // store line
 606     st(O4, O1, 3*sizeof(intptr_t));
 607     add(O0, 1, O0);
 608     and3(O0, JavaThread::jump_ring_buffer_size  - 1, O0);
 609     st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()));
 610     restore();
 611 #endif /* PRODUCT */
 612   }
 613   jmp(r1, offset);
 614 }
 615 
 616 // This code sequence is relocatable to any address, even on LP64.
 617 void MacroAssembler::jumpl( Address& a, Register d, int offset, const char* file, int line ) {
 618   assert_not_delayed();
 619   // Force fixed length sethi because NativeJump and NativeFarCall don't handle
 620   // variable length instruction streams.
 621   sethi(a, /*ForceRelocatable=*/ true);
 622   if (TraceJumps) {
 623 #ifndef PRODUCT
 624     // Must do the add here so relocation can find the remainder of the
 625     // value to be relocated.
 626     add(a.base(), a.disp() + offset, a.base(), a.rspec(offset));
 627     save_frame(0);
 628     verify_thread();
 629     ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0);
 630     add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1);
 631     sll(O0, exact_log2(4*sizeof(intptr_t)), O2);
 632     add(O2, O1, O1);
 633 
 634     set((intptr_t)file, O3);
 635     set(line, O4);
 636     Label L;
 637 
 638     // get nearby pc, store jmp target
 639     call(L, relocInfo::none);  // No relocation for call to pc+0x8
 640     delayed()->st(a.base()->after_save(), O1, 0);
 641     bind(L);
 642 
 643     // store nearby pc
 644     st(O7, O1, sizeof(intptr_t));
 645     // store file
 646     st(O3, O1, 2*sizeof(intptr_t));
 647     // store line
 648     st(O4, O1, 3*sizeof(intptr_t));
 649     add(O0, 1, O0);
 650     and3(O0, JavaThread::jump_ring_buffer_size  - 1, O0);
 651     st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()));
 652     restore();
 653     jmpl(a.base(), G0, d);
 654 #else
 655     jmpl(a, d, offset);
 656 #endif /* PRODUCT */
 657   } else {
 658     jmpl(a, d, offset);
 659   }
 660 }
 661 
 662 void MacroAssembler::jump( Address& a, int offset, const char* file, int line ) {
 663   jumpl( a, G0, offset, file, line );
 664 }
 665 
 666 
 667 // Convert to C varargs format
 668 void MacroAssembler::set_varargs( Argument inArg, Register d ) {
 669   // spill register-resident args to their memory slots
 670   // (SPARC calling convention requires callers to have already preallocated these)
 671   // Note that the inArg might in fact be an outgoing argument,
 672   // if a leaf routine or stub does some tricky argument shuffling.
 673   // This routine must work even though one of the saved arguments
 674   // is in the d register (e.g., set_varargs(Argument(0, false), O0)).
 675   for (Argument savePtr = inArg;
 676        savePtr.is_register();
 677        savePtr = savePtr.successor()) {
 678     st_ptr(savePtr.as_register(), savePtr.address_in_frame());
 679   }
 680   // return the address of the first memory slot
 681   add(inArg.address_in_frame(), d);
 682 }
 683 
 684 // Conditional breakpoint (for assertion checks in assembly code)
 685 void MacroAssembler::breakpoint_trap(Condition c, CC cc) {
 686   trap(c, cc, G0, ST_RESERVED_FOR_USER_0);
 687 }
 688 
 689 // We want to use ST_BREAKPOINT here, but the debugger is confused by it.
 690 void MacroAssembler::breakpoint_trap() {
 691   trap(ST_RESERVED_FOR_USER_0);
 692 }
 693 
 694 // flush windows (except current) using flushw instruction if avail.
 695 void MacroAssembler::flush_windows() {
 696   if (VM_Version::v9_instructions_work())  flushw();
 697   else                                     flush_windows_trap();
 698 }
 699 
 700 // Write serialization page so VM thread can do a pseudo remote membar
 701 // We use the current thread pointer to calculate a thread specific
 702 // offset to write to within the page. This minimizes bus traffic
 703 // due to cache line collision.
 704 void MacroAssembler::serialize_memory(Register thread, Register tmp1, Register tmp2) {
 705   Address mem_serialize_page(tmp1, os::get_memory_serialize_page());
 706   srl(thread, os::get_serialize_page_shift_count(), tmp2);
 707   if (Assembler::is_simm13(os::vm_page_size())) {
 708     and3(tmp2, (os::vm_page_size() - sizeof(int)), tmp2);
 709   }
 710   else {
 711     set((os::vm_page_size() - sizeof(int)), tmp1);
 712     and3(tmp2, tmp1, tmp2);
 713   }
 714   load_address(mem_serialize_page);
 715   st(G0, tmp1, tmp2);
 716 }
 717 
 718 
 719 
 720 void MacroAssembler::enter() {
 721   Unimplemented();
 722 }
 723 
 724 void MacroAssembler::leave() {
 725   Unimplemented();
 726 }
 727 
 728 void MacroAssembler::mult(Register s1, Register s2, Register d) {
 729   if(VM_Version::v9_instructions_work()) {
 730     mulx (s1, s2, d);
 731   } else {
 732     smul (s1, s2, d);
 733   }
 734 }
 735 
 736 void MacroAssembler::mult(Register s1, int simm13a, Register d) {
 737   if(VM_Version::v9_instructions_work()) {
 738     mulx (s1, simm13a, d);
 739   } else {
 740     smul (s1, simm13a, d);
 741   }
 742 }
 743 
 744 
 745 #ifdef ASSERT
 746 void MacroAssembler::read_ccr_v8_assert(Register ccr_save) {
 747   const Register s1 = G3_scratch;
 748   const Register s2 = G4_scratch;
 749   Label get_psr_test;
 750   // Get the condition codes the V8 way.
 751   read_ccr_trap(s1);
 752   mov(ccr_save, s2);
 753   // This is a test of V8 which has icc but not xcc
 754   // so mask off the xcc bits
 755   and3(s2, 0xf, s2);
 756   // Compare condition codes from the V8 and V9 ways.
 757   subcc(s2, s1, G0);
 758   br(Assembler::notEqual, true, Assembler::pt, get_psr_test);
 759   delayed()->breakpoint_trap();
 760   bind(get_psr_test);
 761 }
 762 
 763 void MacroAssembler::write_ccr_v8_assert(Register ccr_save) {
 764   const Register s1 = G3_scratch;
 765   const Register s2 = G4_scratch;
 766   Label set_psr_test;
 767   // Write out the saved condition codes the V8 way
 768   write_ccr_trap(ccr_save, s1, s2);
 769   // Read back the condition codes using the V9 instruction
 770   rdccr(s1);
 771   mov(ccr_save, s2);
 772   // This is a test of V8 which has icc but not xcc
 773   // so mask off the xcc bits
 774   and3(s2, 0xf, s2);
 775   and3(s1, 0xf, s1);
 776   // Compare the V8 way with the V9 way.
 777   subcc(s2, s1, G0);
 778   br(Assembler::notEqual, true, Assembler::pt, set_psr_test);
 779   delayed()->breakpoint_trap();
 780   bind(set_psr_test);
 781 }
 782 #else
 783 #define read_ccr_v8_assert(x)
 784 #define write_ccr_v8_assert(x)
 785 #endif // ASSERT
 786 
 787 void MacroAssembler::read_ccr(Register ccr_save) {
 788   if (VM_Version::v9_instructions_work()) {
 789     rdccr(ccr_save);
 790     // Test code sequence used on V8.  Do not move above rdccr.
 791     read_ccr_v8_assert(ccr_save);
 792   } else {
 793     read_ccr_trap(ccr_save);
 794   }
 795 }
 796 
 797 void MacroAssembler::write_ccr(Register ccr_save) {
 798   if (VM_Version::v9_instructions_work()) {
 799     // Test code sequence used on V8.  Do not move below wrccr.
 800     write_ccr_v8_assert(ccr_save);
 801     wrccr(ccr_save);
 802   } else {
 803     const Register temp_reg1 = G3_scratch;
 804     const Register temp_reg2 = G4_scratch;
 805     write_ccr_trap(ccr_save, temp_reg1, temp_reg2);
 806   }
 807 }
 808 
 809 
 810 // Calls to C land
 811 
 812 #ifdef ASSERT
 813 // a hook for debugging
 814 static Thread* reinitialize_thread() {
 815   return ThreadLocalStorage::thread();
 816 }
 817 #else
 818 #define reinitialize_thread ThreadLocalStorage::thread
 819 #endif
 820 
 821 #ifdef ASSERT
 822 address last_get_thread = NULL;
 823 #endif
 824 
 825 // call this when G2_thread is not known to be valid
 826 void MacroAssembler::get_thread() {
 827   save_frame(0);                // to avoid clobbering O0
 828   mov(G1, L0);                  // avoid clobbering G1
 829   mov(G5_method, L1);           // avoid clobbering G5
 830   mov(G3, L2);                  // avoid clobbering G3 also
 831   mov(G4, L5);                  // avoid clobbering G4
 832 #ifdef ASSERT
 833   Address last_get_thread_addr(L3, (address)&last_get_thread);
 834   sethi(last_get_thread_addr);
 835   inc(L4, get_pc(L4) + 2 * BytesPerInstWord); // skip getpc() code + inc + st_ptr to point L4 at call
 836   st_ptr(L4, last_get_thread_addr);
 837 #endif
 838   call(CAST_FROM_FN_PTR(address, reinitialize_thread), relocInfo::runtime_call_type);
 839   delayed()->nop();
 840   mov(L0, G1);
 841   mov(L1, G5_method);
 842   mov(L2, G3);
 843   mov(L5, G4);
 844   restore(O0, 0, G2_thread);
 845 }
 846 
 847 static Thread* verify_thread_subroutine(Thread* gthread_value) {
 848   Thread* correct_value = ThreadLocalStorage::thread();
 849   guarantee(gthread_value == correct_value, "G2_thread value must be the thread");
 850   return correct_value;
 851 }
 852 
 853 void MacroAssembler::verify_thread() {
 854   if (VerifyThread) {
 855     // NOTE: this chops off the heads of the 64-bit O registers.
 856 #ifdef CC_INTERP
 857     save_frame(0);
 858 #else
 859     // make sure G2_thread contains the right value
 860     save_frame_and_mov(0, Lmethod, Lmethod);   // to avoid clobbering O0 (and propagate Lmethod for -Xprof)
 861     mov(G1, L1);                // avoid clobbering G1
 862     // G2 saved below
 863     mov(G3, L3);                // avoid clobbering G3
 864     mov(G4, L4);                // avoid clobbering G4
 865     mov(G5_method, L5);         // avoid clobbering G5_method
 866 #endif /* CC_INTERP */
 867 #if defined(COMPILER2) && !defined(_LP64)
 868     // Save & restore possible 64-bit Long arguments in G-regs
 869     srlx(G1,32,L0);
 870     srlx(G4,32,L6);
 871 #endif
 872     call(CAST_FROM_FN_PTR(address,verify_thread_subroutine), relocInfo::runtime_call_type);
 873     delayed()->mov(G2_thread, O0);
 874 
 875     mov(L1, G1);                // Restore G1
 876     // G2 restored below
 877     mov(L3, G3);                // restore G3
 878     mov(L4, G4);                // restore G4
 879     mov(L5, G5_method);         // restore G5_method
 880 #if defined(COMPILER2) && !defined(_LP64)
 881     // Save & restore possible 64-bit Long arguments in G-regs
 882     sllx(L0,32,G2);             // Move old high G1 bits high in G2
 883     sllx(G1, 0,G1);             // Clear current high G1 bits
 884     or3 (G1,G2,G1);             // Recover 64-bit G1
 885     sllx(L6,32,G2);             // Move old high G4 bits high in G2
 886     sllx(G4, 0,G4);             // Clear current high G4 bits
 887     or3 (G4,G2,G4);             // Recover 64-bit G4
 888 #endif
 889     restore(O0, 0, G2_thread);
 890   }
 891 }
 892 
 893 
 894 void MacroAssembler::save_thread(const Register thread_cache) {
 895   verify_thread();
 896   if (thread_cache->is_valid()) {
 897     assert(thread_cache->is_local() || thread_cache->is_in(), "bad volatile");
 898     mov(G2_thread, thread_cache);
 899   }
 900   if (VerifyThread) {
 901     // smash G2_thread, as if the VM were about to anyway
 902     set(0x67676767, G2_thread);
 903   }
 904 }
 905 
 906 
 907 void MacroAssembler::restore_thread(const Register thread_cache) {
 908   if (thread_cache->is_valid()) {
 909     assert(thread_cache->is_local() || thread_cache->is_in(), "bad volatile");
 910     mov(thread_cache, G2_thread);
 911     verify_thread();
 912   } else {
 913     // do it the slow way
 914     get_thread();
 915   }
 916 }
 917 
 918 
 919 // %%% maybe get rid of [re]set_last_Java_frame
 920 void MacroAssembler::set_last_Java_frame(Register last_java_sp, Register last_Java_pc) {
 921   assert_not_delayed();
 922   Address flags(G2_thread,
 923                 0,
 924                 in_bytes(JavaThread::frame_anchor_offset()) +
 925                          in_bytes(JavaFrameAnchor::flags_offset()));
 926   Address pc_addr(G2_thread,
 927                   0,
 928                   in_bytes(JavaThread::last_Java_pc_offset()));
 929 
 930   // Always set last_Java_pc and flags first because once last_Java_sp is visible
 931   // has_last_Java_frame is true and users will look at the rest of the fields.
 932   // (Note: flags should always be zero before we get here so doesn't need to be set.)
 933 
 934 #ifdef ASSERT
 935   // Verify that flags was zeroed on return to Java
 936   Label PcOk;
 937   save_frame(0);                // to avoid clobbering O0
 938   ld_ptr(pc_addr, L0);
 939   tst(L0);
 940 #ifdef _LP64
 941   brx(Assembler::zero, false, Assembler::pt, PcOk);
 942 #else
 943   br(Assembler::zero, false, Assembler::pt, PcOk);
 944 #endif // _LP64
 945   delayed() -> nop();
 946   stop("last_Java_pc not zeroed before leaving Java");
 947   bind(PcOk);
 948 
 949   // Verify that flags was zeroed on return to Java
 950   Label FlagsOk;
 951   ld(flags, L0);
 952   tst(L0);
 953   br(Assembler::zero, false, Assembler::pt, FlagsOk);
 954   delayed() -> restore();
 955   stop("flags not zeroed before leaving Java");
 956   bind(FlagsOk);
 957 #endif /* ASSERT */
 958   //
 959   // When returning from calling out from Java mode the frame anchor's last_Java_pc
 960   // will always be set to NULL. It is set here so that if we are doing a call to
 961   // native (not VM) that we capture the known pc and don't have to rely on the
 962   // native call having a standard frame linkage where we can find the pc.
 963 
 964   if (last_Java_pc->is_valid()) {
 965     st_ptr(last_Java_pc, pc_addr);
 966   }
 967 
 968 #ifdef _LP64
 969 #ifdef ASSERT
 970   // Make sure that we have an odd stack
 971   Label StackOk;
 972   andcc(last_java_sp, 0x01, G0);
 973   br(Assembler::notZero, false, Assembler::pt, StackOk);
 974   delayed() -> nop();
 975   stop("Stack Not Biased in set_last_Java_frame");
 976   bind(StackOk);
 977 #endif // ASSERT
 978   assert( last_java_sp != G4_scratch, "bad register usage in set_last_Java_frame");
 979   add( last_java_sp, STACK_BIAS, G4_scratch );
 980   st_ptr(G4_scratch,    Address(G2_thread, 0, in_bytes(JavaThread::last_Java_sp_offset())));
 981 #else
 982   st_ptr(last_java_sp,    Address(G2_thread, 0, in_bytes(JavaThread::last_Java_sp_offset())));
 983 #endif // _LP64
 984 }
 985 
 986 void MacroAssembler::reset_last_Java_frame(void) {
 987   assert_not_delayed();
 988 
 989   Address sp_addr(G2_thread, 0, in_bytes(JavaThread::last_Java_sp_offset()));
 990   Address pc_addr(G2_thread,
 991                   0,
 992                   in_bytes(JavaThread::frame_anchor_offset()) + in_bytes(JavaFrameAnchor::last_Java_pc_offset()));
 993   Address flags(G2_thread,
 994                 0,
 995                 in_bytes(JavaThread::frame_anchor_offset()) + in_bytes(JavaFrameAnchor::flags_offset()));
 996 
 997 #ifdef ASSERT
 998   // check that it WAS previously set
 999 #ifdef CC_INTERP
1000     save_frame(0);
1001 #else
1002     save_frame_and_mov(0, Lmethod, Lmethod);     // Propagate Lmethod to helper frame for -Xprof
1003 #endif /* CC_INTERP */
1004     ld_ptr(sp_addr, L0);
1005     tst(L0);
1006     breakpoint_trap(Assembler::zero, Assembler::ptr_cc);
1007     restore();
1008 #endif // ASSERT
1009 
1010   st_ptr(G0, sp_addr);
1011   // Always return last_Java_pc to zero
1012   st_ptr(G0, pc_addr);
1013   // Always null flags after return to Java
1014   st(G0, flags);
1015 }
1016 
1017 
1018 void MacroAssembler::call_VM_base(
1019   Register        oop_result,
1020   Register        thread_cache,
1021   Register        last_java_sp,
1022   address         entry_point,
1023   int             number_of_arguments,
1024   bool            check_exceptions)
1025 {
1026   assert_not_delayed();
1027 
1028   // determine last_java_sp register
1029   if (!last_java_sp->is_valid()) {
1030     last_java_sp = SP;
1031   }
1032   // debugging support
1033   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
1034 
1035   // 64-bit last_java_sp is biased!
1036   set_last_Java_frame(last_java_sp, noreg);
1037   if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early
1038   save_thread(thread_cache);
1039   // do the call
1040   call(entry_point, relocInfo::runtime_call_type);
1041   if (!VerifyThread)
1042     delayed()->mov(G2_thread, O0);  // pass thread as first argument
1043   else
1044     delayed()->nop();             // (thread already passed)
1045   restore_thread(thread_cache);
1046   reset_last_Java_frame();
1047 
1048   // check for pending exceptions. use Gtemp as scratch register.
1049   if (check_exceptions) {
1050     check_and_forward_exception(Gtemp);
1051   }
1052 
1053   // get oop result if there is one and reset the value in the thread
1054   if (oop_result->is_valid()) {
1055     get_vm_result(oop_result);
1056   }
1057 }
1058 
1059 void MacroAssembler::check_and_forward_exception(Register scratch_reg)
1060 {
1061   Label L;
1062 
1063   check_and_handle_popframe(scratch_reg);
1064   check_and_handle_earlyret(scratch_reg);
1065 
1066   Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset()));
1067   ld_ptr(exception_addr, scratch_reg);
1068   br_null(scratch_reg,false,pt,L);
1069   delayed()->nop();
1070   // we use O7 linkage so that forward_exception_entry has the issuing PC
1071   call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
1072   delayed()->nop();
1073   bind(L);
1074 }
1075 
1076 
1077 void MacroAssembler::check_and_handle_popframe(Register scratch_reg) {
1078 }
1079 
1080 
1081 void MacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
1082 }
1083 
1084 
1085 void MacroAssembler::call_VM(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
1086   call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
1087 }
1088 
1089 
1090 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) {
1091   // O0 is reserved for the thread
1092   mov(arg_1, O1);
1093   call_VM(oop_result, entry_point, 1, check_exceptions);
1094 }
1095 
1096 
1097 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) {
1098   // O0 is reserved for the thread
1099   mov(arg_1, O1);
1100   mov(arg_2, O2); assert(arg_2 != O1, "smashed argument");
1101   call_VM(oop_result, entry_point, 2, check_exceptions);
1102 }
1103 
1104 
1105 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) {
1106   // O0 is reserved for the thread
1107   mov(arg_1, O1);
1108   mov(arg_2, O2); assert(arg_2 != O1,                "smashed argument");
1109   mov(arg_3, O3); assert(arg_3 != O1 && arg_3 != O2, "smashed argument");
1110   call_VM(oop_result, entry_point, 3, check_exceptions);
1111 }
1112 
1113 
1114 
1115 // Note: The following call_VM overloadings are useful when a "save"
1116 // has already been performed by a stub, and the last Java frame is
1117 // the previous one.  In that case, last_java_sp must be passed as FP
1118 // instead of SP.
1119 
1120 
1121 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments, bool check_exceptions) {
1122   call_VM_base(oop_result, noreg, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1123 }
1124 
1125 
1126 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions) {
1127   // O0 is reserved for the thread
1128   mov(arg_1, O1);
1129   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1130 }
1131 
1132 
1133 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) {
1134   // O0 is reserved for the thread
1135   mov(arg_1, O1);
1136   mov(arg_2, O2); assert(arg_2 != O1, "smashed argument");
1137   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1138 }
1139 
1140 
1141 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) {
1142   // O0 is reserved for the thread
1143   mov(arg_1, O1);
1144   mov(arg_2, O2); assert(arg_2 != O1,                "smashed argument");
1145   mov(arg_3, O3); assert(arg_3 != O1 && arg_3 != O2, "smashed argument");
1146   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1147 }
1148 
1149 
1150 
1151 void MacroAssembler::call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments) {
1152   assert_not_delayed();
1153   save_thread(thread_cache);
1154   // do the call
1155   call(entry_point, relocInfo::runtime_call_type);
1156   delayed()->nop();
1157   restore_thread(thread_cache);
1158 }
1159 
1160 
1161 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, int number_of_arguments) {
1162   call_VM_leaf_base(thread_cache, entry_point, number_of_arguments);
1163 }
1164 
1165 
1166 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1) {
1167   mov(arg_1, O0);
1168   call_VM_leaf(thread_cache, entry_point, 1);
1169 }
1170 
1171 
1172 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
1173   mov(arg_1, O0);
1174   mov(arg_2, O1); assert(arg_2 != O0, "smashed argument");
1175   call_VM_leaf(thread_cache, entry_point, 2);
1176 }
1177 
1178 
1179 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3) {
1180   mov(arg_1, O0);
1181   mov(arg_2, O1); assert(arg_2 != O0,                "smashed argument");
1182   mov(arg_3, O2); assert(arg_3 != O0 && arg_3 != O1, "smashed argument");
1183   call_VM_leaf(thread_cache, entry_point, 3);
1184 }
1185 
1186 
1187 void MacroAssembler::get_vm_result(Register oop_result) {
1188   verify_thread();
1189   Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset()));
1190   ld_ptr(    vm_result_addr, oop_result);
1191   st_ptr(G0, vm_result_addr);
1192   verify_oop(oop_result);
1193 }
1194 
1195 
1196 void MacroAssembler::get_vm_result_2(Register oop_result) {
1197   verify_thread();
1198   Address vm_result_addr_2(G2_thread, 0, in_bytes(JavaThread::vm_result_2_offset()));
1199   ld_ptr(vm_result_addr_2, oop_result);
1200   st_ptr(G0, vm_result_addr_2);
1201   verify_oop(oop_result);
1202 }
1203 
1204 
1205 // We require that C code which does not return a value in vm_result will
1206 // leave it undisturbed.
1207 void MacroAssembler::set_vm_result(Register oop_result) {
1208   verify_thread();
1209   Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset()));
1210   verify_oop(oop_result);
1211 
1212 # ifdef ASSERT
1213     // Check that we are not overwriting any other oop.
1214 #ifdef CC_INTERP
1215     save_frame(0);
1216 #else
1217     save_frame_and_mov(0, Lmethod, Lmethod);     // Propagate Lmethod for -Xprof
1218 #endif /* CC_INTERP */
1219     ld_ptr(vm_result_addr, L0);
1220     tst(L0);
1221     restore();
1222     breakpoint_trap(notZero, Assembler::ptr_cc);
1223     // }
1224 # endif
1225 
1226   st_ptr(oop_result, vm_result_addr);
1227 }
1228 
1229 
1230 void MacroAssembler::card_table_write(jbyte* byte_map_base,
1231                                       Register tmp, Register obj) {
1232 #ifdef _LP64
1233   srlx(obj, CardTableModRefBS::card_shift, obj);
1234 #else
1235   srl(obj, CardTableModRefBS::card_shift, obj);
1236 #endif
1237   assert( tmp != obj, "need separate temp reg");
1238   Address rs(tmp, (address)byte_map_base);
1239   load_address(rs);
1240   stb(G0, rs.base(), obj);
1241 }
1242 
1243 // %%% Note:  The following six instructions have been moved,
1244 //            unchanged, from assembler_sparc.inline.hpp.
1245 //            They will be refactored at a later date.
1246 
1247 void MacroAssembler::sethi(intptr_t imm22a,
1248                             Register d,
1249                             bool ForceRelocatable,
1250                             RelocationHolder const& rspec) {
1251   Address adr( d, (address)imm22a, rspec );
1252   MacroAssembler::sethi( adr, ForceRelocatable );
1253 }
1254 
1255 
1256 void MacroAssembler::sethi(Address& a, bool ForceRelocatable) {
1257   address save_pc;
1258   int shiftcnt;
1259   // if addr of local, do not need to load it
1260   assert(a.base() != FP  &&  a.base() != SP, "just use ld or st for locals");
1261 #ifdef _LP64
1262 # ifdef CHECK_DELAY
1263   assert_not_delayed( (char *)"cannot put two instructions in delay slot" );
1264 # endif
1265   v9_dep();
1266 //  ForceRelocatable = 1;
1267   save_pc = pc();
1268   if (a.hi32() == 0 && a.low32() >= 0) {
1269     Assembler::sethi(a.low32(), a.base(), a.rspec());
1270   }
1271   else if (a.hi32() == -1) {
1272     Assembler::sethi(~a.low32(), a.base(), a.rspec());
1273     xor3(a.base(), ~low10(~0), a.base());
1274   }
1275   else {
1276     Assembler::sethi(a.hi32(), a.base(), a.rspec() );   // 22
1277     if ( a.hi32() & 0x3ff )                     // Any bits?
1278       or3( a.base(), a.hi32() & 0x3ff ,a.base() ); // High 32 bits are now in low 32
1279     if ( a.low32() & 0xFFFFFC00 ) {             // done?
1280       if( (a.low32() >> 20) & 0xfff ) {         // Any bits set?
1281         sllx(a.base(), 12, a.base());           // Make room for next 12 bits
1282         or3( a.base(), (a.low32() >> 20) & 0xfff,a.base() ); // Or in next 12
1283         shiftcnt = 0;                           // We already shifted
1284       }
1285       else
1286         shiftcnt = 12;
1287       if( (a.low32() >> 10) & 0x3ff ) {
1288         sllx(a.base(), shiftcnt+10, a.base());// Make room for last 10 bits
1289         or3( a.base(), (a.low32() >> 10) & 0x3ff,a.base() ); // Or in next 10
1290         shiftcnt = 0;
1291       }
1292       else
1293         shiftcnt = 10;
1294       sllx(a.base(), shiftcnt+10 , a.base());           // Shift leaving disp field 0'd
1295     }
1296     else
1297       sllx( a.base(), 32, a.base() );
1298   }
1299   // Pad out the instruction sequence so it can be
1300   // patched later.
1301   if ( ForceRelocatable || (a.rtype() != relocInfo::none &&
1302                             a.rtype() != relocInfo::runtime_call_type) ) {
1303     while ( pc() < (save_pc + (7 * BytesPerInstWord )) )
1304       nop();
1305   }
1306 #else
1307   Assembler::sethi(a.hi(), a.base(), a.rspec());
1308 #endif
1309 
1310 }
1311 
1312 int MacroAssembler::size_of_sethi(address a, bool worst_case) {
1313 #ifdef _LP64
1314   if (worst_case) return 7;
1315   intptr_t iaddr = (intptr_t)a;
1316   int hi32 = (int)(iaddr >> 32);
1317   int lo32 = (int)(iaddr);
1318   int inst_count;
1319   if (hi32 == 0 && lo32 >= 0)
1320     inst_count = 1;
1321   else if (hi32 == -1)
1322     inst_count = 2;
1323   else {
1324     inst_count = 2;
1325     if ( hi32 & 0x3ff )
1326       inst_count++;
1327     if ( lo32 & 0xFFFFFC00 ) {
1328       if( (lo32 >> 20) & 0xfff ) inst_count += 2;
1329       if( (lo32 >> 10) & 0x3ff ) inst_count += 2;
1330     }
1331   }
1332   return BytesPerInstWord * inst_count;
1333 #else
1334   return BytesPerInstWord;
1335 #endif
1336 }
1337 
1338 int MacroAssembler::worst_case_size_of_set() {
1339   return size_of_sethi(NULL, true) + 1;
1340 }
1341 
1342 void MacroAssembler::set(intptr_t value, Register d,
1343                          RelocationHolder const& rspec) {
1344   Address val( d, (address)value, rspec);
1345 
1346   if ( rspec.type() == relocInfo::none ) {
1347     // can optimize
1348     if (-4096 <= value  &&  value <= 4095) {
1349       or3(G0, value, d); // setsw (this leaves upper 32 bits sign-extended)
1350       return;
1351     }
1352     if (inv_hi22(hi22(value)) == value) {
1353       sethi(val);
1354       return;
1355     }
1356   }
1357   assert_not_delayed( (char *)"cannot put two instructions in delay slot" );
1358   sethi( val );
1359   if (rspec.type() != relocInfo::none || (value & 0x3ff) != 0) {
1360     add( d, value &  0x3ff, d, rspec);
1361   }
1362 }
1363 
1364 void MacroAssembler::setsw(int value, Register d,
1365                            RelocationHolder const& rspec) {
1366   Address val( d, (address)value, rspec);
1367   if ( rspec.type() == relocInfo::none ) {
1368     // can optimize
1369     if (-4096 <= value  &&  value <= 4095) {
1370       or3(G0, value, d);
1371       return;
1372     }
1373     if (inv_hi22(hi22(value)) == value) {
1374       sethi( val );
1375 #ifndef _LP64
1376       if ( value < 0 ) {
1377         assert_not_delayed();
1378         sra (d, G0, d);
1379       }
1380 #endif
1381       return;
1382     }
1383   }
1384   assert_not_delayed();
1385   sethi( val );
1386   add( d, value &  0x3ff, d, rspec);
1387 
1388   // (A negative value could be loaded in 2 insns with sethi/xor,
1389   // but it would take a more complex relocation.)
1390 #ifndef _LP64
1391   if ( value < 0)
1392     sra(d, G0, d);
1393 #endif
1394 }
1395 
1396 // %%% End of moved six set instructions.
1397 
1398 
1399 void MacroAssembler::set64(jlong value, Register d, Register tmp) {
1400   assert_not_delayed();
1401   v9_dep();
1402 
1403   int hi = (int)(value >> 32);
1404   int lo = (int)(value & ~0);
1405   // (Matcher::isSimpleConstant64 knows about the following optimizations.)
1406   if (Assembler::is_simm13(lo) && value == lo) {
1407     or3(G0, lo, d);
1408   } else if (hi == 0) {
1409     Assembler::sethi(lo, d);   // hardware version zero-extends to upper 32
1410     if (low10(lo) != 0)
1411       or3(d, low10(lo), d);
1412   }
1413   else if (hi == -1) {
1414     Assembler::sethi(~lo, d);  // hardware version zero-extends to upper 32
1415     xor3(d, low10(lo) ^ ~low10(~0), d);
1416   }
1417   else if (lo == 0) {
1418     if (Assembler::is_simm13(hi)) {
1419       or3(G0, hi, d);
1420     } else {
1421       Assembler::sethi(hi, d);   // hardware version zero-extends to upper 32
1422       if (low10(hi) != 0)
1423         or3(d, low10(hi), d);
1424     }
1425     sllx(d, 32, d);
1426   }
1427   else {
1428     Assembler::sethi(hi, tmp);
1429     Assembler::sethi(lo,   d); // macro assembler version sign-extends
1430     if (low10(hi) != 0)
1431       or3 (tmp, low10(hi), tmp);
1432     if (low10(lo) != 0)
1433       or3 (  d, low10(lo),   d);
1434     sllx(tmp, 32, tmp);
1435     or3 (d, tmp, d);
1436   }
1437 }
1438 
1439 // compute size in bytes of sparc frame, given
1440 // number of extraWords
1441 int MacroAssembler::total_frame_size_in_bytes(int extraWords) {
1442 
1443   int nWords = frame::memory_parameter_word_sp_offset;
1444 
1445   nWords += extraWords;
1446 
1447   if (nWords & 1) ++nWords; // round up to double-word
1448 
1449   return nWords * BytesPerWord;
1450 }
1451 
1452 
1453 // save_frame: given number of "extra" words in frame,
1454 // issue approp. save instruction (p 200, v8 manual)
1455 
1456 void MacroAssembler::save_frame(int extraWords = 0) {
1457   int delta = -total_frame_size_in_bytes(extraWords);
1458   if (is_simm13(delta)) {
1459     save(SP, delta, SP);
1460   } else {
1461     set(delta, G3_scratch);
1462     save(SP, G3_scratch, SP);
1463   }
1464 }
1465 
1466 
1467 void MacroAssembler::save_frame_c1(int size_in_bytes) {
1468   if (is_simm13(-size_in_bytes)) {
1469     save(SP, -size_in_bytes, SP);
1470   } else {
1471     set(-size_in_bytes, G3_scratch);
1472     save(SP, G3_scratch, SP);
1473   }
1474 }
1475 
1476 
1477 void MacroAssembler::save_frame_and_mov(int extraWords,
1478                                         Register s1, Register d1,
1479                                         Register s2, Register d2) {
1480   assert_not_delayed();
1481 
1482   // The trick here is to use precisely the same memory word
1483   // that trap handlers also use to save the register.
1484   // This word cannot be used for any other purpose, but
1485   // it works fine to save the register's value, whether or not
1486   // an interrupt flushes register windows at any given moment!
1487   Address s1_addr;
1488   if (s1->is_valid() && (s1->is_in() || s1->is_local())) {
1489     s1_addr = s1->address_in_saved_window();
1490     st_ptr(s1, s1_addr);
1491   }
1492 
1493   Address s2_addr;
1494   if (s2->is_valid() && (s2->is_in() || s2->is_local())) {
1495     s2_addr = s2->address_in_saved_window();
1496     st_ptr(s2, s2_addr);
1497   }
1498 
1499   save_frame(extraWords);
1500 
1501   if (s1_addr.base() == SP) {
1502     ld_ptr(s1_addr.after_save(), d1);
1503   } else if (s1->is_valid()) {
1504     mov(s1->after_save(), d1);
1505   }
1506 
1507   if (s2_addr.base() == SP) {
1508     ld_ptr(s2_addr.after_save(), d2);
1509   } else if (s2->is_valid()) {
1510     mov(s2->after_save(), d2);
1511   }
1512 }
1513 
1514 
1515 Address MacroAssembler::allocate_oop_address(jobject obj, Register d) {
1516   assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
1517   int oop_index = oop_recorder()->allocate_index(obj);
1518   return Address(d, address(obj), oop_Relocation::spec(oop_index));
1519 }
1520 
1521 
1522 Address MacroAssembler::constant_oop_address(jobject obj, Register d) {
1523   assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
1524   int oop_index = oop_recorder()->find_index(obj);
1525   return Address(d, address(obj), oop_Relocation::spec(oop_index));
1526 }
1527 
1528 void  MacroAssembler::set_narrow_oop(jobject obj, Register d) {
1529   assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
1530   int oop_index = oop_recorder()->find_index(obj);
1531   RelocationHolder rspec = oop_Relocation::spec(oop_index);
1532 
1533   assert_not_delayed();
1534   // Relocation with special format (see relocInfo_sparc.hpp).
1535   relocate(rspec, 1);
1536   // Assembler::sethi(0x3fffff, d);
1537   emit_long( op(branch_op) | rd(d) | op2(sethi_op2) | hi22(0x3fffff) );
1538   // Don't add relocation for 'add'. Do patching during 'sethi' processing.
1539   add(d, 0x3ff, d);
1540 
1541 }
1542 
1543 
1544 void MacroAssembler::align(int modulus) {
1545   while (offset() % modulus != 0) nop();
1546 }
1547 
1548 
1549 void MacroAssembler::safepoint() {
1550   relocate(breakpoint_Relocation::spec(breakpoint_Relocation::safepoint));
1551 }
1552 
1553 
1554 void RegistersForDebugging::print(outputStream* s) {
1555   int j;
1556   for ( j = 0;  j < 8;  ++j )
1557     if ( j != 6 ) s->print_cr("i%d = 0x%.16lx", j, i[j]);
1558     else          s->print_cr( "fp = 0x%.16lx",    i[j]);
1559   s->cr();
1560 
1561   for ( j = 0;  j < 8;  ++j )
1562     s->print_cr("l%d = 0x%.16lx", j, l[j]);
1563   s->cr();
1564 
1565   for ( j = 0;  j < 8;  ++j )
1566     if ( j != 6 ) s->print_cr("o%d = 0x%.16lx", j, o[j]);
1567     else          s->print_cr( "sp = 0x%.16lx",    o[j]);
1568   s->cr();
1569 
1570   for ( j = 0;  j < 8;  ++j )
1571     s->print_cr("g%d = 0x%.16lx", j, g[j]);
1572   s->cr();
1573 
1574   // print out floats with compression
1575   for (j = 0; j < 32; ) {
1576     jfloat val = f[j];
1577     int last = j;
1578     for ( ;  last+1 < 32;  ++last ) {
1579       char b1[1024], b2[1024];
1580       sprintf(b1, "%f", val);
1581       sprintf(b2, "%f", f[last+1]);
1582       if (strcmp(b1, b2))
1583         break;
1584     }
1585     s->print("f%d", j);
1586     if ( j != last )  s->print(" - f%d", last);
1587     s->print(" = %f", val);
1588     s->fill_to(25);
1589     s->print_cr(" (0x%x)", val);
1590     j = last + 1;
1591   }
1592   s->cr();
1593 
1594   // and doubles (evens only)
1595   for (j = 0; j < 32; ) {
1596     jdouble val = d[j];
1597     int last = j;
1598     for ( ;  last+1 < 32;  ++last ) {
1599       char b1[1024], b2[1024];
1600       sprintf(b1, "%f", val);
1601       sprintf(b2, "%f", d[last+1]);
1602       if (strcmp(b1, b2))
1603         break;
1604     }
1605     s->print("d%d", 2 * j);
1606     if ( j != last )  s->print(" - d%d", last);
1607     s->print(" = %f", val);
1608     s->fill_to(30);
1609     s->print("(0x%x)", *(int*)&val);
1610     s->fill_to(42);
1611     s->print_cr("(0x%x)", *(1 + (int*)&val));
1612     j = last + 1;
1613   }
1614   s->cr();
1615 }
1616 
1617 void RegistersForDebugging::save_registers(MacroAssembler* a) {
1618   a->sub(FP, round_to(sizeof(RegistersForDebugging), sizeof(jdouble)) - STACK_BIAS, O0);
1619   a->flush_windows();
1620   int i;
1621   for (i = 0; i < 8; ++i) {
1622     a->ld_ptr(as_iRegister(i)->address_in_saved_window().after_save(), L1);  a->st_ptr( L1, O0, i_offset(i));
1623     a->ld_ptr(as_lRegister(i)->address_in_saved_window().after_save(), L1);  a->st_ptr( L1, O0, l_offset(i));
1624     a->st_ptr(as_oRegister(i)->after_save(), O0, o_offset(i));
1625     a->st_ptr(as_gRegister(i)->after_save(), O0, g_offset(i));
1626   }
1627   for (i = 0;  i < 32; ++i) {
1628     a->stf(FloatRegisterImpl::S, as_FloatRegister(i), O0, f_offset(i));
1629   }
1630   for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) {
1631     a->stf(FloatRegisterImpl::D, as_FloatRegister(i), O0, d_offset(i));
1632   }
1633 }
1634 
1635 void RegistersForDebugging::restore_registers(MacroAssembler* a, Register r) {
1636   for (int i = 1; i < 8;  ++i) {
1637     a->ld_ptr(r, g_offset(i), as_gRegister(i));
1638   }
1639   for (int j = 0; j < 32; ++j) {
1640     a->ldf(FloatRegisterImpl::S, O0, f_offset(j), as_FloatRegister(j));
1641   }
1642   for (int k = 0; k < (VM_Version::v9_instructions_work() ? 64 : 32); k += 2) {
1643     a->ldf(FloatRegisterImpl::D, O0, d_offset(k), as_FloatRegister(k));
1644   }
1645 }
1646 
1647 
1648 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1649 void MacroAssembler::push_fTOS() {
1650   // %%%%%% need to implement this
1651 }
1652 
1653 // pops double TOS element from CPU stack and pushes on FPU stack
1654 void MacroAssembler::pop_fTOS() {
1655   // %%%%%% need to implement this
1656 }
1657 
1658 void MacroAssembler::empty_FPU_stack() {
1659   // %%%%%% need to implement this
1660 }
1661 
1662 void MacroAssembler::_verify_oop(Register reg, const char* msg, const char * file, int line) {
1663   // plausibility check for oops
1664   if (!VerifyOops) return;
1665 
1666   if (reg == G0)  return;       // always NULL, which is always an oop
1667 
1668   char buffer[64];
1669 #ifdef COMPILER1
1670   if (CommentedAssembly) {
1671     snprintf(buffer, sizeof(buffer), "verify_oop at %d", offset());
1672     block_comment(buffer);
1673   }
1674 #endif
1675 
1676   int len = strlen(file) + strlen(msg) + 1 + 4;
1677   sprintf(buffer, "%d", line);
1678   len += strlen(buffer);
1679   sprintf(buffer, " at offset %d ", offset());
1680   len += strlen(buffer);
1681   char * real_msg = new char[len];
1682   sprintf(real_msg, "%s%s(%s:%d)", msg, buffer, file, line);
1683 
1684   // Call indirectly to solve generation ordering problem
1685   Address a(O7, (address)StubRoutines::verify_oop_subroutine_entry_address());
1686 
1687   // Make some space on stack above the current register window.
1688   // Enough to hold 8 64-bit registers.
1689   add(SP,-8*8,SP);
1690 
1691   // Save some 64-bit registers; a normal 'save' chops the heads off
1692   // of 64-bit longs in the 32-bit build.
1693   stx(O0,SP,frame::register_save_words*wordSize+STACK_BIAS+0*8);
1694   stx(O1,SP,frame::register_save_words*wordSize+STACK_BIAS+1*8);
1695   mov(reg,O0); // Move arg into O0; arg might be in O7 which is about to be crushed
1696   stx(O7,SP,frame::register_save_words*wordSize+STACK_BIAS+7*8);
1697 
1698   set((intptr_t)real_msg, O1);
1699   // Load address to call to into O7
1700   load_ptr_contents(a, O7);
1701   // Register call to verify_oop_subroutine
1702   callr(O7, G0);
1703   delayed()->nop();
1704   // recover frame size
1705   add(SP, 8*8,SP);
1706 }
1707 
1708 void MacroAssembler::_verify_oop_addr(Address addr, const char* msg, const char * file, int line) {
1709   // plausibility check for oops
1710   if (!VerifyOops) return;
1711 
1712   char buffer[64];
1713   sprintf(buffer, "%d", line);
1714   int len = strlen(file) + strlen(msg) + 1 + 4 + strlen(buffer);
1715   sprintf(buffer, " at SP+%d ", addr.disp());
1716   len += strlen(buffer);
1717   char * real_msg = new char[len];
1718   sprintf(real_msg, "%s at SP+%d (%s:%d)", msg, addr.disp(), file, line);
1719 
1720   // Call indirectly to solve generation ordering problem
1721   Address a(O7, (address)StubRoutines::verify_oop_subroutine_entry_address());
1722 
1723   // Make some space on stack above the current register window.
1724   // Enough to hold 8 64-bit registers.
1725   add(SP,-8*8,SP);
1726 
1727   // Save some 64-bit registers; a normal 'save' chops the heads off
1728   // of 64-bit longs in the 32-bit build.
1729   stx(O0,SP,frame::register_save_words*wordSize+STACK_BIAS+0*8);
1730   stx(O1,SP,frame::register_save_words*wordSize+STACK_BIAS+1*8);
1731   ld_ptr(addr.base(), addr.disp() + 8*8, O0); // Load arg into O0; arg might be in O7 which is about to be crushed
1732   stx(O7,SP,frame::register_save_words*wordSize+STACK_BIAS+7*8);
1733 
1734   set((intptr_t)real_msg, O1);
1735   // Load address to call to into O7
1736   load_ptr_contents(a, O7);
1737   // Register call to verify_oop_subroutine
1738   callr(O7, G0);
1739   delayed()->nop();
1740   // recover frame size
1741   add(SP, 8*8,SP);
1742 }
1743 
1744 // side-door communication with signalHandler in os_solaris.cpp
1745 address MacroAssembler::_verify_oop_implicit_branch[3] = { NULL };
1746 
1747 // This macro is expanded just once; it creates shared code.  Contract:
1748 // receives an oop in O0.  Must restore O0 & O7 from TLS.  Must not smash ANY
1749 // registers, including flags.  May not use a register 'save', as this blows
1750 // the high bits of the O-regs if they contain Long values.  Acts as a 'leaf'
1751 // call.
1752 void MacroAssembler::verify_oop_subroutine() {
1753   assert( VM_Version::v9_instructions_work(), "VerifyOops not supported for V8" );
1754 
1755   // Leaf call; no frame.
1756   Label succeed, fail, null_or_fail;
1757 
1758   // O0 and O7 were saved already (O0 in O0's TLS home, O7 in O5's TLS home).
1759   // O0 is now the oop to be checked.  O7 is the return address.
1760   Register O0_obj = O0;
1761 
1762   // Save some more registers for temps.
1763   stx(O2,SP,frame::register_save_words*wordSize+STACK_BIAS+2*8);
1764   stx(O3,SP,frame::register_save_words*wordSize+STACK_BIAS+3*8);
1765   stx(O4,SP,frame::register_save_words*wordSize+STACK_BIAS+4*8);
1766   stx(O5,SP,frame::register_save_words*wordSize+STACK_BIAS+5*8);
1767 
1768   // Save flags
1769   Register O5_save_flags = O5;
1770   rdccr( O5_save_flags );
1771 
1772   { // count number of verifies
1773     Register O2_adr   = O2;
1774     Register O3_accum = O3;
1775     Address count_addr( O2_adr, (address) StubRoutines::verify_oop_count_addr() );
1776     sethi(count_addr);
1777     ld(count_addr, O3_accum);
1778     inc(O3_accum);
1779     st(O3_accum, count_addr);
1780   }
1781 
1782   Register O2_mask = O2;
1783   Register O3_bits = O3;
1784   Register O4_temp = O4;
1785 
1786   // mark lower end of faulting range
1787   assert(_verify_oop_implicit_branch[0] == NULL, "set once");
1788   _verify_oop_implicit_branch[0] = pc();
1789 
1790   // We can't check the mark oop because it could be in the process of
1791   // locking or unlocking while this is running.
1792   set(Universe::verify_oop_mask (), O2_mask);
1793   set(Universe::verify_oop_bits (), O3_bits);
1794 
1795   // assert((obj & oop_mask) == oop_bits);
1796   and3(O0_obj, O2_mask, O4_temp);
1797   cmp(O4_temp, O3_bits);
1798   brx(notEqual, false, pn, null_or_fail);
1799   delayed()->nop();
1800 
1801   if ((NULL_WORD & Universe::verify_oop_mask()) == Universe::verify_oop_bits()) {
1802     // the null_or_fail case is useless; must test for null separately
1803     br_null(O0_obj, false, pn, succeed);
1804     delayed()->nop();
1805   }
1806 
1807   // Check the klassOop of this object for being in the right area of memory.
1808   // Cannot do the load in the delay above slot in case O0 is null
1809   load_klass(O0_obj, O0_obj);
1810   // assert((klass & klass_mask) == klass_bits);
1811   if( Universe::verify_klass_mask() != Universe::verify_oop_mask() )
1812     set(Universe::verify_klass_mask(), O2_mask);
1813   if( Universe::verify_klass_bits() != Universe::verify_oop_bits() )
1814     set(Universe::verify_klass_bits(), O3_bits);
1815   and3(O0_obj, O2_mask, O4_temp);
1816   cmp(O4_temp, O3_bits);
1817   brx(notEqual, false, pn, fail);
1818   delayed()->nop();
1819   // Check the klass's klass
1820   load_klass(O0_obj, O0_obj);
1821   and3(O0_obj, O2_mask, O4_temp);
1822   cmp(O4_temp, O3_bits);
1823   brx(notEqual, false, pn, fail);
1824   delayed()->wrccr( O5_save_flags ); // Restore CCR's
1825 
1826   // mark upper end of faulting range
1827   _verify_oop_implicit_branch[1] = pc();
1828 
1829   //-----------------------
1830   // all tests pass
1831   bind(succeed);
1832 
1833   // Restore prior 64-bit registers
1834   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+0*8,O0);
1835   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+1*8,O1);
1836   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+2*8,O2);
1837   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+3*8,O3);
1838   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+4*8,O4);
1839   ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+5*8,O5);
1840 
1841   retl();                       // Leaf return; restore prior O7 in delay slot
1842   delayed()->ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+7*8,O7);
1843 
1844   //-----------------------
1845   bind(null_or_fail);           // nulls are less common but OK
1846   br_null(O0_obj, false, pt, succeed);
1847   delayed()->wrccr( O5_save_flags ); // Restore CCR's
1848 
1849   //-----------------------
1850   // report failure:
1851   bind(fail);
1852   _verify_oop_implicit_branch[2] = pc();
1853 
1854   wrccr( O5_save_flags ); // Restore CCR's
1855 
1856   save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2));
1857 
1858   // stop_subroutine expects message pointer in I1.
1859   mov(I1, O1);
1860 
1861   // Restore prior 64-bit registers
1862   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+0*8,I0);
1863   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+1*8,I1);
1864   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+2*8,I2);
1865   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+3*8,I3);
1866   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+4*8,I4);
1867   ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+5*8,I5);
1868 
1869   // factor long stop-sequence into subroutine to save space
1870   assert(StubRoutines::Sparc::stop_subroutine_entry_address(), "hasn't been generated yet");
1871 
1872   // call indirectly to solve generation ordering problem
1873   Address a(O5, (address)StubRoutines::Sparc::stop_subroutine_entry_address());
1874   load_ptr_contents(a, O5);
1875   jmpl(O5, 0, O7);
1876   delayed()->nop();
1877 }
1878 
1879 
1880 void MacroAssembler::stop(const char* msg) {
1881   // save frame first to get O7 for return address
1882   // add one word to size in case struct is odd number of words long
1883   // It must be doubleword-aligned for storing doubles into it.
1884 
1885     save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2));
1886 
1887     // stop_subroutine expects message pointer in I1.
1888     set((intptr_t)msg, O1);
1889 
1890     // factor long stop-sequence into subroutine to save space
1891     assert(StubRoutines::Sparc::stop_subroutine_entry_address(), "hasn't been generated yet");
1892 
1893     // call indirectly to solve generation ordering problem
1894     Address a(O5, (address)StubRoutines::Sparc::stop_subroutine_entry_address());
1895     load_ptr_contents(a, O5);
1896     jmpl(O5, 0, O7);
1897     delayed()->nop();
1898 
1899     breakpoint_trap();   // make stop actually stop rather than writing
1900                          // unnoticeable results in the output files.
1901 
1902     // restore(); done in callee to save space!
1903 }
1904 
1905 
1906 void MacroAssembler::warn(const char* msg) {
1907   save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2));
1908   RegistersForDebugging::save_registers(this);
1909   mov(O0, L0);
1910   set((intptr_t)msg, O0);
1911   call( CAST_FROM_FN_PTR(address, warning) );
1912   delayed()->nop();
1913 //  ret();
1914 //  delayed()->restore();
1915   RegistersForDebugging::restore_registers(this, L0);
1916   restore();
1917 }
1918 
1919 
1920 void MacroAssembler::untested(const char* what) {
1921   // We must be able to turn interactive prompting off
1922   // in order to run automated test scripts on the VM
1923   // Use the flag ShowMessageBoxOnError
1924 
1925   char* b = new char[1024];
1926   sprintf(b, "untested: %s", what);
1927 
1928   if ( ShowMessageBoxOnError )   stop(b);
1929   else                           warn(b);
1930 }
1931 
1932 
1933 void MacroAssembler::stop_subroutine() {
1934   RegistersForDebugging::save_registers(this);
1935 
1936   // for the sake of the debugger, stick a PC on the current frame
1937   // (this assumes that the caller has performed an extra "save")
1938   mov(I7, L7);
1939   add(O7, -7 * BytesPerInt, I7);
1940 
1941   save_frame(); // one more save to free up another O7 register
1942   mov(I0, O1); // addr of reg save area
1943 
1944   // We expect pointer to message in I1. Caller must set it up in O1
1945   mov(I1, O0); // get msg
1946   call (CAST_FROM_FN_PTR(address, MacroAssembler::debug), relocInfo::runtime_call_type);
1947   delayed()->nop();
1948 
1949   restore();
1950 
1951   RegistersForDebugging::restore_registers(this, O0);
1952 
1953   save_frame(0);
1954   call(CAST_FROM_FN_PTR(address,breakpoint));
1955   delayed()->nop();
1956   restore();
1957 
1958   mov(L7, I7);
1959   retl();
1960   delayed()->restore(); // see stop above
1961 }
1962 
1963 
1964 void MacroAssembler::debug(char* msg, RegistersForDebugging* regs) {
1965   if ( ShowMessageBoxOnError ) {
1966       JavaThreadState saved_state = JavaThread::current()->thread_state();
1967       JavaThread::current()->set_thread_state(_thread_in_vm);
1968       {
1969         // In order to get locks work, we need to fake a in_VM state
1970         ttyLocker ttyl;
1971         ::tty->print_cr("EXECUTION STOPPED: %s\n", msg);
1972         if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
1973           ::tty->print_cr("Interpreter::bytecode_counter = %d", BytecodeCounter::counter_value());
1974         }
1975         if (os::message_box(msg, "Execution stopped, print registers?"))
1976           regs->print(::tty);
1977       }
1978       ThreadStateTransition::transition(JavaThread::current(), _thread_in_vm, saved_state);
1979   }
1980   else
1981      ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
1982   assert(false, "error");
1983 }
1984 
1985 
1986 #ifndef PRODUCT
1987 void MacroAssembler::test() {
1988   ResourceMark rm;
1989 
1990   CodeBuffer cb("test", 10000, 10000);
1991   MacroAssembler* a = new MacroAssembler(&cb);
1992   VM_Version::allow_all();
1993   a->test_v9();
1994   a->test_v8_onlys();
1995   VM_Version::revert();
1996 
1997   StubRoutines::Sparc::test_stop_entry()();
1998 }
1999 #endif
2000 
2001 
2002 void MacroAssembler::calc_mem_param_words(Register Rparam_words, Register Rresult) {
2003   subcc( Rparam_words, Argument::n_register_parameters, Rresult); // how many mem words?
2004   Label no_extras;
2005   br( negative, true, pt, no_extras ); // if neg, clear reg
2006   delayed()->set( 0, Rresult);         // annuled, so only if taken
2007   bind( no_extras );
2008 }
2009 
2010 
2011 void MacroAssembler::calc_frame_size(Register Rextra_words, Register Rresult) {
2012 #ifdef _LP64
2013   add(Rextra_words, frame::memory_parameter_word_sp_offset, Rresult);
2014 #else
2015   add(Rextra_words, frame::memory_parameter_word_sp_offset + 1, Rresult);
2016 #endif
2017   bclr(1, Rresult);
2018   sll(Rresult, LogBytesPerWord, Rresult);  // Rresult has total frame bytes
2019 }
2020 
2021 
2022 void MacroAssembler::calc_frame_size_and_save(Register Rextra_words, Register Rresult) {
2023   calc_frame_size(Rextra_words, Rresult);
2024   neg(Rresult);
2025   save(SP, Rresult, SP);
2026 }
2027 
2028 
2029 // ---------------------------------------------------------
2030 Assembler::RCondition cond2rcond(Assembler::Condition c) {
2031   switch (c) {
2032     /*case zero: */
2033     case Assembler::equal:        return Assembler::rc_z;
2034     case Assembler::lessEqual:    return Assembler::rc_lez;
2035     case Assembler::less:         return Assembler::rc_lz;
2036     /*case notZero:*/
2037     case Assembler::notEqual:     return Assembler::rc_nz;
2038     case Assembler::greater:      return Assembler::rc_gz;
2039     case Assembler::greaterEqual: return Assembler::rc_gez;
2040   }
2041   ShouldNotReachHere();
2042   return Assembler::rc_z;
2043 }
2044 
2045 // compares register with zero and branches.  NOT FOR USE WITH 64-bit POINTERS
2046 void MacroAssembler::br_zero( Condition c, bool a, Predict p, Register s1, Label& L) {
2047   tst(s1);
2048   br (c, a, p, L);
2049 }
2050 
2051 
2052 // Compares a pointer register with zero and branches on null.
2053 // Does a test & branch on 32-bit systems and a register-branch on 64-bit.
2054 void MacroAssembler::br_null( Register s1, bool a, Predict p, Label& L ) {
2055   assert_not_delayed();
2056 #ifdef _LP64
2057   bpr( rc_z, a, p, s1, L );
2058 #else
2059   tst(s1);
2060   br ( zero, a, p, L );
2061 #endif
2062 }
2063 
2064 void MacroAssembler::br_notnull( Register s1, bool a, Predict p, Label& L ) {
2065   assert_not_delayed();
2066 #ifdef _LP64
2067   bpr( rc_nz, a, p, s1, L );
2068 #else
2069   tst(s1);
2070   br ( notZero, a, p, L );
2071 #endif
2072 }
2073 
2074 void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p,
2075                                      Register s1, address d,
2076                                      relocInfo::relocType rt ) {
2077   if (VM_Version::v9_instructions_work()) {
2078     bpr(rc, a, p, s1, d, rt);
2079   } else {
2080     tst(s1);
2081     br(reg_cond_to_cc_cond(rc), a, p, d, rt);
2082   }
2083 }
2084 
2085 void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p,
2086                                      Register s1, Label& L ) {
2087   if (VM_Version::v9_instructions_work()) {
2088     bpr(rc, a, p, s1, L);
2089   } else {
2090     tst(s1);
2091     br(reg_cond_to_cc_cond(rc), a, p, L);
2092   }
2093 }
2094 
2095 
2096 // instruction sequences factored across compiler & interpreter
2097 
2098 
2099 void MacroAssembler::lcmp( Register Ra_hi, Register Ra_low,
2100                            Register Rb_hi, Register Rb_low,
2101                            Register Rresult) {
2102 
2103   Label check_low_parts, done;
2104 
2105   cmp(Ra_hi, Rb_hi );  // compare hi parts
2106   br(equal, true, pt, check_low_parts);
2107   delayed()->cmp(Ra_low, Rb_low); // test low parts
2108 
2109   // And, with an unsigned comparison, it does not matter if the numbers
2110   // are negative or not.
2111   // E.g., -2 cmp -1: the low parts are 0xfffffffe and 0xffffffff.
2112   // The second one is bigger (unsignedly).
2113 
2114   // Other notes:  The first move in each triplet can be unconditional
2115   // (and therefore probably prefetchable).
2116   // And the equals case for the high part does not need testing,
2117   // since that triplet is reached only after finding the high halves differ.
2118 
2119   if (VM_Version::v9_instructions_work()) {
2120 
2121                                     mov  (                     -1, Rresult);
2122     ba( false, done );  delayed()-> movcc(greater, false, icc,  1, Rresult);
2123   }
2124   else {
2125     br(less,    true, pt, done); delayed()-> set(-1, Rresult);
2126     br(greater, true, pt, done); delayed()-> set( 1, Rresult);
2127   }
2128 
2129   bind( check_low_parts );
2130 
2131   if (VM_Version::v9_instructions_work()) {
2132     mov(                               -1, Rresult);
2133     movcc(equal,           false, icc,  0, Rresult);
2134     movcc(greaterUnsigned, false, icc,  1, Rresult);
2135   }
2136   else {
2137                                                     set(-1, Rresult);
2138     br(equal,           true, pt, done); delayed()->set( 0, Rresult);
2139     br(greaterUnsigned, true, pt, done); delayed()->set( 1, Rresult);
2140   }
2141   bind( done );
2142 }
2143 
2144 void MacroAssembler::lneg( Register Rhi, Register Rlow ) {
2145   subcc(  G0, Rlow, Rlow );
2146   subc(   G0, Rhi,  Rhi  );
2147 }
2148 
2149 void MacroAssembler::lshl( Register Rin_high,  Register Rin_low,
2150                            Register Rcount,
2151                            Register Rout_high, Register Rout_low,
2152                            Register Rtemp ) {
2153 
2154 
2155   Register Ralt_count = Rtemp;
2156   Register Rxfer_bits = Rtemp;
2157 
2158   assert( Ralt_count != Rin_high
2159       &&  Ralt_count != Rin_low
2160       &&  Ralt_count != Rcount
2161       &&  Rxfer_bits != Rin_low
2162       &&  Rxfer_bits != Rin_high
2163       &&  Rxfer_bits != Rcount
2164       &&  Rxfer_bits != Rout_low
2165       &&  Rout_low   != Rin_high,
2166         "register alias checks");
2167 
2168   Label big_shift, done;
2169 
2170   // This code can be optimized to use the 64 bit shifts in V9.
2171   // Here we use the 32 bit shifts.
2172 
2173   and3( Rcount,         0x3f,           Rcount);     // take least significant 6 bits
2174   subcc(Rcount,         31,             Ralt_count);
2175   br(greater, true, pn, big_shift);
2176   delayed()->
2177   dec(Ralt_count);
2178 
2179   // shift < 32 bits, Ralt_count = Rcount-31
2180 
2181   // We get the transfer bits by shifting right by 32-count the low
2182   // register. This is done by shifting right by 31-count and then by one
2183   // more to take care of the special (rare) case where count is zero
2184   // (shifting by 32 would not work).
2185 
2186   neg(  Ralt_count                                 );
2187 
2188   // The order of the next two instructions is critical in the case where
2189   // Rin and Rout are the same and should not be reversed.
2190 
2191   srl(  Rin_low,        Ralt_count,     Rxfer_bits ); // shift right by 31-count
2192   if (Rcount != Rout_low) {
2193     sll(        Rin_low,        Rcount,         Rout_low   ); // low half
2194   }
2195   sll(  Rin_high,       Rcount,         Rout_high  );
2196   if (Rcount == Rout_low) {
2197     sll(        Rin_low,        Rcount,         Rout_low   ); // low half
2198   }
2199   srl(  Rxfer_bits,     1,              Rxfer_bits ); // shift right by one more
2200   ba (false, done);
2201   delayed()->
2202   or3(  Rout_high,      Rxfer_bits,     Rout_high);   // new hi value: or in shifted old hi part and xfer from low
2203 
2204   // shift >= 32 bits, Ralt_count = Rcount-32
2205   bind(big_shift);
2206   sll(  Rin_low,        Ralt_count,     Rout_high  );
2207   clr(  Rout_low                                   );
2208 
2209   bind(done);
2210 }
2211 
2212 
2213 void MacroAssembler::lshr( Register Rin_high,  Register Rin_low,
2214                            Register Rcount,
2215                            Register Rout_high, Register Rout_low,
2216                            Register Rtemp ) {
2217 
2218   Register Ralt_count = Rtemp;
2219   Register Rxfer_bits = Rtemp;
2220 
2221   assert( Ralt_count != Rin_high
2222       &&  Ralt_count != Rin_low
2223       &&  Ralt_count != Rcount
2224       &&  Rxfer_bits != Rin_low
2225       &&  Rxfer_bits != Rin_high
2226       &&  Rxfer_bits != Rcount
2227       &&  Rxfer_bits != Rout_high
2228       &&  Rout_high  != Rin_low,
2229         "register alias checks");
2230 
2231   Label big_shift, done;
2232 
2233   // This code can be optimized to use the 64 bit shifts in V9.
2234   // Here we use the 32 bit shifts.
2235 
2236   and3( Rcount,         0x3f,           Rcount);     // take least significant 6 bits
2237   subcc(Rcount,         31,             Ralt_count);
2238   br(greater, true, pn, big_shift);
2239   delayed()->dec(Ralt_count);
2240 
2241   // shift < 32 bits, Ralt_count = Rcount-31
2242 
2243   // We get the transfer bits by shifting left by 32-count the high
2244   // register. This is done by shifting left by 31-count and then by one
2245   // more to take care of the special (rare) case where count is zero
2246   // (shifting by 32 would not work).
2247 
2248   neg(  Ralt_count                                  );
2249   if (Rcount != Rout_low) {
2250     srl(        Rin_low,        Rcount,         Rout_low    );
2251   }
2252 
2253   // The order of the next two instructions is critical in the case where
2254   // Rin and Rout are the same and should not be reversed.
2255 
2256   sll(  Rin_high,       Ralt_count,     Rxfer_bits  ); // shift left by 31-count
2257   sra(  Rin_high,       Rcount,         Rout_high   ); // high half
2258   sll(  Rxfer_bits,     1,              Rxfer_bits  ); // shift left by one more
2259   if (Rcount == Rout_low) {
2260     srl(        Rin_low,        Rcount,         Rout_low    );
2261   }
2262   ba (false, done);
2263   delayed()->
2264   or3(  Rout_low,       Rxfer_bits,     Rout_low    ); // new low value: or shifted old low part and xfer from high
2265 
2266   // shift >= 32 bits, Ralt_count = Rcount-32
2267   bind(big_shift);
2268 
2269   sra(  Rin_high,       Ralt_count,     Rout_low    );
2270   sra(  Rin_high,       31,             Rout_high   ); // sign into hi
2271 
2272   bind( done );
2273 }
2274 
2275 
2276 
2277 void MacroAssembler::lushr( Register Rin_high,  Register Rin_low,
2278                             Register Rcount,
2279                             Register Rout_high, Register Rout_low,
2280                             Register Rtemp ) {
2281 
2282   Register Ralt_count = Rtemp;
2283   Register Rxfer_bits = Rtemp;
2284 
2285   assert( Ralt_count != Rin_high
2286       &&  Ralt_count != Rin_low
2287       &&  Ralt_count != Rcount
2288       &&  Rxfer_bits != Rin_low
2289       &&  Rxfer_bits != Rin_high
2290       &&  Rxfer_bits != Rcount
2291       &&  Rxfer_bits != Rout_high
2292       &&  Rout_high  != Rin_low,
2293         "register alias checks");
2294 
2295   Label big_shift, done;
2296 
2297   // This code can be optimized to use the 64 bit shifts in V9.
2298   // Here we use the 32 bit shifts.
2299 
2300   and3( Rcount,         0x3f,           Rcount);     // take least significant 6 bits
2301   subcc(Rcount,         31,             Ralt_count);
2302   br(greater, true, pn, big_shift);
2303   delayed()->dec(Ralt_count);
2304 
2305   // shift < 32 bits, Ralt_count = Rcount-31
2306 
2307   // We get the transfer bits by shifting left by 32-count the high
2308   // register. This is done by shifting left by 31-count and then by one
2309   // more to take care of the special (rare) case where count is zero
2310   // (shifting by 32 would not work).
2311 
2312   neg(  Ralt_count                                  );
2313   if (Rcount != Rout_low) {
2314     srl(        Rin_low,        Rcount,         Rout_low    );
2315   }
2316 
2317   // The order of the next two instructions is critical in the case where
2318   // Rin and Rout are the same and should not be reversed.
2319 
2320   sll(  Rin_high,       Ralt_count,     Rxfer_bits  ); // shift left by 31-count
2321   srl(  Rin_high,       Rcount,         Rout_high   ); // high half
2322   sll(  Rxfer_bits,     1,              Rxfer_bits  ); // shift left by one more
2323   if (Rcount == Rout_low) {
2324     srl(        Rin_low,        Rcount,         Rout_low    );
2325   }
2326   ba (false, done);
2327   delayed()->
2328   or3(  Rout_low,       Rxfer_bits,     Rout_low    ); // new low value: or shifted old low part and xfer from high
2329 
2330   // shift >= 32 bits, Ralt_count = Rcount-32
2331   bind(big_shift);
2332 
2333   srl(  Rin_high,       Ralt_count,     Rout_low    );
2334   clr(  Rout_high                                   );
2335 
2336   bind( done );
2337 }
2338 
2339 #ifdef _LP64
2340 void MacroAssembler::lcmp( Register Ra, Register Rb, Register Rresult) {
2341   cmp(Ra, Rb);
2342   mov(                       -1, Rresult);
2343   movcc(equal,   false, xcc,  0, Rresult);
2344   movcc(greater, false, xcc,  1, Rresult);
2345 }
2346 #endif
2347 
2348 
2349 void MacroAssembler::float_cmp( bool is_float, int unordered_result,
2350                                 FloatRegister Fa, FloatRegister Fb,
2351                                 Register Rresult) {
2352 
2353   fcmp(is_float ? FloatRegisterImpl::S : FloatRegisterImpl::D, fcc0, Fa, Fb);
2354 
2355   Condition lt = unordered_result == -1 ? f_unorderedOrLess    : f_less;
2356   Condition eq =                          f_equal;
2357   Condition gt = unordered_result ==  1 ? f_unorderedOrGreater : f_greater;
2358 
2359   if (VM_Version::v9_instructions_work()) {
2360 
2361     mov(                   -1, Rresult );
2362     movcc( eq, true, fcc0,  0, Rresult );
2363     movcc( gt, true, fcc0,  1, Rresult );
2364 
2365   } else {
2366     Label done;
2367 
2368                                          set( -1, Rresult );
2369     //fb(lt, true, pn, done); delayed()->set( -1, Rresult );
2370     fb( eq, true, pn, done);  delayed()->set(  0, Rresult );
2371     fb( gt, true, pn, done);  delayed()->set(  1, Rresult );
2372 
2373     bind (done);
2374   }
2375 }
2376 
2377 
2378 void MacroAssembler::fneg( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d)
2379 {
2380   if (VM_Version::v9_instructions_work()) {
2381     Assembler::fneg(w, s, d);
2382   } else {
2383     if (w == FloatRegisterImpl::S) {
2384       Assembler::fneg(w, s, d);
2385     } else if (w == FloatRegisterImpl::D) {
2386       // number() does a sanity check on the alignment.
2387       assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) &&
2388         ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check");
2389 
2390       Assembler::fneg(FloatRegisterImpl::S, s, d);
2391       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2392     } else {
2393       assert(w == FloatRegisterImpl::Q, "Invalid float register width");
2394 
2395       // number() does a sanity check on the alignment.
2396       assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) &&
2397         ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check");
2398 
2399       Assembler::fneg(FloatRegisterImpl::S, s, d);
2400       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2401       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor());
2402       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor());
2403     }
2404   }
2405 }
2406 
2407 void MacroAssembler::fmov( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d)
2408 {
2409   if (VM_Version::v9_instructions_work()) {
2410     Assembler::fmov(w, s, d);
2411   } else {
2412     if (w == FloatRegisterImpl::S) {
2413       Assembler::fmov(w, s, d);
2414     } else if (w == FloatRegisterImpl::D) {
2415       // number() does a sanity check on the alignment.
2416       assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) &&
2417         ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check");
2418 
2419       Assembler::fmov(FloatRegisterImpl::S, s, d);
2420       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2421     } else {
2422       assert(w == FloatRegisterImpl::Q, "Invalid float register width");
2423 
2424       // number() does a sanity check on the alignment.
2425       assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) &&
2426         ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check");
2427 
2428       Assembler::fmov(FloatRegisterImpl::S, s, d);
2429       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2430       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor());
2431       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor());
2432     }
2433   }
2434 }
2435 
2436 void MacroAssembler::fabs( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d)
2437 {
2438   if (VM_Version::v9_instructions_work()) {
2439     Assembler::fabs(w, s, d);
2440   } else {
2441     if (w == FloatRegisterImpl::S) {
2442       Assembler::fabs(w, s, d);
2443     } else if (w == FloatRegisterImpl::D) {
2444       // number() does a sanity check on the alignment.
2445       assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) &&
2446         ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check");
2447 
2448       Assembler::fabs(FloatRegisterImpl::S, s, d);
2449       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2450     } else {
2451       assert(w == FloatRegisterImpl::Q, "Invalid float register width");
2452 
2453       // number() does a sanity check on the alignment.
2454       assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) &&
2455        ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check");
2456 
2457       Assembler::fabs(FloatRegisterImpl::S, s, d);
2458       Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor());
2459       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor());
2460       Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor());
2461     }
2462   }
2463 }
2464 
2465 void MacroAssembler::save_all_globals_into_locals() {
2466   mov(G1,L1);
2467   mov(G2,L2);
2468   mov(G3,L3);
2469   mov(G4,L4);
2470   mov(G5,L5);
2471   mov(G6,L6);
2472   mov(G7,L7);
2473 }
2474 
2475 void MacroAssembler::restore_globals_from_locals() {
2476   mov(L1,G1);
2477   mov(L2,G2);
2478   mov(L3,G3);
2479   mov(L4,G4);
2480   mov(L5,G5);
2481   mov(L6,G6);
2482   mov(L7,G7);
2483 }
2484 
2485 // Use for 64 bit operation.
2486 void MacroAssembler::casx_under_lock(Register top_ptr_reg, Register top_reg, Register ptr_reg, address lock_addr, bool use_call_vm)
2487 {
2488   // store ptr_reg as the new top value
2489 #ifdef _LP64
2490   casx(top_ptr_reg, top_reg, ptr_reg);
2491 #else
2492   cas_under_lock(top_ptr_reg, top_reg, ptr_reg, lock_addr, use_call_vm);
2493 #endif // _LP64
2494 }
2495 
2496 // [RGV] This routine does not handle 64 bit operations.
2497 //       use casx_under_lock() or casx directly!!!
2498 void MacroAssembler::cas_under_lock(Register top_ptr_reg, Register top_reg, Register ptr_reg, address lock_addr, bool use_call_vm)
2499 {
2500   // store ptr_reg as the new top value
2501   if (VM_Version::v9_instructions_work()) {
2502     cas(top_ptr_reg, top_reg, ptr_reg);
2503   } else {
2504 
2505     // If the register is not an out nor global, it is not visible
2506     // after the save.  Allocate a register for it, save its
2507     // value in the register save area (the save may not flush
2508     // registers to the save area).
2509 
2510     Register top_ptr_reg_after_save;
2511     Register top_reg_after_save;
2512     Register ptr_reg_after_save;
2513 
2514     if (top_ptr_reg->is_out() || top_ptr_reg->is_global()) {
2515       top_ptr_reg_after_save = top_ptr_reg->after_save();
2516     } else {
2517       Address reg_save_addr = top_ptr_reg->address_in_saved_window();
2518       top_ptr_reg_after_save = L0;
2519       st(top_ptr_reg, reg_save_addr);
2520     }
2521 
2522     if (top_reg->is_out() || top_reg->is_global()) {
2523       top_reg_after_save = top_reg->after_save();
2524     } else {
2525       Address reg_save_addr = top_reg->address_in_saved_window();
2526       top_reg_after_save = L1;
2527       st(top_reg, reg_save_addr);
2528     }
2529 
2530     if (ptr_reg->is_out() || ptr_reg->is_global()) {
2531       ptr_reg_after_save = ptr_reg->after_save();
2532     } else {
2533       Address reg_save_addr = ptr_reg->address_in_saved_window();
2534       ptr_reg_after_save = L2;
2535       st(ptr_reg, reg_save_addr);
2536     }
2537 
2538     const Register& lock_reg = L3;
2539     const Register& lock_ptr_reg = L4;
2540     const Register& value_reg = L5;
2541     const Register& yield_reg = L6;
2542     const Register& yieldall_reg = L7;
2543 
2544     save_frame();
2545 
2546     if (top_ptr_reg_after_save == L0) {
2547       ld(top_ptr_reg->address_in_saved_window().after_save(), top_ptr_reg_after_save);
2548     }
2549 
2550     if (top_reg_after_save == L1) {
2551       ld(top_reg->address_in_saved_window().after_save(), top_reg_after_save);
2552     }
2553 
2554     if (ptr_reg_after_save == L2) {
2555       ld(ptr_reg->address_in_saved_window().after_save(), ptr_reg_after_save);
2556     }
2557 
2558     Label(retry_get_lock);
2559     Label(not_same);
2560     Label(dont_yield);
2561 
2562     assert(lock_addr, "lock_address should be non null for v8");
2563     set((intptr_t)lock_addr, lock_ptr_reg);
2564     // Initialize yield counter
2565     mov(G0,yield_reg);
2566     mov(G0, yieldall_reg);
2567     set(StubRoutines::Sparc::locked, lock_reg);
2568 
2569     bind(retry_get_lock);
2570     cmp(yield_reg, V8AtomicOperationUnderLockSpinCount);
2571     br(Assembler::less, false, Assembler::pt, dont_yield);
2572     delayed()->nop();
2573 
2574     if(use_call_vm) {
2575       Untested("Need to verify global reg consistancy");
2576       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::yield_all), yieldall_reg);
2577     } else {
2578       // Save the regs and make space for a C call
2579       save(SP, -96, SP);
2580       save_all_globals_into_locals();
2581       call(CAST_FROM_FN_PTR(address,os::yield_all));
2582       delayed()->mov(yieldall_reg, O0);
2583       restore_globals_from_locals();
2584       restore();
2585     }
2586 
2587     // reset the counter
2588     mov(G0,yield_reg);
2589     add(yieldall_reg, 1, yieldall_reg);
2590 
2591     bind(dont_yield);
2592     // try to get lock
2593     swap(lock_ptr_reg, 0, lock_reg);
2594 
2595     // did we get the lock?
2596     cmp(lock_reg, StubRoutines::Sparc::unlocked);
2597     br(Assembler::notEqual, true, Assembler::pn, retry_get_lock);
2598     delayed()->add(yield_reg,1,yield_reg);
2599 
2600     // yes, got lock.  do we have the same top?
2601     ld(top_ptr_reg_after_save, 0, value_reg);
2602     cmp(value_reg, top_reg_after_save);
2603     br(Assembler::notEqual, false, Assembler::pn, not_same);
2604     delayed()->nop();
2605 
2606     // yes, same top.
2607     st(ptr_reg_after_save, top_ptr_reg_after_save, 0);
2608     membar(Assembler::StoreStore);
2609 
2610     bind(not_same);
2611     mov(value_reg, ptr_reg_after_save);
2612     st(lock_reg, lock_ptr_reg, 0); // unlock
2613 
2614     restore();
2615   }
2616 }
2617 
2618 RegisterConstant MacroAssembler::delayed_value(intptr_t* delayed_value_addr,
2619                                                Register tmp,
2620                                                int offset) {
2621   intptr_t value = *delayed_value_addr;
2622   if (value != 0)
2623     return value + offset;
2624 
2625   // load indirectly to solve generation ordering problem
2626   Address a(tmp, (address) delayed_value_addr);
2627   load_ptr_contents(a, tmp);
2628 
2629 #ifdef ASSERT
2630   tst(tmp);
2631   breakpoint_trap(zero, xcc);
2632 #endif
2633 
2634   if (offset != 0)
2635     add(tmp, offset, tmp);
2636 
2637   return tmp;
2638 }
2639 
2640 
2641 void MacroAssembler::lookup_interface_method(Register recv_klass,
2642                                              Register intf_klass,
2643                                              RegisterConstant itable_index,
2644                                              Register method_result,
2645                                              Register scan_temp,
2646                                              Label& L_no_such_interface) {
2647   Unimplemented();
2648 }
2649 
2650 
2651 // Test sub_klass against super_klass.
2652 // Fall through on failure, but branch to L_success if there is a match.
2653 // Use up the given temp_reg, but don't kill any other register.
2654 // Update the sub's secondary super cache if necesary.
2655 void MacroAssembler::check_klass_subtype(Register sub_klass,
2656                                          Register super_klass,
2657                                          Register temp_reg,
2658                                          Label& L_success) {
2659   Unimplemented();
2660 }
2661 
2662 
2663 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg,
2664                                               Register temp_reg,
2665                                               Label& wrong_method_type) {
2666   assert_different_registers(mtype_reg, mh_reg, temp_reg);
2667   // compare method type against that of the receiver
2668   RegisterConstant mhtype_offset = delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg);
2669   ld_ptr(mh_reg, mhtype_offset, temp_reg);
2670   cmp(temp_reg, mtype_reg);
2671   br(Assembler::notEqual, false, Assembler::pn, wrong_method_type);
2672   delayed()->nop();
2673 }
2674 
2675 
2676 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) {
2677   assert(mh_reg == G3_method_handle, "caller must put MH object in G3");
2678   assert_different_registers(mh_reg, temp_reg);
2679 
2680   // pick out the interpreted side of the handler
2681   ld_ptr(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg), temp_reg);
2682 
2683   // off we go...
2684   ld_ptr(temp_reg, MethodEntry::from_interpreted_entry_offset_in_bytes(), temp_reg);
2685   jmp(temp_reg, 0);
2686 
2687   // for the various stubs which take control at this point,
2688   // see MethodHandles::generate_method_handle_stub
2689 }
2690 
2691 RegisterConstant MacroAssembler::argument_offset(RegisterConstant arg_slot,
2692                                                  int extra_slot_offset) {
2693   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2694   int stackElementSize = Interpreter::stackElementWords() * wordSize;
2695   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2696   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2697   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2698   if (arg_slot.is_constant()) {
2699     offset += arg_slot.as_constant() * stackElementSize;
2700     return offset;
2701   } else {
2702     Register temp = arg_slot.as_register();
2703     sll_ptr(temp, exact_log2(stackElementSize), temp);
2704     if (offset != 0)
2705       add(temp, offset, temp);
2706     return temp;
2707   }
2708 }
2709 
2710 
2711 
2712 void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg,
2713                                           Register temp_reg,
2714                                           Label& done, Label* slow_case,
2715                                           BiasedLockingCounters* counters) {
2716   assert(UseBiasedLocking, "why call this otherwise?");
2717 
2718   if (PrintBiasedLockingStatistics) {
2719     assert_different_registers(obj_reg, mark_reg, temp_reg, O7);
2720     if (counters == NULL)
2721       counters = BiasedLocking::counters();
2722   }
2723 
2724   Label cas_label;
2725 
2726   // Biased locking
2727   // See whether the lock is currently biased toward our thread and
2728   // whether the epoch is still valid
2729   // Note that the runtime guarantees sufficient alignment of JavaThread
2730   // pointers to allow age to be placed into low bits
2731   assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
2732   and3(mark_reg, markOopDesc::biased_lock_mask_in_place, temp_reg);
2733   cmp(temp_reg, markOopDesc::biased_lock_pattern);
2734   brx(Assembler::notEqual, false, Assembler::pn, cas_label);
2735   delayed()->nop();
2736 
2737   load_klass(obj_reg, temp_reg);
2738   ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
2739   or3(G2_thread, temp_reg, temp_reg);
2740   xor3(mark_reg, temp_reg, temp_reg);
2741   andcc(temp_reg, ~((int) markOopDesc::age_mask_in_place), temp_reg);
2742   if (counters != NULL) {
2743     cond_inc(Assembler::equal, (address) counters->biased_lock_entry_count_addr(), mark_reg, temp_reg);
2744     // Reload mark_reg as we may need it later
2745     ld_ptr(Address(obj_reg, 0, oopDesc::mark_offset_in_bytes()), mark_reg);
2746   }
2747   brx(Assembler::equal, true, Assembler::pt, done);
2748   delayed()->nop();
2749 
2750   Label try_revoke_bias;
2751   Label try_rebias;
2752   Address mark_addr = Address(obj_reg, 0, oopDesc::mark_offset_in_bytes());
2753   assert(mark_addr.disp() == 0, "cas must take a zero displacement");
2754 
2755   // At this point we know that the header has the bias pattern and
2756   // that we are not the bias owner in the current epoch. We need to
2757   // figure out more details about the state of the header in order to
2758   // know what operations can be legally performed on the object's
2759   // header.
2760 
2761   // If the low three bits in the xor result aren't clear, that means
2762   // the prototype header is no longer biased and we have to revoke
2763   // the bias on this object.
2764   btst(markOopDesc::biased_lock_mask_in_place, temp_reg);
2765   brx(Assembler::notZero, false, Assembler::pn, try_revoke_bias);
2766 
2767   // Biasing is still enabled for this data type. See whether the
2768   // epoch of the current bias is still valid, meaning that the epoch
2769   // bits of the mark word are equal to the epoch bits of the
2770   // prototype header. (Note that the prototype header's epoch bits
2771   // only change at a safepoint.) If not, attempt to rebias the object
2772   // toward the current thread. Note that we must be absolutely sure
2773   // that the current epoch is invalid in order to do this because
2774   // otherwise the manipulations it performs on the mark word are
2775   // illegal.
2776   delayed()->btst(markOopDesc::epoch_mask_in_place, temp_reg);
2777   brx(Assembler::notZero, false, Assembler::pn, try_rebias);
2778 
2779   // The epoch of the current bias is still valid but we know nothing
2780   // about the owner; it might be set or it might be clear. Try to
2781   // acquire the bias of the object using an atomic operation. If this
2782   // fails we will go in to the runtime to revoke the object's bias.
2783   // Note that we first construct the presumed unbiased header so we
2784   // don't accidentally blow away another thread's valid bias.
2785   delayed()->and3(mark_reg,
2786                   markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place,
2787                   mark_reg);
2788   or3(G2_thread, mark_reg, temp_reg);
2789   casn(mark_addr.base(), mark_reg, temp_reg);
2790   // If the biasing toward our thread failed, this means that
2791   // another thread succeeded in biasing it toward itself and we
2792   // need to revoke that bias. The revocation will occur in the
2793   // interpreter runtime in the slow case.
2794   cmp(mark_reg, temp_reg);
2795   if (counters != NULL) {
2796     cond_inc(Assembler::zero, (address) counters->anonymously_biased_lock_entry_count_addr(), mark_reg, temp_reg);
2797   }
2798   if (slow_case != NULL) {
2799     brx(Assembler::notEqual, true, Assembler::pn, *slow_case);
2800     delayed()->nop();
2801   }
2802   br(Assembler::always, false, Assembler::pt, done);
2803   delayed()->nop();
2804 
2805   bind(try_rebias);
2806   // At this point we know the epoch has expired, meaning that the
2807   // current "bias owner", if any, is actually invalid. Under these
2808   // circumstances _only_, we are allowed to use the current header's
2809   // value as the comparison value when doing the cas to acquire the
2810   // bias in the current epoch. In other words, we allow transfer of
2811   // the bias from one thread to another directly in this situation.
2812   //
2813   // FIXME: due to a lack of registers we currently blow away the age
2814   // bits in this situation. Should attempt to preserve them.
2815   load_klass(obj_reg, temp_reg);
2816   ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
2817   or3(G2_thread, temp_reg, temp_reg);
2818   casn(mark_addr.base(), mark_reg, temp_reg);
2819   // If the biasing toward our thread failed, this means that
2820   // another thread succeeded in biasing it toward itself and we
2821   // need to revoke that bias. The revocation will occur in the
2822   // interpreter runtime in the slow case.
2823   cmp(mark_reg, temp_reg);
2824   if (counters != NULL) {
2825     cond_inc(Assembler::zero, (address) counters->rebiased_lock_entry_count_addr(), mark_reg, temp_reg);
2826   }
2827   if (slow_case != NULL) {
2828     brx(Assembler::notEqual, true, Assembler::pn, *slow_case);
2829     delayed()->nop();
2830   }
2831   br(Assembler::always, false, Assembler::pt, done);
2832   delayed()->nop();
2833 
2834   bind(try_revoke_bias);
2835   // The prototype mark in the klass doesn't have the bias bit set any
2836   // more, indicating that objects of this data type are not supposed
2837   // to be biased any more. We are going to try to reset the mark of
2838   // this object to the prototype value and fall through to the
2839   // CAS-based locking scheme. Note that if our CAS fails, it means
2840   // that another thread raced us for the privilege of revoking the
2841   // bias of this particular object, so it's okay to continue in the
2842   // normal locking code.
2843   //
2844   // FIXME: due to a lack of registers we currently blow away the age
2845   // bits in this situation. Should attempt to preserve them.
2846   load_klass(obj_reg, temp_reg);
2847   ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
2848   casn(mark_addr.base(), mark_reg, temp_reg);
2849   // Fall through to the normal CAS-based lock, because no matter what
2850   // the result of the above CAS, some thread must have succeeded in
2851   // removing the bias bit from the object's header.
2852   if (counters != NULL) {
2853     cmp(mark_reg, temp_reg);
2854     cond_inc(Assembler::zero, (address) counters->revoked_lock_entry_count_addr(), mark_reg, temp_reg);
2855   }
2856 
2857   bind(cas_label);
2858 }
2859 
2860 void MacroAssembler::biased_locking_exit (Address mark_addr, Register temp_reg, Label& done,
2861                                           bool allow_delay_slot_filling) {
2862   // Check for biased locking unlock case, which is a no-op
2863   // Note: we do not have to check the thread ID for two reasons.
2864   // First, the interpreter checks for IllegalMonitorStateException at
2865   // a higher level. Second, if the bias was revoked while we held the
2866   // lock, the object could not be rebiased toward another thread, so
2867   // the bias bit would be clear.
2868   ld_ptr(mark_addr, temp_reg);
2869   and3(temp_reg, markOopDesc::biased_lock_mask_in_place, temp_reg);
2870   cmp(temp_reg, markOopDesc::biased_lock_pattern);
2871   brx(Assembler::equal, allow_delay_slot_filling, Assembler::pt, done);
2872   delayed();
2873   if (!allow_delay_slot_filling) {
2874     nop();
2875   }
2876 }
2877 
2878 
2879 // CASN -- 32-64 bit switch hitter similar to the synthetic CASN provided by
2880 // Solaris/SPARC's "as".  Another apt name would be cas_ptr()
2881 
2882 void MacroAssembler::casn (Register addr_reg, Register cmp_reg, Register set_reg ) {
2883   casx_under_lock (addr_reg, cmp_reg, set_reg, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()) ;
2884 }
2885 
2886 
2887 
2888 // compiler_lock_object() and compiler_unlock_object() are direct transliterations
2889 // of i486.ad fast_lock() and fast_unlock().  See those methods for detailed comments.
2890 // The code could be tightened up considerably.
2891 //
2892 // box->dhw disposition - post-conditions at DONE_LABEL.
2893 // -   Successful inflated lock:  box->dhw != 0.
2894 //     Any non-zero value suffices.
2895 //     Consider G2_thread, rsp, boxReg, or unused_mark()
2896 // -   Successful Stack-lock: box->dhw == mark.
2897 //     box->dhw must contain the displaced mark word value
2898 // -   Failure -- icc.ZFlag == 0 and box->dhw is undefined.
2899 //     The slow-path fast_enter() and slow_enter() operators
2900 //     are responsible for setting box->dhw = NonZero (typically ::unused_mark).
2901 // -   Biased: box->dhw is undefined
2902 //
2903 // SPARC refworkload performance - specifically jetstream and scimark - are
2904 // extremely sensitive to the size of the code emitted by compiler_lock_object
2905 // and compiler_unlock_object.  Critically, the key factor is code size, not path
2906 // length.  (Simply experiments to pad CLO with unexecuted NOPs demonstrte the
2907 // effect).
2908 
2909 
2910 void MacroAssembler::compiler_lock_object(Register Roop, Register Rmark,
2911                                           Register Rbox, Register Rscratch,
2912                                           BiasedLockingCounters* counters,
2913                                           bool try_bias) {
2914    Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
2915 
2916    verify_oop(Roop);
2917    Label done ;
2918 
2919    if (counters != NULL) {
2920      inc_counter((address) counters->total_entry_count_addr(), Rmark, Rscratch);
2921    }
2922 
2923    if (EmitSync & 1) {
2924      mov    (3, Rscratch) ;
2925      st_ptr (Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
2926      cmp    (SP, G0) ;
2927      return ;
2928    }
2929 
2930    if (EmitSync & 2) {
2931 
2932      // Fetch object's markword
2933      ld_ptr(mark_addr, Rmark);
2934 
2935      if (try_bias) {
2936         biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
2937      }
2938 
2939      // Save Rbox in Rscratch to be used for the cas operation
2940      mov(Rbox, Rscratch);
2941 
2942      // set Rmark to markOop | markOopDesc::unlocked_value
2943      or3(Rmark, markOopDesc::unlocked_value, Rmark);
2944 
2945      // Initialize the box.  (Must happen before we update the object mark!)
2946      st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes());
2947 
2948      // compare object markOop with Rmark and if equal exchange Rscratch with object markOop
2949      assert(mark_addr.disp() == 0, "cas must take a zero displacement");
2950      casx_under_lock(mark_addr.base(), Rmark, Rscratch,
2951         (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
2952 
2953      // if compare/exchange succeeded we found an unlocked object and we now have locked it
2954      // hence we are done
2955      cmp(Rmark, Rscratch);
2956 #ifdef _LP64
2957      sub(Rscratch, STACK_BIAS, Rscratch);
2958 #endif
2959      brx(Assembler::equal, false, Assembler::pt, done);
2960      delayed()->sub(Rscratch, SP, Rscratch);  //pull next instruction into delay slot
2961 
2962      // we did not find an unlocked object so see if this is a recursive case
2963      // sub(Rscratch, SP, Rscratch);
2964      assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
2965      andcc(Rscratch, 0xfffff003, Rscratch);
2966      st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
2967      bind (done) ;
2968      return ;
2969    }
2970 
2971    Label Egress ;
2972 
2973    if (EmitSync & 256) {
2974       Label IsInflated ;
2975 
2976       ld_ptr (mark_addr, Rmark);           // fetch obj->mark
2977       // Triage: biased, stack-locked, neutral, inflated
2978       if (try_bias) {
2979         biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
2980         // Invariant: if control reaches this point in the emitted stream
2981         // then Rmark has not been modified.
2982       }
2983 
2984       // Store mark into displaced mark field in the on-stack basic-lock "box"
2985       // Critically, this must happen before the CAS
2986       // Maximize the ST-CAS distance to minimize the ST-before-CAS penalty.
2987       st_ptr (Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes());
2988       andcc  (Rmark, 2, G0) ;
2989       brx    (Assembler::notZero, false, Assembler::pn, IsInflated) ;
2990       delayed() ->
2991 
2992       // Try stack-lock acquisition.
2993       // Beware: the 1st instruction is in a delay slot
2994       mov    (Rbox,  Rscratch);
2995       or3    (Rmark, markOopDesc::unlocked_value, Rmark);
2996       assert (mark_addr.disp() == 0, "cas must take a zero displacement");
2997       casn   (mark_addr.base(), Rmark, Rscratch) ;
2998       cmp    (Rmark, Rscratch);
2999       brx    (Assembler::equal, false, Assembler::pt, done);
3000       delayed()->sub(Rscratch, SP, Rscratch);
3001 
3002       // Stack-lock attempt failed - check for recursive stack-lock.
3003       // See the comments below about how we might remove this case.
3004 #ifdef _LP64
3005       sub    (Rscratch, STACK_BIAS, Rscratch);
3006 #endif
3007       assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
3008       andcc  (Rscratch, 0xfffff003, Rscratch);
3009       br     (Assembler::always, false, Assembler::pt, done) ;
3010       delayed()-> st_ptr (Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
3011 
3012       bind   (IsInflated) ;
3013       if (EmitSync & 64) {
3014          // If m->owner != null goto IsLocked
3015          // Pessimistic form: Test-and-CAS vs CAS
3016          // The optimistic form avoids RTS->RTO cache line upgrades.
3017          ld_ptr (Address (Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2), Rscratch) ;
3018          andcc  (Rscratch, Rscratch, G0) ;
3019          brx    (Assembler::notZero, false, Assembler::pn, done) ;
3020          delayed()->nop() ;
3021          // m->owner == null : it's unlocked.
3022       }
3023 
3024       // Try to CAS m->owner from null to Self
3025       // Invariant: if we acquire the lock then _recursions should be 0.
3026       add    (Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark) ;
3027       mov    (G2_thread, Rscratch) ;
3028       casn   (Rmark, G0, Rscratch) ;
3029       cmp    (Rscratch, G0) ;
3030       // Intentional fall-through into done
3031    } else {
3032       // Aggressively avoid the Store-before-CAS penalty
3033       // Defer the store into box->dhw until after the CAS
3034       Label IsInflated, Recursive ;
3035 
3036 // Anticipate CAS -- Avoid RTS->RTO upgrade
3037 // prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads) ;
3038 
3039       ld_ptr (mark_addr, Rmark);           // fetch obj->mark
3040       // Triage: biased, stack-locked, neutral, inflated
3041 
3042       if (try_bias) {
3043         biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
3044         // Invariant: if control reaches this point in the emitted stream
3045         // then Rmark has not been modified.
3046       }
3047       andcc  (Rmark, 2, G0) ;
3048       brx    (Assembler::notZero, false, Assembler::pn, IsInflated) ;
3049       delayed()->                         // Beware - dangling delay-slot
3050 
3051       // Try stack-lock acquisition.
3052       // Transiently install BUSY (0) encoding in the mark word.
3053       // if the CAS of 0 into the mark was successful then we execute:
3054       //   ST box->dhw  = mark   -- save fetched mark in on-stack basiclock box
3055       //   ST obj->mark = box    -- overwrite transient 0 value
3056       // This presumes TSO, of course.
3057 
3058       mov    (0, Rscratch) ;
3059       or3    (Rmark, markOopDesc::unlocked_value, Rmark);
3060       assert (mark_addr.disp() == 0, "cas must take a zero displacement");
3061       casn   (mark_addr.base(), Rmark, Rscratch) ;
3062 // prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads) ;
3063       cmp    (Rscratch, Rmark) ;
3064       brx    (Assembler::notZero, false, Assembler::pn, Recursive) ;
3065       delayed() ->
3066         st_ptr (Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes());
3067       if (counters != NULL) {
3068         cond_inc(Assembler::equal, (address) counters->fast_path_entry_count_addr(), Rmark, Rscratch);
3069       }
3070       br     (Assembler::always, false, Assembler::pt, done);
3071       delayed() ->
3072         st_ptr (Rbox, mark_addr) ;
3073 
3074       bind   (Recursive) ;
3075       // Stack-lock attempt failed - check for recursive stack-lock.
3076       // Tests show that we can remove the recursive case with no impact
3077       // on refworkload 0.83.  If we need to reduce the size of the code
3078       // emitted by compiler_lock_object() the recursive case is perfect
3079       // candidate.
3080       //
3081       // A more extreme idea is to always inflate on stack-lock recursion.
3082       // This lets us eliminate the recursive checks in compiler_lock_object
3083       // and compiler_unlock_object and the (box->dhw == 0) encoding.
3084       // A brief experiment - requiring changes to synchronizer.cpp, interpreter,
3085       // and showed a performance *increase*.  In the same experiment I eliminated
3086       // the fast-path stack-lock code from the interpreter and always passed
3087       // control to the "slow" operators in synchronizer.cpp.
3088 
3089       // RScratch contains the fetched obj->mark value from the failed CASN.
3090 #ifdef _LP64
3091       sub    (Rscratch, STACK_BIAS, Rscratch);
3092 #endif
3093       sub(Rscratch, SP, Rscratch);
3094       assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
3095       andcc  (Rscratch, 0xfffff003, Rscratch);
3096       if (counters != NULL) {
3097         // Accounting needs the Rscratch register
3098         st_ptr (Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
3099         cond_inc(Assembler::equal, (address) counters->fast_path_entry_count_addr(), Rmark, Rscratch);
3100         br     (Assembler::always, false, Assembler::pt, done) ;
3101         delayed()->nop() ;
3102       } else {
3103         br     (Assembler::always, false, Assembler::pt, done) ;
3104         delayed()-> st_ptr (Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
3105       }
3106 
3107       bind   (IsInflated) ;
3108       if (EmitSync & 64) {
3109          // If m->owner != null goto IsLocked
3110          // Test-and-CAS vs CAS
3111          // Pessimistic form avoids futile (doomed) CAS attempts
3112          // The optimistic form avoids RTS->RTO cache line upgrades.
3113          ld_ptr (Address (Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2), Rscratch) ;
3114          andcc  (Rscratch, Rscratch, G0) ;
3115          brx    (Assembler::notZero, false, Assembler::pn, done) ;
3116          delayed()->nop() ;
3117          // m->owner == null : it's unlocked.
3118       }
3119 
3120       // Try to CAS m->owner from null to Self
3121       // Invariant: if we acquire the lock then _recursions should be 0.
3122       add    (Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark) ;
3123       mov    (G2_thread, Rscratch) ;
3124       casn   (Rmark, G0, Rscratch) ;
3125       cmp    (Rscratch, G0) ;
3126       // ST box->displaced_header = NonZero.
3127       // Any non-zero value suffices:
3128       //    unused_mark(), G2_thread, RBox, RScratch, rsp, etc.
3129       st_ptr (Rbox, Rbox, BasicLock::displaced_header_offset_in_bytes());
3130       // Intentional fall-through into done
3131    }
3132 
3133    bind   (done) ;
3134 }
3135 
3136 void MacroAssembler::compiler_unlock_object(Register Roop, Register Rmark,
3137                                             Register Rbox, Register Rscratch,
3138                                             bool try_bias) {
3139    Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
3140 
3141    Label done ;
3142 
3143    if (EmitSync & 4) {
3144      cmp  (SP, G0) ;
3145      return ;
3146    }
3147 
3148    if (EmitSync & 8) {
3149      if (try_bias) {
3150         biased_locking_exit(mark_addr, Rscratch, done);
3151      }
3152 
3153      // Test first if it is a fast recursive unlock
3154      ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
3155      cmp(Rmark, G0);
3156      brx(Assembler::equal, false, Assembler::pt, done);
3157      delayed()->nop();
3158 
3159      // Check if it is still a light weight lock, this is is true if we see
3160      // the stack address of the basicLock in the markOop of the object
3161      assert(mark_addr.disp() == 0, "cas must take a zero displacement");
3162      casx_under_lock(mark_addr.base(), Rbox, Rmark,
3163        (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
3164      br (Assembler::always, false, Assembler::pt, done);
3165      delayed()->cmp(Rbox, Rmark);
3166      bind (done) ;
3167      return ;
3168    }
3169 
3170    // Beware ... If the aggregate size of the code emitted by CLO and CUO is
3171    // is too large performance rolls abruptly off a cliff.
3172    // This could be related to inlining policies, code cache management, or
3173    // I$ effects.
3174    Label LStacked ;
3175 
3176    if (try_bias) {
3177       // TODO: eliminate redundant LDs of obj->mark
3178       biased_locking_exit(mark_addr, Rscratch, done);
3179    }
3180 
3181    ld_ptr (Roop, oopDesc::mark_offset_in_bytes(), Rmark) ;
3182    ld_ptr (Rbox, BasicLock::displaced_header_offset_in_bytes(), Rscratch);
3183    andcc  (Rscratch, Rscratch, G0);
3184    brx    (Assembler::zero, false, Assembler::pn, done);
3185    delayed()-> nop() ;      // consider: relocate fetch of mark, above, into this DS
3186    andcc  (Rmark, 2, G0) ;
3187    brx    (Assembler::zero, false, Assembler::pt, LStacked) ;
3188    delayed()-> nop() ;
3189 
3190    // It's inflated
3191    // Conceptually we need a #loadstore|#storestore "release" MEMBAR before
3192    // the ST of 0 into _owner which releases the lock.  This prevents loads
3193    // and stores within the critical section from reordering (floating)
3194    // past the store that releases the lock.  But TSO is a strong memory model
3195    // and that particular flavor of barrier is a noop, so we can safely elide it.
3196    // Note that we use 1-0 locking by default for the inflated case.  We
3197    // close the resultant (and rare) race by having contented threads in
3198    // monitorenter periodically poll _owner.
3199    ld_ptr (Address(Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2), Rscratch) ;
3200    ld_ptr (Address(Rmark, 0, ObjectMonitor::recursions_offset_in_bytes()-2), Rbox) ;
3201    xor3   (Rscratch, G2_thread, Rscratch) ;
3202    orcc   (Rbox, Rscratch, Rbox) ;
3203    brx    (Assembler::notZero, false, Assembler::pn, done) ;
3204    delayed()->
3205    ld_ptr (Address (Rmark, 0, ObjectMonitor::EntryList_offset_in_bytes()-2), Rscratch) ;
3206    ld_ptr (Address (Rmark, 0, ObjectMonitor::cxq_offset_in_bytes()-2), Rbox) ;
3207    orcc   (Rbox, Rscratch, G0) ;
3208    if (EmitSync & 65536) {
3209       Label LSucc ;
3210       brx    (Assembler::notZero, false, Assembler::pn, LSucc) ;
3211       delayed()->nop() ;
3212       br     (Assembler::always, false, Assembler::pt, done) ;
3213       delayed()->
3214       st_ptr (G0, Address (Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3215 
3216       bind   (LSucc) ;
3217       st_ptr (G0, Address (Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3218       if (os::is_MP()) { membar (StoreLoad) ; }
3219       ld_ptr (Address (Rmark, 0, ObjectMonitor::succ_offset_in_bytes()-2), Rscratch) ;
3220       andcc  (Rscratch, Rscratch, G0) ;
3221       brx    (Assembler::notZero, false, Assembler::pt, done) ;
3222       delayed()-> andcc (G0, G0, G0) ;
3223       add    (Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark) ;
3224       mov    (G2_thread, Rscratch) ;
3225       casn   (Rmark, G0, Rscratch) ;
3226       cmp    (Rscratch, G0) ;
3227       // invert icc.zf and goto done
3228       brx    (Assembler::notZero, false, Assembler::pt, done) ;
3229       delayed() -> cmp (G0, G0) ;
3230       br     (Assembler::always, false, Assembler::pt, done);
3231       delayed() -> cmp (G0, 1) ;
3232    } else {
3233       brx    (Assembler::notZero, false, Assembler::pn, done) ;
3234       delayed()->nop() ;
3235       br     (Assembler::always, false, Assembler::pt, done) ;
3236       delayed()->
3237       st_ptr (G0, Address (Rmark, 0, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3238    }
3239 
3240    bind   (LStacked) ;
3241    // Consider: we could replace the expensive CAS in the exit
3242    // path with a simple ST of the displaced mark value fetched from
3243    // the on-stack basiclock box.  That admits a race where a thread T2
3244    // in the slow lock path -- inflating with monitor M -- could race a
3245    // thread T1 in the fast unlock path, resulting in a missed wakeup for T2.
3246    // More precisely T1 in the stack-lock unlock path could "stomp" the
3247    // inflated mark value M installed by T2, resulting in an orphan
3248    // object monitor M and T2 becoming stranded.  We can remedy that situation
3249    // by having T2 periodically poll the object's mark word using timed wait
3250    // operations.  If T2 discovers that a stomp has occurred it vacates
3251    // the monitor M and wakes any other threads stranded on the now-orphan M.
3252    // In addition the monitor scavenger, which performs deflation,
3253    // would also need to check for orpan monitors and stranded threads.
3254    //
3255    // Finally, inflation is also used when T2 needs to assign a hashCode
3256    // to O and O is stack-locked by T1.  The "stomp" race could cause
3257    // an assigned hashCode value to be lost.  We can avoid that condition
3258    // and provide the necessary hashCode stability invariants by ensuring
3259    // that hashCode generation is idempotent between copying GCs.
3260    // For example we could compute the hashCode of an object O as
3261    // O's heap address XOR some high quality RNG value that is refreshed
3262    // at GC-time.  The monitor scavenger would install the hashCode
3263    // found in any orphan monitors.  Again, the mechanism admits a
3264    // lost-update "stomp" WAW race but detects and recovers as needed.
3265    //
3266    // A prototype implementation showed excellent results, although
3267    // the scavenger and timeout code was rather involved.
3268 
3269    casn   (mark_addr.base(), Rbox, Rscratch) ;
3270    cmp    (Rbox, Rscratch);
3271    // Intentional fall through into done ...
3272 
3273    bind   (done) ;
3274 }
3275 
3276 
3277 
3278 void MacroAssembler::print_CPU_state() {
3279   // %%%%% need to implement this
3280 }
3281 
3282 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
3283   // %%%%% need to implement this
3284 }
3285 
3286 void MacroAssembler::push_IU_state() {
3287   // %%%%% need to implement this
3288 }
3289 
3290 
3291 void MacroAssembler::pop_IU_state() {
3292   // %%%%% need to implement this
3293 }
3294 
3295 
3296 void MacroAssembler::push_FPU_state() {
3297   // %%%%% need to implement this
3298 }
3299 
3300 
3301 void MacroAssembler::pop_FPU_state() {
3302   // %%%%% need to implement this
3303 }
3304 
3305 
3306 void MacroAssembler::push_CPU_state() {
3307   // %%%%% need to implement this
3308 }
3309 
3310 
3311 void MacroAssembler::pop_CPU_state() {
3312   // %%%%% need to implement this
3313 }
3314 
3315 
3316 
3317 void MacroAssembler::verify_tlab() {
3318 #ifdef ASSERT
3319   if (UseTLAB && VerifyOops) {
3320     Label next, next2, ok;
3321     Register t1 = L0;
3322     Register t2 = L1;
3323     Register t3 = L2;
3324 
3325     save_frame(0);
3326     ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), t1);
3327     ld_ptr(G2_thread, in_bytes(JavaThread::tlab_start_offset()), t2);
3328     or3(t1, t2, t3);
3329     cmp(t1, t2);
3330     br(Assembler::greaterEqual, false, Assembler::pn, next);
3331     delayed()->nop();
3332     stop("assert(top >= start)");
3333     should_not_reach_here();
3334 
3335     bind(next);
3336     ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), t1);
3337     ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), t2);
3338     or3(t3, t2, t3);
3339     cmp(t1, t2);
3340     br(Assembler::lessEqual, false, Assembler::pn, next2);
3341     delayed()->nop();
3342     stop("assert(top <= end)");
3343     should_not_reach_here();
3344 
3345     bind(next2);
3346     and3(t3, MinObjAlignmentInBytesMask, t3);
3347     cmp(t3, 0);
3348     br(Assembler::lessEqual, false, Assembler::pn, ok);
3349     delayed()->nop();
3350     stop("assert(aligned)");
3351     should_not_reach_here();
3352 
3353     bind(ok);
3354     restore();
3355   }
3356 #endif
3357 }
3358 
3359 
3360 void MacroAssembler::eden_allocate(
3361   Register obj,                        // result: pointer to object after successful allocation
3362   Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
3363   int      con_size_in_bytes,          // object size in bytes if   known at compile time
3364   Register t1,                         // temp register
3365   Register t2,                         // temp register
3366   Label&   slow_case                   // continuation point if fast allocation fails
3367 ){
3368   // make sure arguments make sense
3369   assert_different_registers(obj, var_size_in_bytes, t1, t2);
3370   assert(0 <= con_size_in_bytes && Assembler::is_simm13(con_size_in_bytes), "illegal object size");
3371   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, "object size is not multiple of alignment");
3372 
3373   if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
3374     // No allocation in the shared eden.
3375     br(Assembler::always, false, Assembler::pt, slow_case);
3376     delayed()->nop();
3377   } else {
3378     // get eden boundaries
3379     // note: we need both top & top_addr!
3380     const Register top_addr = t1;
3381     const Register end      = t2;
3382 
3383     CollectedHeap* ch = Universe::heap();
3384     set((intx)ch->top_addr(), top_addr);
3385     intx delta = (intx)ch->end_addr() - (intx)ch->top_addr();
3386     ld_ptr(top_addr, delta, end);
3387     ld_ptr(top_addr, 0, obj);
3388 
3389     // try to allocate
3390     Label retry;
3391     bind(retry);
3392 #ifdef ASSERT
3393     // make sure eden top is properly aligned
3394     {
3395       Label L;
3396       btst(MinObjAlignmentInBytesMask, obj);
3397       br(Assembler::zero, false, Assembler::pt, L);
3398       delayed()->nop();
3399       stop("eden top is not properly aligned");
3400       bind(L);
3401     }
3402 #endif // ASSERT
3403     const Register free = end;
3404     sub(end, obj, free);                                   // compute amount of free space
3405     if (var_size_in_bytes->is_valid()) {
3406       // size is unknown at compile time
3407       cmp(free, var_size_in_bytes);
3408       br(Assembler::lessUnsigned, false, Assembler::pn, slow_case); // if there is not enough space go the slow case
3409       delayed()->add(obj, var_size_in_bytes, end);
3410     } else {
3411       // size is known at compile time
3412       cmp(free, con_size_in_bytes);
3413       br(Assembler::lessUnsigned, false, Assembler::pn, slow_case); // if there is not enough space go the slow case
3414       delayed()->add(obj, con_size_in_bytes, end);
3415     }
3416     // Compare obj with the value at top_addr; if still equal, swap the value of
3417     // end with the value at top_addr. If not equal, read the value at top_addr
3418     // into end.
3419     casx_under_lock(top_addr, obj, end, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
3420     // if someone beat us on the allocation, try again, otherwise continue
3421     cmp(obj, end);
3422     brx(Assembler::notEqual, false, Assembler::pn, retry);
3423     delayed()->mov(end, obj);                              // nop if successfull since obj == end
3424 
3425 #ifdef ASSERT
3426     // make sure eden top is properly aligned
3427     {
3428       Label L;
3429       const Register top_addr = t1;
3430 
3431       set((intx)ch->top_addr(), top_addr);
3432       ld_ptr(top_addr, 0, top_addr);
3433       btst(MinObjAlignmentInBytesMask, top_addr);
3434       br(Assembler::zero, false, Assembler::pt, L);
3435       delayed()->nop();
3436       stop("eden top is not properly aligned");
3437       bind(L);
3438     }
3439 #endif // ASSERT
3440   }
3441 }
3442 
3443 
3444 void MacroAssembler::tlab_allocate(
3445   Register obj,                        // result: pointer to object after successful allocation
3446   Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
3447   int      con_size_in_bytes,          // object size in bytes if   known at compile time
3448   Register t1,                         // temp register
3449   Label&   slow_case                   // continuation point if fast allocation fails
3450 ){
3451   // make sure arguments make sense
3452   assert_different_registers(obj, var_size_in_bytes, t1);
3453   assert(0 <= con_size_in_bytes && is_simm13(con_size_in_bytes), "illegal object size");
3454   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, "object size is not multiple of alignment");
3455 
3456   const Register free  = t1;
3457 
3458   verify_tlab();
3459 
3460   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), obj);
3461 
3462   // calculate amount of free space
3463   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), free);
3464   sub(free, obj, free);
3465 
3466   Label done;
3467   if (var_size_in_bytes == noreg) {
3468     cmp(free, con_size_in_bytes);
3469   } else {
3470     cmp(free, var_size_in_bytes);
3471   }
3472   br(Assembler::less, false, Assembler::pn, slow_case);
3473   // calculate the new top pointer
3474   if (var_size_in_bytes == noreg) {
3475     delayed()->add(obj, con_size_in_bytes, free);
3476   } else {
3477     delayed()->add(obj, var_size_in_bytes, free);
3478   }
3479 
3480   bind(done);
3481 
3482 #ifdef ASSERT
3483   // make sure new free pointer is properly aligned
3484   {
3485     Label L;
3486     btst(MinObjAlignmentInBytesMask, free);
3487     br(Assembler::zero, false, Assembler::pt, L);
3488     delayed()->nop();
3489     stop("updated TLAB free is not properly aligned");
3490     bind(L);
3491   }
3492 #endif // ASSERT
3493 
3494   // update the tlab top pointer
3495   st_ptr(free, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3496   verify_tlab();
3497 }
3498 
3499 
3500 void MacroAssembler::tlab_refill(Label& retry, Label& try_eden, Label& slow_case) {
3501   Register top = O0;
3502   Register t1 = G1;
3503   Register t2 = G3;
3504   Register t3 = O1;
3505   assert_different_registers(top, t1, t2, t3, G4, G5 /* preserve G4 and G5 */);
3506   Label do_refill, discard_tlab;
3507 
3508   if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
3509     // No allocation in the shared eden.
3510     br(Assembler::always, false, Assembler::pt, slow_case);
3511     delayed()->nop();
3512   }
3513 
3514   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), top);
3515   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), t1);
3516   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), t2);
3517 
3518   // calculate amount of free space
3519   sub(t1, top, t1);
3520   srl_ptr(t1, LogHeapWordSize, t1);
3521 
3522   // Retain tlab and allocate object in shared space if
3523   // the amount free in the tlab is too large to discard.
3524   cmp(t1, t2);
3525   brx(Assembler::lessEqual, false, Assembler::pt, discard_tlab);
3526 
3527   // increment waste limit to prevent getting stuck on this slow path
3528   delayed()->add(t2, ThreadLocalAllocBuffer::refill_waste_limit_increment(), t2);
3529   st_ptr(t2, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
3530   if (TLABStats) {
3531     // increment number of slow_allocations
3532     ld(G2_thread, in_bytes(JavaThread::tlab_slow_allocations_offset()), t2);
3533     add(t2, 1, t2);
3534     stw(t2, G2_thread, in_bytes(JavaThread::tlab_slow_allocations_offset()));
3535   }
3536   br(Assembler::always, false, Assembler::pt, try_eden);
3537   delayed()->nop();
3538 
3539   bind(discard_tlab);
3540   if (TLABStats) {
3541     // increment number of refills
3542     ld(G2_thread, in_bytes(JavaThread::tlab_number_of_refills_offset()), t2);
3543     add(t2, 1, t2);
3544     stw(t2, G2_thread, in_bytes(JavaThread::tlab_number_of_refills_offset()));
3545     // accumulate wastage
3546     ld(G2_thread, in_bytes(JavaThread::tlab_fast_refill_waste_offset()), t2);
3547     add(t2, t1, t2);
3548     stw(t2, G2_thread, in_bytes(JavaThread::tlab_fast_refill_waste_offset()));
3549   }
3550 
3551   // if tlab is currently allocated (top or end != null) then
3552   // fill [top, end + alignment_reserve) with array object
3553   br_null(top, false, Assembler::pn, do_refill);
3554   delayed()->nop();
3555 
3556   set((intptr_t)markOopDesc::prototype()->copy_set_hash(0x2), t2);
3557   st_ptr(t2, top, oopDesc::mark_offset_in_bytes()); // set up the mark word
3558   // set klass to intArrayKlass
3559   sub(t1, typeArrayOopDesc::header_size(T_INT), t1);
3560   add(t1, ThreadLocalAllocBuffer::alignment_reserve(), t1);
3561   sll_ptr(t1, log2_intptr(HeapWordSize/sizeof(jint)), t1);
3562   st(t1, top, arrayOopDesc::length_offset_in_bytes());
3563   set((intptr_t)Universe::intArrayKlassObj_addr(), t2);
3564   ld_ptr(t2, 0, t2);
3565   // store klass last.  concurrent gcs assumes klass length is valid if
3566   // klass field is not null.
3567   store_klass(t2, top);
3568   verify_oop(top);
3569 
3570   // refill the tlab with an eden allocation
3571   bind(do_refill);
3572   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t1);
3573   sll_ptr(t1, LogHeapWordSize, t1);
3574   // add object_size ??
3575   eden_allocate(top, t1, 0, t2, t3, slow_case);
3576 
3577   st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_start_offset()));
3578   st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3579 #ifdef ASSERT
3580   // check that tlab_size (t1) is still valid
3581   {
3582     Label ok;
3583     ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t2);
3584     sll_ptr(t2, LogHeapWordSize, t2);
3585     cmp(t1, t2);
3586     br(Assembler::equal, false, Assembler::pt, ok);
3587     delayed()->nop();
3588     stop("assert(t1 == tlab_size)");
3589     should_not_reach_here();
3590 
3591     bind(ok);
3592   }
3593 #endif // ASSERT
3594   add(top, t1, top); // t1 is tlab_size
3595   sub(top, ThreadLocalAllocBuffer::alignment_reserve_in_bytes(), top);
3596   st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_end_offset()));
3597   verify_tlab();
3598   br(Assembler::always, false, Assembler::pt, retry);
3599   delayed()->nop();
3600 }
3601 
3602 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
3603   switch (cond) {
3604     // Note some conditions are synonyms for others
3605     case Assembler::never:                return Assembler::always;
3606     case Assembler::zero:                 return Assembler::notZero;
3607     case Assembler::lessEqual:            return Assembler::greater;
3608     case Assembler::less:                 return Assembler::greaterEqual;
3609     case Assembler::lessEqualUnsigned:    return Assembler::greaterUnsigned;
3610     case Assembler::lessUnsigned:         return Assembler::greaterEqualUnsigned;
3611     case Assembler::negative:             return Assembler::positive;
3612     case Assembler::overflowSet:          return Assembler::overflowClear;
3613     case Assembler::always:               return Assembler::never;
3614     case Assembler::notZero:              return Assembler::zero;
3615     case Assembler::greater:              return Assembler::lessEqual;
3616     case Assembler::greaterEqual:         return Assembler::less;
3617     case Assembler::greaterUnsigned:      return Assembler::lessEqualUnsigned;
3618     case Assembler::greaterEqualUnsigned: return Assembler::lessUnsigned;
3619     case Assembler::positive:             return Assembler::negative;
3620     case Assembler::overflowClear:        return Assembler::overflowSet;
3621   }
3622 
3623   ShouldNotReachHere(); return Assembler::overflowClear;
3624 }
3625 
3626 void MacroAssembler::cond_inc(Assembler::Condition cond, address counter_ptr,
3627                               Register Rtmp1, Register Rtmp2 /*, Register Rtmp3, Register Rtmp4 */) {
3628   Condition negated_cond = negate_condition(cond);
3629   Label L;
3630   brx(negated_cond, false, Assembler::pt, L);
3631   delayed()->nop();
3632   inc_counter(counter_ptr, Rtmp1, Rtmp2);
3633   bind(L);
3634 }
3635 
3636 void MacroAssembler::inc_counter(address counter_ptr, Register Rtmp1, Register Rtmp2) {
3637   Address counter_addr(Rtmp1, counter_ptr);
3638   load_contents(counter_addr, Rtmp2);
3639   inc(Rtmp2);
3640   store_contents(Rtmp2, counter_addr);
3641 }
3642 
3643 SkipIfEqual::SkipIfEqual(
3644     MacroAssembler* masm, Register temp, const bool* flag_addr,
3645     Assembler::Condition condition) {
3646   _masm = masm;
3647   Address flag(temp, (address)flag_addr, relocInfo::none);
3648   _masm->sethi(flag);
3649   _masm->ldub(flag, temp);
3650   _masm->tst(temp);
3651   _masm->br(condition, false, Assembler::pt, _label);
3652   _masm->delayed()->nop();
3653 }
3654 
3655 SkipIfEqual::~SkipIfEqual() {
3656   _masm->bind(_label);
3657 }
3658 
3659 
3660 // Writes to stack successive pages until offset reached to check for
3661 // stack overflow + shadow pages.  This clobbers tsp and scratch.
3662 void MacroAssembler::bang_stack_size(Register Rsize, Register Rtsp,
3663                                      Register Rscratch) {
3664   // Use stack pointer in temp stack pointer
3665   mov(SP, Rtsp);
3666 
3667   // Bang stack for total size given plus stack shadow page size.
3668   // Bang one page at a time because a large size can overflow yellow and
3669   // red zones (the bang will fail but stack overflow handling can't tell that
3670   // it was a stack overflow bang vs a regular segv).
3671   int offset = os::vm_page_size();
3672   Register Roffset = Rscratch;
3673 
3674   Label loop;
3675   bind(loop);
3676   set((-offset)+STACK_BIAS, Rscratch);
3677   st(G0, Rtsp, Rscratch);
3678   set(offset, Roffset);
3679   sub(Rsize, Roffset, Rsize);
3680   cmp(Rsize, G0);
3681   br(Assembler::greater, false, Assembler::pn, loop);
3682   delayed()->sub(Rtsp, Roffset, Rtsp);
3683 
3684   // Bang down shadow pages too.
3685   // The -1 because we already subtracted 1 page.
3686   for (int i = 0; i< StackShadowPages-1; i++) {
3687     set((-i*offset)+STACK_BIAS, Rscratch);
3688     st(G0, Rtsp, Rscratch);
3689   }
3690 }
3691 
3692 ///////////////////////////////////////////////////////////////////////////////////
3693 #ifndef SERIALGC
3694 
3695 static uint num_stores = 0;
3696 static uint num_null_pre_stores = 0;
3697 
3698 static void count_null_pre_vals(void* pre_val) {
3699   num_stores++;
3700   if (pre_val == NULL) num_null_pre_stores++;
3701   if ((num_stores % 1000000) == 0) {
3702     tty->print_cr(UINT32_FORMAT " stores, " UINT32_FORMAT " (%5.2f%%) with null pre-vals.",
3703                   num_stores, num_null_pre_stores,
3704                   100.0*(float)num_null_pre_stores/(float)num_stores);
3705   }
3706 }
3707 
3708 static address satb_log_enqueue_with_frame = 0;
3709 static u_char* satb_log_enqueue_with_frame_end = 0;
3710 
3711 static address satb_log_enqueue_frameless = 0;
3712 static u_char* satb_log_enqueue_frameless_end = 0;
3713 
3714 static int EnqueueCodeSize = 128 DEBUG_ONLY( + 256); // Instructions?
3715 
3716 // The calls to this don't work.  We'd need to do a fair amount of work to
3717 // make it work.
3718 static void check_index(int ind) {
3719   assert(0 <= ind && ind <= 64*K && ((ind % oopSize) == 0),
3720          "Invariants.")
3721 }
3722 
3723 static void generate_satb_log_enqueue(bool with_frame) {
3724   BufferBlob* bb = BufferBlob::create("enqueue_with_frame", EnqueueCodeSize);
3725   CodeBuffer buf(bb->instructions_begin(), bb->instructions_size());
3726   MacroAssembler masm(&buf);
3727   address start = masm.pc();
3728   Register pre_val;
3729 
3730   Label refill, restart;
3731   if (with_frame) {
3732     masm.save_frame(0);
3733     pre_val = I0;  // Was O0 before the save.
3734   } else {
3735     pre_val = O0;
3736   }
3737   int satb_q_index_byte_offset =
3738     in_bytes(JavaThread::satb_mark_queue_offset() +
3739              PtrQueue::byte_offset_of_index());
3740   int satb_q_buf_byte_offset =
3741     in_bytes(JavaThread::satb_mark_queue_offset() +
3742              PtrQueue::byte_offset_of_buf());
3743   assert(in_bytes(PtrQueue::byte_width_of_index()) == sizeof(intptr_t) &&
3744          in_bytes(PtrQueue::byte_width_of_buf()) == sizeof(intptr_t),
3745          "check sizes in assembly below");
3746 
3747   masm.bind(restart);
3748   masm.ld_ptr(G2_thread, satb_q_index_byte_offset, L0);
3749 
3750   masm.br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn, L0, refill);
3751   // If the branch is taken, no harm in executing this in the delay slot.
3752   masm.delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, L1);
3753   masm.sub(L0, oopSize, L0);
3754 
3755   masm.st_ptr(pre_val, L1, L0);  // [_buf + index] := I0
3756   if (!with_frame) {
3757     // Use return-from-leaf
3758     masm.retl();
3759     masm.delayed()->st_ptr(L0, G2_thread, satb_q_index_byte_offset);
3760   } else {
3761     // Not delayed.
3762     masm.st_ptr(L0, G2_thread, satb_q_index_byte_offset);
3763   }
3764   if (with_frame) {
3765     masm.ret();
3766     masm.delayed()->restore();
3767   }
3768   masm.bind(refill);
3769 
3770   address handle_zero =
3771     CAST_FROM_FN_PTR(address,
3772                      &SATBMarkQueueSet::handle_zero_index_for_thread);
3773   // This should be rare enough that we can afford to save all the
3774   // scratch registers that the calling context might be using.
3775   masm.mov(G1_scratch, L0);
3776   masm.mov(G3_scratch, L1);
3777   masm.mov(G4, L2);
3778   // We need the value of O0 above (for the write into the buffer), so we
3779   // save and restore it.
3780   masm.mov(O0, L3);
3781   // Since the call will overwrite O7, we save and restore that, as well.
3782   masm.mov(O7, L4);
3783   masm.call_VM_leaf(L5, handle_zero, G2_thread);
3784   masm.mov(L0, G1_scratch);
3785   masm.mov(L1, G3_scratch);
3786   masm.mov(L2, G4);
3787   masm.mov(L3, O0);
3788   masm.br(Assembler::always, /*annul*/false, Assembler::pt, restart);
3789   masm.delayed()->mov(L4, O7);
3790 
3791   if (with_frame) {
3792     satb_log_enqueue_with_frame = start;
3793     satb_log_enqueue_with_frame_end = masm.pc();
3794   } else {
3795     satb_log_enqueue_frameless = start;
3796     satb_log_enqueue_frameless_end = masm.pc();
3797   }
3798 }
3799 
3800 static inline void generate_satb_log_enqueue_if_necessary(bool with_frame) {
3801   if (with_frame) {
3802     if (satb_log_enqueue_with_frame == 0) {
3803       generate_satb_log_enqueue(with_frame);
3804       assert(satb_log_enqueue_with_frame != 0, "postcondition.");
3805       if (G1SATBPrintStubs) {
3806         tty->print_cr("Generated with-frame satb enqueue:");
3807         Disassembler::decode((u_char*)satb_log_enqueue_with_frame,
3808                              satb_log_enqueue_with_frame_end,
3809                              tty);
3810       }
3811     }
3812   } else {
3813     if (satb_log_enqueue_frameless == 0) {
3814       generate_satb_log_enqueue(with_frame);
3815       assert(satb_log_enqueue_frameless != 0, "postcondition.");
3816       if (G1SATBPrintStubs) {
3817         tty->print_cr("Generated frameless satb enqueue:");
3818         Disassembler::decode((u_char*)satb_log_enqueue_frameless,
3819                              satb_log_enqueue_frameless_end,
3820                              tty);
3821       }
3822     }
3823   }
3824 }
3825 
3826 void MacroAssembler::g1_write_barrier_pre(Register obj, Register index, int offset, Register tmp, bool preserve_o_regs) {
3827   assert(offset == 0 || index == noreg, "choose one");
3828 
3829   if (G1DisablePreBarrier) return;
3830   // satb_log_barrier(tmp, obj, offset, preserve_o_regs);
3831   Label filtered;
3832   // satb_log_barrier_work0(tmp, filtered);
3833   if (in_bytes(PtrQueue::byte_width_of_active()) == 4) {
3834     ld(G2,
3835        in_bytes(JavaThread::satb_mark_queue_offset() +
3836                 PtrQueue::byte_offset_of_active()),
3837        tmp);
3838   } else {
3839     guarantee(in_bytes(PtrQueue::byte_width_of_active()) == 1,
3840               "Assumption");
3841     ldsb(G2,
3842          in_bytes(JavaThread::satb_mark_queue_offset() +
3843                   PtrQueue::byte_offset_of_active()),
3844          tmp);
3845   }
3846   // Check on whether to annul.
3847   br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered);
3848   delayed() -> nop();
3849 
3850   // satb_log_barrier_work1(tmp, offset);
3851   if (index == noreg) {
3852     if (Assembler::is_simm13(offset)) {
3853       ld_ptr(obj, offset, tmp);
3854     } else {
3855       set(offset, tmp);
3856       ld_ptr(obj, tmp, tmp);
3857     }
3858   } else {
3859     ld_ptr(obj, index, tmp);
3860   }
3861 
3862   // satb_log_barrier_work2(obj, tmp, offset);
3863 
3864   // satb_log_barrier_work3(tmp, filtered, preserve_o_regs);
3865 
3866   const Register pre_val = tmp;
3867 
3868   if (G1SATBBarrierPrintNullPreVals) {
3869     save_frame(0);
3870     mov(pre_val, O0);
3871     // Save G-regs that target may use.
3872     mov(G1, L1);
3873     mov(G2, L2);
3874     mov(G3, L3);
3875     mov(G4, L4);
3876     mov(G5, L5);
3877     call(CAST_FROM_FN_PTR(address, &count_null_pre_vals));
3878     delayed()->nop();
3879     // Restore G-regs that target may have used.
3880     mov(L1, G1);
3881     mov(L2, G2);
3882     mov(L3, G3);
3883     mov(L4, G4);
3884     mov(L5, G5);
3885     restore(G0, G0, G0);
3886   }
3887 
3888   // Check on whether to annul.
3889   br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, pre_val, filtered);
3890   delayed() -> nop();
3891 
3892   // OK, it's not filtered, so we'll need to call enqueue.  In the normal
3893   // case, pre_val will be a scratch G-reg, but there's some cases in which
3894   // it's an O-reg.  In the first case, do a normal call.  In the latter,
3895   // do a save here and call the frameless version.
3896 
3897   guarantee(pre_val->is_global() || pre_val->is_out(),
3898             "Or we need to think harder.");
3899   if (pre_val->is_global() && !preserve_o_regs) {
3900     generate_satb_log_enqueue_if_necessary(true); // with frame.
3901     call(satb_log_enqueue_with_frame);
3902     delayed()->mov(pre_val, O0);
3903   } else {
3904     generate_satb_log_enqueue_if_necessary(false); // with frameless.
3905     save_frame(0);
3906     call(satb_log_enqueue_frameless);
3907     delayed()->mov(pre_val->after_save(), O0);
3908     restore();
3909   }
3910 
3911   bind(filtered);
3912 }
3913 
3914 static jint num_ct_writes = 0;
3915 static jint num_ct_writes_filtered_in_hr = 0;
3916 static jint num_ct_writes_filtered_null = 0;
3917 static jint num_ct_writes_filtered_pop = 0;
3918 static G1CollectedHeap* g1 = NULL;
3919 
3920 static Thread* count_ct_writes(void* filter_val, void* new_val) {
3921   Atomic::inc(&num_ct_writes);
3922   if (filter_val == NULL) {
3923     Atomic::inc(&num_ct_writes_filtered_in_hr);
3924   } else if (new_val == NULL) {
3925     Atomic::inc(&num_ct_writes_filtered_null);
3926   } else {
3927     if (g1 == NULL) {
3928       g1 = G1CollectedHeap::heap();
3929     }
3930     if ((HeapWord*)new_val < g1->popular_object_boundary()) {
3931       Atomic::inc(&num_ct_writes_filtered_pop);
3932     }
3933   }
3934   if ((num_ct_writes % 1000000) == 0) {
3935     jint num_ct_writes_filtered =
3936       num_ct_writes_filtered_in_hr +
3937       num_ct_writes_filtered_null +
3938       num_ct_writes_filtered_pop;
3939 
3940     tty->print_cr("%d potential CT writes: %5.2f%% filtered\n"
3941                   "   (%5.2f%% intra-HR, %5.2f%% null, %5.2f%% popular).",
3942                   num_ct_writes,
3943                   100.0*(float)num_ct_writes_filtered/(float)num_ct_writes,
3944                   100.0*(float)num_ct_writes_filtered_in_hr/
3945                   (float)num_ct_writes,
3946                   100.0*(float)num_ct_writes_filtered_null/
3947                   (float)num_ct_writes,
3948                   100.0*(float)num_ct_writes_filtered_pop/
3949                   (float)num_ct_writes);
3950   }
3951   return Thread::current();
3952 }
3953 
3954 static address dirty_card_log_enqueue = 0;
3955 static u_char* dirty_card_log_enqueue_end = 0;
3956 
3957 // This gets to assume that o0 contains the object address.
3958 static void generate_dirty_card_log_enqueue(jbyte* byte_map_base) {
3959   BufferBlob* bb = BufferBlob::create("dirty_card_enqueue", EnqueueCodeSize*2);
3960   CodeBuffer buf(bb->instructions_begin(), bb->instructions_size());
3961   MacroAssembler masm(&buf);
3962   address start = masm.pc();
3963 
3964   Label not_already_dirty, restart, refill;
3965 
3966 #ifdef _LP64
3967   masm.srlx(O0, CardTableModRefBS::card_shift, O0);
3968 #else
3969   masm.srl(O0, CardTableModRefBS::card_shift, O0);
3970 #endif
3971   Address rs(O1, (address)byte_map_base);
3972   masm.load_address(rs); // O1 := <card table base>
3973   masm.ldub(O0, O1, O2); // O2 := [O0 + O1]
3974 
3975   masm.br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
3976                       O2, not_already_dirty);
3977   // Get O1 + O2 into a reg by itself -- useful in the take-the-branch
3978   // case, harmless if not.
3979   masm.delayed()->add(O0, O1, O3);
3980 
3981   // We didn't take the branch, so we're already dirty: return.
3982   // Use return-from-leaf
3983   masm.retl();
3984   masm.delayed()->nop();
3985 
3986   // Not dirty.
3987   masm.bind(not_already_dirty);
3988   // First, dirty it.
3989   masm.stb(G0, O3, G0);  // [cardPtr] := 0  (i.e., dirty).
3990   int dirty_card_q_index_byte_offset =
3991     in_bytes(JavaThread::dirty_card_queue_offset() +
3992              PtrQueue::byte_offset_of_index());
3993   int dirty_card_q_buf_byte_offset =
3994     in_bytes(JavaThread::dirty_card_queue_offset() +
3995              PtrQueue::byte_offset_of_buf());
3996   masm.bind(restart);
3997   masm.ld_ptr(G2_thread, dirty_card_q_index_byte_offset, L0);
3998 
3999   masm.br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
4000                       L0, refill);
4001   // If the branch is taken, no harm in executing this in the delay slot.
4002   masm.delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, L1);
4003   masm.sub(L0, oopSize, L0);
4004 
4005   masm.st_ptr(O3, L1, L0);  // [_buf + index] := I0
4006   // Use return-from-leaf
4007   masm.retl();
4008   masm.delayed()->st_ptr(L0, G2_thread, dirty_card_q_index_byte_offset);
4009 
4010   masm.bind(refill);
4011   address handle_zero =
4012     CAST_FROM_FN_PTR(address,
4013                      &DirtyCardQueueSet::handle_zero_index_for_thread);
4014   // This should be rare enough that we can afford to save all the
4015   // scratch registers that the calling context might be using.
4016   masm.mov(G1_scratch, L3);
4017   masm.mov(G3_scratch, L5);
4018   // We need the value of O3 above (for the write into the buffer), so we
4019   // save and restore it.
4020   masm.mov(O3, L6);
4021   // Since the call will overwrite O7, we save and restore that, as well.
4022   masm.mov(O7, L4);
4023 
4024   masm.call_VM_leaf(L7_thread_cache, handle_zero, G2_thread);
4025   masm.mov(L3, G1_scratch);
4026   masm.mov(L5, G3_scratch);
4027   masm.mov(L6, O3);
4028   masm.br(Assembler::always, /*annul*/false, Assembler::pt, restart);
4029   masm.delayed()->mov(L4, O7);
4030 
4031   dirty_card_log_enqueue = start;
4032   dirty_card_log_enqueue_end = masm.pc();
4033   // XXX Should have a guarantee here about not going off the end!
4034   // Does it already do so?  Do an experiment...
4035 }
4036 
4037 static inline void
4038 generate_dirty_card_log_enqueue_if_necessary(jbyte* byte_map_base) {
4039   if (dirty_card_log_enqueue == 0) {
4040     generate_dirty_card_log_enqueue(byte_map_base);
4041     assert(dirty_card_log_enqueue != 0, "postcondition.");
4042     if (G1SATBPrintStubs) {
4043       tty->print_cr("Generated dirty_card enqueue:");
4044       Disassembler::decode((u_char*)dirty_card_log_enqueue,
4045                            dirty_card_log_enqueue_end,
4046                            tty);
4047     }
4048   }
4049 }
4050 
4051 
4052 void MacroAssembler::g1_write_barrier_post(Register store_addr, Register new_val, Register tmp) {
4053 
4054   Label filtered;
4055   MacroAssembler* post_filter_masm = this;
4056 
4057   if (new_val == G0) return;
4058   if (G1DisablePostBarrier) return;
4059 
4060   G1SATBCardTableModRefBS* bs = (G1SATBCardTableModRefBS*) Universe::heap()->barrier_set();
4061   assert(bs->kind() == BarrierSet::G1SATBCT ||
4062          bs->kind() == BarrierSet::G1SATBCTLogging, "wrong barrier");
4063   if (G1RSBarrierRegionFilter) {
4064     xor3(store_addr, new_val, tmp);
4065 #ifdef _LP64
4066     srlx(tmp, HeapRegion::LogOfHRGrainBytes, tmp);
4067 #else
4068     srl(tmp, HeapRegion::LogOfHRGrainBytes, tmp);
4069 #endif
4070     if (G1PrintCTFilterStats) {
4071       guarantee(tmp->is_global(), "Or stats won't work...");
4072       // This is a sleazy hack: I'm temporarily hijacking G2, which I
4073       // promise to restore.
4074       mov(new_val, G2);
4075       save_frame(0);
4076       mov(tmp, O0);
4077       mov(G2, O1);
4078       // Save G-regs that target may use.
4079       mov(G1, L1);
4080       mov(G2, L2);
4081       mov(G3, L3);
4082       mov(G4, L4);
4083       mov(G5, L5);
4084       call(CAST_FROM_FN_PTR(address, &count_ct_writes));
4085       delayed()->nop();
4086       mov(O0, G2);
4087       // Restore G-regs that target may have used.
4088       mov(L1, G1);
4089       mov(L3, G3);
4090       mov(L4, G4);
4091       mov(L5, G5);
4092       restore(G0, G0, G0);
4093     }
4094     // XXX Should I predict this taken or not?  Does it mattern?
4095     br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered);
4096     delayed()->nop();
4097   }
4098 
4099   // Now we decide how to generate the card table write.  If we're
4100   // enqueueing, we call out to a generated function.  Otherwise, we do it
4101   // inline here.
4102 
4103   if (G1RSBarrierUseQueue) {
4104     // If the "store_addr" register is an "in" or "local" register, move it to
4105     // a scratch reg so we can pass it as an argument.
4106     bool use_scr = !(store_addr->is_global() || store_addr->is_out());
4107     // Pick a scratch register different from "tmp".
4108     Register scr = (tmp == G1_scratch ? G3_scratch : G1_scratch);
4109     // Make sure we use up the delay slot!
4110     if (use_scr) {
4111       post_filter_masm->mov(store_addr, scr);
4112     } else {
4113       post_filter_masm->nop();
4114     }
4115     generate_dirty_card_log_enqueue_if_necessary(bs->byte_map_base);
4116     save_frame(0);
4117     call(dirty_card_log_enqueue);
4118     if (use_scr) {
4119       delayed()->mov(scr, O0);
4120     } else {
4121       delayed()->mov(store_addr->after_save(), O0);
4122     }
4123     restore();
4124 
4125   } else {
4126 
4127 #ifdef _LP64
4128     post_filter_masm->srlx(store_addr, CardTableModRefBS::card_shift, store_addr);
4129 #else
4130     post_filter_masm->srl(store_addr, CardTableModRefBS::card_shift, store_addr);
4131 #endif
4132     assert( tmp != store_addr, "need separate temp reg");
4133     Address rs(tmp, (address)bs->byte_map_base);
4134     load_address(rs);
4135     stb(G0, rs.base(), store_addr);
4136   }
4137 
4138   bind(filtered);
4139 
4140 }
4141 
4142 #endif  // SERIALGC
4143 ///////////////////////////////////////////////////////////////////////////////////
4144 
4145 void MacroAssembler::card_write_barrier_post(Register store_addr, Register new_val, Register tmp) {
4146   // If we're writing constant NULL, we can skip the write barrier.
4147   if (new_val == G0) return;
4148   CardTableModRefBS* bs = (CardTableModRefBS*) Universe::heap()->barrier_set();
4149   assert(bs->kind() == BarrierSet::CardTableModRef ||
4150          bs->kind() == BarrierSet::CardTableExtension, "wrong barrier");
4151   card_table_write(bs->byte_map_base, tmp, store_addr);
4152 }
4153 
4154 void MacroAssembler::load_klass(Register src_oop, Register klass) {
4155   // The number of bytes in this code is used by
4156   // MachCallDynamicJavaNode::ret_addr_offset()
4157   // if this changes, change that.
4158   if (UseCompressedOops) {
4159     lduw(src_oop, oopDesc::klass_offset_in_bytes(), klass);
4160     decode_heap_oop_not_null(klass);
4161   } else {
4162     ld_ptr(src_oop, oopDesc::klass_offset_in_bytes(), klass);
4163   }
4164 }
4165 
4166 void MacroAssembler::store_klass(Register klass, Register dst_oop) {
4167   if (UseCompressedOops) {
4168     assert(dst_oop != klass, "not enough registers");
4169     encode_heap_oop_not_null(klass);
4170     st(klass, dst_oop, oopDesc::klass_offset_in_bytes());
4171   } else {
4172     st_ptr(klass, dst_oop, oopDesc::klass_offset_in_bytes());
4173   }
4174 }
4175 
4176 void MacroAssembler::store_klass_gap(Register s, Register d) {
4177   if (UseCompressedOops) {
4178     assert(s != d, "not enough registers");
4179     st(s, d, oopDesc::klass_gap_offset_in_bytes());
4180   }
4181 }
4182 
4183 void MacroAssembler::load_heap_oop(const Address& s, Register d, int offset) {
4184   if (UseCompressedOops) {
4185     lduw(s, d, offset);
4186     decode_heap_oop(d);
4187   } else {
4188     ld_ptr(s, d, offset);
4189   }
4190 }
4191 
4192 void MacroAssembler::load_heap_oop(Register s1, Register s2, Register d) {
4193    if (UseCompressedOops) {
4194     lduw(s1, s2, d);
4195     decode_heap_oop(d, d);
4196   } else {
4197     ld_ptr(s1, s2, d);
4198   }
4199 }
4200 
4201 void MacroAssembler::load_heap_oop(Register s1, int simm13a, Register d) {
4202    if (UseCompressedOops) {
4203     lduw(s1, simm13a, d);
4204     decode_heap_oop(d, d);
4205   } else {
4206     ld_ptr(s1, simm13a, d);
4207   }
4208 }
4209 
4210 void MacroAssembler::store_heap_oop(Register d, Register s1, Register s2) {
4211   if (UseCompressedOops) {
4212     assert(s1 != d && s2 != d, "not enough registers");
4213     encode_heap_oop(d);
4214     st(d, s1, s2);
4215   } else {
4216     st_ptr(d, s1, s2);
4217   }
4218 }
4219 
4220 void MacroAssembler::store_heap_oop(Register d, Register s1, int simm13a) {
4221   if (UseCompressedOops) {
4222     assert(s1 != d, "not enough registers");
4223     encode_heap_oop(d);
4224     st(d, s1, simm13a);
4225   } else {
4226     st_ptr(d, s1, simm13a);
4227   }
4228 }
4229 
4230 void MacroAssembler::store_heap_oop(Register d, const Address& a, int offset) {
4231   if (UseCompressedOops) {
4232     assert(a.base() != d, "not enough registers");
4233     encode_heap_oop(d);
4234     st(d, a, offset);
4235   } else {
4236     st_ptr(d, a, offset);
4237   }
4238 }
4239 
4240 
4241 void MacroAssembler::encode_heap_oop(Register src, Register dst) {
4242   assert (UseCompressedOops, "must be compressed");
4243   verify_oop(src);
4244   Label done;
4245   if (src == dst) {
4246     // optimize for frequent case src == dst
4247     bpr(rc_nz, true, Assembler::pt, src, done);
4248     delayed() -> sub(src, G6_heapbase, dst); // annuled if not taken
4249     bind(done);
4250     srlx(src, LogMinObjAlignmentInBytes, dst);
4251   } else {
4252     bpr(rc_z, false, Assembler::pn, src, done);
4253     delayed() -> mov(G0, dst);
4254     // could be moved before branch, and annulate delay,
4255     // but may add some unneeded work decoding null
4256     sub(src, G6_heapbase, dst);
4257     srlx(dst, LogMinObjAlignmentInBytes, dst);
4258     bind(done);
4259   }
4260 }
4261 
4262 
4263 void MacroAssembler::encode_heap_oop_not_null(Register r) {
4264   assert (UseCompressedOops, "must be compressed");
4265   verify_oop(r);
4266   sub(r, G6_heapbase, r);
4267   srlx(r, LogMinObjAlignmentInBytes, r);
4268 }
4269 
4270 void MacroAssembler::encode_heap_oop_not_null(Register src, Register dst) {
4271   assert (UseCompressedOops, "must be compressed");
4272   verify_oop(src);
4273   sub(src, G6_heapbase, dst);
4274   srlx(dst, LogMinObjAlignmentInBytes, dst);
4275 }
4276 
4277 // Same algorithm as oops.inline.hpp decode_heap_oop.
4278 void  MacroAssembler::decode_heap_oop(Register src, Register dst) {
4279   assert (UseCompressedOops, "must be compressed");
4280   Label done;
4281   sllx(src, LogMinObjAlignmentInBytes, dst);
4282   bpr(rc_nz, true, Assembler::pt, dst, done);
4283   delayed() -> add(dst, G6_heapbase, dst); // annuled if not taken
4284   bind(done);
4285   verify_oop(dst);
4286 }
4287 
4288 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
4289   // Do not add assert code to this unless you change vtableStubs_sparc.cpp
4290   // pd_code_size_limit.
4291   // Also do not verify_oop as this is called by verify_oop.
4292   assert (UseCompressedOops, "must be compressed");
4293   sllx(r, LogMinObjAlignmentInBytes, r);
4294   add(r, G6_heapbase, r);
4295 }
4296 
4297 void  MacroAssembler::decode_heap_oop_not_null(Register src, Register dst) {
4298   // Do not add assert code to this unless you change vtableStubs_sparc.cpp
4299   // pd_code_size_limit.
4300   // Also do not verify_oop as this is called by verify_oop.
4301   assert (UseCompressedOops, "must be compressed");
4302   sllx(src, LogMinObjAlignmentInBytes, dst);
4303   add(dst, G6_heapbase, dst);
4304 }
4305 
4306 void MacroAssembler::reinit_heapbase() {
4307   if (UseCompressedOops) {
4308     // call indirectly to solve generation ordering problem
4309     Address base(G6_heapbase, (address)Universe::heap_base_addr());
4310     load_ptr_contents(base, G6_heapbase);
4311   }
4312 }