1 /*
   2  * Copyright 1997-2007 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/_stubGenerator_sparc.cpp.incl"
  27 
  28 // Declaration and definition of StubGenerator (no .hpp file).
  29 // For a more detailed description of the stub routine structure
  30 // see the comment in stubRoutines.hpp.
  31 
  32 #define __ _masm->
  33 
  34 #ifdef PRODUCT
  35 #define BLOCK_COMMENT(str) /* nothing */
  36 #else
  37 #define BLOCK_COMMENT(str) __ block_comment(str)
  38 #endif
  39 
  40 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  41 
  42 // Note:  The register L7 is used as L7_thread_cache, and may not be used
  43 //        any other way within this module.
  44 
  45 
  46 static const Register& Lstub_temp = L2;
  47 
  48 // -------------------------------------------------------------------------------------------------------------------------
  49 // Stub Code definitions
  50 
  51 static address handle_unsafe_access() {
  52   JavaThread* thread = JavaThread::current();
  53   address pc  = thread->saved_exception_pc();
  54   address npc = thread->saved_exception_npc();
  55   // pc is the instruction which we must emulate
  56   // doing a no-op is fine:  return garbage from the load
  57 
  58   // request an async exception
  59   thread->set_pending_unsafe_access_error();
  60 
  61   // return address of next instruction to execute
  62   return npc;
  63 }
  64 
  65 class StubGenerator: public StubCodeGenerator {
  66  private:
  67 
  68 #ifdef PRODUCT
  69 #define inc_counter_np(a,b,c) (0)
  70 #else
  71   void inc_counter_np_(int& counter, Register t1, Register t2) {
  72     Address counter_addr(t2, (address) &counter);
  73     __ sethi(counter_addr);
  74     __ ld(counter_addr, t1);
  75     __ inc(t1);
  76     __ st(t1, counter_addr);
  77   }
  78 #define inc_counter_np(counter, t1, t2) \
  79   BLOCK_COMMENT("inc_counter " #counter); \
  80   inc_counter_np_(counter, t1, t2);
  81 #endif
  82 
  83   //----------------------------------------------------------------------------------------------------
  84   // Call stubs are used to call Java from C
  85 
  86   address generate_call_stub(address& return_pc) {
  87     StubCodeMark mark(this, "StubRoutines", "call_stub");
  88     address start = __ pc();
  89 
  90     // Incoming arguments:
  91     //
  92     // o0         : call wrapper address
  93     // o1         : result (address)
  94     // o2         : result type
  95     // o3         : method
  96     // o4         : (interpreter) entry point
  97     // o5         : parameters (address)
  98     // [sp + 0x5c]: parameter size (in words)
  99     // [sp + 0x60]: thread
 100     //
 101     // +---------------+ <--- sp + 0
 102     // |               |
 103     // . reg save area .
 104     // |               |
 105     // +---------------+ <--- sp + 0x40
 106     // |               |
 107     // . extra 7 slots .
 108     // |               |
 109     // +---------------+ <--- sp + 0x5c
 110     // |  param. size  |
 111     // +---------------+ <--- sp + 0x60
 112     // |    thread     |
 113     // +---------------+
 114     // |               |
 115 
 116     // note: if the link argument position changes, adjust
 117     //       the code in frame::entry_frame_call_wrapper()
 118 
 119     const Argument link           = Argument(0, false); // used only for GC
 120     const Argument result         = Argument(1, false);
 121     const Argument result_type    = Argument(2, false);
 122     const Argument method         = Argument(3, false);
 123     const Argument entry_point    = Argument(4, false);
 124     const Argument parameters     = Argument(5, false);
 125     const Argument parameter_size = Argument(6, false);
 126     const Argument thread         = Argument(7, false);
 127 
 128     // setup thread register
 129     __ ld_ptr(thread.as_address(), G2_thread);
 130     __ reinit_heapbase();
 131 
 132 #ifdef ASSERT
 133     // make sure we have no pending exceptions
 134     { const Register t = G3_scratch;
 135       Label L;
 136       __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), t);
 137       __ br_null(t, false, Assembler::pt, L);
 138       __ delayed()->nop();
 139       __ stop("StubRoutines::call_stub: entered with pending exception");
 140       __ bind(L);
 141     }
 142 #endif
 143 
 144     // create activation frame & allocate space for parameters
 145     { const Register t = G3_scratch;
 146       __ ld_ptr(parameter_size.as_address(), t);                // get parameter size (in words)
 147       __ add(t, frame::memory_parameter_word_sp_offset, t);     // add space for save area (in words)
 148       __ round_to(t, WordsPerLong);                             // make sure it is multiple of 2 (in words)
 149       __ sll(t, Interpreter::logStackElementSize(), t);                    // compute number of bytes
 150       __ neg(t);                                                // negate so it can be used with save
 151       __ save(SP, t, SP);                                       // setup new frame
 152     }
 153 
 154     // +---------------+ <--- sp + 0
 155     // |               |
 156     // . reg save area .
 157     // |               |
 158     // +---------------+ <--- sp + 0x40
 159     // |               |
 160     // . extra 7 slots .
 161     // |               |
 162     // +---------------+ <--- sp + 0x5c
 163     // |  empty slot   |      (only if parameter size is even)
 164     // +---------------+
 165     // |               |
 166     // .  parameters   .
 167     // |               |
 168     // +---------------+ <--- fp + 0
 169     // |               |
 170     // . reg save area .
 171     // |               |
 172     // +---------------+ <--- fp + 0x40
 173     // |               |
 174     // . extra 7 slots .
 175     // |               |
 176     // +---------------+ <--- fp + 0x5c
 177     // |  param. size  |
 178     // +---------------+ <--- fp + 0x60
 179     // |    thread     |
 180     // +---------------+
 181     // |               |
 182 
 183     // pass parameters if any
 184     BLOCK_COMMENT("pass parameters if any");
 185     { const Register src = parameters.as_in().as_register();
 186       const Register dst = Lentry_args;
 187       const Register tmp = G3_scratch;
 188       const Register cnt = G4_scratch;
 189 
 190       // test if any parameters & setup of Lentry_args
 191       Label exit;
 192       __ ld_ptr(parameter_size.as_in().as_address(), cnt);      // parameter counter
 193       __ add( FP, STACK_BIAS, dst );
 194       __ tst(cnt);
 195       __ br(Assembler::zero, false, Assembler::pn, exit);
 196       __ delayed()->sub(dst, BytesPerWord, dst);                 // setup Lentry_args
 197 
 198       // copy parameters if any
 199       Label loop;
 200       __ BIND(loop);
 201       // Store tag first.
 202       if (TaggedStackInterpreter) {
 203         __ ld_ptr(src, 0, tmp);
 204         __ add(src, BytesPerWord, src);  // get next
 205         __ st_ptr(tmp, dst, Interpreter::tag_offset_in_bytes());
 206       }
 207       // Store parameter value
 208       __ ld_ptr(src, 0, tmp);
 209       __ add(src, BytesPerWord, src);
 210       __ st_ptr(tmp, dst, Interpreter::value_offset_in_bytes());
 211       __ deccc(cnt);
 212       __ br(Assembler::greater, false, Assembler::pt, loop);
 213       __ delayed()->sub(dst, Interpreter::stackElementSize(), dst);
 214 
 215       // done
 216       __ BIND(exit);
 217     }
 218 
 219     // setup parameters, method & call Java function
 220 #ifdef ASSERT
 221     // layout_activation_impl checks it's notion of saved SP against
 222     // this register, so if this changes update it as well.
 223     const Register saved_SP = Lscratch;
 224     __ mov(SP, saved_SP);                               // keep track of SP before call
 225 #endif
 226 
 227     // setup parameters
 228     const Register t = G3_scratch;
 229     __ ld_ptr(parameter_size.as_in().as_address(), t); // get parameter size (in words)
 230     __ sll(t, Interpreter::logStackElementSize(), t);            // compute number of bytes
 231     __ sub(FP, t, Gargs);                              // setup parameter pointer
 232 #ifdef _LP64
 233     __ add( Gargs, STACK_BIAS, Gargs );                // Account for LP64 stack bias
 234 #endif
 235     __ mov(SP, O5_savedSP);
 236 
 237 
 238     // do the call
 239     //
 240     // the following register must be setup:
 241     //
 242     // G2_thread
 243     // G5_method
 244     // Gargs
 245     BLOCK_COMMENT("call Java function");
 246     __ jmpl(entry_point.as_in().as_register(), G0, O7);
 247     __ delayed()->mov(method.as_in().as_register(), G5_method);   // setup method
 248 
 249     BLOCK_COMMENT("call_stub_return_address:");
 250     return_pc = __ pc();
 251 
 252     // The callee, if it wasn't interpreted, can return with SP changed so
 253     // we can no longer assert of change of SP.
 254 
 255     // store result depending on type
 256     // (everything that is not T_OBJECT, T_LONG, T_FLOAT, or T_DOUBLE
 257     //  is treated as T_INT)
 258     { const Register addr = result     .as_in().as_register();
 259       const Register type = result_type.as_in().as_register();
 260       Label is_long, is_float, is_double, is_object, exit;
 261       __            cmp(type, T_OBJECT);  __ br(Assembler::equal, false, Assembler::pn, is_object);
 262       __ delayed()->cmp(type, T_FLOAT);   __ br(Assembler::equal, false, Assembler::pn, is_float);
 263       __ delayed()->cmp(type, T_DOUBLE);  __ br(Assembler::equal, false, Assembler::pn, is_double);
 264       __ delayed()->cmp(type, T_LONG);    __ br(Assembler::equal, false, Assembler::pn, is_long);
 265       __ delayed()->nop();
 266 
 267       // store int result
 268       __ st(O0, addr, G0);
 269 
 270       __ BIND(exit);
 271       __ ret();
 272       __ delayed()->restore();
 273 
 274       __ BIND(is_object);
 275       __ ba(false, exit);
 276       __ delayed()->st_ptr(O0, addr, G0);
 277 
 278       __ BIND(is_float);
 279       __ ba(false, exit);
 280       __ delayed()->stf(FloatRegisterImpl::S, F0, addr, G0);
 281 
 282       __ BIND(is_double);
 283       __ ba(false, exit);
 284       __ delayed()->stf(FloatRegisterImpl::D, F0, addr, G0);
 285 
 286       __ BIND(is_long);
 287 #ifdef _LP64
 288       __ ba(false, exit);
 289       __ delayed()->st_long(O0, addr, G0);      // store entire long
 290 #else
 291 #if defined(COMPILER2)
 292   // All return values are where we want them, except for Longs.  C2 returns
 293   // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
 294   // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
 295   // build we simply always use G1.
 296   // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
 297   // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
 298   // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
 299 
 300       __ ba(false, exit);
 301       __ delayed()->stx(G1, addr, G0);  // store entire long
 302 #else
 303       __ st(O1, addr, BytesPerInt);
 304       __ ba(false, exit);
 305       __ delayed()->st(O0, addr, G0);
 306 #endif /* COMPILER2 */
 307 #endif /* _LP64 */
 308      }
 309      return start;
 310   }
 311 
 312 
 313   //----------------------------------------------------------------------------------------------------
 314   // Return point for a Java call if there's an exception thrown in Java code.
 315   // The exception is caught and transformed into a pending exception stored in
 316   // JavaThread that can be tested from within the VM.
 317   //
 318   // Oexception: exception oop
 319 
 320   address generate_catch_exception() {
 321     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 322 
 323     address start = __ pc();
 324     // verify that thread corresponds
 325     __ verify_thread();
 326 
 327     const Register& temp_reg = Gtemp;
 328     Address pending_exception_addr    (G2_thread, 0, in_bytes(Thread::pending_exception_offset()));
 329     Address exception_file_offset_addr(G2_thread, 0, in_bytes(Thread::exception_file_offset   ()));
 330     Address exception_line_offset_addr(G2_thread, 0, in_bytes(Thread::exception_line_offset   ()));
 331 
 332     // set pending exception
 333     __ verify_oop(Oexception);
 334     __ st_ptr(Oexception, pending_exception_addr);
 335     __ set((intptr_t)__FILE__, temp_reg);
 336     __ st_ptr(temp_reg, exception_file_offset_addr);
 337     __ set((intptr_t)__LINE__, temp_reg);
 338     __ st(temp_reg, exception_line_offset_addr);
 339 
 340     // complete return to VM
 341     assert(StubRoutines::_call_stub_return_address != NULL, "must have been generated before");
 342 
 343     Address stub_ret(temp_reg, StubRoutines::_call_stub_return_address);
 344     __ jump_to(stub_ret);
 345     __ delayed()->nop();
 346 
 347     return start;
 348   }
 349 
 350 
 351   //----------------------------------------------------------------------------------------------------
 352   // Continuation point for runtime calls returning with a pending exception
 353   // The pending exception check happened in the runtime or native call stub
 354   // The pending exception in Thread is converted into a Java-level exception
 355   //
 356   // Contract with Java-level exception handler: O0 = exception
 357   //                                             O1 = throwing pc
 358 
 359   address generate_forward_exception() {
 360     StubCodeMark mark(this, "StubRoutines", "forward_exception");
 361     address start = __ pc();
 362 
 363     // Upon entry, O7 has the return address returning into Java
 364     // (interpreted or compiled) code; i.e. the return address
 365     // becomes the throwing pc.
 366 
 367     const Register& handler_reg = Gtemp;
 368 
 369     Address exception_addr (G2_thread, 0, in_bytes(Thread::pending_exception_offset()));
 370 
 371 #ifdef ASSERT
 372     // make sure that this code is only executed if there is a pending exception
 373     { Label L;
 374       __ ld_ptr(exception_addr, Gtemp);
 375       __ br_notnull(Gtemp, false, Assembler::pt, L);
 376       __ delayed()->nop();
 377       __ stop("StubRoutines::forward exception: no pending exception (1)");
 378       __ bind(L);
 379     }
 380 #endif
 381 
 382     // compute exception handler into handler_reg
 383     __ get_thread();
 384     __ ld_ptr(exception_addr, Oexception);
 385     __ verify_oop(Oexception);
 386     __ save_frame(0);             // compensates for compiler weakness
 387     __ add(O7->after_save(), frame::pc_return_offset, Lscratch); // save the issuing PC
 388     BLOCK_COMMENT("call exception_handler_for_return_address");
 389     __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), Lscratch);
 390     __ mov(O0, handler_reg);
 391     __ restore();                 // compensates for compiler weakness
 392 
 393     __ ld_ptr(exception_addr, Oexception);
 394     __ add(O7, frame::pc_return_offset, Oissuing_pc); // save the issuing PC
 395 
 396 #ifdef ASSERT
 397     // make sure exception is set
 398     { Label L;
 399       __ br_notnull(Oexception, false, Assembler::pt, L);
 400       __ delayed()->nop();
 401       __ stop("StubRoutines::forward exception: no pending exception (2)");
 402       __ bind(L);
 403     }
 404 #endif
 405     // jump to exception handler
 406     __ jmp(handler_reg, 0);
 407     // clear pending exception
 408     __ delayed()->st_ptr(G0, exception_addr);
 409 
 410     return start;
 411   }
 412 
 413 
 414   //------------------------------------------------------------------------------------------------------------------------
 415   // Continuation point for throwing of implicit exceptions that are not handled in
 416   // the current activation. Fabricates an exception oop and initiates normal
 417   // exception dispatching in this frame. Only callee-saved registers are preserved
 418   // (through the normal register window / RegisterMap handling).
 419   // If the compiler needs all registers to be preserved between the fault
 420   // point and the exception handler then it must assume responsibility for that in
 421   // AbstractCompiler::continuation_for_implicit_null_exception or
 422   // continuation_for_implicit_division_by_zero_exception. All other implicit
 423   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
 424   // either at call sites or otherwise assume that stack unwinding will be initiated,
 425   // so caller saved registers were assumed volatile in the compiler.
 426 
 427   // Note that we generate only this stub into a RuntimeStub, because it needs to be
 428   // properly traversed and ignored during GC, so we change the meaning of the "__"
 429   // macro within this method.
 430 #undef __
 431 #define __ masm->
 432 
 433   address generate_throw_exception(const char* name, address runtime_entry, bool restore_saved_exception_pc) {
 434 #ifdef ASSERT
 435     int insts_size = VerifyThread ? 1 * K : 600;
 436 #else
 437     int insts_size = VerifyThread ? 1 * K : 256;
 438 #endif /* ASSERT */
 439     int locs_size  = 32;
 440 
 441     CodeBuffer      code(name, insts_size, locs_size);
 442     MacroAssembler* masm = new MacroAssembler(&code);
 443 
 444     __ verify_thread();
 445 
 446     // This is an inlined and slightly modified version of call_VM
 447     // which has the ability to fetch the return PC out of thread-local storage
 448     __ assert_not_delayed();
 449 
 450     // Note that we always push a frame because on the SPARC
 451     // architecture, for all of our implicit exception kinds at call
 452     // sites, the implicit exception is taken before the callee frame
 453     // is pushed.
 454     __ save_frame(0);
 455 
 456     int frame_complete = __ offset();
 457 
 458     if (restore_saved_exception_pc) {
 459       Address saved_exception_pc(G2_thread, 0, in_bytes(JavaThread::saved_exception_pc_offset()));
 460       __ ld_ptr(saved_exception_pc, I7);
 461       __ sub(I7, frame::pc_return_offset, I7);
 462     }
 463 
 464     // Note that we always have a runtime stub frame on the top of stack by this point
 465     Register last_java_sp = SP;
 466     // 64-bit last_java_sp is biased!
 467     __ set_last_Java_frame(last_java_sp, G0);
 468     if (VerifyThread)  __ mov(G2_thread, O0); // about to be smashed; pass early
 469     __ save_thread(noreg);
 470     // do the call
 471     BLOCK_COMMENT("call runtime_entry");
 472     __ call(runtime_entry, relocInfo::runtime_call_type);
 473     if (!VerifyThread)
 474       __ delayed()->mov(G2_thread, O0);  // pass thread as first argument
 475     else
 476       __ delayed()->nop();             // (thread already passed)
 477     __ restore_thread(noreg);
 478     __ reset_last_Java_frame();
 479 
 480     // check for pending exceptions. use Gtemp as scratch register.
 481 #ifdef ASSERT
 482     Label L;
 483 
 484     Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset()));
 485     Register scratch_reg = Gtemp;
 486     __ ld_ptr(exception_addr, scratch_reg);
 487     __ br_notnull(scratch_reg, false, Assembler::pt, L);
 488     __ delayed()->nop();
 489     __ should_not_reach_here();
 490     __ bind(L);
 491 #endif // ASSERT
 492     BLOCK_COMMENT("call forward_exception_entry");
 493     __ call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
 494     // we use O7 linkage so that forward_exception_entry has the issuing PC
 495     __ delayed()->restore();
 496 
 497     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, masm->total_frame_size_in_bytes(0), NULL, false);
 498     return stub->entry_point();
 499   }
 500 
 501 #undef __
 502 #define __ _masm->
 503 
 504 
 505   // Generate a routine that sets all the registers so we
 506   // can tell if the stop routine prints them correctly.
 507   address generate_test_stop() {
 508     StubCodeMark mark(this, "StubRoutines", "test_stop");
 509     address start = __ pc();
 510 
 511     int i;
 512 
 513     __ save_frame(0);
 514 
 515     static jfloat zero = 0.0, one = 1.0;
 516 
 517     // put addr in L0, then load through L0 to F0
 518     __ set((intptr_t)&zero, L0);  __ ldf( FloatRegisterImpl::S, L0, 0, F0);
 519     __ set((intptr_t)&one,  L0);  __ ldf( FloatRegisterImpl::S, L0, 0, F1); // 1.0 to F1
 520 
 521     // use add to put 2..18 in F2..F18
 522     for ( i = 2;  i <= 18;  ++i ) {
 523       __ fadd( FloatRegisterImpl::S, F1, as_FloatRegister(i-1),  as_FloatRegister(i));
 524     }
 525 
 526     // Now put double 2 in F16, double 18 in F18
 527     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F2, F16 );
 528     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F18, F18 );
 529 
 530     // use add to put 20..32 in F20..F32
 531     for (i = 20; i < 32; i += 2) {
 532       __ fadd( FloatRegisterImpl::D, F16, as_FloatRegister(i-2),  as_FloatRegister(i));
 533     }
 534 
 535     // put 0..7 in i's, 8..15 in l's, 16..23 in o's, 24..31 in g's
 536     for ( i = 0; i < 8; ++i ) {
 537       if (i < 6) {
 538         __ set(     i, as_iRegister(i));
 539         __ set(16 + i, as_oRegister(i));
 540         __ set(24 + i, as_gRegister(i));
 541       }
 542       __ set( 8 + i, as_lRegister(i));
 543     }
 544 
 545     __ stop("testing stop");
 546 
 547 
 548     __ ret();
 549     __ delayed()->restore();
 550 
 551     return start;
 552   }
 553 
 554 
 555   address generate_stop_subroutine() {
 556     StubCodeMark mark(this, "StubRoutines", "stop_subroutine");
 557     address start = __ pc();
 558 
 559     __ stop_subroutine();
 560 
 561     return start;
 562   }
 563 
 564   address generate_flush_callers_register_windows() {
 565     StubCodeMark mark(this, "StubRoutines", "flush_callers_register_windows");
 566     address start = __ pc();
 567 
 568     __ flush_windows();
 569     __ retl(false);
 570     __ delayed()->add( FP, STACK_BIAS, O0 );
 571     // The returned value must be a stack pointer whose register save area
 572     // is flushed, and will stay flushed while the caller executes.
 573 
 574     return start;
 575   }
 576 
 577   // Helper functions for v8 atomic operations.
 578   //
 579   void get_v8_oop_lock_ptr(Register lock_ptr_reg, Register mark_oop_reg, Register scratch_reg) {
 580     if (mark_oop_reg == noreg) {
 581       address lock_ptr = (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr();
 582       __ set((intptr_t)lock_ptr, lock_ptr_reg);
 583     } else {
 584       assert(scratch_reg != noreg, "just checking");
 585       address lock_ptr = (address)StubRoutines::Sparc::_v8_oop_lock_cache;
 586       __ set((intptr_t)lock_ptr, lock_ptr_reg);
 587       __ and3(mark_oop_reg, StubRoutines::Sparc::v8_oop_lock_mask_in_place, scratch_reg);
 588       __ add(lock_ptr_reg, scratch_reg, lock_ptr_reg);
 589     }
 590   }
 591 
 592   void generate_v8_lock_prologue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) {
 593 
 594     get_v8_oop_lock_ptr(lock_ptr_reg, mark_oop_reg, scratch_reg);
 595     __ set(StubRoutines::Sparc::locked, lock_reg);
 596     // Initialize yield counter
 597     __ mov(G0,yield_reg);
 598 
 599     __ BIND(retry);
 600     __ cmp(yield_reg, V8AtomicOperationUnderLockSpinCount);
 601     __ br(Assembler::less, false, Assembler::pt, dontyield);
 602     __ delayed()->nop();
 603 
 604     // This code can only be called from inside the VM, this
 605     // stub is only invoked from Atomic::add().  We do not
 606     // want to use call_VM, because _last_java_sp and such
 607     // must already be set.
 608     //
 609     // Save the regs and make space for a C call
 610     __ save(SP, -96, SP);
 611     __ save_all_globals_into_locals();
 612     BLOCK_COMMENT("call os::naked_sleep");
 613     __ call(CAST_FROM_FN_PTR(address, os::naked_sleep));
 614     __ delayed()->nop();
 615     __ restore_globals_from_locals();
 616     __ restore();
 617     // reset the counter
 618     __ mov(G0,yield_reg);
 619 
 620     __ BIND(dontyield);
 621 
 622     // try to get lock
 623     __ swap(lock_ptr_reg, 0, lock_reg);
 624 
 625     // did we get the lock?
 626     __ cmp(lock_reg, StubRoutines::Sparc::unlocked);
 627     __ br(Assembler::notEqual, true, Assembler::pn, retry);
 628     __ delayed()->add(yield_reg,1,yield_reg);
 629 
 630     // yes, got lock. do the operation here.
 631   }
 632 
 633   void generate_v8_lock_epilogue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) {
 634     __ st(lock_reg, lock_ptr_reg, 0); // unlock
 635   }
 636 
 637   // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest).
 638   //
 639   // Arguments :
 640   //
 641   //      exchange_value: O0
 642   //      dest:           O1
 643   //
 644   // Results:
 645   //
 646   //     O0: the value previously stored in dest
 647   //
 648   address generate_atomic_xchg() {
 649     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 650     address start = __ pc();
 651 
 652     if (UseCASForSwap) {
 653       // Use CAS instead of swap, just in case the MP hardware
 654       // prefers to work with just one kind of synch. instruction.
 655       Label retry;
 656       __ BIND(retry);
 657       __ mov(O0, O3);       // scratch copy of exchange value
 658       __ ld(O1, 0, O2);     // observe the previous value
 659       // try to replace O2 with O3
 660       __ cas_under_lock(O1, O2, O3,
 661       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false);
 662       __ cmp(O2, O3);
 663       __ br(Assembler::notEqual, false, Assembler::pn, retry);
 664       __ delayed()->nop();
 665 
 666       __ retl(false);
 667       __ delayed()->mov(O2, O0);  // report previous value to caller
 668 
 669     } else {
 670       if (VM_Version::v9_instructions_work()) {
 671         __ retl(false);
 672         __ delayed()->swap(O1, 0, O0);
 673       } else {
 674         const Register& lock_reg = O2;
 675         const Register& lock_ptr_reg = O3;
 676         const Register& yield_reg = O4;
 677 
 678         Label retry;
 679         Label dontyield;
 680 
 681         generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
 682         // got the lock, do the swap
 683         __ swap(O1, 0, O0);
 684 
 685         generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
 686         __ retl(false);
 687         __ delayed()->nop();
 688       }
 689     }
 690 
 691     return start;
 692   }
 693 
 694 
 695   // Support for jint Atomic::cmpxchg(jint exchange_value, volatile jint* dest, jint compare_value)
 696   //
 697   // Arguments :
 698   //
 699   //      exchange_value: O0
 700   //      dest:           O1
 701   //      compare_value:  O2
 702   //
 703   // Results:
 704   //
 705   //     O0: the value previously stored in dest
 706   //
 707   // Overwrites (v8): O3,O4,O5
 708   //
 709   address generate_atomic_cmpxchg() {
 710     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
 711     address start = __ pc();
 712 
 713     // cmpxchg(dest, compare_value, exchange_value)
 714     __ cas_under_lock(O1, O2, O0,
 715       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false);
 716     __ retl(false);
 717     __ delayed()->nop();
 718 
 719     return start;
 720   }
 721 
 722   // Support for jlong Atomic::cmpxchg(jlong exchange_value, volatile jlong *dest, jlong compare_value)
 723   //
 724   // Arguments :
 725   //
 726   //      exchange_value: O1:O0
 727   //      dest:           O2
 728   //      compare_value:  O4:O3
 729   //
 730   // Results:
 731   //
 732   //     O1:O0: the value previously stored in dest
 733   //
 734   // This only works on V9, on V8 we don't generate any
 735   // code and just return NULL.
 736   //
 737   // Overwrites: G1,G2,G3
 738   //
 739   address generate_atomic_cmpxchg_long() {
 740     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
 741     address start = __ pc();
 742 
 743     if (!VM_Version::supports_cx8())
 744         return NULL;;
 745     __ sllx(O0, 32, O0);
 746     __ srl(O1, 0, O1);
 747     __ or3(O0,O1,O0);      // O0 holds 64-bit value from compare_value
 748     __ sllx(O3, 32, O3);
 749     __ srl(O4, 0, O4);
 750     __ or3(O3,O4,O3);     // O3 holds 64-bit value from exchange_value
 751     __ casx(O2, O3, O0);
 752     __ srl(O0, 0, O1);    // unpacked return value in O1:O0
 753     __ retl(false);
 754     __ delayed()->srlx(O0, 32, O0);
 755 
 756     return start;
 757   }
 758 
 759 
 760   // Support for jint Atomic::add(jint add_value, volatile jint* dest).
 761   //
 762   // Arguments :
 763   //
 764   //      add_value: O0   (e.g., +1 or -1)
 765   //      dest:      O1
 766   //
 767   // Results:
 768   //
 769   //     O0: the new value stored in dest
 770   //
 771   // Overwrites (v9): O3
 772   // Overwrites (v8): O3,O4,O5
 773   //
 774   address generate_atomic_add() {
 775     StubCodeMark mark(this, "StubRoutines", "atomic_add");
 776     address start = __ pc();
 777     __ BIND(_atomic_add_stub);
 778 
 779     if (VM_Version::v9_instructions_work()) {
 780       Label(retry);
 781       __ BIND(retry);
 782 
 783       __ lduw(O1, 0, O2);
 784       __ add(O0,   O2, O3);
 785       __ cas(O1,   O2, O3);
 786       __ cmp(      O2, O3);
 787       __ br(Assembler::notEqual, false, Assembler::pn, retry);
 788       __ delayed()->nop();
 789       __ retl(false);
 790       __ delayed()->add(O0, O2, O0); // note that cas made O2==O3
 791     } else {
 792       const Register& lock_reg = O2;
 793       const Register& lock_ptr_reg = O3;
 794       const Register& value_reg = O4;
 795       const Register& yield_reg = O5;
 796 
 797       Label(retry);
 798       Label(dontyield);
 799 
 800       generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
 801       // got lock, do the increment
 802       __ ld(O1, 0, value_reg);
 803       __ add(O0, value_reg, value_reg);
 804       __ st(value_reg, O1, 0);
 805 
 806       // %%% only for RMO and PSO
 807       __ membar(Assembler::StoreStore);
 808 
 809       generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
 810 
 811       __ retl(false);
 812       __ delayed()->mov(value_reg, O0);
 813     }
 814 
 815     return start;
 816   }
 817   Label _atomic_add_stub;  // called from other stubs
 818 
 819 
 820   // Support for void OrderAccess::fence().
 821   //
 822   address generate_fence() {
 823     StubCodeMark mark(this, "StubRoutines", "fence");
 824     address start = __ pc();
 825 
 826     __ membar(Assembler::Membar_mask_bits(Assembler::LoadLoad  | Assembler::LoadStore |
 827                                           Assembler::StoreLoad | Assembler::StoreStore));
 828     __ retl(false);
 829     __ delayed()->nop();
 830 
 831     return start;
 832   }
 833 
 834 
 835   //------------------------------------------------------------------------------------------------------------------------
 836   // The following routine generates a subroutine to throw an asynchronous
 837   // UnknownError when an unsafe access gets a fault that could not be
 838   // reasonably prevented by the programmer.  (Example: SIGBUS/OBJERR.)
 839   //
 840   // Arguments :
 841   //
 842   //      trapping PC:    O7
 843   //
 844   // Results:
 845   //     posts an asynchronous exception, skips the trapping instruction
 846   //
 847 
 848   address generate_handler_for_unsafe_access() {
 849     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 850     address start = __ pc();
 851 
 852     const int preserve_register_words = (64 * 2);
 853     Address preserve_addr(FP, 0, (-preserve_register_words * wordSize) + STACK_BIAS);
 854 
 855     Register Lthread = L7_thread_cache;
 856     int i;
 857 
 858     __ save_frame(0);
 859     __ mov(G1, L1);
 860     __ mov(G2, L2);
 861     __ mov(G3, L3);
 862     __ mov(G4, L4);
 863     __ mov(G5, L5);
 864     for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) {
 865       __ stf(FloatRegisterImpl::D, as_FloatRegister(i), preserve_addr, i * wordSize);
 866     }
 867 
 868     address entry_point = CAST_FROM_FN_PTR(address, handle_unsafe_access);
 869     BLOCK_COMMENT("call handle_unsafe_access");
 870     __ call(entry_point, relocInfo::runtime_call_type);
 871     __ delayed()->nop();
 872 
 873     __ mov(L1, G1);
 874     __ mov(L2, G2);
 875     __ mov(L3, G3);
 876     __ mov(L4, G4);
 877     __ mov(L5, G5);
 878     for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) {
 879       __ ldf(FloatRegisterImpl::D, preserve_addr, as_FloatRegister(i), i * wordSize);
 880     }
 881 
 882     __ verify_thread();
 883 
 884     __ jmp(O0, 0);
 885     __ delayed()->restore();
 886 
 887     return start;
 888   }
 889 
 890 
 891   // Support for uint StubRoutine::Sparc::partial_subtype_check( Klass sub, Klass super );
 892   // Arguments :
 893   //
 894   //      ret  : O0, returned
 895   //      icc/xcc: set as O0 (depending on wordSize)
 896   //      sub  : O1, argument, not changed
 897   //      super: O2, argument, not changed
 898   //      raddr: O7, blown by call
 899   address generate_partial_subtype_check() {
 900     __ align(CodeEntryAlignment);
 901     StubCodeMark mark(this, "StubRoutines", "partial_subtype_check");
 902     address start = __ pc();
 903     Label loop, miss;
 904 
 905     // Compare super with sub directly, since super is not in its own SSA.
 906     // The compiler used to emit this test, but we fold it in here,
 907     // to increase overall code density, with no real loss of speed.
 908     { Label L;
 909       __ cmp(O1, O2);
 910       __ brx(Assembler::notEqual, false, Assembler::pt, L);
 911       __ delayed()->nop();
 912       __ retl();
 913       __ delayed()->addcc(G0,0,O0); // set Z flags, zero result
 914       __ bind(L);
 915     }
 916 
 917 #if defined(COMPILER2) && !defined(_LP64)
 918     // Do not use a 'save' because it blows the 64-bit O registers.
 919     __ add(SP,-4*wordSize,SP);  // Make space for 4 temps (stack must be 2 words aligned)
 920     __ st_ptr(L0,SP,(frame::register_save_words+0)*wordSize);
 921     __ st_ptr(L1,SP,(frame::register_save_words+1)*wordSize);
 922     __ st_ptr(L2,SP,(frame::register_save_words+2)*wordSize);
 923     __ st_ptr(L3,SP,(frame::register_save_words+3)*wordSize);
 924     Register Rret   = O0;
 925     Register Rsub   = O1;
 926     Register Rsuper = O2;
 927 #else
 928     __ save_frame(0);
 929     Register Rret   = I0;
 930     Register Rsub   = I1;
 931     Register Rsuper = I2;
 932 #endif
 933 
 934     Register L0_ary_len = L0;
 935     Register L1_ary_ptr = L1;
 936     Register L2_super   = L2;
 937     Register L3_index   = L3;
 938 
 939 #ifdef _LP64
 940     Register L4_ooptmp  = L4;
 941 
 942     if (UseCompressedOops) {
 943       // this must be under UseCompressedOops check, as we rely upon fact
 944       // that L4 not clobbered in C2 on 32-bit platforms, where we do explicit save
 945       // on stack, see several lines above
 946       __ encode_heap_oop(Rsuper, L4_ooptmp);
 947     }
 948 #endif
 949 
 950     inc_counter_np(SharedRuntime::_partial_subtype_ctr, L0, L1);
 951 
 952     __ ld_ptr( Rsub, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 );
 953     __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0_ary_len);
 954     __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1_ary_ptr);
 955     __ clr(L3_index);           // zero index
 956     // Load a little early; will load 1 off the end of the array.
 957     // Ok for now; revisit if we have other uses of this routine.
 958     if (UseCompressedOops) {
 959       __ ld(L1_ary_ptr,0,L2_super);// Will load a little early
 960     } else {
 961       __ ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early
 962     }
 963 
 964     assert(heapOopSize != 0, "heapOopSize should be initialized");
 965     // The scan loop
 966     __ BIND(loop);
 967     __ add(L1_ary_ptr, heapOopSize, L1_ary_ptr); // Bump by OOP size
 968     __ cmp(L3_index,L0_ary_len);
 969     __ br(Assembler::equal,false,Assembler::pn,miss);
 970     __ delayed()->inc(L3_index); // Bump index
 971 
 972     if (UseCompressedOops) {
 973 #ifdef  _LP64
 974       __ subcc(L2_super,L4_ooptmp,Rret);   // Check for match; zero in Rret for a hit
 975       __ br( Assembler::notEqual, false, Assembler::pt, loop );
 976       __ delayed()->ld(L1_ary_ptr,0,L2_super);// Will load a little early
 977 #else
 978       ShouldNotReachHere();
 979 #endif
 980     } else {
 981       __ subcc(L2_super,Rsuper,Rret);   // Check for match; zero in Rret for a hit
 982       __ brx( Assembler::notEqual, false, Assembler::pt, loop );
 983       __ delayed()->ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early
 984     }
 985 
 986     // Got a hit; report success; set cache.  Cache load doesn't
 987     // happen here; for speed it is directly emitted by the compiler.
 988     __ st_ptr( Rsuper, Rsub, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() );
 989 
 990 #if defined(COMPILER2) && !defined(_LP64)
 991     __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0);
 992     __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1);
 993     __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2);
 994     __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3);
 995     __ retl();                  // Result in Rret is zero; flags set to Z
 996     __ delayed()->add(SP,4*wordSize,SP);
 997 #else
 998     __ ret();                   // Result in Rret is zero; flags set to Z
 999     __ delayed()->restore();
1000 #endif
1001 
1002     // Hit or miss falls through here
1003     __ BIND(miss);
1004     __ addcc(G0,1,Rret);        // set NZ flags, NZ result
1005 
1006 #if defined(COMPILER2) && !defined(_LP64)
1007     __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0);
1008     __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1);
1009     __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2);
1010     __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3);
1011     __ retl();                  // Result in Rret is != 0; flags set to NZ
1012     __ delayed()->add(SP,4*wordSize,SP);
1013 #else
1014     __ ret();                   // Result in Rret is != 0; flags set to NZ
1015     __ delayed()->restore();
1016 #endif
1017 
1018     return start;
1019   }
1020 
1021 
1022   // Called from MacroAssembler::verify_oop
1023   //
1024   address generate_verify_oop_subroutine() {
1025     StubCodeMark mark(this, "StubRoutines", "verify_oop_stub");
1026 
1027     address start = __ pc();
1028 
1029     __ verify_oop_subroutine();
1030 
1031     return start;
1032   }
1033 
1034   static address disjoint_byte_copy_entry;
1035   static address disjoint_short_copy_entry;
1036   static address disjoint_int_copy_entry;
1037   static address disjoint_long_copy_entry;
1038   static address disjoint_oop_copy_entry;
1039 
1040   static address byte_copy_entry;
1041   static address short_copy_entry;
1042   static address int_copy_entry;
1043   static address long_copy_entry;
1044   static address oop_copy_entry;
1045 
1046   static address checkcast_copy_entry;
1047 
1048   //
1049   // Verify that a register contains clean 32-bits positive value
1050   // (high 32-bits are 0) so it could be used in 64-bits shifts (sllx, srax).
1051   //
1052   //  Input:
1053   //    Rint  -  32-bits value
1054   //    Rtmp  -  scratch
1055   //
1056   void assert_clean_int(Register Rint, Register Rtmp) {
1057 #if defined(ASSERT) && defined(_LP64)
1058     __ signx(Rint, Rtmp);
1059     __ cmp(Rint, Rtmp);
1060     __ breakpoint_trap(Assembler::notEqual, Assembler::xcc);
1061 #endif
1062   }
1063 
1064   //
1065   //  Generate overlap test for array copy stubs
1066   //
1067   //  Input:
1068   //    O0    -  array1
1069   //    O1    -  array2
1070   //    O2    -  element count
1071   //
1072   //  Kills temps:  O3, O4
1073   //
1074   void array_overlap_test(address no_overlap_target, int log2_elem_size) {
1075     assert(no_overlap_target != NULL, "must be generated");
1076     array_overlap_test(no_overlap_target, NULL, log2_elem_size);
1077   }
1078   void array_overlap_test(Label& L_no_overlap, int log2_elem_size) {
1079     array_overlap_test(NULL, &L_no_overlap, log2_elem_size);
1080   }
1081   void array_overlap_test(address no_overlap_target, Label* NOLp, int log2_elem_size) {
1082     const Register from       = O0;
1083     const Register to         = O1;
1084     const Register count      = O2;
1085     const Register to_from    = O3; // to - from
1086     const Register byte_count = O4; // count << log2_elem_size
1087 
1088       __ subcc(to, from, to_from);
1089       __ sll_ptr(count, log2_elem_size, byte_count);
1090       if (NOLp == NULL)
1091         __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, no_overlap_target);
1092       else
1093         __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, (*NOLp));
1094       __ delayed()->cmp(to_from, byte_count);
1095       if (NOLp == NULL)
1096         __ brx(Assembler::greaterEqual, false, Assembler::pt, no_overlap_target);
1097       else
1098         __ brx(Assembler::greaterEqual, false, Assembler::pt, (*NOLp));
1099       __ delayed()->nop();
1100   }
1101 
1102   //
1103   //  Generate pre-write barrier for array.
1104   //
1105   //  Input:
1106   //     addr     - register containing starting address
1107   //     count    - register containing element count
1108   //     tmp      - scratch register
1109   //
1110   //  The input registers are overwritten.
1111   //
1112   void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1113 #if 0 // G1 only
1114     BarrierSet* bs = Universe::heap()->barrier_set();
1115     if (bs->has_write_ref_pre_barrier()) {
1116       assert(bs->has_write_ref_array_pre_opt(),
1117              "Else unsupported barrier set.");
1118 
1119       assert(addr->is_global() && count->is_global(),
1120              "If not, then we have to fix this code to handle more "
1121              "general cases.");
1122       // Get some new fresh output registers.
1123       __ save_frame(0);
1124       // Save the necessary global regs... will be used after.
1125       __ mov(addr, L0);
1126       __ mov(count, L1);
1127 
1128       __ mov(addr, O0);
1129       // Get the count into O1
1130       __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre));
1131       __ delayed()->mov(count, O1);
1132       __ mov(L0, addr);
1133       __ mov(L1, count);
1134       __ restore();
1135     }
1136 #endif // 0
1137   }
1138   //
1139   //  Generate post-write barrier for array.
1140   //
1141   //  Input:
1142   //     addr     - register containing starting address
1143   //     count    - register containing element count
1144   //     tmp      - scratch register
1145   //
1146   //  The input registers are overwritten.
1147   //
1148   void gen_write_ref_array_post_barrier(Register addr, Register count,
1149                                    Register tmp) {
1150     BarrierSet* bs = Universe::heap()->barrier_set();
1151 
1152     switch (bs->kind()) {
1153 #if 0 // G1 - only
1154       case BarrierSet::G1SATBCT:
1155       case BarrierSet::G1SATBCTLogging:
1156         {
1157           assert(addr->is_global() && count->is_global(),
1158                  "If not, then we have to fix this code to handle more "
1159                  "general cases.");
1160           // Get some new fresh output registers.
1161           __ save_frame(0);
1162           __ mov(addr, O0);
1163           __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
1164           __ delayed()->mov(count, O1);
1165           __ restore();
1166         }
1167         break;
1168 #endif // 0 G1 - only
1169       case BarrierSet::CardTableModRef:
1170       case BarrierSet::CardTableExtension:
1171         {
1172           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1173           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1174           assert_different_registers(addr, count, tmp);
1175 
1176           Label L_loop;
1177 
1178           __ sll_ptr(count, LogBytesPerHeapOop, count);
1179           __ sub(count, BytesPerHeapOop, count);
1180           __ add(count, addr, count);
1181           // Use two shifts to clear out those low order two bits! (Cannot opt. into 1.)
1182           __ srl_ptr(addr, CardTableModRefBS::card_shift, addr);
1183           __ srl_ptr(count, CardTableModRefBS::card_shift, count);
1184           __ sub(count, addr, count);
1185           Address rs(tmp, (address)ct->byte_map_base);
1186           __ load_address(rs);
1187         __ BIND(L_loop);
1188           __ stb(G0, rs.base(), addr);
1189           __ subcc(count, 1, count);
1190           __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1191           __ delayed()->add(addr, 1, addr);
1192 
1193           }
1194         break;
1195       case BarrierSet::ModRef:
1196         break;
1197       default      :
1198         ShouldNotReachHere();
1199 
1200     }
1201   }
1202 
1203 
1204   // Copy big chunks forward with shift
1205   //
1206   // Inputs:
1207   //   from      - source arrays
1208   //   to        - destination array aligned to 8-bytes
1209   //   count     - elements count to copy >= the count equivalent to 16 bytes
1210   //   count_dec - elements count's decrement equivalent to 16 bytes
1211   //   L_copy_bytes - copy exit label
1212   //
1213   void copy_16_bytes_forward_with_shift(Register from, Register to,
1214                      Register count, int count_dec, Label& L_copy_bytes) {
1215     Label L_loop, L_aligned_copy, L_copy_last_bytes;
1216 
1217     // if both arrays have the same alignment mod 8, do 8 bytes aligned copy
1218       __ andcc(from, 7, G1); // misaligned bytes
1219       __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1220       __ delayed()->nop();
1221 
1222     const Register left_shift  = G1; // left  shift bit counter
1223     const Register right_shift = G5; // right shift bit counter
1224 
1225       __ sll(G1, LogBitsPerByte, left_shift);
1226       __ mov(64, right_shift);
1227       __ sub(right_shift, left_shift, right_shift);
1228 
1229     //
1230     // Load 2 aligned 8-bytes chunks and use one from previous iteration
1231     // to form 2 aligned 8-bytes chunks to store.
1232     //
1233       __ deccc(count, count_dec); // Pre-decrement 'count'
1234       __ andn(from, 7, from);     // Align address
1235       __ ldx(from, 0, O3);
1236       __ inc(from, 8);
1237       __ align(16);
1238     __ BIND(L_loop);
1239       __ ldx(from, 0, O4);
1240       __ deccc(count, count_dec); // Can we do next iteration after this one?
1241       __ ldx(from, 8, G4);
1242       __ inc(to, 16);
1243       __ inc(from, 16);
1244       __ sllx(O3, left_shift,  O3);
1245       __ srlx(O4, right_shift, G3);
1246       __ bset(G3, O3);
1247       __ stx(O3, to, -16);
1248       __ sllx(O4, left_shift,  O4);
1249       __ srlx(G4, right_shift, G3);
1250       __ bset(G3, O4);
1251       __ stx(O4, to, -8);
1252       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1253       __ delayed()->mov(G4, O3);
1254 
1255       __ inccc(count, count_dec>>1 ); // + 8 bytes
1256       __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes);
1257       __ delayed()->inc(count, count_dec>>1); // restore 'count'
1258 
1259       // copy 8 bytes, part of them already loaded in O3
1260       __ ldx(from, 0, O4);
1261       __ inc(to, 8);
1262       __ inc(from, 8);
1263       __ sllx(O3, left_shift,  O3);
1264       __ srlx(O4, right_shift, G3);
1265       __ bset(O3, G3);
1266       __ stx(G3, to, -8);
1267 
1268     __ BIND(L_copy_last_bytes);
1269       __ srl(right_shift, LogBitsPerByte, right_shift); // misaligned bytes
1270       __ br(Assembler::always, false, Assembler::pt, L_copy_bytes);
1271       __ delayed()->sub(from, right_shift, from);       // restore address
1272 
1273     __ BIND(L_aligned_copy);
1274   }
1275 
1276   // Copy big chunks backward with shift
1277   //
1278   // Inputs:
1279   //   end_from  - source arrays end address
1280   //   end_to    - destination array end address aligned to 8-bytes
1281   //   count     - elements count to copy >= the count equivalent to 16 bytes
1282   //   count_dec - elements count's decrement equivalent to 16 bytes
1283   //   L_aligned_copy - aligned copy exit label
1284   //   L_copy_bytes   - copy exit label
1285   //
1286   void copy_16_bytes_backward_with_shift(Register end_from, Register end_to,
1287                      Register count, int count_dec,
1288                      Label& L_aligned_copy, Label& L_copy_bytes) {
1289     Label L_loop, L_copy_last_bytes;
1290 
1291     // if both arrays have the same alignment mod 8, do 8 bytes aligned copy
1292       __ andcc(end_from, 7, G1); // misaligned bytes
1293       __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1294       __ delayed()->deccc(count, count_dec); // Pre-decrement 'count'
1295 
1296     const Register left_shift  = G1; // left  shift bit counter
1297     const Register right_shift = G5; // right shift bit counter
1298 
1299       __ sll(G1, LogBitsPerByte, left_shift);
1300       __ mov(64, right_shift);
1301       __ sub(right_shift, left_shift, right_shift);
1302 
1303     //
1304     // Load 2 aligned 8-bytes chunks and use one from previous iteration
1305     // to form 2 aligned 8-bytes chunks to store.
1306     //
1307       __ andn(end_from, 7, end_from);     // Align address
1308       __ ldx(end_from, 0, O3);
1309       __ align(16);
1310     __ BIND(L_loop);
1311       __ ldx(end_from, -8, O4);
1312       __ deccc(count, count_dec); // Can we do next iteration after this one?
1313       __ ldx(end_from, -16, G4);
1314       __ dec(end_to, 16);
1315       __ dec(end_from, 16);
1316       __ srlx(O3, right_shift, O3);
1317       __ sllx(O4, left_shift,  G3);
1318       __ bset(G3, O3);
1319       __ stx(O3, end_to, 8);
1320       __ srlx(O4, right_shift, O4);
1321       __ sllx(G4, left_shift,  G3);
1322       __ bset(G3, O4);
1323       __ stx(O4, end_to, 0);
1324       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1325       __ delayed()->mov(G4, O3);
1326 
1327       __ inccc(count, count_dec>>1 ); // + 8 bytes
1328       __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes);
1329       __ delayed()->inc(count, count_dec>>1); // restore 'count'
1330 
1331       // copy 8 bytes, part of them already loaded in O3
1332       __ ldx(end_from, -8, O4);
1333       __ dec(end_to, 8);
1334       __ dec(end_from, 8);
1335       __ srlx(O3, right_shift, O3);
1336       __ sllx(O4, left_shift,  G3);
1337       __ bset(O3, G3);
1338       __ stx(G3, end_to, 0);
1339 
1340     __ BIND(L_copy_last_bytes);
1341       __ srl(left_shift, LogBitsPerByte, left_shift);    // misaligned bytes
1342       __ br(Assembler::always, false, Assembler::pt, L_copy_bytes);
1343       __ delayed()->add(end_from, left_shift, end_from); // restore address
1344   }
1345 
1346   //
1347   //  Generate stub for disjoint byte copy.  If "aligned" is true, the
1348   //  "from" and "to" addresses are assumed to be heapword aligned.
1349   //
1350   // Arguments for generated stub:
1351   //      from:  O0
1352   //      to:    O1
1353   //      count: O2 treated as signed
1354   //
1355   address generate_disjoint_byte_copy(bool aligned, const char * name) {
1356     __ align(CodeEntryAlignment);
1357     StubCodeMark mark(this, "StubRoutines", name);
1358     address start = __ pc();
1359 
1360     Label L_skip_alignment, L_align;
1361     Label L_copy_byte, L_copy_byte_loop, L_exit;
1362 
1363     const Register from      = O0;   // source array address
1364     const Register to        = O1;   // destination array address
1365     const Register count     = O2;   // elements count
1366     const Register offset    = O5;   // offset from start of arrays
1367     // O3, O4, G3, G4 are used as temp registers
1368 
1369     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
1370 
1371     if (!aligned)  disjoint_byte_copy_entry = __ pc();
1372     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1373     if (!aligned)  BLOCK_COMMENT("Entry:");
1374 
1375     // for short arrays, just do single element copy
1376     __ cmp(count, 23); // 16 + 7
1377     __ brx(Assembler::less, false, Assembler::pn, L_copy_byte);
1378     __ delayed()->mov(G0, offset);
1379 
1380     if (aligned) {
1381       // 'aligned' == true when it is known statically during compilation
1382       // of this arraycopy call site that both 'from' and 'to' addresses
1383       // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1384       //
1385       // Aligned arrays have 4 bytes alignment in 32-bits VM
1386       // and 8 bytes - in 64-bits VM. So we do it only for 32-bits VM
1387       //
1388 #ifndef _LP64
1389       // copy a 4-bytes word if necessary to align 'to' to 8 bytes
1390       __ andcc(to, 7, G0);
1391       __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment);
1392       __ delayed()->ld(from, 0, O3);
1393       __ inc(from, 4);
1394       __ inc(to, 4);
1395       __ dec(count, 4);
1396       __ st(O3, to, -4);
1397     __ BIND(L_skip_alignment);
1398 #endif
1399     } else {
1400       // copy bytes to align 'to' on 8 byte boundary
1401       __ andcc(to, 7, G1); // misaligned bytes
1402       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1403       __ delayed()->neg(G1);
1404       __ inc(G1, 8);       // bytes need to copy to next 8-bytes alignment
1405       __ sub(count, G1, count);
1406     __ BIND(L_align);
1407       __ ldub(from, 0, O3);
1408       __ deccc(G1);
1409       __ inc(from);
1410       __ stb(O3, to, 0);
1411       __ br(Assembler::notZero, false, Assembler::pt, L_align);
1412       __ delayed()->inc(to);
1413     __ BIND(L_skip_alignment);
1414     }
1415 #ifdef _LP64
1416     if (!aligned)
1417 #endif
1418     {
1419       // Copy with shift 16 bytes per iteration if arrays do not have
1420       // the same alignment mod 8, otherwise fall through to the next
1421       // code for aligned copy.
1422       // The compare above (count >= 23) guarantes 'count' >= 16 bytes.
1423       // Also jump over aligned copy after the copy with shift completed.
1424 
1425       copy_16_bytes_forward_with_shift(from, to, count, 16, L_copy_byte);
1426     }
1427 
1428     // Both array are 8 bytes aligned, copy 16 bytes at a time
1429       __ and3(count, 7, G4); // Save count
1430       __ srl(count, 3, count);
1431      generate_disjoint_long_copy_core(aligned);
1432       __ mov(G4, count);     // Restore count
1433 
1434     // copy tailing bytes
1435     __ BIND(L_copy_byte);
1436       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1437       __ delayed()->nop();
1438       __ align(16);
1439     __ BIND(L_copy_byte_loop);
1440       __ ldub(from, offset, O3);
1441       __ deccc(count);
1442       __ stb(O3, to, offset);
1443       __ brx(Assembler::notZero, false, Assembler::pt, L_copy_byte_loop);
1444       __ delayed()->inc(offset);
1445 
1446     __ BIND(L_exit);
1447       // O3, O4 are used as temp registers
1448       inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4);
1449       __ retl();
1450       __ delayed()->mov(G0, O0); // return 0
1451     return start;
1452   }
1453 
1454   //
1455   //  Generate stub for conjoint byte copy.  If "aligned" is true, the
1456   //  "from" and "to" addresses are assumed to be heapword aligned.
1457   //
1458   // Arguments for generated stub:
1459   //      from:  O0
1460   //      to:    O1
1461   //      count: O2 treated as signed
1462   //
1463   address generate_conjoint_byte_copy(bool aligned, const char * name) {
1464     // Do reverse copy.
1465 
1466     __ align(CodeEntryAlignment);
1467     StubCodeMark mark(this, "StubRoutines", name);
1468     address start = __ pc();
1469     address nooverlap_target = aligned ?
1470         StubRoutines::arrayof_jbyte_disjoint_arraycopy() :
1471         disjoint_byte_copy_entry;
1472 
1473     Label L_skip_alignment, L_align, L_aligned_copy;
1474     Label L_copy_byte, L_copy_byte_loop, L_exit;
1475 
1476     const Register from      = O0;   // source array address
1477     const Register to        = O1;   // destination array address
1478     const Register count     = O2;   // elements count
1479     const Register end_from  = from; // source array end address
1480     const Register end_to    = to;   // destination array end address
1481 
1482     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
1483 
1484     if (!aligned)  byte_copy_entry = __ pc();
1485     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1486     if (!aligned)  BLOCK_COMMENT("Entry:");
1487 
1488     array_overlap_test(nooverlap_target, 0);
1489 
1490     __ add(to, count, end_to);       // offset after last copied element
1491 
1492     // for short arrays, just do single element copy
1493     __ cmp(count, 23); // 16 + 7
1494     __ brx(Assembler::less, false, Assembler::pn, L_copy_byte);
1495     __ delayed()->add(from, count, end_from);
1496 
1497     {
1498       // Align end of arrays since they could be not aligned even
1499       // when arrays itself are aligned.
1500 
1501       // copy bytes to align 'end_to' on 8 byte boundary
1502       __ andcc(end_to, 7, G1); // misaligned bytes
1503       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1504       __ delayed()->nop();
1505       __ sub(count, G1, count);
1506     __ BIND(L_align);
1507       __ dec(end_from);
1508       __ dec(end_to);
1509       __ ldub(end_from, 0, O3);
1510       __ deccc(G1);
1511       __ brx(Assembler::notZero, false, Assembler::pt, L_align);
1512       __ delayed()->stb(O3, end_to, 0);
1513     __ BIND(L_skip_alignment);
1514     }
1515 #ifdef _LP64
1516     if (aligned) {
1517       // Both arrays are aligned to 8-bytes in 64-bits VM.
1518       // The 'count' is decremented in copy_16_bytes_backward_with_shift()
1519       // in unaligned case.
1520       __ dec(count, 16);
1521     } else
1522 #endif
1523     {
1524       // Copy with shift 16 bytes per iteration if arrays do not have
1525       // the same alignment mod 8, otherwise jump to the next
1526       // code for aligned copy (and substracting 16 from 'count' before jump).
1527       // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1528       // Also jump over aligned copy after the copy with shift completed.
1529 
1530       copy_16_bytes_backward_with_shift(end_from, end_to, count, 16,
1531                                         L_aligned_copy, L_copy_byte);
1532     }
1533     // copy 4 elements (16 bytes) at a time
1534       __ align(16);
1535     __ BIND(L_aligned_copy);
1536       __ dec(end_from, 16);
1537       __ ldx(end_from, 8, O3);
1538       __ ldx(end_from, 0, O4);
1539       __ dec(end_to, 16);
1540       __ deccc(count, 16);
1541       __ stx(O3, end_to, 8);
1542       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
1543       __ delayed()->stx(O4, end_to, 0);
1544       __ inc(count, 16);
1545 
1546     // copy 1 element (2 bytes) at a time
1547     __ BIND(L_copy_byte);
1548       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1549       __ delayed()->nop();
1550       __ align(16);
1551     __ BIND(L_copy_byte_loop);
1552       __ dec(end_from);
1553       __ dec(end_to);
1554       __ ldub(end_from, 0, O4);
1555       __ deccc(count);
1556       __ brx(Assembler::greater, false, Assembler::pt, L_copy_byte_loop);
1557       __ delayed()->stb(O4, end_to, 0);
1558 
1559     __ BIND(L_exit);
1560     // O3, O4 are used as temp registers
1561     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4);
1562     __ retl();
1563     __ delayed()->mov(G0, O0); // return 0
1564     return start;
1565   }
1566 
1567   //
1568   //  Generate stub for disjoint short copy.  If "aligned" is true, the
1569   //  "from" and "to" addresses are assumed to be heapword aligned.
1570   //
1571   // Arguments for generated stub:
1572   //      from:  O0
1573   //      to:    O1
1574   //      count: O2 treated as signed
1575   //
1576   address generate_disjoint_short_copy(bool aligned, const char * name) {
1577     __ align(CodeEntryAlignment);
1578     StubCodeMark mark(this, "StubRoutines", name);
1579     address start = __ pc();
1580 
1581     Label L_skip_alignment, L_skip_alignment2;
1582     Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit;
1583 
1584     const Register from      = O0;   // source array address
1585     const Register to        = O1;   // destination array address
1586     const Register count     = O2;   // elements count
1587     const Register offset    = O5;   // offset from start of arrays
1588     // O3, O4, G3, G4 are used as temp registers
1589 
1590     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
1591 
1592     if (!aligned)  disjoint_short_copy_entry = __ pc();
1593     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1594     if (!aligned)  BLOCK_COMMENT("Entry:");
1595 
1596     // for short arrays, just do single element copy
1597     __ cmp(count, 11); // 8 + 3  (22 bytes)
1598     __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes);
1599     __ delayed()->mov(G0, offset);
1600 
1601     if (aligned) {
1602       // 'aligned' == true when it is known statically during compilation
1603       // of this arraycopy call site that both 'from' and 'to' addresses
1604       // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1605       //
1606       // Aligned arrays have 4 bytes alignment in 32-bits VM
1607       // and 8 bytes - in 64-bits VM.
1608       //
1609 #ifndef _LP64
1610       // copy a 2-elements word if necessary to align 'to' to 8 bytes
1611       __ andcc(to, 7, G0);
1612       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1613       __ delayed()->ld(from, 0, O3);
1614       __ inc(from, 4);
1615       __ inc(to, 4);
1616       __ dec(count, 2);
1617       __ st(O3, to, -4);
1618     __ BIND(L_skip_alignment);
1619 #endif
1620     } else {
1621       // copy 1 element if necessary to align 'to' on an 4 bytes
1622       __ andcc(to, 3, G0);
1623       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1624       __ delayed()->lduh(from, 0, O3);
1625       __ inc(from, 2);
1626       __ inc(to, 2);
1627       __ dec(count);
1628       __ sth(O3, to, -2);
1629     __ BIND(L_skip_alignment);
1630 
1631       // copy 2 elements to align 'to' on an 8 byte boundary
1632       __ andcc(to, 7, G0);
1633       __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2);
1634       __ delayed()->lduh(from, 0, O3);
1635       __ dec(count, 2);
1636       __ lduh(from, 2, O4);
1637       __ inc(from, 4);
1638       __ inc(to, 4);
1639       __ sth(O3, to, -4);
1640       __ sth(O4, to, -2);
1641     __ BIND(L_skip_alignment2);
1642     }
1643 #ifdef _LP64
1644     if (!aligned)
1645 #endif
1646     {
1647       // Copy with shift 16 bytes per iteration if arrays do not have
1648       // the same alignment mod 8, otherwise fall through to the next
1649       // code for aligned copy.
1650       // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1651       // Also jump over aligned copy after the copy with shift completed.
1652 
1653       copy_16_bytes_forward_with_shift(from, to, count, 8, L_copy_2_bytes);
1654     }
1655 
1656     // Both array are 8 bytes aligned, copy 16 bytes at a time
1657       __ and3(count, 3, G4); // Save
1658       __ srl(count, 2, count);
1659      generate_disjoint_long_copy_core(aligned);
1660       __ mov(G4, count); // restore
1661 
1662     // copy 1 element at a time
1663     __ BIND(L_copy_2_bytes);
1664       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1665       __ delayed()->nop();
1666       __ align(16);
1667     __ BIND(L_copy_2_bytes_loop);
1668       __ lduh(from, offset, O3);
1669       __ deccc(count);
1670       __ sth(O3, to, offset);
1671       __ brx(Assembler::notZero, false, Assembler::pt, L_copy_2_bytes_loop);
1672       __ delayed()->inc(offset, 2);
1673 
1674     __ BIND(L_exit);
1675       // O3, O4 are used as temp registers
1676       inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4);
1677       __ retl();
1678       __ delayed()->mov(G0, O0); // return 0
1679     return start;
1680   }
1681 
1682   //
1683   //  Generate stub for conjoint short copy.  If "aligned" is true, the
1684   //  "from" and "to" addresses are assumed to be heapword aligned.
1685   //
1686   // Arguments for generated stub:
1687   //      from:  O0
1688   //      to:    O1
1689   //      count: O2 treated as signed
1690   //
1691   address generate_conjoint_short_copy(bool aligned, const char * name) {
1692     // Do reverse copy.
1693 
1694     __ align(CodeEntryAlignment);
1695     StubCodeMark mark(this, "StubRoutines", name);
1696     address start = __ pc();
1697     address nooverlap_target = aligned ?
1698         StubRoutines::arrayof_jshort_disjoint_arraycopy() :
1699         disjoint_short_copy_entry;
1700 
1701     Label L_skip_alignment, L_skip_alignment2, L_aligned_copy;
1702     Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit;
1703 
1704     const Register from      = O0;   // source array address
1705     const Register to        = O1;   // destination array address
1706     const Register count     = O2;   // elements count
1707     const Register end_from  = from; // source array end address
1708     const Register end_to    = to;   // destination array end address
1709 
1710     const Register byte_count = O3;  // bytes count to copy
1711 
1712     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
1713 
1714     if (!aligned)  short_copy_entry = __ pc();
1715     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1716     if (!aligned)  BLOCK_COMMENT("Entry:");
1717 
1718     array_overlap_test(nooverlap_target, 1);
1719 
1720     __ sllx(count, LogBytesPerShort, byte_count);
1721     __ add(to, byte_count, end_to);  // offset after last copied element
1722 
1723     // for short arrays, just do single element copy
1724     __ cmp(count, 11); // 8 + 3  (22 bytes)
1725     __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes);
1726     __ delayed()->add(from, byte_count, end_from);
1727 
1728     {
1729       // Align end of arrays since they could be not aligned even
1730       // when arrays itself are aligned.
1731 
1732       // copy 1 element if necessary to align 'end_to' on an 4 bytes
1733       __ andcc(end_to, 3, G0);
1734       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1735       __ delayed()->lduh(end_from, -2, O3);
1736       __ dec(end_from, 2);
1737       __ dec(end_to, 2);
1738       __ dec(count);
1739       __ sth(O3, end_to, 0);
1740     __ BIND(L_skip_alignment);
1741 
1742       // copy 2 elements to align 'end_to' on an 8 byte boundary
1743       __ andcc(end_to, 7, G0);
1744       __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2);
1745       __ delayed()->lduh(end_from, -2, O3);
1746       __ dec(count, 2);
1747       __ lduh(end_from, -4, O4);
1748       __ dec(end_from, 4);
1749       __ dec(end_to, 4);
1750       __ sth(O3, end_to, 2);
1751       __ sth(O4, end_to, 0);
1752     __ BIND(L_skip_alignment2);
1753     }
1754 #ifdef _LP64
1755     if (aligned) {
1756       // Both arrays are aligned to 8-bytes in 64-bits VM.
1757       // The 'count' is decremented in copy_16_bytes_backward_with_shift()
1758       // in unaligned case.
1759       __ dec(count, 8);
1760     } else
1761 #endif
1762     {
1763       // Copy with shift 16 bytes per iteration if arrays do not have
1764       // the same alignment mod 8, otherwise jump to the next
1765       // code for aligned copy (and substracting 8 from 'count' before jump).
1766       // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1767       // Also jump over aligned copy after the copy with shift completed.
1768 
1769       copy_16_bytes_backward_with_shift(end_from, end_to, count, 8,
1770                                         L_aligned_copy, L_copy_2_bytes);
1771     }
1772     // copy 4 elements (16 bytes) at a time
1773       __ align(16);
1774     __ BIND(L_aligned_copy);
1775       __ dec(end_from, 16);
1776       __ ldx(end_from, 8, O3);
1777       __ ldx(end_from, 0, O4);
1778       __ dec(end_to, 16);
1779       __ deccc(count, 8);
1780       __ stx(O3, end_to, 8);
1781       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
1782       __ delayed()->stx(O4, end_to, 0);
1783       __ inc(count, 8);
1784 
1785     // copy 1 element (2 bytes) at a time
1786     __ BIND(L_copy_2_bytes);
1787       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1788       __ delayed()->nop();
1789     __ BIND(L_copy_2_bytes_loop);
1790       __ dec(end_from, 2);
1791       __ dec(end_to, 2);
1792       __ lduh(end_from, 0, O4);
1793       __ deccc(count);
1794       __ brx(Assembler::greater, false, Assembler::pt, L_copy_2_bytes_loop);
1795       __ delayed()->sth(O4, end_to, 0);
1796 
1797     __ BIND(L_exit);
1798     // O3, O4 are used as temp registers
1799     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4);
1800     __ retl();
1801     __ delayed()->mov(G0, O0); // return 0
1802     return start;
1803   }
1804 
1805   //
1806   //  Generate core code for disjoint int copy (and oop copy on 32-bit).
1807   //  If "aligned" is true, the "from" and "to" addresses are assumed
1808   //  to be heapword aligned.
1809   //
1810   // Arguments:
1811   //      from:  O0
1812   //      to:    O1
1813   //      count: O2 treated as signed
1814   //
1815   void generate_disjoint_int_copy_core(bool aligned) {
1816 
1817     Label L_skip_alignment, L_aligned_copy;
1818     Label L_copy_16_bytes,  L_copy_4_bytes, L_copy_4_bytes_loop, L_exit;
1819 
1820     const Register from      = O0;   // source array address
1821     const Register to        = O1;   // destination array address
1822     const Register count     = O2;   // elements count
1823     const Register offset    = O5;   // offset from start of arrays
1824     // O3, O4, G3, G4 are used as temp registers
1825 
1826     // 'aligned' == true when it is known statically during compilation
1827     // of this arraycopy call site that both 'from' and 'to' addresses
1828     // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1829     //
1830     // Aligned arrays have 4 bytes alignment in 32-bits VM
1831     // and 8 bytes - in 64-bits VM.
1832     //
1833 #ifdef _LP64
1834     if (!aligned)
1835 #endif
1836     {
1837       // The next check could be put under 'ifndef' since the code in
1838       // generate_disjoint_long_copy_core() has own checks and set 'offset'.
1839 
1840       // for short arrays, just do single element copy
1841       __ cmp(count, 5); // 4 + 1 (20 bytes)
1842       __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes);
1843       __ delayed()->mov(G0, offset);
1844 
1845       // copy 1 element to align 'to' on an 8 byte boundary
1846       __ andcc(to, 7, G0);
1847       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1848       __ delayed()->ld(from, 0, O3);
1849       __ inc(from, 4);
1850       __ inc(to, 4);
1851       __ dec(count);
1852       __ st(O3, to, -4);
1853     __ BIND(L_skip_alignment);
1854 
1855     // if arrays have same alignment mod 8, do 4 elements copy
1856       __ andcc(from, 7, G0);
1857       __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1858       __ delayed()->ld(from, 0, O3);
1859 
1860     //
1861     // Load 2 aligned 8-bytes chunks and use one from previous iteration
1862     // to form 2 aligned 8-bytes chunks to store.
1863     //
1864     // copy_16_bytes_forward_with_shift() is not used here since this
1865     // code is more optimal.
1866 
1867     // copy with shift 4 elements (16 bytes) at a time
1868       __ dec(count, 4);   // The cmp at the beginning guaranty count >= 4
1869 
1870       __ align(16);
1871     __ BIND(L_copy_16_bytes);
1872       __ ldx(from, 4, O4);
1873       __ deccc(count, 4); // Can we do next iteration after this one?
1874       __ ldx(from, 12, G4);
1875       __ inc(to, 16);
1876       __ inc(from, 16);
1877       __ sllx(O3, 32, O3);
1878       __ srlx(O4, 32, G3);
1879       __ bset(G3, O3);
1880       __ stx(O3, to, -16);
1881       __ sllx(O4, 32, O4);
1882       __ srlx(G4, 32, G3);
1883       __ bset(G3, O4);
1884       __ stx(O4, to, -8);
1885       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
1886       __ delayed()->mov(G4, O3);
1887 
1888       __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes);
1889       __ delayed()->inc(count, 4); // restore 'count'
1890 
1891     __ BIND(L_aligned_copy);
1892     }
1893     // copy 4 elements (16 bytes) at a time
1894       __ and3(count, 1, G4); // Save
1895       __ srl(count, 1, count);
1896      generate_disjoint_long_copy_core(aligned);
1897       __ mov(G4, count);     // Restore
1898 
1899     // copy 1 element at a time
1900     __ BIND(L_copy_4_bytes);
1901       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1902       __ delayed()->nop();
1903     __ BIND(L_copy_4_bytes_loop);
1904       __ ld(from, offset, O3);
1905       __ deccc(count);
1906       __ st(O3, to, offset);
1907       __ brx(Assembler::notZero, false, Assembler::pt, L_copy_4_bytes_loop);
1908       __ delayed()->inc(offset, 4);
1909     __ BIND(L_exit);
1910   }
1911 
1912   //
1913   //  Generate stub for disjoint int copy.  If "aligned" is true, the
1914   //  "from" and "to" addresses are assumed to be heapword aligned.
1915   //
1916   // Arguments for generated stub:
1917   //      from:  O0
1918   //      to:    O1
1919   //      count: O2 treated as signed
1920   //
1921   address generate_disjoint_int_copy(bool aligned, const char * name) {
1922     __ align(CodeEntryAlignment);
1923     StubCodeMark mark(this, "StubRoutines", name);
1924     address start = __ pc();
1925 
1926     const Register count = O2;
1927     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
1928 
1929     if (!aligned)  disjoint_int_copy_entry = __ pc();
1930     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1931     if (!aligned)  BLOCK_COMMENT("Entry:");
1932 
1933     generate_disjoint_int_copy_core(aligned);
1934 
1935     // O3, O4 are used as temp registers
1936     inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4);
1937     __ retl();
1938     __ delayed()->mov(G0, O0); // return 0
1939     return start;
1940   }
1941 
1942   //
1943   //  Generate core code for conjoint int copy (and oop copy on 32-bit).
1944   //  If "aligned" is true, the "from" and "to" addresses are assumed
1945   //  to be heapword aligned.
1946   //
1947   // Arguments:
1948   //      from:  O0
1949   //      to:    O1
1950   //      count: O2 treated as signed
1951   //
1952   void generate_conjoint_int_copy_core(bool aligned) {
1953     // Do reverse copy.
1954 
1955     Label L_skip_alignment, L_aligned_copy;
1956     Label L_copy_16_bytes,  L_copy_4_bytes, L_copy_4_bytes_loop, L_exit;
1957 
1958     const Register from      = O0;   // source array address
1959     const Register to        = O1;   // destination array address
1960     const Register count     = O2;   // elements count
1961     const Register end_from  = from; // source array end address
1962     const Register end_to    = to;   // destination array end address
1963     // O3, O4, O5, G3 are used as temp registers
1964 
1965     const Register byte_count = O3;  // bytes count to copy
1966 
1967       __ sllx(count, LogBytesPerInt, byte_count);
1968       __ add(to, byte_count, end_to); // offset after last copied element
1969 
1970       __ cmp(count, 5); // for short arrays, just do single element copy
1971       __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes);
1972       __ delayed()->add(from, byte_count, end_from);
1973 
1974     // copy 1 element to align 'to' on an 8 byte boundary
1975       __ andcc(end_to, 7, G0);
1976       __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1977       __ delayed()->nop();
1978       __ dec(count);
1979       __ dec(end_from, 4);
1980       __ dec(end_to,   4);
1981       __ ld(end_from, 0, O4);
1982       __ st(O4, end_to, 0);
1983     __ BIND(L_skip_alignment);
1984 
1985     // Check if 'end_from' and 'end_to' has the same alignment.
1986       __ andcc(end_from, 7, G0);
1987       __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1988       __ delayed()->dec(count, 4); // The cmp at the start guaranty cnt >= 4
1989 
1990     // copy with shift 4 elements (16 bytes) at a time
1991     //
1992     // Load 2 aligned 8-bytes chunks and use one from previous iteration
1993     // to form 2 aligned 8-bytes chunks to store.
1994     //
1995       __ ldx(end_from, -4, O3);
1996       __ align(16);
1997     __ BIND(L_copy_16_bytes);
1998       __ ldx(end_from, -12, O4);
1999       __ deccc(count, 4);
2000       __ ldx(end_from, -20, O5);
2001       __ dec(end_to, 16);
2002       __ dec(end_from, 16);
2003       __ srlx(O3, 32, O3);
2004       __ sllx(O4, 32, G3);
2005       __ bset(G3, O3);
2006       __ stx(O3, end_to, 8);
2007       __ srlx(O4, 32, O4);
2008       __ sllx(O5, 32, G3);
2009       __ bset(O4, G3);
2010       __ stx(G3, end_to, 0);
2011       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
2012       __ delayed()->mov(O5, O3);
2013 
2014       __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes);
2015       __ delayed()->inc(count, 4);
2016 
2017     // copy 4 elements (16 bytes) at a time
2018       __ align(16);
2019     __ BIND(L_aligned_copy);
2020       __ dec(end_from, 16);
2021       __ ldx(end_from, 8, O3);
2022       __ ldx(end_from, 0, O4);
2023       __ dec(end_to, 16);
2024       __ deccc(count, 4);
2025       __ stx(O3, end_to, 8);
2026       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
2027       __ delayed()->stx(O4, end_to, 0);
2028       __ inc(count, 4);
2029 
2030     // copy 1 element (4 bytes) at a time
2031     __ BIND(L_copy_4_bytes);
2032       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
2033       __ delayed()->nop();
2034     __ BIND(L_copy_4_bytes_loop);
2035       __ dec(end_from, 4);
2036       __ dec(end_to, 4);
2037       __ ld(end_from, 0, O4);
2038       __ deccc(count);
2039       __ brx(Assembler::greater, false, Assembler::pt, L_copy_4_bytes_loop);
2040       __ delayed()->st(O4, end_to, 0);
2041     __ BIND(L_exit);
2042   }
2043 
2044   //
2045   //  Generate stub for conjoint int copy.  If "aligned" is true, the
2046   //  "from" and "to" addresses are assumed to be heapword aligned.
2047   //
2048   // Arguments for generated stub:
2049   //      from:  O0
2050   //      to:    O1
2051   //      count: O2 treated as signed
2052   //
2053   address generate_conjoint_int_copy(bool aligned, const char * name) {
2054     __ align(CodeEntryAlignment);
2055     StubCodeMark mark(this, "StubRoutines", name);
2056     address start = __ pc();
2057 
2058     address nooverlap_target = aligned ?
2059         StubRoutines::arrayof_jint_disjoint_arraycopy() :
2060         disjoint_int_copy_entry;
2061 
2062     assert_clean_int(O2, O3);     // Make sure 'count' is clean int.
2063 
2064     if (!aligned)  int_copy_entry = __ pc();
2065     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2066     if (!aligned)  BLOCK_COMMENT("Entry:");
2067 
2068     array_overlap_test(nooverlap_target, 2);
2069 
2070     generate_conjoint_int_copy_core(aligned);
2071 
2072     // O3, O4 are used as temp registers
2073     inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4);
2074     __ retl();
2075     __ delayed()->mov(G0, O0); // return 0
2076     return start;
2077   }
2078 
2079   //
2080   //  Generate core code for disjoint long copy (and oop copy on 64-bit).
2081   //  "aligned" is ignored, because we must make the stronger
2082   //  assumption that both addresses are always 64-bit aligned.
2083   //
2084   // Arguments:
2085   //      from:  O0
2086   //      to:    O1
2087   //      count: O2 treated as signed
2088   //
2089   void generate_disjoint_long_copy_core(bool aligned) {
2090     Label L_copy_8_bytes, L_copy_16_bytes, L_exit;
2091     const Register from    = O0;  // source array address
2092     const Register to      = O1;  // destination array address
2093     const Register count   = O2;  // elements count
2094     const Register offset0 = O4;  // element offset
2095     const Register offset8 = O5;  // next element offset
2096 
2097       __ deccc(count, 2);
2098       __ mov(G0, offset0);   // offset from start of arrays (0)
2099       __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes );
2100       __ delayed()->add(offset0, 8, offset8);
2101       __ align(16);
2102     __ BIND(L_copy_16_bytes);
2103       __ ldx(from, offset0, O3);
2104       __ ldx(from, offset8, G3);
2105       __ deccc(count, 2);
2106       __ stx(O3, to, offset0);
2107       __ inc(offset0, 16);
2108       __ stx(G3, to, offset8);
2109       __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
2110       __ delayed()->inc(offset8, 16);
2111 
2112     __ BIND(L_copy_8_bytes);
2113       __ inccc(count, 2);
2114       __ brx(Assembler::zero, true, Assembler::pn, L_exit );
2115       __ delayed()->mov(offset0, offset8); // Set O5 used by other stubs
2116       __ ldx(from, offset0, O3);
2117       __ stx(O3, to, offset0);
2118     __ BIND(L_exit);
2119   }
2120 
2121   //
2122   //  Generate stub for disjoint long copy.
2123   //  "aligned" is ignored, because we must make the stronger
2124   //  assumption that both addresses are always 64-bit aligned.
2125   //
2126   // Arguments for generated stub:
2127   //      from:  O0
2128   //      to:    O1
2129   //      count: O2 treated as signed
2130   //
2131   address generate_disjoint_long_copy(bool aligned, const char * name) {
2132     __ align(CodeEntryAlignment);
2133     StubCodeMark mark(this, "StubRoutines", name);
2134     address start = __ pc();
2135 
2136     assert_clean_int(O2, O3);     // Make sure 'count' is clean int.
2137 
2138     if (!aligned)  disjoint_long_copy_entry = __ pc();
2139     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2140     if (!aligned)  BLOCK_COMMENT("Entry:");
2141 
2142     generate_disjoint_long_copy_core(aligned);
2143 
2144     // O3, O4 are used as temp registers
2145     inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4);
2146     __ retl();
2147     __ delayed()->mov(G0, O0); // return 0
2148     return start;
2149   }
2150 
2151   //
2152   //  Generate core code for conjoint long copy (and oop copy on 64-bit).
2153   //  "aligned" is ignored, because we must make the stronger
2154   //  assumption that both addresses are always 64-bit aligned.
2155   //
2156   // Arguments:
2157   //      from:  O0
2158   //      to:    O1
2159   //      count: O2 treated as signed
2160   //
2161   void generate_conjoint_long_copy_core(bool aligned) {
2162     // Do reverse copy.
2163     Label L_copy_8_bytes, L_copy_16_bytes, L_exit;
2164     const Register from    = O0;  // source array address
2165     const Register to      = O1;  // destination array address
2166     const Register count   = O2;  // elements count
2167     const Register offset8 = O4;  // element offset
2168     const Register offset0 = O5;  // previous element offset
2169 
2170       __ subcc(count, 1, count);
2171       __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_8_bytes );
2172       __ delayed()->sllx(count, LogBytesPerLong, offset8);
2173       __ sub(offset8, 8, offset0);
2174       __ align(16);
2175     __ BIND(L_copy_16_bytes);
2176       __ ldx(from, offset8, O2);
2177       __ ldx(from, offset0, O3);
2178       __ stx(O2, to, offset8);
2179       __ deccc(offset8, 16);      // use offset8 as counter
2180       __ stx(O3, to, offset0);
2181       __ brx(Assembler::greater, false, Assembler::pt, L_copy_16_bytes);
2182       __ delayed()->dec(offset0, 16);
2183 
2184     __ BIND(L_copy_8_bytes);
2185       __ brx(Assembler::negative, false, Assembler::pn, L_exit );
2186       __ delayed()->nop();
2187       __ ldx(from, 0, O3);
2188       __ stx(O3, to, 0);
2189     __ BIND(L_exit);
2190   }
2191 
2192   //  Generate stub for conjoint long copy.
2193   //  "aligned" is ignored, because we must make the stronger
2194   //  assumption that both addresses are always 64-bit aligned.
2195   //
2196   // Arguments for generated stub:
2197   //      from:  O0
2198   //      to:    O1
2199   //      count: O2 treated as signed
2200   //
2201   address generate_conjoint_long_copy(bool aligned, const char * name) {
2202     __ align(CodeEntryAlignment);
2203     StubCodeMark mark(this, "StubRoutines", name);
2204     address start = __ pc();
2205 
2206     assert(!aligned, "usage");
2207     address nooverlap_target = disjoint_long_copy_entry;
2208 
2209     assert_clean_int(O2, O3);     // Make sure 'count' is clean int.
2210 
2211     if (!aligned)  long_copy_entry = __ pc();
2212     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2213     if (!aligned)  BLOCK_COMMENT("Entry:");
2214 
2215     array_overlap_test(nooverlap_target, 3);
2216 
2217     generate_conjoint_long_copy_core(aligned);
2218 
2219     // O3, O4 are used as temp registers
2220     inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4);
2221     __ retl();
2222     __ delayed()->mov(G0, O0); // return 0
2223     return start;
2224   }
2225 
2226   //  Generate stub for disjoint oop copy.  If "aligned" is true, the
2227   //  "from" and "to" addresses are assumed to be heapword aligned.
2228   //
2229   // Arguments for generated stub:
2230   //      from:  O0
2231   //      to:    O1
2232   //      count: O2 treated as signed
2233   //
2234   address generate_disjoint_oop_copy(bool aligned, const char * name) {
2235 
2236     const Register from  = O0;  // source array address
2237     const Register to    = O1;  // destination array address
2238     const Register count = O2;  // elements count
2239 
2240     __ align(CodeEntryAlignment);
2241     StubCodeMark mark(this, "StubRoutines", name);
2242     address start = __ pc();
2243 
2244     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
2245 
2246     if (!aligned)  disjoint_oop_copy_entry = __ pc();
2247     // caller can pass a 64-bit byte count here
2248     if (!aligned)  BLOCK_COMMENT("Entry:");
2249 
2250     // save arguments for barrier generation
2251     __ mov(to, G1);
2252     __ mov(count, G5);
2253     gen_write_ref_array_pre_barrier(G1, G5);
2254   #ifdef _LP64
2255     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
2256     if (UseCompressedOops) {
2257       generate_disjoint_int_copy_core(aligned);
2258     } else {
2259       generate_disjoint_long_copy_core(aligned);
2260     }
2261   #else
2262     generate_disjoint_int_copy_core(aligned);
2263   #endif
2264     // O0 is used as temp register
2265     gen_write_ref_array_post_barrier(G1, G5, O0);
2266 
2267     // O3, O4 are used as temp registers
2268     inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4);
2269     __ retl();
2270     __ delayed()->mov(G0, O0); // return 0
2271     return start;
2272   }
2273 
2274   //  Generate stub for conjoint oop copy.  If "aligned" is true, the
2275   //  "from" and "to" addresses are assumed to be heapword aligned.
2276   //
2277   // Arguments for generated stub:
2278   //      from:  O0
2279   //      to:    O1
2280   //      count: O2 treated as signed
2281   //
2282   address generate_conjoint_oop_copy(bool aligned, const char * name) {
2283 
2284     const Register from  = O0;  // source array address
2285     const Register to    = O1;  // destination array address
2286     const Register count = O2;  // elements count
2287 
2288     __ align(CodeEntryAlignment);
2289     StubCodeMark mark(this, "StubRoutines", name);
2290     address start = __ pc();
2291 
2292     assert_clean_int(count, O3);     // Make sure 'count' is clean int.
2293 
2294     if (!aligned)  oop_copy_entry = __ pc();
2295     // caller can pass a 64-bit byte count here
2296     if (!aligned)  BLOCK_COMMENT("Entry:");
2297 
2298     // save arguments for barrier generation
2299     __ mov(to, G1);
2300     __ mov(count, G5);
2301 
2302     gen_write_ref_array_pre_barrier(G1, G5);
2303 
2304     address nooverlap_target = aligned ?
2305         StubRoutines::arrayof_oop_disjoint_arraycopy() :
2306         disjoint_oop_copy_entry;
2307 
2308     array_overlap_test(nooverlap_target, LogBytesPerHeapOop);
2309 
2310   #ifdef _LP64
2311     if (UseCompressedOops) {
2312       generate_conjoint_int_copy_core(aligned);
2313     } else {
2314       generate_conjoint_long_copy_core(aligned);
2315     }
2316   #else
2317     generate_conjoint_int_copy_core(aligned);
2318   #endif
2319 
2320     // O0 is used as temp register
2321     gen_write_ref_array_post_barrier(G1, G5, O0);
2322 
2323     // O3, O4 are used as temp registers
2324     inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4);
2325     __ retl();
2326     __ delayed()->mov(G0, O0); // return 0
2327     return start;
2328   }
2329 
2330 
2331   // Helper for generating a dynamic type check.
2332   // Smashes only the given temp registers.
2333   void generate_type_check(Register sub_klass,
2334                            Register super_check_offset,
2335                            Register super_klass,
2336                            Register temp,
2337                            Label& L_success,
2338                            Register deccc_hack = noreg) {
2339     assert_different_registers(sub_klass, super_check_offset, super_klass, temp);
2340 
2341     BLOCK_COMMENT("type_check:");
2342 
2343     Label L_miss;
2344 
2345     assert_clean_int(super_check_offset, temp);
2346 
2347     // maybe decrement caller's trip count:
2348 #define DELAY_SLOT delayed();   \
2349     { if (deccc_hack == noreg) __ nop(); else __ deccc(deccc_hack); }
2350 
2351     // if the pointers are equal, we are done (e.g., String[] elements)
2352     __ cmp(sub_klass, super_klass);
2353     __ brx(Assembler::equal, true, Assembler::pt, L_success);
2354     __ DELAY_SLOT;
2355 
2356     // check the supertype display:
2357     __ ld_ptr(sub_klass, super_check_offset, temp); // query the super type
2358     __ cmp(super_klass,                      temp); // test the super type
2359     __ brx(Assembler::equal, true, Assembler::pt, L_success);
2360     __ DELAY_SLOT;
2361 
2362     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
2363                      Klass::secondary_super_cache_offset_in_bytes());
2364     __ cmp(super_klass, sc_offset);
2365     __ brx(Assembler::notEqual, true, Assembler::pt, L_miss);
2366     __ delayed()->nop();
2367 
2368     __ save_frame(0);
2369     __ mov(sub_klass->after_save(), O1);
2370     // mov(super_klass->after_save(), O2); //fill delay slot
2371     assert(StubRoutines::Sparc::_partial_subtype_check != NULL, "order of generation");
2372     __ call(StubRoutines::Sparc::_partial_subtype_check);
2373     __ delayed()->mov(super_klass->after_save(), O2);
2374     __ restore();
2375 
2376     // Upon return, the condition codes are already set.
2377     __ brx(Assembler::equal, true, Assembler::pt, L_success);
2378     __ DELAY_SLOT;
2379 
2380 #undef DELAY_SLOT
2381 
2382     // Fall through on failure!
2383     __ BIND(L_miss);
2384   }
2385 
2386 
2387   //  Generate stub for checked oop copy.
2388   //
2389   // Arguments for generated stub:
2390   //      from:  O0
2391   //      to:    O1
2392   //      count: O2 treated as signed
2393   //      ckoff: O3 (super_check_offset)
2394   //      ckval: O4 (super_klass)
2395   //      ret:   O0 zero for success; (-1^K) where K is partial transfer count
2396   //
2397   address generate_checkcast_copy(const char* name) {
2398 
2399     const Register O0_from   = O0;      // source array address
2400     const Register O1_to     = O1;      // destination array address
2401     const Register O2_count  = O2;      // elements count
2402     const Register O3_ckoff  = O3;      // super_check_offset
2403     const Register O4_ckval  = O4;      // super_klass
2404 
2405     const Register O5_offset = O5;      // loop var, with stride wordSize
2406     const Register G1_remain = G1;      // loop var, with stride -1
2407     const Register G3_oop    = G3;      // actual oop copied
2408     const Register G4_klass  = G4;      // oop._klass
2409     const Register G5_super  = G5;      // oop._klass._primary_supers[ckval]
2410 
2411     __ align(CodeEntryAlignment);
2412     StubCodeMark mark(this, "StubRoutines", name);
2413     address start = __ pc();
2414 
2415     gen_write_ref_array_pre_barrier(G1, G5);
2416 
2417 
2418 #ifdef ASSERT
2419     // We sometimes save a frame (see partial_subtype_check below).
2420     // If this will cause trouble, let's fail now instead of later.
2421     __ save_frame(0);
2422     __ restore();
2423 #endif
2424 
2425 #ifdef ASSERT
2426     // caller guarantees that the arrays really are different
2427     // otherwise, we would have to make conjoint checks
2428     { Label L;
2429       __ mov(O3, G1);           // spill: overlap test smashes O3
2430       __ mov(O4, G4);           // spill: overlap test smashes O4
2431       array_overlap_test(L, LogBytesPerHeapOop);
2432       __ stop("checkcast_copy within a single array");
2433       __ bind(L);
2434       __ mov(G1, O3);
2435       __ mov(G4, O4);
2436     }
2437 #endif //ASSERT
2438 
2439     assert_clean_int(O2_count, G1);     // Make sure 'count' is clean int.
2440 
2441     checkcast_copy_entry = __ pc();
2442     // caller can pass a 64-bit byte count here (from generic stub)
2443     BLOCK_COMMENT("Entry:");
2444 
2445     Label load_element, store_element, do_card_marks, fail, done;
2446     __ addcc(O2_count, 0, G1_remain);   // initialize loop index, and test it
2447     __ brx(Assembler::notZero, false, Assembler::pt, load_element);
2448     __ delayed()->mov(G0, O5_offset);   // offset from start of arrays
2449 
2450     // Empty array:  Nothing to do.
2451     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4);
2452     __ retl();
2453     __ delayed()->set(0, O0);           // return 0 on (trivial) success
2454 
2455     // ======== begin loop ========
2456     // (Loop is rotated; its entry is load_element.)
2457     // Loop variables:
2458     //   (O5 = 0; ; O5 += wordSize) --- offset from src, dest arrays
2459     //   (O2 = len; O2 != 0; O2--) --- number of oops *remaining*
2460     //   G3, G4, G5 --- current oop, oop.klass, oop.klass.super
2461     __ align(16);
2462 
2463     __ bind(store_element);
2464     // deccc(G1_remain);                // decrement the count (hoisted)
2465     __ store_heap_oop(G3_oop, O1_to, O5_offset); // store the oop
2466     __ inc(O5_offset, heapOopSize);     // step to next offset
2467     __ brx(Assembler::zero, true, Assembler::pt, do_card_marks);
2468     __ delayed()->set(0, O0);           // return -1 on success
2469 
2470     // ======== loop entry is here ========
2471     __ bind(load_element);
2472     __ load_heap_oop(O0_from, O5_offset, G3_oop);  // load the oop
2473     __ br_null(G3_oop, true, Assembler::pt, store_element);
2474     __ delayed()->deccc(G1_remain);     // decrement the count
2475 
2476     __ load_klass(G3_oop, G4_klass); // query the object klass
2477 
2478     generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super,
2479                         // branch to this on success:
2480                         store_element,
2481                         // decrement this on success:
2482                         G1_remain);
2483     // ======== end loop ========
2484 
2485     // It was a real error; we must depend on the caller to finish the job.
2486     // Register G1 has number of *remaining* oops, O2 number of *total* oops.
2487     // Emit GC store barriers for the oops we have copied (O2 minus G1),
2488     // and report their number to the caller.
2489     __ bind(fail);
2490     __ subcc(O2_count, G1_remain, O2_count);
2491     __ brx(Assembler::zero, false, Assembler::pt, done);
2492     __ delayed()->not1(O2_count, O0);   // report (-1^K) to caller
2493 
2494     __ bind(do_card_marks);
2495     gen_write_ref_array_post_barrier(O1_to, O2_count, O3);   // store check on O1[0..O2]
2496 
2497     __ bind(done);
2498     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4);
2499     __ retl();
2500     __ delayed()->nop();             // return value in 00
2501 
2502     return start;
2503   }
2504 
2505 
2506   //  Generate 'unsafe' array copy stub
2507   //  Though just as safe as the other stubs, it takes an unscaled
2508   //  size_t argument instead of an element count.
2509   //
2510   // Arguments for generated stub:
2511   //      from:  O0
2512   //      to:    O1
2513   //      count: O2 byte count, treated as ssize_t, can be zero
2514   //
2515   // Examines the alignment of the operands and dispatches
2516   // to a long, int, short, or byte copy loop.
2517   //
2518   address generate_unsafe_copy(const char* name) {
2519 
2520     const Register O0_from   = O0;      // source array address
2521     const Register O1_to     = O1;      // destination array address
2522     const Register O2_count  = O2;      // elements count
2523 
2524     const Register G1_bits   = G1;      // test copy of low bits
2525 
2526     __ align(CodeEntryAlignment);
2527     StubCodeMark mark(this, "StubRoutines", name);
2528     address start = __ pc();
2529 
2530     // bump this on entry, not on exit:
2531     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr, G1, G3);
2532 
2533     __ or3(O0_from, O1_to, G1_bits);
2534     __ or3(O2_count,       G1_bits, G1_bits);
2535 
2536     __ btst(BytesPerLong-1, G1_bits);
2537     __ br(Assembler::zero, true, Assembler::pt,
2538           long_copy_entry, relocInfo::runtime_call_type);
2539     // scale the count on the way out:
2540     __ delayed()->srax(O2_count, LogBytesPerLong, O2_count);
2541 
2542     __ btst(BytesPerInt-1, G1_bits);
2543     __ br(Assembler::zero, true, Assembler::pt,
2544           int_copy_entry, relocInfo::runtime_call_type);
2545     // scale the count on the way out:
2546     __ delayed()->srax(O2_count, LogBytesPerInt, O2_count);
2547 
2548     __ btst(BytesPerShort-1, G1_bits);
2549     __ br(Assembler::zero, true, Assembler::pt,
2550           short_copy_entry, relocInfo::runtime_call_type);
2551     // scale the count on the way out:
2552     __ delayed()->srax(O2_count, LogBytesPerShort, O2_count);
2553 
2554     __ br(Assembler::always, false, Assembler::pt,
2555           byte_copy_entry, relocInfo::runtime_call_type);
2556     __ delayed()->nop();
2557 
2558     return start;
2559   }
2560 
2561 
2562   // Perform range checks on the proposed arraycopy.
2563   // Kills the two temps, but nothing else.
2564   // Also, clean the sign bits of src_pos and dst_pos.
2565   void arraycopy_range_checks(Register src,     // source array oop (O0)
2566                               Register src_pos, // source position (O1)
2567                               Register dst,     // destination array oo (O2)
2568                               Register dst_pos, // destination position (O3)
2569                               Register length,  // length of copy (O4)
2570                               Register temp1, Register temp2,
2571                               Label& L_failed) {
2572     BLOCK_COMMENT("arraycopy_range_checks:");
2573 
2574     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;
2575 
2576     const Register array_length = temp1;  // scratch
2577     const Register end_pos      = temp2;  // scratch
2578 
2579     // Note:  This next instruction may be in the delay slot of a branch:
2580     __ add(length, src_pos, end_pos);  // src_pos + length
2581     __ lduw(src, arrayOopDesc::length_offset_in_bytes(), array_length);
2582     __ cmp(end_pos, array_length);
2583     __ br(Assembler::greater, false, Assembler::pn, L_failed);
2584 
2585     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
2586     __ delayed()->add(length, dst_pos, end_pos); // dst_pos + length
2587     __ lduw(dst, arrayOopDesc::length_offset_in_bytes(), array_length);
2588     __ cmp(end_pos, array_length);
2589     __ br(Assembler::greater, false, Assembler::pn, L_failed);
2590 
2591     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2592     // Move with sign extension can be used since they are positive.
2593     __ delayed()->signx(src_pos, src_pos);
2594     __ signx(dst_pos, dst_pos);
2595 
2596     BLOCK_COMMENT("arraycopy_range_checks done");
2597   }
2598 
2599 
2600   //
2601   //  Generate generic array copy stubs
2602   //
2603   //  Input:
2604   //    O0    -  src oop
2605   //    O1    -  src_pos
2606   //    O2    -  dst oop
2607   //    O3    -  dst_pos
2608   //    O4    -  element count
2609   //
2610   //  Output:
2611   //    O0 ==  0  -  success
2612   //    O0 == -1  -  need to call System.arraycopy
2613   //
2614   address generate_generic_copy(const char *name) {
2615 
2616     Label L_failed, L_objArray;
2617 
2618     // Input registers
2619     const Register src      = O0;  // source array oop
2620     const Register src_pos  = O1;  // source position
2621     const Register dst      = O2;  // destination array oop
2622     const Register dst_pos  = O3;  // destination position
2623     const Register length   = O4;  // elements count
2624 
2625     // registers used as temp
2626     const Register G3_src_klass = G3; // source array klass
2627     const Register G4_dst_klass = G4; // destination array klass
2628     const Register G5_lh        = G5; // layout handler
2629     const Register O5_temp      = O5;
2630 
2631     __ align(CodeEntryAlignment);
2632     StubCodeMark mark(this, "StubRoutines", name);
2633     address start = __ pc();
2634 
2635     // bump this on entry, not on exit:
2636     inc_counter_np(SharedRuntime::_generic_array_copy_ctr, G1, G3);
2637 
2638     // In principle, the int arguments could be dirty.
2639     //assert_clean_int(src_pos, G1);
2640     //assert_clean_int(dst_pos, G1);
2641     //assert_clean_int(length, G1);
2642 
2643     //-----------------------------------------------------------------------
2644     // Assembler stubs will be used for this call to arraycopy
2645     // if the following conditions are met:
2646     //
2647     // (1) src and dst must not be null.
2648     // (2) src_pos must not be negative.
2649     // (3) dst_pos must not be negative.
2650     // (4) length  must not be negative.
2651     // (5) src klass and dst klass should be the same and not NULL.
2652     // (6) src and dst should be arrays.
2653     // (7) src_pos + length must not exceed length of src.
2654     // (8) dst_pos + length must not exceed length of dst.
2655     BLOCK_COMMENT("arraycopy initial argument checks");
2656 
2657     //  if (src == NULL) return -1;
2658     __ br_null(src, false, Assembler::pn, L_failed);
2659 
2660     //  if (src_pos < 0) return -1;
2661     __ delayed()->tst(src_pos);
2662     __ br(Assembler::negative, false, Assembler::pn, L_failed);
2663     __ delayed()->nop();
2664 
2665     //  if (dst == NULL) return -1;
2666     __ br_null(dst, false, Assembler::pn, L_failed);
2667 
2668     //  if (dst_pos < 0) return -1;
2669     __ delayed()->tst(dst_pos);
2670     __ br(Assembler::negative, false, Assembler::pn, L_failed);
2671 
2672     //  if (length < 0) return -1;
2673     __ delayed()->tst(length);
2674     __ br(Assembler::negative, false, Assembler::pn, L_failed);
2675 
2676     BLOCK_COMMENT("arraycopy argument klass checks");
2677     //  get src->klass()
2678     if (UseCompressedOops) {
2679       __ delayed()->nop(); // ??? not good
2680       __ load_klass(src, G3_src_klass);
2681     } else {
2682       __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), G3_src_klass);
2683     }
2684 
2685 #ifdef ASSERT
2686     //  assert(src->klass() != NULL);
2687     BLOCK_COMMENT("assert klasses not null");
2688     { Label L_a, L_b;
2689       __ br_notnull(G3_src_klass, false, Assembler::pt, L_b); // it is broken if klass is NULL
2690       __ delayed()->nop();
2691       __ bind(L_a);
2692       __ stop("broken null klass");
2693       __ bind(L_b);
2694       __ load_klass(dst, G4_dst_klass);
2695       __ br_null(G4_dst_klass, false, Assembler::pn, L_a); // this would be broken also
2696       __ delayed()->mov(G0, G4_dst_klass);      // scribble the temp
2697       BLOCK_COMMENT("assert done");
2698     }
2699 #endif
2700 
2701     // Load layout helper
2702     //
2703     //  | array_tag | element_type | log2_element_size | header_size |
2704     // 32          28             24                  16             0
2705     //
2706     //   array_tag: typeArray = 0xF, objArray = 0xD, normal instance = 0x0
2707     //
2708 
2709     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2710                     Klass::layout_helper_offset_in_bytes();
2711 
2712     // Load 32-bits signed value. Use br() instruction with it to check icc.
2713     __ lduw(G3_src_klass, lh_offset, G5_lh);
2714 
2715     if (UseCompressedOops) {
2716       __ load_klass(dst, G4_dst_klass);
2717     }
2718     // Handle objArrays completely differently...
2719     // This is necessary because they have covariant element types.
2720     // The next type check (for src.klass == dst.klass), can get a
2721     // false negative in the case of object arrays.
2722     jint objArray_lh = LayoutHelper::for_array(T_OBJECT).as_int();
2723     __ set(objArray_lh, O5_temp);
2724     __ cmp(G5_lh,       O5_temp);
2725     __ br(Assembler::equal, false, Assembler::pt, L_objArray);
2726     if (UseCompressedOops) {
2727       __ delayed()->nop();
2728     } else {
2729       __ delayed()->ld_ptr(dst, oopDesc::klass_offset_in_bytes(), G4_dst_klass);
2730     }
2731 
2732     //  if (src->klass() != dst->klass()) return -1;
2733     __ cmp(G3_src_klass, G4_dst_klass);
2734     __ brx(Assembler::notEqual, false, Assembler::pn, L_failed);
2735     __ delayed()->nop();
2736 
2737     //  if (!src->has_length_field()) return -1;
2738     __ cmp(G5_lh, LayoutHelper::_neutral_value); // < 0
2739     __ br(Assembler::greaterEqual, false, Assembler::pn, L_failed);
2740 
2741     __ delayed();                               // match next insn to prev branch
2742 
2743     arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
2744                            O5_temp, G4_dst_klass, L_failed);
2745 
2746     // typeArrayKlass
2747     //
2748     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2749     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2750     //
2751 
2752     const Register G4_offset = G4_dst_klass;    // array offset
2753     const Register G3_elsize = G3_src_klass;    // log2 element size
2754 
2755     assert(LayoutHelper::_header_size_shift == 0, "");
2756     __ set(LayoutHelper::_header_size_mask, G4_offset);
2757     __ and3(G5_lh, G4_offset, G4_offset); // array_offset
2758     __ add(src, G4_offset, src);       // src array offset
2759     __ add(dst, G4_offset, dst);       // dst array offset
2760     __ srl( G5_lh,     LayoutHelper::_element_size_shift, G3_elsize);
2761     __ and3(G3_elsize, LayoutHelper::_element_size_mask,  G3_elsize); // log2 element size
2762 
2763     // next registers should be set before the jump to corresponding stub
2764     const Register from     = O0;  // source array address
2765     const Register to       = O1;  // destination array address
2766     const Register count    = O2;  // elements count
2767 
2768     // 'from', 'to', 'count' registers should be set in this order
2769     // since they are the same as 'src', 'src_pos', 'dst'.
2770 
2771     BLOCK_COMMENT("scale indexes to element size");
2772     __ sll_ptr(src_pos, G3_elsize, src_pos);
2773     __ sll_ptr(dst_pos, G3_elsize, dst_pos);
2774     __ add(src, src_pos, from);       // src_addr
2775     __ add(dst, dst_pos, to);         // dst_addr
2776 
2777     BLOCK_COMMENT("choose copy loop based on element size");
2778     __ cmp(G3_elsize, 0);
2779     __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jbyte_arraycopy);
2780     __ delayed()->signx(length, count); // length
2781 
2782     __ cmp(G3_elsize, LogBytesPerShort);
2783     __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jshort_arraycopy);
2784     __ delayed()->signx(length, count); // length
2785 
2786     __ cmp(G3_elsize, LogBytesPerInt);
2787     __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jint_arraycopy);
2788     __ delayed()->signx(length, count); // length
2789 #ifdef ASSERT
2790     { Label L;
2791       __ cmp(G3_elsize, LogBytesPerLong);
2792       __ br(Assembler::equal, false, Assembler::pt, L);
2793       __ delayed()->nop();
2794       __ stop("must be long copy, but elsize is wrong");
2795       __ bind(L);
2796     }
2797 #endif
2798     __ br(Assembler::always,false,Assembler::pt,StubRoutines::_jlong_arraycopy);
2799     __ delayed()->signx(length, count); // length
2800 
2801     Label L_different_objArray, L_plain_copy, L_checkcast_copy;
2802 
2803     // objArrayKlass
2804   __ BIND(L_objArray);
2805     // live at this point:  G3_src_klass, G4_dst_klass, src[_pos], dst[_pos], length
2806 
2807     //  test array classes for subtyping
2808     __ cmp(G3_src_klass, G4_dst_klass);         // usual case is exact equality
2809     __ brx(Assembler::notEqual, true, Assembler::pn, L_different_objArray);
2810     __ delayed()->lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted from below
2811 
2812     // Identically typed arrays can be copied without element-wise checks.
2813     arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
2814                            O5_temp, G5_lh, L_failed);
2815 
2816     __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset
2817     __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset
2818     __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos);
2819     __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos);
2820     __ add(src, src_pos, from);       // src_addr
2821     __ add(dst, dst_pos, to);         // dst_addr
2822   __ BIND(L_plain_copy);
2823     __ br(Assembler::always, false, Assembler::pt,StubRoutines::_oop_arraycopy);
2824     __ delayed()->signx(length, count); // length
2825 
2826   __ BIND(L_different_objArray);
2827     // live at this point:  G3_src_klass, G4_dst_klass
2828     {
2829       // Before looking at dst.length, make sure dst is also an objArray.
2830       // lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted to delay slot
2831       __ cmp(G5_lh,                    O5_temp);
2832       __ br(Assembler::notEqual, false, Assembler::pn, L_failed);
2833 
2834       // It is safe to examine both src.length and dst.length.
2835       __ delayed();                             // match next insn to prev branch
2836       arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
2837                              O5_temp, G5_lh, L_failed);
2838 
2839       // Marshal the base address arguments now, freeing registers.
2840       __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset
2841       __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset
2842       __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos);
2843       __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos);
2844       __ add(src, src_pos, from);               // src_addr
2845       __ add(dst, dst_pos, to);                 // dst_addr
2846       __ signx(length, count);                  // length (reloaded)
2847 
2848       Register sco_temp = O3;                   // this register is free now
2849       assert_different_registers(from, to, count, sco_temp,
2850                                  G4_dst_klass, G3_src_klass);
2851 
2852       // Generate the type check.
2853       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2854                         Klass::super_check_offset_offset_in_bytes());
2855       __ lduw(G4_dst_klass, sco_offset, sco_temp);
2856       generate_type_check(G3_src_klass, sco_temp, G4_dst_klass,
2857                           O5_temp, L_plain_copy);
2858 
2859       // Fetch destination element klass from the objArrayKlass header.
2860       int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2861                        objArrayKlass::element_klass_offset_in_bytes());
2862 
2863       // the checkcast_copy loop needs two extra arguments:
2864       __ ld_ptr(G4_dst_klass, ek_offset, O4);   // dest elem klass
2865       // lduw(O4, sco_offset, O3);              // sco of elem klass
2866 
2867       __ br(Assembler::always, false, Assembler::pt, checkcast_copy_entry);
2868       __ delayed()->lduw(O4, sco_offset, O3);
2869     }
2870 
2871   __ BIND(L_failed);
2872     __ retl();
2873     __ delayed()->sub(G0, 1, O0); // return -1
2874     return start;
2875   }
2876 
2877   void generate_arraycopy_stubs() {
2878 
2879     // Note:  the disjoint stubs must be generated first, some of
2880     //        the conjoint stubs use them.
2881     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2882     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2883     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_copy(false, "jint_disjoint_arraycopy");
2884     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_copy(false, "jlong_disjoint_arraycopy");
2885     StubRoutines::_oop_disjoint_arraycopy    = generate_disjoint_oop_copy(false, "oop_disjoint_arraycopy");
2886     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(true, "arrayof_jbyte_disjoint_arraycopy");
2887     StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy");
2888     StubRoutines::_arrayof_jint_disjoint_arraycopy   = generate_disjoint_int_copy(true, "arrayof_jint_disjoint_arraycopy");
2889     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = generate_disjoint_long_copy(true, "arrayof_jlong_disjoint_arraycopy");
2890     StubRoutines::_arrayof_oop_disjoint_arraycopy    =  generate_disjoint_oop_copy(true, "arrayof_oop_disjoint_arraycopy");
2891 
2892     StubRoutines::_jbyte_arraycopy  = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2893     StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
2894     StubRoutines::_jint_arraycopy   = generate_conjoint_int_copy(false, "jint_arraycopy");
2895     StubRoutines::_jlong_arraycopy  = generate_conjoint_long_copy(false, "jlong_arraycopy");
2896     StubRoutines::_oop_arraycopy    = generate_conjoint_oop_copy(false, "oop_arraycopy");
2897     StubRoutines::_arrayof_jbyte_arraycopy    = generate_conjoint_byte_copy(true, "arrayof_jbyte_arraycopy");
2898     StubRoutines::_arrayof_jshort_arraycopy   = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy");
2899 #ifdef _LP64
2900     // since sizeof(jint) < sizeof(HeapWord), there's a different flavor:
2901     StubRoutines::_arrayof_jint_arraycopy     = generate_conjoint_int_copy(true, "arrayof_jint_arraycopy");
2902   #else
2903     StubRoutines::_arrayof_jint_arraycopy     = StubRoutines::_jint_arraycopy;
2904 #endif
2905     StubRoutines::_arrayof_jlong_arraycopy    = StubRoutines::_jlong_arraycopy;
2906     StubRoutines::_arrayof_oop_arraycopy      = StubRoutines::_oop_arraycopy;
2907 
2908     StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2909     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy");
2910     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy");
2911   }
2912 
2913   void generate_initial() {
2914     // Generates all stubs and initializes the entry points
2915 
2916     //------------------------------------------------------------------------------------------------------------------------
2917     // entry points that exist in all platforms
2918     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2919     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2920     StubRoutines::_forward_exception_entry                 = generate_forward_exception();
2921 
2922     StubRoutines::_call_stub_entry                         = generate_call_stub(StubRoutines::_call_stub_return_address);
2923     StubRoutines::_catch_exception_entry                   = generate_catch_exception();
2924 
2925     //------------------------------------------------------------------------------------------------------------------------
2926     // entry points that are platform specific
2927     StubRoutines::Sparc::_test_stop_entry                  = generate_test_stop();
2928 
2929     StubRoutines::Sparc::_stop_subroutine_entry            = generate_stop_subroutine();
2930     StubRoutines::Sparc::_flush_callers_register_windows_entry = generate_flush_callers_register_windows();
2931 
2932 #if !defined(COMPILER2) && !defined(_LP64)
2933     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
2934     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
2935     StubRoutines::_atomic_add_entry          = generate_atomic_add();
2936     StubRoutines::_atomic_xchg_ptr_entry     = StubRoutines::_atomic_xchg_entry;
2937     StubRoutines::_atomic_cmpxchg_ptr_entry  = StubRoutines::_atomic_cmpxchg_entry;
2938     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2939     StubRoutines::_atomic_add_ptr_entry      = StubRoutines::_atomic_add_entry;
2940     StubRoutines::_fence_entry               = generate_fence();
2941 #endif  // COMPILER2 !=> _LP64
2942 
2943     StubRoutines::Sparc::_partial_subtype_check                = generate_partial_subtype_check();
2944   }
2945 
2946 
2947   void generate_all() {
2948     // Generates all stubs and initializes the entry points
2949 
2950     // These entry points require SharedInfo::stack0 to be set up in non-core builds
2951     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError),  false);
2952     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError),  false);
2953     StubRoutines::_throw_ArithmeticException_entry         = generate_throw_exception("ArithmeticException throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException),  true);
2954     StubRoutines::_throw_NullPointerException_entry        = generate_throw_exception("NullPointerException throw_exception",         CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2955     StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
2956     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",           CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError),   false);
2957 
2958     StubRoutines::_handler_for_unsafe_access_entry =
2959       generate_handler_for_unsafe_access();
2960 
2961     // support for verify_oop (must happen after universe_init)
2962     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop_subroutine();
2963 
2964     // arraycopy stubs used by compilers
2965     generate_arraycopy_stubs();
2966   }
2967 
2968 
2969  public:
2970   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2971     // replace the standard masm with a special one:
2972     _masm = new MacroAssembler(code);
2973 
2974     _stub_count = !all ? 0x100 : 0x200;
2975     if (all) {
2976       generate_all();
2977     } else {
2978       generate_initial();
2979     }
2980 
2981     // make sure this stub is available for all local calls
2982     if (_atomic_add_stub.is_unbound()) {
2983       // generate a second time, if necessary
2984       (void) generate_atomic_add();
2985     }
2986   }
2987 
2988 
2989  private:
2990   int _stub_count;
2991   void stub_prolog(StubCodeDesc* cdesc) {
2992     # ifdef ASSERT
2993       // put extra information in the stub code, to make it more readable
2994 #ifdef _LP64
2995 // Write the high part of the address
2996 // [RGV] Check if there is a dependency on the size of this prolog
2997       __ emit_data((intptr_t)cdesc >> 32,    relocInfo::none);
2998 #endif
2999       __ emit_data((intptr_t)cdesc,    relocInfo::none);
3000       __ emit_data(++_stub_count, relocInfo::none);
3001     # endif
3002     align(true);
3003   }
3004 
3005   void align(bool at_header = false) {
3006     // %%%%% move this constant somewhere else
3007     // UltraSPARC cache line size is 8 instructions:
3008     const unsigned int icache_line_size = 32;
3009     const unsigned int icache_half_line_size = 16;
3010 
3011     if (at_header) {
3012       while ((intptr_t)(__ pc()) % icache_line_size != 0) {
3013         __ emit_data(0, relocInfo::none);
3014       }
3015     } else {
3016       while ((intptr_t)(__ pc()) % icache_half_line_size != 0) {
3017         __ nop();
3018       }
3019     }
3020   }
3021 
3022 }; // end class declaration
3023 
3024 
3025 address StubGenerator::disjoint_byte_copy_entry  = NULL;
3026 address StubGenerator::disjoint_short_copy_entry = NULL;
3027 address StubGenerator::disjoint_int_copy_entry   = NULL;
3028 address StubGenerator::disjoint_long_copy_entry  = NULL;
3029 address StubGenerator::disjoint_oop_copy_entry   = NULL;
3030 
3031 address StubGenerator::byte_copy_entry  = NULL;
3032 address StubGenerator::short_copy_entry = NULL;
3033 address StubGenerator::int_copy_entry   = NULL;
3034 address StubGenerator::long_copy_entry  = NULL;
3035 address StubGenerator::oop_copy_entry   = NULL;
3036 
3037 address StubGenerator::checkcast_copy_entry = NULL;
3038 
3039 void StubGenerator_generate(CodeBuffer* code, bool all) {
3040   StubGenerator g(code, all);
3041 }