1 /*
   2  * Copyright 2003-2007 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_x86_64.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 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
  34 
  35 #ifdef PRODUCT
  36 #define BLOCK_COMMENT(str) /* nothing */
  37 #else
  38 #define BLOCK_COMMENT(str) __ block_comment(str)
  39 #endif
  40 
  41 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  42 const int MXCSR_MASK = 0xFFC0;  // Mask out any pending exceptions
  43 
  44 // Stub Code definitions
  45 
  46 static address handle_unsafe_access() {
  47   JavaThread* thread = JavaThread::current();
  48   address pc = thread->saved_exception_pc();
  49   // pc is the instruction which we must emulate
  50   // doing a no-op is fine:  return garbage from the load
  51   // therefore, compute npc
  52   address npc = Assembler::locate_next_instruction(pc);
  53 
  54   // request an async exception
  55   thread->set_pending_unsafe_access_error();
  56 
  57   // return address of next instruction to execute
  58   return npc;
  59 }
  60 
  61 class StubGenerator: public StubCodeGenerator {
  62  private:
  63 
  64 #ifdef PRODUCT
  65 #define inc_counter_np(counter) (0)
  66 #else
  67   void inc_counter_np_(int& counter) {
  68     __ incrementl(ExternalAddress((address)&counter));
  69   }
  70 #define inc_counter_np(counter) \
  71   BLOCK_COMMENT("inc_counter " #counter); \
  72   inc_counter_np_(counter);
  73 #endif
  74 
  75   // Call stubs are used to call Java from C
  76   //
  77   // Linux Arguments:
  78   //    c_rarg0:   call wrapper address                   address
  79   //    c_rarg1:   result                                 address
  80   //    c_rarg2:   result type                            BasicType
  81   //    c_rarg3:   method                                 methodOop
  82   //    c_rarg4:   (interpreter) entry point              address
  83   //    c_rarg5:   parameters                             intptr_t*
  84   //    16(rbp): parameter size (in words)              int
  85   //    24(rbp): thread                                 Thread*
  86   //
  87   //     [ return_from_Java     ] <--- rsp
  88   //     [ argument word n      ]
  89   //      ...
  90   // -12 [ argument word 1      ]
  91   // -11 [ saved r15            ] <--- rsp_after_call
  92   // -10 [ saved r14            ]
  93   //  -9 [ saved r13            ]
  94   //  -8 [ saved r12            ]
  95   //  -7 [ saved rbx            ]
  96   //  -6 [ call wrapper         ]
  97   //  -5 [ result               ]
  98   //  -4 [ result type          ]
  99   //  -3 [ method               ]
 100   //  -2 [ entry point          ]
 101   //  -1 [ parameters           ]
 102   //   0 [ saved rbp            ] <--- rbp
 103   //   1 [ return address       ]
 104   //   2 [ parameter size       ]
 105   //   3 [ thread               ]
 106   //
 107   // Windows Arguments:
 108   //    c_rarg0:   call wrapper address                   address
 109   //    c_rarg1:   result                                 address
 110   //    c_rarg2:   result type                            BasicType
 111   //    c_rarg3:   method                                 methodOop
 112   //    48(rbp): (interpreter) entry point              address
 113   //    56(rbp): parameters                             intptr_t*
 114   //    64(rbp): parameter size (in words)              int
 115   //    72(rbp): thread                                 Thread*
 116   //
 117   //     [ return_from_Java     ] <--- rsp
 118   //     [ argument word n      ]
 119   //      ...
 120   //  -8 [ argument word 1      ]
 121   //  -7 [ saved r15            ] <--- rsp_after_call
 122   //  -6 [ saved r14            ]
 123   //  -5 [ saved r13            ]
 124   //  -4 [ saved r12            ]
 125   //  -3 [ saved rdi            ]
 126   //  -2 [ saved rsi            ]
 127   //  -1 [ saved rbx            ]
 128   //   0 [ saved rbp            ] <--- rbp
 129   //   1 [ return address       ]
 130   //   2 [ call wrapper         ]
 131   //   3 [ result               ]
 132   //   4 [ result type          ]
 133   //   5 [ method               ]
 134   //   6 [ entry point          ]
 135   //   7 [ parameters           ]
 136   //   8 [ parameter size       ]
 137   //   9 [ thread               ]
 138   //
 139   //    Windows reserves the callers stack space for arguments 1-4.
 140   //    We spill c_rarg0-c_rarg3 to this space.
 141 
 142   // Call stub stack layout word offsets from rbp
 143   enum call_stub_layout {
 144 #ifdef _WIN64
 145     rsp_after_call_off = -7,
 146     r15_off            = rsp_after_call_off,
 147     r14_off            = -6,
 148     r13_off            = -5,
 149     r12_off            = -4,
 150     rdi_off            = -3,
 151     rsi_off            = -2,
 152     rbx_off            = -1,
 153     rbp_off            =  0,
 154     retaddr_off        =  1,
 155     call_wrapper_off   =  2,
 156     result_off         =  3,
 157     result_type_off    =  4,
 158     method_off         =  5,
 159     entry_point_off    =  6,
 160     parameters_off     =  7,
 161     parameter_size_off =  8,
 162     thread_off         =  9
 163 #else
 164     rsp_after_call_off = -12,
 165     mxcsr_off          = rsp_after_call_off,
 166     r15_off            = -11,
 167     r14_off            = -10,
 168     r13_off            = -9,
 169     r12_off            = -8,
 170     rbx_off            = -7,
 171     call_wrapper_off   = -6,
 172     result_off         = -5,
 173     result_type_off    = -4,
 174     method_off         = -3,
 175     entry_point_off    = -2,
 176     parameters_off     = -1,
 177     rbp_off            =  0,
 178     retaddr_off        =  1,
 179     parameter_size_off =  2,
 180     thread_off         =  3
 181 #endif
 182   };
 183 
 184   address generate_call_stub(address& return_address) {
 185     assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 186            (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 187            "adjust this code");
 188     StubCodeMark mark(this, "StubRoutines", "call_stub");
 189     address start = __ pc();
 190 
 191     // same as in generate_catch_exception()!
 192     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 193 
 194     const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 195     const Address result        (rbp, result_off         * wordSize);
 196     const Address result_type   (rbp, result_type_off    * wordSize);
 197     const Address method        (rbp, method_off         * wordSize);
 198     const Address entry_point   (rbp, entry_point_off    * wordSize);
 199     const Address parameters    (rbp, parameters_off     * wordSize);
 200     const Address parameter_size(rbp, parameter_size_off * wordSize);
 201 
 202     // same as in generate_catch_exception()!
 203     const Address thread        (rbp, thread_off         * wordSize);
 204 
 205     const Address r15_save(rbp, r15_off * wordSize);
 206     const Address r14_save(rbp, r14_off * wordSize);
 207     const Address r13_save(rbp, r13_off * wordSize);
 208     const Address r12_save(rbp, r12_off * wordSize);
 209     const Address rbx_save(rbp, rbx_off * wordSize);
 210 
 211     // stub code
 212     __ enter();
 213     __ subq(rsp, -rsp_after_call_off * wordSize);
 214 
 215     // save register parameters
 216 #ifndef _WIN64
 217     __ movq(parameters,   c_rarg5); // parameters
 218     __ movq(entry_point,  c_rarg4); // entry_point
 219 #endif
 220 
 221     __ movq(method,       c_rarg3); // method
 222     __ movl(result_type,  c_rarg2); // result type
 223     __ movq(result,       c_rarg1); // result
 224     __ movq(call_wrapper, c_rarg0); // call wrapper
 225 
 226     // save regs belonging to calling function
 227     __ movq(rbx_save, rbx);
 228     __ movq(r12_save, r12);
 229     __ movq(r13_save, r13);
 230     __ movq(r14_save, r14);
 231     __ movq(r15_save, r15);
 232 
 233 #ifdef _WIN64
 234     const Address rdi_save(rbp, rdi_off * wordSize);
 235     const Address rsi_save(rbp, rsi_off * wordSize);
 236 
 237     __ movq(rsi_save, rsi);
 238     __ movq(rdi_save, rdi);
 239 #else
 240     const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 241     {
 242       Label skip_ldmx;
 243       __ stmxcsr(mxcsr_save);
 244       __ movl(rax, mxcsr_save);
 245       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 246       ExternalAddress mxcsr_std(StubRoutines::amd64::mxcsr_std());
 247       __ cmp32(rax, mxcsr_std);
 248       __ jcc(Assembler::equal, skip_ldmx);
 249       __ ldmxcsr(mxcsr_std);
 250       __ bind(skip_ldmx);
 251     }
 252 #endif
 253 
 254     // Load up thread register
 255     __ movq(r15_thread, thread);
 256     __ reinit_heapbase();
 257 
 258 #ifdef ASSERT
 259     // make sure we have no pending exceptions
 260     {
 261       Label L;
 262       __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD);
 263       __ jcc(Assembler::equal, L);
 264       __ stop("StubRoutines::call_stub: entered with pending exception");
 265       __ bind(L);
 266     }
 267 #endif
 268 
 269     // pass parameters if any
 270     BLOCK_COMMENT("pass parameters if any");
 271     Label parameters_done;
 272     __ movl(c_rarg3, parameter_size);
 273     __ testl(c_rarg3, c_rarg3);
 274     __ jcc(Assembler::zero, parameters_done);
 275 
 276     Label loop;
 277     __ movq(c_rarg2, parameters);     // parameter pointer
 278     __ movl(c_rarg1, c_rarg3);        // parameter counter is in c_rarg1
 279     __ BIND(loop);
 280     if (TaggedStackInterpreter) {
 281       __ movq(rax, Address(c_rarg2, 0)); // get tag
 282       __ addq(c_rarg2, wordSize);     // advance to next tag
 283       __ pushq(rax);                  // pass tag
 284     }
 285     __ movq(rax, Address(c_rarg2, 0));  // get parameter
 286     __ addq(c_rarg2, wordSize);       // advance to next parameter
 287     __ decrementl(c_rarg1);           // decrement counter
 288     __ pushq(rax);                    // pass parameter
 289     __ jcc(Assembler::notZero, loop);
 290 
 291     // call Java function
 292     __ BIND(parameters_done);
 293     __ movq(rbx, method);             // get methodOop
 294     __ movq(c_rarg1, entry_point);    // get entry_point
 295     __ movq(r13, rsp);                // set sender sp
 296     BLOCK_COMMENT("call Java function");
 297     __ call(c_rarg1);
 298 
 299     BLOCK_COMMENT("call_stub_return_address:");
 300     return_address = __ pc();
 301 
 302     // store result depending on type (everything that is not
 303     // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 304     __ movq(c_rarg0, result);
 305     Label is_long, is_float, is_double, exit;
 306     __ movl(c_rarg1, result_type);
 307     __ cmpl(c_rarg1, T_OBJECT);
 308     __ jcc(Assembler::equal, is_long);
 309     __ cmpl(c_rarg1, T_LONG);
 310     __ jcc(Assembler::equal, is_long);
 311     __ cmpl(c_rarg1, T_FLOAT);
 312     __ jcc(Assembler::equal, is_float);
 313     __ cmpl(c_rarg1, T_DOUBLE);
 314     __ jcc(Assembler::equal, is_double);
 315 
 316     // handle T_INT case
 317     __ movl(Address(c_rarg0, 0), rax);
 318 
 319     __ BIND(exit);
 320 
 321     // pop parameters
 322     __ leaq(rsp, rsp_after_call);
 323 
 324 #ifdef ASSERT
 325     // verify that threads correspond
 326     {
 327       Label L, S;
 328       __ cmpq(r15_thread, thread);
 329       __ jcc(Assembler::notEqual, S);
 330       __ get_thread(rbx);
 331       __ cmpq(r15_thread, rbx);
 332       __ jcc(Assembler::equal, L);
 333       __ bind(S);
 334       __ jcc(Assembler::equal, L);
 335       __ stop("StubRoutines::call_stub: threads must correspond");
 336       __ bind(L);
 337     }
 338 #endif
 339 
 340     // restore regs belonging to calling function
 341     __ movq(r15, r15_save);
 342     __ movq(r14, r14_save);
 343     __ movq(r13, r13_save);
 344     __ movq(r12, r12_save);
 345     __ movq(rbx, rbx_save);
 346 
 347 #ifdef _WIN64
 348     __ movq(rdi, rdi_save);
 349     __ movq(rsi, rsi_save);
 350 #else
 351     __ ldmxcsr(mxcsr_save);
 352 #endif
 353 
 354     // restore rsp
 355     __ addq(rsp, -rsp_after_call_off * wordSize);
 356 
 357     // return
 358     __ popq(rbp);
 359     __ ret(0);
 360 
 361     // handle return types different from T_INT
 362     __ BIND(is_long);
 363     __ movq(Address(c_rarg0, 0), rax);
 364     __ jmp(exit);
 365 
 366     __ BIND(is_float);
 367     __ movflt(Address(c_rarg0, 0), xmm0);
 368     __ jmp(exit);
 369 
 370     __ BIND(is_double);
 371     __ movdbl(Address(c_rarg0, 0), xmm0);
 372     __ jmp(exit);
 373 
 374     return start;
 375   }
 376 
 377   // Return point for a Java call if there's an exception thrown in
 378   // Java code.  The exception is caught and transformed into a
 379   // pending exception stored in JavaThread that can be tested from
 380   // within the VM.
 381   //
 382   // Note: Usually the parameters are removed by the callee. In case
 383   // of an exception crossing an activation frame boundary, that is
 384   // not the case if the callee is compiled code => need to setup the
 385   // rsp.
 386   //
 387   // rax: exception oop
 388 
 389   address generate_catch_exception() {
 390     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 391     address start = __ pc();
 392 
 393     // same as in generate_call_stub():
 394     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 395     const Address thread        (rbp, thread_off         * wordSize);
 396 
 397 #ifdef ASSERT
 398     // verify that threads correspond
 399     {
 400       Label L, S;
 401       __ cmpq(r15_thread, thread);
 402       __ jcc(Assembler::notEqual, S);
 403       __ get_thread(rbx);
 404       __ cmpq(r15_thread, rbx);
 405       __ jcc(Assembler::equal, L);
 406       __ bind(S);
 407       __ stop("StubRoutines::catch_exception: threads must correspond");
 408       __ bind(L);
 409     }
 410 #endif
 411 
 412     // set pending exception
 413     __ verify_oop(rax);
 414 
 415     __ movq(Address(r15_thread, Thread::pending_exception_offset()), rax);
 416     __ lea(rscratch1, ExternalAddress((address)__FILE__));
 417     __ movq(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 418     __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 419 
 420     // complete return to VM
 421     assert(StubRoutines::_call_stub_return_address != NULL,
 422            "_call_stub_return_address must have been generated before");
 423     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 424 
 425     return start;
 426   }
 427 
 428   // Continuation point for runtime calls returning with a pending
 429   // exception.  The pending exception check happened in the runtime
 430   // or native call stub.  The pending exception in Thread is
 431   // converted into a Java-level exception.
 432   //
 433   // Contract with Java-level exception handlers:
 434   // rax: exception
 435   // rdx: throwing pc
 436   //
 437   // NOTE: At entry of this stub, exception-pc must be on stack !!
 438 
 439   address generate_forward_exception() {
 440     StubCodeMark mark(this, "StubRoutines", "forward exception");
 441     address start = __ pc();
 442 
 443     // Upon entry, the sp points to the return address returning into
 444     // Java (interpreted or compiled) code; i.e., the return address
 445     // becomes the throwing pc.
 446     //
 447     // Arguments pushed before the runtime call are still on the stack
 448     // but the exception handler will reset the stack pointer ->
 449     // ignore them.  A potential result in registers can be ignored as
 450     // well.
 451 
 452 #ifdef ASSERT
 453     // make sure this code is only executed if there is a pending exception
 454     {
 455       Label L;
 456       __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int) NULL);
 457       __ jcc(Assembler::notEqual, L);
 458       __ stop("StubRoutines::forward exception: no pending exception (1)");
 459       __ bind(L);
 460     }
 461 #endif
 462 
 463     // compute exception handler into rbx
 464     __ movq(c_rarg0, Address(rsp, 0));
 465     BLOCK_COMMENT("call exception_handler_for_return_address");
 466     __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 467                          SharedRuntime::exception_handler_for_return_address),
 468                     c_rarg0);
 469     __ movq(rbx, rax);
 470 
 471     // setup rax & rdx, remove return address & clear pending exception
 472     __ popq(rdx);
 473     __ movq(rax, Address(r15_thread, Thread::pending_exception_offset()));
 474     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD);
 475 
 476 #ifdef ASSERT
 477     // make sure exception is set
 478     {
 479       Label L;
 480       __ testq(rax, rax);
 481       __ jcc(Assembler::notEqual, L);
 482       __ stop("StubRoutines::forward exception: no pending exception (2)");
 483       __ bind(L);
 484     }
 485 #endif
 486 
 487     // continue at exception handler (return address removed)
 488     // rax: exception
 489     // rbx: exception handler
 490     // rdx: throwing pc
 491     __ verify_oop(rax);
 492     __ jmp(rbx);
 493 
 494     return start;
 495   }
 496 
 497   // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
 498   //
 499   // Arguments :
 500   //    c_rarg0: exchange_value
 501   //    c_rarg0: dest
 502   //
 503   // Result:
 504   //    *dest <- ex, return (orig *dest)
 505   address generate_atomic_xchg() {
 506     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 507     address start = __ pc();
 508 
 509     __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
 510     __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
 511     __ ret(0);
 512 
 513     return start;
 514   }
 515 
 516   // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
 517   //
 518   // Arguments :
 519   //    c_rarg0: exchange_value
 520   //    c_rarg1: dest
 521   //
 522   // Result:
 523   //    *dest <- ex, return (orig *dest)
 524   address generate_atomic_xchg_ptr() {
 525     StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
 526     address start = __ pc();
 527 
 528     __ movq(rax, c_rarg0); // Copy to eax we need a return value anyhow
 529     __ xchgq(rax, Address(c_rarg1, 0)); // automatic LOCK
 530     __ ret(0);
 531 
 532     return start;
 533   }
 534 
 535   // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
 536   //                                         jint compare_value)
 537   //
 538   // Arguments :
 539   //    c_rarg0: exchange_value
 540   //    c_rarg1: dest
 541   //    c_rarg2: compare_value
 542   //
 543   // Result:
 544   //    if ( compare_value == *dest ) {
 545   //       *dest = exchange_value
 546   //       return compare_value;
 547   //    else
 548   //       return *dest;
 549   address generate_atomic_cmpxchg() {
 550     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
 551     address start = __ pc();
 552 
 553     __ movl(rax, c_rarg2);
 554    if ( os::is_MP() ) __ lock();
 555     __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
 556     __ ret(0);
 557 
 558     return start;
 559   }
 560 
 561   // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
 562   //                                             volatile jlong* dest,
 563   //                                             jlong compare_value)
 564   // Arguments :
 565   //    c_rarg0: exchange_value
 566   //    c_rarg1: dest
 567   //    c_rarg2: compare_value
 568   //
 569   // Result:
 570   //    if ( compare_value == *dest ) {
 571   //       *dest = exchange_value
 572   //       return compare_value;
 573   //    else
 574   //       return *dest;
 575   address generate_atomic_cmpxchg_long() {
 576     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
 577     address start = __ pc();
 578 
 579     __ movq(rax, c_rarg2);
 580    if ( os::is_MP() ) __ lock();
 581     __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
 582     __ ret(0);
 583 
 584     return start;
 585   }
 586 
 587   // Support for jint atomic::add(jint add_value, volatile jint* dest)
 588   //
 589   // Arguments :
 590   //    c_rarg0: add_value
 591   //    c_rarg1: dest
 592   //
 593   // Result:
 594   //    *dest += add_value
 595   //    return *dest;
 596   address generate_atomic_add() {
 597     StubCodeMark mark(this, "StubRoutines", "atomic_add");
 598     address start = __ pc();
 599 
 600     __ movl(rax, c_rarg0);
 601    if ( os::is_MP() ) __ lock();
 602     __ xaddl(Address(c_rarg1, 0), c_rarg0);
 603     __ addl(rax, c_rarg0);
 604     __ ret(0);
 605 
 606     return start;
 607   }
 608 
 609   // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
 610   //
 611   // Arguments :
 612   //    c_rarg0: add_value
 613   //    c_rarg1: dest
 614   //
 615   // Result:
 616   //    *dest += add_value
 617   //    return *dest;
 618   address generate_atomic_add_ptr() {
 619     StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
 620     address start = __ pc();
 621 
 622     __ movq(rax, c_rarg0); // Copy to eax we need a return value anyhow
 623    if ( os::is_MP() ) __ lock();
 624     __ xaddl(Address(c_rarg1, 0), c_rarg0);
 625     __ addl(rax, c_rarg0);
 626     __ ret(0);
 627 
 628     return start;
 629   }
 630 
 631   // Support for intptr_t OrderAccess::fence()
 632   //
 633   // Arguments :
 634   //
 635   // Result:
 636   address generate_orderaccess_fence() {
 637     StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 638     address start = __ pc();
 639     __ mfence();
 640     __ ret(0);
 641 
 642     return start;
 643   }
 644 
 645   // Support for intptr_t get_previous_fp()
 646   //
 647   // This routine is used to find the previous frame pointer for the
 648   // caller (current_frame_guess). This is used as part of debugging
 649   // ps() is seemingly lost trying to find frames.
 650   // This code assumes that caller current_frame_guess) has a frame.
 651   address generate_get_previous_fp() {
 652     StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
 653     const Address old_fp(rbp, 0);
 654     const Address older_fp(rax, 0);
 655     address start = __ pc();
 656 
 657     __ enter();
 658     __ movq(rax, old_fp); // callers fp
 659     __ movq(rax, older_fp); // the frame for ps()
 660     __ popq(rbp);
 661     __ ret(0);
 662 
 663     return start;
 664   }
 665 
 666   //----------------------------------------------------------------------------------------------------
 667   // Support for void verify_mxcsr()
 668   //
 669   // This routine is used with -Xcheck:jni to verify that native
 670   // JNI code does not return to Java code without restoring the
 671   // MXCSR register to our expected state.
 672 
 673   address generate_verify_mxcsr() {
 674     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 675     address start = __ pc();
 676 
 677     const Address mxcsr_save(rsp, 0);
 678 
 679     if (CheckJNICalls) {
 680       Label ok_ret;
 681       __ pushq(rax);
 682       __ subq(rsp, wordSize);      // allocate a temp location
 683       __ stmxcsr(mxcsr_save);
 684       __ movl(rax, mxcsr_save);
 685       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 686       __ cmpl(rax, *(int *)(StubRoutines::amd64::mxcsr_std()));
 687       __ jcc(Assembler::equal, ok_ret);
 688 
 689       __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 690 
 691       __ ldmxcsr(ExternalAddress(StubRoutines::amd64::mxcsr_std()));
 692 
 693       __ bind(ok_ret);
 694       __ addq(rsp, wordSize);
 695       __ popq(rax);
 696     }
 697 
 698     __ ret(0);
 699 
 700     return start;
 701   }
 702 
 703   address generate_f2i_fixup() {
 704     StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 705     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 706 
 707     address start = __ pc();
 708 
 709     Label L;
 710 
 711     __ pushq(rax);
 712     __ pushq(c_rarg3);
 713     __ pushq(c_rarg2);
 714     __ pushq(c_rarg1);
 715 
 716     __ movl(rax, 0x7f800000);
 717     __ xorl(c_rarg3, c_rarg3);
 718     __ movl(c_rarg2, inout);
 719     __ movl(c_rarg1, c_rarg2);
 720     __ andl(c_rarg1, 0x7fffffff);
 721     __ cmpl(rax, c_rarg1); // NaN? -> 0
 722     __ jcc(Assembler::negative, L);
 723     __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 724     __ movl(c_rarg3, 0x80000000);
 725     __ movl(rax, 0x7fffffff);
 726     __ cmovl(Assembler::positive, c_rarg3, rax);
 727 
 728     __ bind(L);
 729     __ movq(inout, c_rarg3);
 730 
 731     __ popq(c_rarg1);
 732     __ popq(c_rarg2);
 733     __ popq(c_rarg3);
 734     __ popq(rax);
 735 
 736     __ ret(0);
 737 
 738     return start;
 739   }
 740 
 741   address generate_f2l_fixup() {
 742     StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 743     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 744     address start = __ pc();
 745 
 746     Label L;
 747 
 748     __ pushq(rax);
 749     __ pushq(c_rarg3);
 750     __ pushq(c_rarg2);
 751     __ pushq(c_rarg1);
 752 
 753     __ movl(rax, 0x7f800000);
 754     __ xorl(c_rarg3, c_rarg3);
 755     __ movl(c_rarg2, inout);
 756     __ movl(c_rarg1, c_rarg2);
 757     __ andl(c_rarg1, 0x7fffffff);
 758     __ cmpl(rax, c_rarg1); // NaN? -> 0
 759     __ jcc(Assembler::negative, L);
 760     __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 761     __ mov64(c_rarg3, 0x8000000000000000);
 762     __ mov64(rax, 0x7fffffffffffffff);
 763     __ cmovq(Assembler::positive, c_rarg3, rax);
 764 
 765     __ bind(L);
 766     __ movq(inout, c_rarg3);
 767 
 768     __ popq(c_rarg1);
 769     __ popq(c_rarg2);
 770     __ popq(c_rarg3);
 771     __ popq(rax);
 772 
 773     __ ret(0);
 774 
 775     return start;
 776   }
 777 
 778   address generate_d2i_fixup() {
 779     StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 780     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 781 
 782     address start = __ pc();
 783 
 784     Label L;
 785 
 786     __ pushq(rax);
 787     __ pushq(c_rarg3);
 788     __ pushq(c_rarg2);
 789     __ pushq(c_rarg1);
 790     __ pushq(c_rarg0);
 791 
 792     __ movl(rax, 0x7ff00000);
 793     __ movq(c_rarg2, inout);
 794     __ movl(c_rarg3, c_rarg2);
 795     __ movq(c_rarg1, c_rarg2);
 796     __ movq(c_rarg0, c_rarg2);
 797     __ negl(c_rarg3);
 798     __ shrq(c_rarg1, 0x20);
 799     __ orl(c_rarg3, c_rarg2);
 800     __ andl(c_rarg1, 0x7fffffff);
 801     __ xorl(c_rarg2, c_rarg2);
 802     __ shrl(c_rarg3, 0x1f);
 803     __ orl(c_rarg1, c_rarg3);
 804     __ cmpl(rax, c_rarg1);
 805     __ jcc(Assembler::negative, L); // NaN -> 0
 806     __ testq(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 807     __ movl(c_rarg2, 0x80000000);
 808     __ movl(rax, 0x7fffffff);
 809     __ cmovl(Assembler::positive, c_rarg2, rax);
 810 
 811     __ bind(L);
 812     __ movq(inout, c_rarg2);
 813 
 814     __ popq(c_rarg0);
 815     __ popq(c_rarg1);
 816     __ popq(c_rarg2);
 817     __ popq(c_rarg3);
 818     __ popq(rax);
 819 
 820     __ ret(0);
 821 
 822     return start;
 823   }
 824 
 825   address generate_d2l_fixup() {
 826     StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 827     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 828 
 829     address start = __ pc();
 830 
 831     Label L;
 832 
 833     __ pushq(rax);
 834     __ pushq(c_rarg3);
 835     __ pushq(c_rarg2);
 836     __ pushq(c_rarg1);
 837     __ pushq(c_rarg0);
 838 
 839     __ movl(rax, 0x7ff00000);
 840     __ movq(c_rarg2, inout);
 841     __ movl(c_rarg3, c_rarg2);
 842     __ movq(c_rarg1, c_rarg2);
 843     __ movq(c_rarg0, c_rarg2);
 844     __ negl(c_rarg3);
 845     __ shrq(c_rarg1, 0x20);
 846     __ orl(c_rarg3, c_rarg2);
 847     __ andl(c_rarg1, 0x7fffffff);
 848     __ xorl(c_rarg2, c_rarg2);
 849     __ shrl(c_rarg3, 0x1f);
 850     __ orl(c_rarg1, c_rarg3);
 851     __ cmpl(rax, c_rarg1);
 852     __ jcc(Assembler::negative, L); // NaN -> 0
 853     __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 854     __ mov64(c_rarg2, 0x8000000000000000);
 855     __ mov64(rax, 0x7fffffffffffffff);
 856     __ cmovq(Assembler::positive, c_rarg2, rax);
 857 
 858     __ bind(L);
 859     __ movq(inout, c_rarg2);
 860 
 861     __ popq(c_rarg0);
 862     __ popq(c_rarg1);
 863     __ popq(c_rarg2);
 864     __ popq(c_rarg3);
 865     __ popq(rax);
 866 
 867     __ ret(0);
 868 
 869     return start;
 870   }
 871 
 872   address generate_fp_mask(const char *stub_name, int64_t mask) {
 873     StubCodeMark mark(this, "StubRoutines", stub_name);
 874 
 875     __ align(16);
 876     address start = __ pc();
 877 
 878     __ emit_data64( mask, relocInfo::none );
 879     __ emit_data64( mask, relocInfo::none );
 880 
 881     return start;
 882   }
 883 
 884   // The following routine generates a subroutine to throw an
 885   // asynchronous UnknownError when an unsafe access gets a fault that
 886   // could not be reasonably prevented by the programmer.  (Example:
 887   // SIGBUS/OBJERR.)
 888   address generate_handler_for_unsafe_access() {
 889     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 890     address start = __ pc();
 891 
 892     __ pushq(0);                      // hole for return address-to-be
 893     __ pushaq();                      // push registers
 894     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 895 
 896     __ subq(rsp, frame::arg_reg_save_area_bytes);
 897     BLOCK_COMMENT("call handle_unsafe_access");
 898     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 899     __ addq(rsp, frame::arg_reg_save_area_bytes);
 900 
 901     __ movq(next_pc, rax);            // stuff next address
 902     __ popaq();
 903     __ ret(0);                        // jump to next address
 904 
 905     return start;
 906   }
 907 
 908   // Non-destructive plausibility checks for oops
 909   //
 910   // Arguments:
 911   //    all args on stack!
 912   //
 913   // Stack after saving c_rarg3:
 914   //    [tos + 0]: saved c_rarg3
 915   //    [tos + 1]: saved c_rarg2
 916   //    [tos + 2]: saved r12 (several TemplateTable methods use it)
 917   //    [tos + 3]: saved flags
 918   //    [tos + 4]: return address
 919   //  * [tos + 5]: error message (char*)
 920   //  * [tos + 6]: object to verify (oop)
 921   //  * [tos + 7]: saved rax - saved by caller and bashed
 922   //  * = popped on exit
 923   address generate_verify_oop() {
 924     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 925     address start = __ pc();
 926 
 927     Label exit, error;
 928 
 929     __ pushfq();
 930     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 931 
 932     __ pushq(r12);
 933 
 934     // save c_rarg2 and c_rarg3
 935     __ pushq(c_rarg2);
 936     __ pushq(c_rarg3);
 937 
 938     enum {
 939            // After previous pushes.
 940            oop_to_verify = 6 * wordSize,
 941            saved_rax     = 7 * wordSize,
 942 
 943            // Before the call to MacroAssembler::debug(), see below.
 944            return_addr   = 16 * wordSize,
 945            error_msg     = 17 * wordSize
 946     };
 947 
 948     // get object
 949     __ movq(rax, Address(rsp, oop_to_verify));
 950 
 951     // make sure object is 'reasonable'
 952     __ testq(rax, rax);
 953     __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
 954     // Check if the oop is in the right area of memory
 955     __ movq(c_rarg2, rax);
 956     __ movptr(c_rarg3, (int64_t) Universe::verify_oop_mask());
 957     __ andq(c_rarg2, c_rarg3);
 958     __ movptr(c_rarg3, (int64_t) Universe::verify_oop_bits());
 959     __ cmpq(c_rarg2, c_rarg3);
 960     __ jcc(Assembler::notZero, error);
 961 
 962     // set r12 to heapbase for load_klass()
 963     __ reinit_heapbase();
 964 
 965     // make sure klass is 'reasonable'
 966     __ load_klass(rax, rax);  // get klass
 967     __ testq(rax, rax);
 968     __ jcc(Assembler::zero, error); // if klass is NULL it is broken
 969     // Check if the klass is in the right area of memory
 970     __ movq(c_rarg2, rax);
 971     __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask());
 972     __ andq(c_rarg2, c_rarg3);
 973     __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits());
 974     __ cmpq(c_rarg2, c_rarg3);
 975     __ jcc(Assembler::notZero, error);
 976 
 977     // make sure klass' klass is 'reasonable'
 978     __ load_klass(rax, rax);
 979     __ testq(rax, rax);
 980     __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
 981     // Check if the klass' klass is in the right area of memory
 982     __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask());
 983     __ andq(rax, c_rarg3);
 984     __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits());
 985     __ cmpq(rax, c_rarg3);
 986     __ jcc(Assembler::notZero, error);
 987 
 988     // return if everything seems ok
 989     __ bind(exit);
 990     __ movq(rax, Address(rsp, saved_rax));       // get saved rax back
 991     __ popq(c_rarg3);                            // restore c_rarg3
 992     __ popq(c_rarg2);                            // restore c_rarg2
 993     __ popq(r12);                                // restore r12
 994     __ popfq();                                  // restore flags
 995     __ ret(3 * wordSize);                        // pop caller saved stuff
 996 
 997     // handle errors
 998     __ bind(error);
 999     __ movq(rax, Address(rsp, saved_rax));       // get saved rax back
1000     __ popq(c_rarg3);                            // get saved c_rarg3 back
1001     __ popq(c_rarg2);                            // get saved c_rarg2 back
1002     __ popq(r12);                                // get saved r12 back
1003     __ popfq();                                  // get saved flags off stack --
1004                                                  // will be ignored
1005 
1006     __ pushaq();                                 // push registers
1007                                                  // (rip is already
1008                                                  // already pushed)
1009     // debug(char* msg, int64_t pc, int64_t regs[])
1010     // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1011     // pushed all the registers, so now the stack looks like:
1012     //     [tos +  0] 16 saved registers
1013     //     [tos + 16] return address
1014     //   * [tos + 17] error message (char*)
1015     //   * [tos + 18] object to verify (oop)
1016     //   * [tos + 19] saved rax - saved by caller and bashed
1017     //   * = popped on exit
1018 
1019     __ movq(c_rarg0, Address(rsp, error_msg));   // pass address of error message
1020     __ movq(c_rarg1, Address(rsp, return_addr)); // pass return address
1021     __ movq(c_rarg2, rsp);                       // pass address of regs on stack
1022     __ movq(r12, rsp);                           // remember rsp
1023     __ subq(rsp, frame::arg_reg_save_area_bytes);// windows
1024     __ andq(rsp, -16);                           // align stack as required by ABI
1025     BLOCK_COMMENT("call MacroAssembler::debug");
1026     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug)));
1027     __ movq(rsp, r12);                           // restore rsp
1028     __ popaq();                                  // pop registers (includes r12)
1029     __ ret(3 * wordSize);                        // pop caller saved stuff
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.
1051   //
1052   //  Input:
1053   //    Rint  -  32-bits value
1054   //    Rtmp  -  scratch
1055   //
1056   void assert_clean_int(Register Rint, Register Rtmp) {
1057 #ifdef ASSERT
1058     Label L;
1059     assert_different_registers(Rtmp, Rint);
1060     __ movslq(Rtmp, Rint);
1061     __ cmpq(Rtmp, Rint);
1062     __ jcc(Assembler::equal, L);
1063     __ stop("high 32-bits of int value are not 0");
1064     __ bind(L);
1065 #endif
1066   }
1067 
1068   //  Generate overlap test for array copy stubs
1069   //
1070   //  Input:
1071   //     c_rarg0 - from
1072   //     c_rarg1 - to
1073   //     c_rarg2 - element count
1074   //
1075   //  Output:
1076   //     rax   - &from[element count - 1]
1077   //
1078   void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1079     assert(no_overlap_target != NULL, "must be generated");
1080     array_overlap_test(no_overlap_target, NULL, sf);
1081   }
1082   void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1083     array_overlap_test(NULL, &L_no_overlap, sf);
1084   }
1085   void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1086     const Register from     = c_rarg0;
1087     const Register to       = c_rarg1;
1088     const Register count    = c_rarg2;
1089     const Register end_from = rax;
1090 
1091     __ cmpq(to, from);
1092     __ leaq(end_from, Address(from, count, sf, 0));
1093     if (NOLp == NULL) {
1094       ExternalAddress no_overlap(no_overlap_target);
1095       __ jump_cc(Assembler::belowEqual, no_overlap);
1096       __ cmpq(to, end_from);
1097       __ jump_cc(Assembler::aboveEqual, no_overlap);
1098     } else {
1099       __ jcc(Assembler::belowEqual, (*NOLp));
1100       __ cmpq(to, end_from);
1101       __ jcc(Assembler::aboveEqual, (*NOLp));
1102     }
1103   }
1104 
1105   // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1106   //
1107   // Outputs:
1108   //    rdi - rcx
1109   //    rsi - rdx
1110   //    rdx - r8
1111   //    rcx - r9
1112   //
1113   // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1114   // are non-volatile.  r9 and r10 should not be used by the caller.
1115   //
1116   void setup_arg_regs(int nargs = 3) {
1117     const Register saved_rdi = r9;
1118     const Register saved_rsi = r10;
1119     assert(nargs == 3 || nargs == 4, "else fix");
1120 #ifdef _WIN64
1121     assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1122            "unexpected argument registers");
1123     if (nargs >= 4)
1124       __ movq(rax, r9);  // r9 is also saved_rdi
1125     __ movq(saved_rdi, rdi);
1126     __ movq(saved_rsi, rsi);
1127     __ movq(rdi, rcx); // c_rarg0
1128     __ movq(rsi, rdx); // c_rarg1
1129     __ movq(rdx, r8);  // c_rarg2
1130     if (nargs >= 4)
1131       __ movq(rcx, rax); // c_rarg3 (via rax)
1132 #else
1133     assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1134            "unexpected argument registers");
1135 #endif
1136   }
1137 
1138   void restore_arg_regs() {
1139     const Register saved_rdi = r9;
1140     const Register saved_rsi = r10;
1141 #ifdef _WIN64
1142     __ movq(rdi, saved_rdi);
1143     __ movq(rsi, saved_rsi);
1144 #endif
1145   }
1146 
1147   // Generate code for an array write pre barrier
1148   //
1149   //     addr    -  starting address
1150   //     count    -  element count
1151   //
1152   //     Destroy no registers!
1153   //
1154   void  gen_write_ref_array_pre_barrier(Register addr, Register count) {
1155 #if 0 // G1 - only
1156     assert_different_registers(addr, c_rarg1);
1157     assert_different_registers(count, c_rarg0);
1158     BarrierSet* bs = Universe::heap()->barrier_set();
1159     switch (bs->kind()) {
1160       case BarrierSet::G1SATBCT:
1161       case BarrierSet::G1SATBCTLogging:
1162         {
1163           __ pushaq();                      // push registers
1164           __ movq(c_rarg0, addr);
1165           __ movq(c_rarg1, count);
1166           __ call(RuntimeAddress(BarrierSet::static_write_ref_array_pre));
1167           __ popaq();
1168         }
1169         break;
1170       case BarrierSet::CardTableModRef:
1171       case BarrierSet::CardTableExtension:
1172       case BarrierSet::ModRef:
1173         break;
1174       default      :
1175         ShouldNotReachHere();
1176 
1177     }
1178 #endif // 0 G1 - only
1179   }
1180 
1181   //
1182   // Generate code for an array write post barrier
1183   //
1184   //  Input:
1185   //     start    - register containing starting address of destination array
1186   //     end      - register containing ending address of destination array
1187   //     scratch  - scratch register
1188   //
1189   //  The input registers are overwritten.
1190   //  The ending address is inclusive.
1191   void  gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1192     assert_different_registers(start, end, scratch);
1193     BarrierSet* bs = Universe::heap()->barrier_set();
1194     switch (bs->kind()) {
1195 #if 0 // G1 - only
1196       case BarrierSet::G1SATBCT:
1197       case BarrierSet::G1SATBCTLogging:
1198 
1199         {
1200           __ pushaq();                      // push registers (overkill)
1201           // must compute element count unless barrier set interface is changed (other platforms supply count)
1202           assert_different_registers(start, end, scratch);
1203           __ leaq(scratch, Address(end, wordSize));
1204           __ subq(scratch, start);
1205           __ shrq(scratch, LogBytesPerWord);
1206           __ movq(c_rarg0, start);
1207           __ movq(c_rarg1, scratch);
1208           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
1209           __ popaq();
1210         }
1211         break;
1212 #endif // 0 G1 - only
1213       case BarrierSet::CardTableModRef:
1214       case BarrierSet::CardTableExtension:
1215         {
1216           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1217           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1218 
1219           Label L_loop;
1220 
1221            __ shrq(start, CardTableModRefBS::card_shift);
1222            __ shrq(end, CardTableModRefBS::card_shift);
1223            __ subq(end, start); // number of bytes to copy
1224 
1225           const Register count = end; // 'end' register contains bytes count now
1226           __ lea(scratch, ExternalAddress((address)ct->byte_map_base));
1227           __ addq(start, scratch);
1228         __ BIND(L_loop);
1229           __ movb(Address(start, count, Address::times_1), 0);
1230           __ decrementq(count);
1231           __ jcc(Assembler::greaterEqual, L_loop);
1232         }
1233       }
1234    }
1235 
1236   // Copy big chunks forward
1237   //
1238   // Inputs:
1239   //   end_from     - source arrays end address
1240   //   end_to       - destination array end address
1241   //   qword_count  - 64-bits element count, negative
1242   //   to           - scratch
1243   //   L_copy_32_bytes - entry label
1244   //   L_copy_8_bytes  - exit  label
1245   //
1246   void copy_32_bytes_forward(Register end_from, Register end_to,
1247                              Register qword_count, Register to,
1248                              Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1249     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1250     Label L_loop;
1251     __ align(16);
1252   __ BIND(L_loop);
1253     __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1254     __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1255     __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1256     __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1257     __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1258     __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1259     __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1260     __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1261   __ BIND(L_copy_32_bytes);
1262     __ addq(qword_count, 4);
1263     __ jcc(Assembler::lessEqual, L_loop);
1264     __ subq(qword_count, 4);
1265     __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1266   }
1267 
1268 
1269   // Copy big chunks backward
1270   //
1271   // Inputs:
1272   //   from         - source arrays address
1273   //   dest         - destination array address
1274   //   qword_count  - 64-bits element count
1275   //   to           - scratch
1276   //   L_copy_32_bytes - entry label
1277   //   L_copy_8_bytes  - exit  label
1278   //
1279   void copy_32_bytes_backward(Register from, Register dest,
1280                               Register qword_count, Register to,
1281                               Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1282     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1283     Label L_loop;
1284     __ align(16);
1285   __ BIND(L_loop);
1286     __ movq(to, Address(from, qword_count, Address::times_8, 24));
1287     __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1288     __ movq(to, Address(from, qword_count, Address::times_8, 16));
1289     __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1290     __ movq(to, Address(from, qword_count, Address::times_8,  8));
1291     __ movq(Address(dest, qword_count, Address::times_8,  8), to);
1292     __ movq(to, Address(from, qword_count, Address::times_8,  0));
1293     __ movq(Address(dest, qword_count, Address::times_8,  0), to);
1294   __ BIND(L_copy_32_bytes);
1295     __ subq(qword_count, 4);
1296     __ jcc(Assembler::greaterEqual, L_loop);
1297     __ addq(qword_count, 4);
1298     __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1299   }
1300 
1301 
1302   // Arguments:
1303   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1304   //             ignored
1305   //   name    - stub name string
1306   //
1307   // Inputs:
1308   //   c_rarg0   - source array address
1309   //   c_rarg1   - destination array address
1310   //   c_rarg2   - element count, treated as ssize_t, can be zero
1311   //
1312   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1313   // we let the hardware handle it.  The one to eight bytes within words,
1314   // dwords or qwords that span cache line boundaries will still be loaded
1315   // and stored atomically.
1316   //
1317   // Side Effects:
1318   //   disjoint_byte_copy_entry is set to the no-overlap entry point
1319   //   used by generate_conjoint_byte_copy().
1320   //
1321   address generate_disjoint_byte_copy(bool aligned, const char *name) {
1322     __ align(CodeEntryAlignment);
1323     StubCodeMark mark(this, "StubRoutines", name);
1324     address start = __ pc();
1325 
1326     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1327     Label L_copy_byte, L_exit;
1328     const Register from        = rdi;  // source array address
1329     const Register to          = rsi;  // destination array address
1330     const Register count       = rdx;  // elements count
1331     const Register byte_count  = rcx;
1332     const Register qword_count = count;
1333     const Register end_from    = from; // source array end address
1334     const Register end_to      = to;   // destination array end address
1335     // End pointers are inclusive, and if count is not zero they point
1336     // to the last unit copied:  end_to[0] := end_from[0]
1337 
1338     __ enter(); // required for proper stackwalking of RuntimeStub frame
1339     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1340 
1341     disjoint_byte_copy_entry = __ pc();
1342     BLOCK_COMMENT("Entry:");
1343     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1344 
1345     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1346                       // r9 and r10 may be used to save non-volatile registers
1347 
1348     // 'from', 'to' and 'count' are now valid
1349     __ movq(byte_count, count);
1350     __ shrq(count, 3); // count => qword_count
1351 
1352     // Copy from low to high addresses.  Use 'to' as scratch.
1353     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1354     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1355     __ negq(qword_count); // make the count negative
1356     __ jmp(L_copy_32_bytes);
1357 
1358     // Copy trailing qwords
1359   __ BIND(L_copy_8_bytes);
1360     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1361     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1362     __ incrementq(qword_count);
1363     __ jcc(Assembler::notZero, L_copy_8_bytes);
1364 
1365     // Check for and copy trailing dword
1366   __ BIND(L_copy_4_bytes);
1367     __ testq(byte_count, 4);
1368     __ jccb(Assembler::zero, L_copy_2_bytes);
1369     __ movl(rax, Address(end_from, 8));
1370     __ movl(Address(end_to, 8), rax);
1371 
1372     __ addq(end_from, 4);
1373     __ addq(end_to, 4);
1374 
1375     // Check for and copy trailing word
1376   __ BIND(L_copy_2_bytes);
1377     __ testq(byte_count, 2);
1378     __ jccb(Assembler::zero, L_copy_byte);
1379     __ movw(rax, Address(end_from, 8));
1380     __ movw(Address(end_to, 8), rax);
1381 
1382     __ addq(end_from, 2);
1383     __ addq(end_to, 2);
1384 
1385     // Check for and copy trailing byte
1386   __ BIND(L_copy_byte);
1387     __ testq(byte_count, 1);
1388     __ jccb(Assembler::zero, L_exit);
1389     __ movb(rax, Address(end_from, 8));
1390     __ movb(Address(end_to, 8), rax);
1391 
1392   __ BIND(L_exit);
1393     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1394     restore_arg_regs();
1395     __ xorq(rax, rax); // return 0
1396     __ leave(); // required for proper stackwalking of RuntimeStub frame
1397     __ ret(0);
1398 
1399     // Copy in 32-bytes chunks
1400     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1401     __ jmp(L_copy_4_bytes);
1402 
1403     return start;
1404   }
1405 
1406   // Arguments:
1407   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1408   //             ignored
1409   //   name    - stub name string
1410   //
1411   // Inputs:
1412   //   c_rarg0   - source array address
1413   //   c_rarg1   - destination array address
1414   //   c_rarg2   - element count, treated as ssize_t, can be zero
1415   //
1416   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1417   // we let the hardware handle it.  The one to eight bytes within words,
1418   // dwords or qwords that span cache line boundaries will still be loaded
1419   // and stored atomically.
1420   //
1421   address generate_conjoint_byte_copy(bool aligned, const char *name) {
1422     __ align(CodeEntryAlignment);
1423     StubCodeMark mark(this, "StubRoutines", name);
1424     address start = __ pc();
1425 
1426     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1427     const Register from        = rdi;  // source array address
1428     const Register to          = rsi;  // destination array address
1429     const Register count       = rdx;  // elements count
1430     const Register byte_count  = rcx;
1431     const Register qword_count = count;
1432 
1433     __ enter(); // required for proper stackwalking of RuntimeStub frame
1434     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1435 
1436     byte_copy_entry = __ pc();
1437     BLOCK_COMMENT("Entry:");
1438     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1439 
1440     array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1441     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1442                       // r9 and r10 may be used to save non-volatile registers
1443 
1444     // 'from', 'to' and 'count' are now valid
1445     __ movq(byte_count, count);
1446     __ shrq(count, 3);   // count => qword_count
1447 
1448     // Copy from high to low addresses.
1449 
1450     // Check for and copy trailing byte
1451     __ testq(byte_count, 1);
1452     __ jcc(Assembler::zero, L_copy_2_bytes);
1453     __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1454     __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1455     __ decrementq(byte_count); // Adjust for possible trailing word
1456 
1457     // Check for and copy trailing word
1458   __ BIND(L_copy_2_bytes);
1459     __ testq(byte_count, 2);
1460     __ jcc(Assembler::zero, L_copy_4_bytes);
1461     __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1462     __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1463 
1464     // Check for and copy trailing dword
1465   __ BIND(L_copy_4_bytes);
1466     __ testq(byte_count, 4);
1467     __ jcc(Assembler::zero, L_copy_32_bytes);
1468     __ movl(rax, Address(from, qword_count, Address::times_8));
1469     __ movl(Address(to, qword_count, Address::times_8), rax);
1470     __ jmp(L_copy_32_bytes);
1471 
1472     // Copy trailing qwords
1473   __ BIND(L_copy_8_bytes);
1474     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1475     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1476     __ decrementq(qword_count);
1477     __ jcc(Assembler::notZero, L_copy_8_bytes);
1478 
1479     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1480     restore_arg_regs();
1481     __ xorq(rax, rax); // return 0
1482     __ leave(); // required for proper stackwalking of RuntimeStub frame
1483     __ ret(0);
1484 
1485     // Copy in 32-bytes chunks
1486     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1487 
1488     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1489     restore_arg_regs();
1490     __ xorq(rax, rax); // return 0
1491     __ leave(); // required for proper stackwalking of RuntimeStub frame
1492     __ ret(0);
1493 
1494     return start;
1495   }
1496 
1497   // Arguments:
1498   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1499   //             ignored
1500   //   name    - stub name string
1501   //
1502   // Inputs:
1503   //   c_rarg0   - source array address
1504   //   c_rarg1   - destination array address
1505   //   c_rarg2   - element count, treated as ssize_t, can be zero
1506   //
1507   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1508   // let the hardware handle it.  The two or four words within dwords
1509   // or qwords that span cache line boundaries will still be loaded
1510   // and stored atomically.
1511   //
1512   // Side Effects:
1513   //   disjoint_short_copy_entry is set to the no-overlap entry point
1514   //   used by generate_conjoint_short_copy().
1515   //
1516   address generate_disjoint_short_copy(bool aligned, const char *name) {
1517     __ align(CodeEntryAlignment);
1518     StubCodeMark mark(this, "StubRoutines", name);
1519     address start = __ pc();
1520 
1521     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1522     const Register from        = rdi;  // source array address
1523     const Register to          = rsi;  // destination array address
1524     const Register count       = rdx;  // elements count
1525     const Register word_count  = rcx;
1526     const Register qword_count = count;
1527     const Register end_from    = from; // source array end address
1528     const Register end_to      = to;   // destination array end address
1529     // End pointers are inclusive, and if count is not zero they point
1530     // to the last unit copied:  end_to[0] := end_from[0]
1531 
1532     __ enter(); // required for proper stackwalking of RuntimeStub frame
1533     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1534 
1535     disjoint_short_copy_entry = __ pc();
1536     BLOCK_COMMENT("Entry:");
1537     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1538 
1539     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1540                       // r9 and r10 may be used to save non-volatile registers
1541 
1542     // 'from', 'to' and 'count' are now valid
1543     __ movq(word_count, count);
1544     __ shrq(count, 2); // count => qword_count
1545 
1546     // Copy from low to high addresses.  Use 'to' as scratch.
1547     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1548     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1549     __ negq(qword_count);
1550     __ jmp(L_copy_32_bytes);
1551 
1552     // Copy trailing qwords
1553   __ BIND(L_copy_8_bytes);
1554     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1555     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1556     __ incrementq(qword_count);
1557     __ jcc(Assembler::notZero, L_copy_8_bytes);
1558 
1559     // Original 'dest' is trashed, so we can't use it as a
1560     // base register for a possible trailing word copy
1561 
1562     // Check for and copy trailing dword
1563   __ BIND(L_copy_4_bytes);
1564     __ testq(word_count, 2);
1565     __ jccb(Assembler::zero, L_copy_2_bytes);
1566     __ movl(rax, Address(end_from, 8));
1567     __ movl(Address(end_to, 8), rax);
1568 
1569     __ addq(end_from, 4);
1570     __ addq(end_to, 4);
1571 
1572     // Check for and copy trailing word
1573   __ BIND(L_copy_2_bytes);
1574     __ testq(word_count, 1);
1575     __ jccb(Assembler::zero, L_exit);
1576     __ movw(rax, Address(end_from, 8));
1577     __ movw(Address(end_to, 8), rax);
1578 
1579   __ BIND(L_exit);
1580     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1581     restore_arg_regs();
1582     __ xorq(rax, rax); // return 0
1583     __ leave(); // required for proper stackwalking of RuntimeStub frame
1584     __ ret(0);
1585 
1586     // Copy in 32-bytes chunks
1587     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1588     __ jmp(L_copy_4_bytes);
1589 
1590     return start;
1591   }
1592 
1593   // Arguments:
1594   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1595   //             ignored
1596   //   name    - stub name string
1597   //
1598   // Inputs:
1599   //   c_rarg0   - source array address
1600   //   c_rarg1   - destination array address
1601   //   c_rarg2   - element count, treated as ssize_t, can be zero
1602   //
1603   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1604   // let the hardware handle it.  The two or four words within dwords
1605   // or qwords that span cache line boundaries will still be loaded
1606   // and stored atomically.
1607   //
1608   address generate_conjoint_short_copy(bool aligned, const char *name) {
1609     __ align(CodeEntryAlignment);
1610     StubCodeMark mark(this, "StubRoutines", name);
1611     address start = __ pc();
1612 
1613     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1614     const Register from        = rdi;  // source array address
1615     const Register to          = rsi;  // destination array address
1616     const Register count       = rdx;  // elements count
1617     const Register word_count  = rcx;
1618     const Register qword_count = count;
1619 
1620     __ enter(); // required for proper stackwalking of RuntimeStub frame
1621     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1622 
1623     short_copy_entry = __ pc();
1624     BLOCK_COMMENT("Entry:");
1625     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1626 
1627     array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1628     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1629                       // r9 and r10 may be used to save non-volatile registers
1630 
1631     // 'from', 'to' and 'count' are now valid
1632     __ movq(word_count, count);
1633     __ shrq(count, 2); // count => qword_count
1634 
1635     // Copy from high to low addresses.  Use 'to' as scratch.
1636 
1637     // Check for and copy trailing word
1638     __ testq(word_count, 1);
1639     __ jccb(Assembler::zero, L_copy_4_bytes);
1640     __ movw(rax, Address(from, word_count, Address::times_2, -2));
1641     __ movw(Address(to, word_count, Address::times_2, -2), rax);
1642 
1643     // Check for and copy trailing dword
1644   __ BIND(L_copy_4_bytes);
1645     __ testq(word_count, 2);
1646     __ jcc(Assembler::zero, L_copy_32_bytes);
1647     __ movl(rax, Address(from, qword_count, Address::times_8));
1648     __ movl(Address(to, qword_count, Address::times_8), rax);
1649     __ jmp(L_copy_32_bytes);
1650 
1651     // Copy trailing qwords
1652   __ BIND(L_copy_8_bytes);
1653     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1654     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1655     __ decrementq(qword_count);
1656     __ jcc(Assembler::notZero, L_copy_8_bytes);
1657 
1658     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1659     restore_arg_regs();
1660     __ xorq(rax, rax); // return 0
1661     __ leave(); // required for proper stackwalking of RuntimeStub frame
1662     __ ret(0);
1663 
1664     // Copy in 32-bytes chunks
1665     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1666 
1667     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1668     restore_arg_regs();
1669     __ xorq(rax, rax); // return 0
1670     __ leave(); // required for proper stackwalking of RuntimeStub frame
1671     __ ret(0);
1672 
1673     return start;
1674   }
1675 
1676   // Arguments:
1677   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1678   //             ignored
1679   //   is_oop  - true => oop array, so generate store check code
1680   //   name    - stub name string
1681   //
1682   // Inputs:
1683   //   c_rarg0   - source array address
1684   //   c_rarg1   - destination array address
1685   //   c_rarg2   - element count, treated as ssize_t, can be zero
1686   //
1687   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1688   // the hardware handle it.  The two dwords within qwords that span
1689   // cache line boundaries will still be loaded and stored atomicly.
1690   //
1691   // Side Effects:
1692   //   disjoint_int_copy_entry is set to the no-overlap entry point
1693   //   used by generate_conjoint_int_oop_copy().
1694   //
1695   address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1696     __ align(CodeEntryAlignment);
1697     StubCodeMark mark(this, "StubRoutines", name);
1698     address start = __ pc();
1699 
1700     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1701     const Register from        = rdi;  // source array address
1702     const Register to          = rsi;  // destination array address
1703     const Register count       = rdx;  // elements count
1704     const Register dword_count = rcx;
1705     const Register qword_count = count;
1706     const Register end_from    = from; // source array end address
1707     const Register end_to      = to;   // destination array end address
1708     const Register saved_to    = r11;  // saved destination array address
1709     // End pointers are inclusive, and if count is not zero they point
1710     // to the last unit copied:  end_to[0] := end_from[0]
1711 
1712     __ enter(); // required for proper stackwalking of RuntimeStub frame
1713     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1714 
1715     (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1716 
1717     if (is_oop) {
1718       // no registers are destroyed by this call
1719       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1720     }
1721 
1722     BLOCK_COMMENT("Entry:");
1723     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1724 
1725     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1726                       // r9 and r10 may be used to save non-volatile registers
1727 
1728     if (is_oop) {
1729       __ movq(saved_to, to);
1730     }
1731 
1732     // 'from', 'to' and 'count' are now valid
1733     __ movq(dword_count, count);
1734     __ shrq(count, 1); // count => qword_count
1735 
1736     // Copy from low to high addresses.  Use 'to' as scratch.
1737     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1738     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1739     __ negq(qword_count);
1740     __ jmp(L_copy_32_bytes);
1741 
1742     // Copy trailing qwords
1743   __ BIND(L_copy_8_bytes);
1744     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1745     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1746     __ incrementq(qword_count);
1747     __ jcc(Assembler::notZero, L_copy_8_bytes);
1748 
1749     // Check for and copy trailing dword
1750   __ BIND(L_copy_4_bytes);
1751     __ testq(dword_count, 1); // Only byte test since the value is 0 or 1
1752     __ jccb(Assembler::zero, L_exit);
1753     __ movl(rax, Address(end_from, 8));
1754     __ movl(Address(end_to, 8), rax);
1755 
1756   __ BIND(L_exit);
1757     if (is_oop) {
1758       __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1759       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1760     }
1761     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1762     restore_arg_regs();
1763     __ xorq(rax, rax); // return 0
1764     __ leave(); // required for proper stackwalking of RuntimeStub frame
1765     __ ret(0);
1766 
1767     // Copy 32-bytes chunks
1768     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1769     __ jmp(L_copy_4_bytes);
1770 
1771     return start;
1772   }
1773 
1774   // Arguments:
1775   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1776   //             ignored
1777   //   is_oop  - true => oop array, so generate store check code
1778   //   name    - stub name string
1779   //
1780   // Inputs:
1781   //   c_rarg0   - source array address
1782   //   c_rarg1   - destination array address
1783   //   c_rarg2   - element count, treated as ssize_t, can be zero
1784   //
1785   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1786   // the hardware handle it.  The two dwords within qwords that span
1787   // cache line boundaries will still be loaded and stored atomicly.
1788   //
1789   address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1790     __ align(CodeEntryAlignment);
1791     StubCodeMark mark(this, "StubRoutines", name);
1792     address start = __ pc();
1793 
1794     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1795     const Register from        = rdi;  // source array address
1796     const Register to          = rsi;  // destination array address
1797     const Register count       = rdx;  // elements count
1798     const Register dword_count = rcx;
1799     const Register qword_count = count;
1800 
1801     __ enter(); // required for proper stackwalking of RuntimeStub frame
1802     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1803 
1804     if (is_oop) {
1805       // no registers are destroyed by this call
1806       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1807     }
1808 
1809     (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1810     BLOCK_COMMENT("Entry:");
1811     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1812 
1813     array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1814                        Address::times_4);
1815     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1816                       // r9 and r10 may be used to save non-volatile registers
1817 
1818     assert_clean_int(count, rax); // Make sure 'count' is clean int.
1819     // 'from', 'to' and 'count' are now valid
1820     __ movq(dword_count, count);
1821     __ shrq(count, 1); // count => qword_count
1822 
1823     // Copy from high to low addresses.  Use 'to' as scratch.
1824 
1825     // Check for and copy trailing dword
1826     __ testq(dword_count, 1);
1827     __ jcc(Assembler::zero, L_copy_32_bytes);
1828     __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1829     __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1830     __ jmp(L_copy_32_bytes);
1831 
1832     // Copy trailing qwords
1833   __ BIND(L_copy_8_bytes);
1834     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1835     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1836     __ decrementq(qword_count);
1837     __ jcc(Assembler::notZero, L_copy_8_bytes);
1838 
1839     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1840     if (is_oop) {
1841       __ jmp(L_exit);
1842     }
1843     restore_arg_regs();
1844     __ xorq(rax, rax); // return 0
1845     __ leave(); // required for proper stackwalking of RuntimeStub frame
1846     __ ret(0);
1847 
1848     // Copy in 32-bytes chunks
1849     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1850 
1851    inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1852    __ bind(L_exit);
1853      if (is_oop) {
1854        Register end_to = rdx;
1855        __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1856        gen_write_ref_array_post_barrier(to, end_to, rax);
1857      }
1858     restore_arg_regs();
1859     __ xorq(rax, rax); // return 0
1860     __ leave(); // required for proper stackwalking of RuntimeStub frame
1861     __ ret(0);
1862 
1863     return start;
1864   }
1865 
1866   // Arguments:
1867   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1868   //             ignored
1869   //   is_oop  - true => oop array, so generate store check code
1870   //   name    - stub name string
1871   //
1872   // Inputs:
1873   //   c_rarg0   - source array address
1874   //   c_rarg1   - destination array address
1875   //   c_rarg2   - element count, treated as ssize_t, can be zero
1876   //
1877  // Side Effects:
1878   //   disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1879   //   no-overlap entry point used by generate_conjoint_long_oop_copy().
1880   //
1881   address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1882     __ align(CodeEntryAlignment);
1883     StubCodeMark mark(this, "StubRoutines", name);
1884     address start = __ pc();
1885 
1886     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1887     const Register from        = rdi;  // source array address
1888     const Register to          = rsi;  // destination array address
1889     const Register qword_count = rdx;  // elements count
1890     const Register end_from    = from; // source array end address
1891     const Register end_to      = rcx;  // destination array end address
1892     const Register saved_to    = to;
1893     // End pointers are inclusive, and if count is not zero they point
1894     // to the last unit copied:  end_to[0] := end_from[0]
1895 
1896     __ enter(); // required for proper stackwalking of RuntimeStub frame
1897     // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1898     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1899 
1900     if (is_oop) {
1901       disjoint_oop_copy_entry  = __ pc();
1902       // no registers are destroyed by this call
1903       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1904     } else {
1905       disjoint_long_copy_entry = __ pc();
1906     }
1907     BLOCK_COMMENT("Entry:");
1908     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1909 
1910     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1911                       // r9 and r10 may be used to save non-volatile registers
1912 
1913     // 'from', 'to' and 'qword_count' are now valid
1914 
1915     // Copy from low to high addresses.  Use 'to' as scratch.
1916     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1917     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1918     __ negq(qword_count);
1919     __ jmp(L_copy_32_bytes);
1920 
1921     // Copy trailing qwords
1922   __ BIND(L_copy_8_bytes);
1923     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1924     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1925     __ incrementq(qword_count);
1926     __ jcc(Assembler::notZero, L_copy_8_bytes);
1927 
1928     if (is_oop) {
1929       __ jmp(L_exit);
1930     } else {
1931       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1932       restore_arg_regs();
1933       __ xorq(rax, rax); // return 0
1934       __ leave(); // required for proper stackwalking of RuntimeStub frame
1935       __ ret(0);
1936     }
1937 
1938     // Copy 64-byte chunks
1939     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1940 
1941     if (is_oop) {
1942     __ BIND(L_exit);
1943       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1944       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1945     } else {
1946       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1947     }
1948     restore_arg_regs();
1949     __ xorq(rax, rax); // return 0
1950     __ leave(); // required for proper stackwalking of RuntimeStub frame
1951     __ ret(0);
1952 
1953     return start;
1954   }
1955 
1956   // Arguments:
1957   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1958   //             ignored
1959   //   is_oop  - true => oop array, so generate store check code
1960   //   name    - stub name string
1961   //
1962   // Inputs:
1963   //   c_rarg0   - source array address
1964   //   c_rarg1   - destination array address
1965   //   c_rarg2   - element count, treated as ssize_t, can be zero
1966   //
1967   address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1968     __ align(CodeEntryAlignment);
1969     StubCodeMark mark(this, "StubRoutines", name);
1970     address start = __ pc();
1971 
1972     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1973     const Register from        = rdi;  // source array address
1974     const Register to          = rsi;  // destination array address
1975     const Register qword_count = rdx;  // elements count
1976     const Register saved_count = rcx;
1977 
1978     __ enter(); // required for proper stackwalking of RuntimeStub frame
1979     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1980 
1981     address disjoint_copy_entry = NULL;
1982     if (is_oop) {
1983       assert(!UseCompressedOops, "shouldn't be called for compressed oops");
1984       disjoint_copy_entry = disjoint_oop_copy_entry;
1985       oop_copy_entry  = __ pc();
1986       array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
1987     } else {
1988       disjoint_copy_entry = disjoint_long_copy_entry;
1989       long_copy_entry = __ pc();
1990       array_overlap_test(disjoint_long_copy_entry, Address::times_8);
1991     }
1992     BLOCK_COMMENT("Entry:");
1993     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1994 
1995     array_overlap_test(disjoint_copy_entry, Address::times_8);
1996     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1997                       // r9 and r10 may be used to save non-volatile registers
1998 
1999     // 'from', 'to' and 'qword_count' are now valid
2000 
2001     if (is_oop) {
2002       // Save to and count for store barrier
2003       __ movq(saved_count, qword_count);
2004       // No registers are destroyed by this call
2005       gen_write_ref_array_pre_barrier(to, saved_count);
2006     }
2007 
2008     __ jmp(L_copy_32_bytes);
2009 
2010     // Copy trailing qwords
2011   __ BIND(L_copy_8_bytes);
2012     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2013     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2014     __ decrementq(qword_count);
2015     __ jcc(Assembler::notZero, L_copy_8_bytes);
2016 
2017     if (is_oop) {
2018       __ jmp(L_exit);
2019     } else {
2020       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2021       restore_arg_regs();
2022       __ xorq(rax, rax); // return 0
2023       __ leave(); // required for proper stackwalking of RuntimeStub frame
2024       __ ret(0);
2025     }
2026 
2027     // Copy in 32-bytes chunks
2028     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2029 
2030     if (is_oop) {
2031     __ BIND(L_exit);
2032       __ leaq(rcx, Address(to, saved_count, Address::times_8, -8));
2033       gen_write_ref_array_post_barrier(to, rcx, rax);
2034       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2035     } else {
2036       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2037     }
2038     restore_arg_regs();
2039     __ xorq(rax, rax); // return 0
2040     __ leave(); // required for proper stackwalking of RuntimeStub frame
2041     __ ret(0);
2042 
2043     return start;
2044   }
2045 
2046 
2047   // Helper for generating a dynamic type check.
2048   // Smashes no registers.
2049   void generate_type_check(Register sub_klass,
2050                            Register super_check_offset,
2051                            Register super_klass,
2052                            Label& L_success) {
2053     assert_different_registers(sub_klass, super_check_offset, super_klass);
2054 
2055     BLOCK_COMMENT("type_check:");
2056 
2057     Label L_miss;
2058 
2059     // a couple of useful fields in sub_klass:
2060     int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
2061                      Klass::secondary_supers_offset_in_bytes());
2062     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
2063                      Klass::secondary_super_cache_offset_in_bytes());
2064     Address secondary_supers_addr(sub_klass, ss_offset);
2065     Address super_cache_addr(     sub_klass, sc_offset);
2066 
2067     // if the pointers are equal, we are done (e.g., String[] elements)
2068     __ cmpq(super_klass, sub_klass);
2069     __ jcc(Assembler::equal, L_success);
2070 
2071     // check the supertype display:
2072     Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
2073     __ cmpq(super_klass, super_check_addr); // test the super type
2074     __ jcc(Assembler::equal, L_success);
2075 
2076     // if it was a primary super, we can just fail immediately
2077     __ cmpl(super_check_offset, sc_offset);
2078     __ jcc(Assembler::notEqual, L_miss);
2079 
2080     // Now do a linear scan of the secondary super-klass chain.
2081     // The repne_scan instruction uses fixed registers, which we must spill.
2082     // (We need a couple more temps in any case.)
2083     // This code is rarely used, so simplicity is a virtue here.
2084     inc_counter_np(SharedRuntime::_partial_subtype_ctr);
2085     {
2086       __ pushq(rax);
2087       __ pushq(rcx);
2088       __ pushq(rdi);
2089       assert_different_registers(sub_klass, super_klass, rax, rcx, rdi);
2090 
2091       __ movq(rdi, secondary_supers_addr);
2092       // Load the array length.
2093       __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
2094       // Skip to start of data.
2095       __ addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
2096       // Scan rcx words at [rdi] for occurance of rax
2097       // Set NZ/Z based on last compare
2098       __ movq(rax, super_klass);
2099       if (UseCompressedOops) {
2100         // Compare against compressed form.  Don't need to uncompress because
2101         // looks like orig rax is restored in popq below.
2102         __ encode_heap_oop(rax);
2103         __ repne_scanl();
2104       } else {
2105          __ repne_scanq();
2106       }
2107 
2108       // Unspill the temp. registers:
2109       __ popq(rdi);
2110       __ popq(rcx);
2111       __ popq(rax);
2112 
2113       __ jcc(Assembler::notEqual, L_miss);
2114     }
2115 
2116     // Success.  Cache the super we found and proceed in triumph.
2117     __ movq(super_cache_addr, super_klass); // note: rax is dead
2118     __ jmp(L_success);
2119 
2120     // Fall through on failure!
2121     __ BIND(L_miss);
2122   }
2123 
2124   //
2125   //  Generate checkcasting array copy stub
2126   //
2127   //  Input:
2128   //    c_rarg0   - source array address
2129   //    c_rarg1   - destination array address
2130   //    c_rarg2   - element count, treated as ssize_t, can be zero
2131   //    c_rarg3   - size_t ckoff (super_check_offset)
2132   // not Win64
2133   //    c_rarg4   - oop ckval (super_klass)
2134   // Win64
2135   //    rsp+40    - oop ckval (super_klass)
2136   //
2137   //  Output:
2138   //    rax ==  0  -  success
2139   //    rax == -1^K - failure, where K is partial transfer count
2140   //
2141   address generate_checkcast_copy(const char *name) {
2142 
2143     Label L_load_element, L_store_element, L_do_card_marks, L_done;
2144 
2145     // Input registers (after setup_arg_regs)
2146     const Register from        = rdi;   // source array address
2147     const Register to          = rsi;   // destination array address
2148     const Register length      = rdx;   // elements count
2149     const Register ckoff       = rcx;   // super_check_offset
2150     const Register ckval       = r8;    // super_klass
2151 
2152     // Registers used as temps (r13, r14 are save-on-entry)
2153     const Register end_from    = from;  // source array end address
2154     const Register end_to      = r13;   // destination array end address
2155     const Register count       = rdx;   // -(count_remaining)
2156     const Register r14_length  = r14;   // saved copy of length
2157     // End pointers are inclusive, and if length is not zero they point
2158     // to the last unit copied:  end_to[0] := end_from[0]
2159 
2160     const Register rax_oop    = rax;    // actual oop copied
2161     const Register r11_klass  = r11;    // oop._klass
2162 
2163     //---------------------------------------------------------------
2164     // Assembler stub will be used for this call to arraycopy
2165     // if the two arrays are subtypes of Object[] but the
2166     // destination array type is not equal to or a supertype
2167     // of the source type.  Each element must be separately
2168     // checked.
2169 
2170     __ align(CodeEntryAlignment);
2171     StubCodeMark mark(this, "StubRoutines", name);
2172     address start = __ pc();
2173 
2174     __ enter(); // required for proper stackwalking of RuntimeStub frame
2175 
2176     checkcast_copy_entry  = __ pc();
2177     BLOCK_COMMENT("Entry:");
2178 
2179 #ifdef ASSERT
2180     // caller guarantees that the arrays really are different
2181     // otherwise, we would have to make conjoint checks
2182     { Label L;
2183       array_overlap_test(L, TIMES_OOP);
2184       __ stop("checkcast_copy within a single array");
2185       __ bind(L);
2186     }
2187 #endif //ASSERT
2188 
2189     // allocate spill slots for r13, r14
2190     enum {
2191       saved_r13_offset,
2192       saved_r14_offset,
2193       saved_rbp_offset,
2194       saved_rip_offset,
2195       saved_rarg0_offset
2196     };
2197     __ subq(rsp, saved_rbp_offset * wordSize);
2198     __ movq(Address(rsp, saved_r13_offset * wordSize), r13);
2199     __ movq(Address(rsp, saved_r14_offset * wordSize), r14);
2200     setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2201                        // ckoff => rcx, ckval => r8
2202                        // r9 and r10 may be used to save non-volatile registers
2203 #ifdef _WIN64
2204     // last argument (#4) is on stack on Win64
2205     const int ckval_offset = saved_rarg0_offset + 4;
2206     __ movq(ckval, Address(rsp, ckval_offset * wordSize));
2207 #endif
2208 
2209     // check that int operands are properly extended to size_t
2210     assert_clean_int(length, rax);
2211     assert_clean_int(ckoff, rax);
2212 
2213 #ifdef ASSERT
2214     BLOCK_COMMENT("assert consistent ckoff/ckval");
2215     // The ckoff and ckval must be mutually consistent,
2216     // even though caller generates both.
2217     { Label L;
2218       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2219                         Klass::super_check_offset_offset_in_bytes());
2220       __ cmpl(ckoff, Address(ckval, sco_offset));
2221       __ jcc(Assembler::equal, L);
2222       __ stop("super_check_offset inconsistent");
2223       __ bind(L);
2224     }
2225 #endif //ASSERT
2226 
2227     // Loop-invariant addresses.  They are exclusive end pointers.
2228     Address end_from_addr(from, length, TIMES_OOP, 0);
2229     Address   end_to_addr(to,   length, TIMES_OOP, 0);
2230     // Loop-variant addresses.  They assume post-incremented count < 0.
2231     Address from_element_addr(end_from, count, TIMES_OOP, 0);
2232     Address   to_element_addr(end_to,   count, TIMES_OOP, 0);
2233 
2234     gen_write_ref_array_pre_barrier(to, count);
2235 
2236     // Copy from low to high addresses, indexed from the end of each array.
2237     __ leaq(end_from, end_from_addr);
2238     __ leaq(end_to,   end_to_addr);
2239     __ movq(r14_length, length);        // save a copy of the length
2240     assert(length == count, "");        // else fix next line:
2241     __ negq(count);                     // negate and test the length
2242     __ jcc(Assembler::notZero, L_load_element);
2243 
2244     // Empty array:  Nothing to do.
2245     __ xorq(rax, rax);                  // return 0 on (trivial) success
2246     __ jmp(L_done);
2247 
2248     // ======== begin loop ========
2249     // (Loop is rotated; its entry is L_load_element.)
2250     // Loop control:
2251     //   for (count = -count; count != 0; count++)
2252     // Base pointers src, dst are biased by 8*(count-1),to last element.
2253     __ align(16);
2254 
2255     __ BIND(L_store_element);
2256     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop
2257     __ incrementq(count);               // increment the count toward zero
2258     __ jcc(Assembler::zero, L_do_card_marks);
2259 
2260     // ======== loop entry is here ========
2261     __ BIND(L_load_element);
2262     __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2263     __ testq(rax_oop, rax_oop);
2264     __ jcc(Assembler::zero, L_store_element);
2265 
2266     __ load_klass(r11_klass, rax_oop);// query the object klass
2267     generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2268     // ======== end loop ========
2269 
2270     // It was a real error; we must depend on the caller to finish the job.
2271     // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2272     // Emit GC store barriers for the oops we have copied (r14 + rdx),
2273     // and report their number to the caller.
2274     assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2275     __ leaq(end_to, to_element_addr);
2276     gen_write_ref_array_post_barrier(to, end_to, rcx);
2277     __ movq(rax, r14_length);           // original oops
2278     __ addq(rax, count);                // K = (original - remaining) oops
2279     __ notq(rax);                       // report (-1^K) to caller
2280     __ jmp(L_done);
2281 
2282     // Come here on success only.
2283     __ BIND(L_do_card_marks);
2284     __ addq(end_to, -wordSize);         // make an inclusive end pointer
2285     gen_write_ref_array_post_barrier(to, end_to, rcx);
2286     __ xorq(rax, rax);                  // return 0 on success
2287 
2288     // Common exit point (success or failure).
2289     __ BIND(L_done);
2290     __ movq(r13, Address(rsp, saved_r13_offset * wordSize));
2291     __ movq(r14, Address(rsp, saved_r14_offset * wordSize));
2292     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2293     restore_arg_regs();
2294     __ leave(); // required for proper stackwalking of RuntimeStub frame
2295     __ ret(0);
2296 
2297     return start;
2298   }
2299 
2300   //
2301   //  Generate 'unsafe' array copy stub
2302   //  Though just as safe as the other stubs, it takes an unscaled
2303   //  size_t argument instead of an element count.
2304   //
2305   //  Input:
2306   //    c_rarg0   - source array address
2307   //    c_rarg1   - destination array address
2308   //    c_rarg2   - byte count, treated as ssize_t, can be zero
2309   //
2310   // Examines the alignment of the operands and dispatches
2311   // to a long, int, short, or byte copy loop.
2312   //
2313   address generate_unsafe_copy(const char *name) {
2314 
2315     Label L_long_aligned, L_int_aligned, L_short_aligned;
2316 
2317     // Input registers (before setup_arg_regs)
2318     const Register from        = c_rarg0;  // source array address
2319     const Register to          = c_rarg1;  // destination array address
2320     const Register size        = c_rarg2;  // byte count (size_t)
2321 
2322     // Register used as a temp
2323     const Register bits        = rax;      // test copy of low bits
2324 
2325     __ align(CodeEntryAlignment);
2326     StubCodeMark mark(this, "StubRoutines", name);
2327     address start = __ pc();
2328 
2329     __ enter(); // required for proper stackwalking of RuntimeStub frame
2330 
2331     // bump this on entry, not on exit:
2332     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2333 
2334     __ movq(bits, from);
2335     __ orq(bits, to);
2336     __ orq(bits, size);
2337 
2338     __ testb(bits, BytesPerLong-1);
2339     __ jccb(Assembler::zero, L_long_aligned);
2340 
2341     __ testb(bits, BytesPerInt-1);
2342     __ jccb(Assembler::zero, L_int_aligned);
2343 
2344     __ testb(bits, BytesPerShort-1);
2345     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2346 
2347     __ BIND(L_short_aligned);
2348     __ shrq(size, LogBytesPerShort); // size => short_count
2349     __ jump(RuntimeAddress(short_copy_entry));
2350 
2351     __ BIND(L_int_aligned);
2352     __ shrq(size, LogBytesPerInt); // size => int_count
2353     __ jump(RuntimeAddress(int_copy_entry));
2354 
2355     __ BIND(L_long_aligned);
2356     __ shrq(size, LogBytesPerLong); // size => qword_count
2357     __ jump(RuntimeAddress(long_copy_entry));
2358 
2359     return start;
2360   }
2361 
2362   // Perform range checks on the proposed arraycopy.
2363   // Kills temp, but nothing else.
2364   // Also, clean the sign bits of src_pos and dst_pos.
2365   void arraycopy_range_checks(Register src,     // source array oop (c_rarg0)
2366                               Register src_pos, // source position (c_rarg1)
2367                               Register dst,     // destination array oo (c_rarg2)
2368                               Register dst_pos, // destination position (c_rarg3)
2369                               Register length,
2370                               Register temp,
2371                               Label& L_failed) {
2372     BLOCK_COMMENT("arraycopy_range_checks:");
2373 
2374     //  if (src_pos + length > arrayOop(src)->length())  FAIL;
2375     __ movl(temp, length);
2376     __ addl(temp, src_pos);             // src_pos + length
2377     __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2378     __ jcc(Assembler::above, L_failed);
2379 
2380     //  if (dst_pos + length > arrayOop(dst)->length())  FAIL;
2381     __ movl(temp, length);
2382     __ addl(temp, dst_pos);             // dst_pos + length
2383     __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2384     __ jcc(Assembler::above, L_failed);
2385 
2386     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2387     // Move with sign extension can be used since they are positive.
2388     __ movslq(src_pos, src_pos);
2389     __ movslq(dst_pos, dst_pos);
2390 
2391     BLOCK_COMMENT("arraycopy_range_checks done");
2392   }
2393 
2394   //
2395   //  Generate generic array copy stubs
2396   //
2397   //  Input:
2398   //    c_rarg0    -  src oop
2399   //    c_rarg1    -  src_pos (32-bits)
2400   //    c_rarg2    -  dst oop
2401   //    c_rarg3    -  dst_pos (32-bits)
2402   // not Win64
2403   //    c_rarg4    -  element count (32-bits)
2404   // Win64
2405   //    rsp+40     -  element count (32-bits)
2406   //
2407   //  Output:
2408   //    rax ==  0  -  success
2409   //    rax == -1^K - failure, where K is partial transfer count
2410   //
2411   address generate_generic_copy(const char *name) {
2412 
2413     Label L_failed, L_failed_0, L_objArray;
2414     Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2415 
2416     // Input registers
2417     const Register src        = c_rarg0;  // source array oop
2418     const Register src_pos    = c_rarg1;  // source position
2419     const Register dst        = c_rarg2;  // destination array oop
2420     const Register dst_pos    = c_rarg3;  // destination position
2421     // elements count is on stack on Win64
2422 #ifdef _WIN64
2423 #define C_RARG4 Address(rsp, 6 * wordSize)
2424 #else
2425 #define C_RARG4 c_rarg4
2426 #endif
2427 
2428     { int modulus = CodeEntryAlignment;
2429       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
2430       int advance = target - (__ offset() % modulus);
2431       if (advance < 0)  advance += modulus;
2432       if (advance > 0)  __ nop(advance);
2433     }
2434     StubCodeMark mark(this, "StubRoutines", name);
2435 
2436     // Short-hop target to L_failed.  Makes for denser prologue code.
2437     __ BIND(L_failed_0);
2438     __ jmp(L_failed);
2439     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2440 
2441     __ align(CodeEntryAlignment);
2442     address start = __ pc();
2443 
2444     __ enter(); // required for proper stackwalking of RuntimeStub frame
2445 
2446     // bump this on entry, not on exit:
2447     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2448 
2449     //-----------------------------------------------------------------------
2450     // Assembler stub will be used for this call to arraycopy
2451     // if the following conditions are met:
2452     //
2453     // (1) src and dst must not be null.
2454     // (2) src_pos must not be negative.
2455     // (3) dst_pos must not be negative.
2456     // (4) length  must not be negative.
2457     // (5) src klass and dst klass should be the same and not NULL.
2458     // (6) src and dst should be arrays.
2459     // (7) src_pos + length must not exceed length of src.
2460     // (8) dst_pos + length must not exceed length of dst.
2461     //
2462 
2463     //  if (src == NULL) return -1;
2464     __ testq(src, src);         // src oop
2465     size_t j1off = __ offset();
2466     __ jccb(Assembler::zero, L_failed_0);
2467 
2468     //  if (src_pos < 0) return -1;
2469     __ testl(src_pos, src_pos); // src_pos (32-bits)
2470     __ jccb(Assembler::negative, L_failed_0);
2471 
2472     //  if (dst == NULL) return -1;
2473     __ testq(dst, dst);         // dst oop
2474     __ jccb(Assembler::zero, L_failed_0);
2475 
2476     //  if (dst_pos < 0) return -1;
2477     __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2478     size_t j4off = __ offset();
2479     __ jccb(Assembler::negative, L_failed_0);
2480 
2481     // The first four tests are very dense code,
2482     // but not quite dense enough to put four
2483     // jumps in a 16-byte instruction fetch buffer.
2484     // That's good, because some branch predicters
2485     // do not like jumps so close together.
2486     // Make sure of this.
2487     guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2488 
2489     // registers used as temp
2490     const Register r11_length    = r11; // elements count to copy
2491     const Register r10_src_klass = r10; // array klass
2492     const Register r9_dst_klass  = r9;  // dest array klass
2493 
2494     //  if (length < 0) return -1;
2495     __ movl(r11_length, C_RARG4);       // length (elements count, 32-bits value)
2496     __ testl(r11_length, r11_length);
2497     __ jccb(Assembler::negative, L_failed_0);
2498 
2499     __ load_klass(r10_src_klass, src);
2500 #ifdef ASSERT
2501     //  assert(src->klass() != NULL);
2502     BLOCK_COMMENT("assert klasses not null");
2503     { Label L1, L2;
2504       __ testq(r10_src_klass, r10_src_klass);
2505       __ jcc(Assembler::notZero, L2);   // it is broken if klass is NULL
2506       __ bind(L1);
2507       __ stop("broken null klass");
2508       __ bind(L2);
2509       __ load_klass(r9_dst_klass, dst);
2510       __ cmpq(r9_dst_klass, 0);
2511       __ jcc(Assembler::equal, L1);     // this would be broken also
2512       BLOCK_COMMENT("assert done");
2513     }
2514 #endif
2515 
2516     // Load layout helper (32-bits)
2517     //
2518     //  |array_tag|     | header_size | element_type |     |log2_element_size|
2519     // 32        30    24            16              8     2                 0
2520     //
2521     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2522     //
2523 
2524     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2525                     Klass::layout_helper_offset_in_bytes();
2526 
2527     const Register rax_lh = rax;  // layout helper
2528 
2529     __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2530 
2531     // Handle objArrays completely differently...
2532     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2533     __ cmpl(rax_lh, objArray_lh);
2534     __ jcc(Assembler::equal, L_objArray);
2535 
2536     //  if (src->klass() != dst->klass()) return -1;
2537     __ load_klass(r9_dst_klass, dst);
2538     __ cmpq(r10_src_klass, r9_dst_klass);
2539     __ jcc(Assembler::notEqual, L_failed);
2540 
2541     //  if (!src->is_Array()) return -1;
2542     __ cmpl(rax_lh, Klass::_lh_neutral_value);
2543     __ jcc(Assembler::greaterEqual, L_failed);
2544 
2545     // At this point, it is known to be a typeArray (array_tag 0x3).
2546 #ifdef ASSERT
2547     { Label L;
2548       __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2549       __ jcc(Assembler::greaterEqual, L);
2550       __ stop("must be a primitive array");
2551       __ bind(L);
2552     }
2553 #endif
2554 
2555     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2556                            r10, L_failed);
2557 
2558     // typeArrayKlass
2559     //
2560     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2561     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2562     //
2563 
2564     const Register r10_offset = r10;    // array offset
2565     const Register rax_elsize = rax_lh; // element size
2566 
2567     __ movl(r10_offset, rax_lh);
2568     __ shrl(r10_offset, Klass::_lh_header_size_shift);
2569     __ andq(r10_offset, Klass::_lh_header_size_mask);   // array_offset
2570     __ addq(src, r10_offset);           // src array offset
2571     __ addq(dst, r10_offset);           // dst array offset
2572     BLOCK_COMMENT("choose copy loop based on element size");
2573     __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2574 
2575     // next registers should be set before the jump to corresponding stub
2576     const Register from     = c_rarg0;  // source array address
2577     const Register to       = c_rarg1;  // destination array address
2578     const Register count    = c_rarg2;  // elements count
2579 
2580     // 'from', 'to', 'count' registers should be set in such order
2581     // since they are the same as 'src', 'src_pos', 'dst'.
2582 
2583   __ BIND(L_copy_bytes);
2584     __ cmpl(rax_elsize, 0);
2585     __ jccb(Assembler::notEqual, L_copy_shorts);
2586     __ leaq(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2587     __ leaq(to,   Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2588     __ movslq(count, r11_length); // length
2589     __ jump(RuntimeAddress(byte_copy_entry));
2590 
2591   __ BIND(L_copy_shorts);
2592     __ cmpl(rax_elsize, LogBytesPerShort);
2593     __ jccb(Assembler::notEqual, L_copy_ints);
2594     __ leaq(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2595     __ leaq(to,   Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2596     __ movslq(count, r11_length); // length
2597     __ jump(RuntimeAddress(short_copy_entry));
2598 
2599   __ BIND(L_copy_ints);
2600     __ cmpl(rax_elsize, LogBytesPerInt);
2601     __ jccb(Assembler::notEqual, L_copy_longs);
2602     __ leaq(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2603     __ leaq(to,   Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2604     __ movslq(count, r11_length); // length
2605     __ jump(RuntimeAddress(int_copy_entry));
2606 
2607   __ BIND(L_copy_longs);
2608 #ifdef ASSERT
2609     { Label L;
2610       __ cmpl(rax_elsize, LogBytesPerLong);
2611       __ jcc(Assembler::equal, L);
2612       __ stop("must be long copy, but elsize is wrong");
2613       __ bind(L);
2614     }
2615 #endif
2616     __ leaq(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2617     __ leaq(to,   Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2618     __ movslq(count, r11_length); // length
2619     __ jump(RuntimeAddress(long_copy_entry));
2620 
2621     // objArrayKlass
2622   __ BIND(L_objArray);
2623     // live at this point:  r10_src_klass, src[_pos], dst[_pos]
2624 
2625     Label L_plain_copy, L_checkcast_copy;
2626     //  test array classes for subtyping
2627     __ load_klass(r9_dst_klass, dst);
2628     __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2629     __ jcc(Assembler::notEqual, L_checkcast_copy);
2630 
2631     // Identically typed arrays can be copied without element-wise checks.
2632     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2633                            r10, L_failed);
2634 
2635     __ leaq(from, Address(src, src_pos, TIMES_OOP,
2636                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2637     __ leaq(to,   Address(dst, dst_pos, TIMES_OOP,
2638                   arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2639     __ movslq(count, r11_length); // length
2640   __ BIND(L_plain_copy);
2641     __ jump(RuntimeAddress(oop_copy_entry));
2642 
2643   __ BIND(L_checkcast_copy);
2644     // live at this point:  r10_src_klass, !r11_length
2645     {
2646       // assert(r11_length == C_RARG4); // will reload from here
2647       Register r11_dst_klass = r11;
2648       __ load_klass(r11_dst_klass, dst);
2649 
2650       // Before looking at dst.length, make sure dst is also an objArray.
2651       __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2652       __ jcc(Assembler::notEqual, L_failed);
2653 
2654       // It is safe to examine both src.length and dst.length.
2655 #ifndef _WIN64
2656       arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2657                              rax, L_failed);
2658 #else
2659       __ movl(r11_length, C_RARG4);     // reload
2660       arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2661                              rax, L_failed);
2662       __ load_klass(r11_dst_klass, dst); // reload
2663 #endif
2664 
2665       // Marshal the base address arguments now, freeing registers.
2666       __ leaq(from, Address(src, src_pos, TIMES_OOP,
2667                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2668       __ leaq(to,   Address(dst, dst_pos, TIMES_OOP,
2669                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2670       __ movl(count, C_RARG4);          // length (reloaded)
2671       Register sco_temp = c_rarg3;      // this register is free now
2672       assert_different_registers(from, to, count, sco_temp,
2673                                  r11_dst_klass, r10_src_klass);
2674       assert_clean_int(count, sco_temp);
2675 
2676       // Generate the type check.
2677       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2678                         Klass::super_check_offset_offset_in_bytes());
2679       __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2680       assert_clean_int(sco_temp, rax);
2681       generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2682 
2683       // Fetch destination element klass from the objArrayKlass header.
2684       int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2685                        objArrayKlass::element_klass_offset_in_bytes());
2686       __ movq(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2687       __ movl(sco_temp,      Address(r11_dst_klass, sco_offset));
2688       assert_clean_int(sco_temp, rax);
2689 
2690       // the checkcast_copy loop needs two extra arguments:
2691       assert(c_rarg3 == sco_temp, "#3 already in place");
2692       __ movq(C_RARG4, r11_dst_klass);  // dst.klass.element_klass
2693       __ jump(RuntimeAddress(checkcast_copy_entry));
2694     }
2695 
2696   __ BIND(L_failed);
2697     __ xorq(rax, rax);
2698     __ notq(rax); // return -1
2699     __ leave();   // required for proper stackwalking of RuntimeStub frame
2700     __ ret(0);
2701 
2702     return start;
2703   }
2704 
2705 #undef length_arg
2706 
2707   void generate_arraycopy_stubs() {
2708     // Call the conjoint generation methods immediately after
2709     // the disjoint ones so that short branches from the former
2710     // to the latter can be generated.
2711     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2712     StubRoutines::_jbyte_arraycopy           = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2713 
2714     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2715     StubRoutines::_jshort_arraycopy          = generate_conjoint_short_copy(false, "jshort_arraycopy");
2716 
2717     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2718     StubRoutines::_jint_arraycopy            = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2719 
2720     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2721     StubRoutines::_jlong_arraycopy           = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2722 
2723 
2724     if (UseCompressedOops) {
2725       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2726       StubRoutines::_oop_arraycopy           = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2727     } else {
2728       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2729       StubRoutines::_oop_arraycopy           = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2730     }
2731 
2732     StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2733     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy");
2734     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy");
2735 
2736     // We don't generate specialized code for HeapWord-aligned source
2737     // arrays, so just use the code we've already generated
2738     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = StubRoutines::_jbyte_disjoint_arraycopy;
2739     StubRoutines::_arrayof_jbyte_arraycopy           = StubRoutines::_jbyte_arraycopy;
2740 
2741     StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2742     StubRoutines::_arrayof_jshort_arraycopy          = StubRoutines::_jshort_arraycopy;
2743 
2744     StubRoutines::_arrayof_jint_disjoint_arraycopy   = StubRoutines::_jint_disjoint_arraycopy;
2745     StubRoutines::_arrayof_jint_arraycopy            = StubRoutines::_jint_arraycopy;
2746 
2747     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = StubRoutines::_jlong_disjoint_arraycopy;
2748     StubRoutines::_arrayof_jlong_arraycopy           = StubRoutines::_jlong_arraycopy;
2749 
2750     StubRoutines::_arrayof_oop_disjoint_arraycopy    = StubRoutines::_oop_disjoint_arraycopy;
2751     StubRoutines::_arrayof_oop_arraycopy             = StubRoutines::_oop_arraycopy;
2752   }
2753 
2754 #undef __
2755 #define __ masm->
2756 
2757   // Continuation point for throwing of implicit exceptions that are
2758   // not handled in the current activation. Fabricates an exception
2759   // oop and initiates normal exception dispatching in this
2760   // frame. Since we need to preserve callee-saved values (currently
2761   // only for C2, but done for C1 as well) we need a callee-saved oop
2762   // map and therefore have to make these stubs into RuntimeStubs
2763   // rather than BufferBlobs.  If the compiler needs all registers to
2764   // be preserved between the fault point and the exception handler
2765   // then it must assume responsibility for that in
2766   // AbstractCompiler::continuation_for_implicit_null_exception or
2767   // continuation_for_implicit_division_by_zero_exception. All other
2768   // implicit exceptions (e.g., NullPointerException or
2769   // AbstractMethodError on entry) are either at call sites or
2770   // otherwise assume that stack unwinding will be initiated, so
2771   // caller saved registers were assumed volatile in the compiler.
2772   address generate_throw_exception(const char* name,
2773                                    address runtime_entry,
2774                                    bool restore_saved_exception_pc) {
2775     // Information about frame layout at time of blocking runtime call.
2776     // Note that we only have to preserve callee-saved registers since
2777     // the compilers are responsible for supplying a continuation point
2778     // if they expect all registers to be preserved.
2779     enum layout {
2780       rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2781       rbp_off2,
2782       return_off,
2783       return_off2,
2784       framesize // inclusive of return address
2785     };
2786 
2787     int insts_size = 512;
2788     int locs_size  = 64;
2789 
2790     CodeBuffer code(name, insts_size, locs_size);
2791     OopMapSet* oop_maps  = new OopMapSet();
2792     MacroAssembler* masm = new MacroAssembler(&code);
2793 
2794     address start = __ pc();
2795 
2796     // This is an inlined and slightly modified version of call_VM
2797     // which has the ability to fetch the return PC out of
2798     // thread-local storage and also sets up last_Java_sp slightly
2799     // differently than the real call_VM
2800     if (restore_saved_exception_pc) {
2801       __ movq(rax,
2802               Address(r15_thread,
2803                       in_bytes(JavaThread::saved_exception_pc_offset())));
2804       __ pushq(rax);
2805     }
2806 
2807     __ enter(); // required for proper stackwalking of RuntimeStub frame
2808 
2809     assert(is_even(framesize/2), "sp not 16-byte aligned");
2810 
2811     // return address and rbp are already in place
2812     __ subq(rsp, (framesize-4) << LogBytesPerInt); // prolog
2813 
2814     int frame_complete = __ pc() - start;
2815 
2816     // Set up last_Java_sp and last_Java_fp
2817     __ set_last_Java_frame(rsp, rbp, NULL);
2818 
2819     // Call runtime
2820     __ movq(c_rarg0, r15_thread);
2821     BLOCK_COMMENT("call runtime_entry");
2822     __ call(RuntimeAddress(runtime_entry));
2823 
2824     // Generate oop map
2825     OopMap* map = new OopMap(framesize, 0);
2826 
2827     oop_maps->add_gc_map(__ pc() - start, map);
2828 
2829     __ reset_last_Java_frame(true, false);
2830 
2831     __ leave(); // required for proper stackwalking of RuntimeStub frame
2832 
2833     // check for pending exceptions
2834 #ifdef ASSERT
2835     Label L;
2836     __ cmpq(Address(r15_thread, Thread::pending_exception_offset()),
2837             (int) NULL);
2838     __ jcc(Assembler::notEqual, L);
2839     __ should_not_reach_here();
2840     __ bind(L);
2841 #endif // ASSERT
2842     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2843 
2844 
2845     // codeBlob framesize is in words (not VMRegImpl::slot_size)
2846     RuntimeStub* stub =
2847       RuntimeStub::new_runtime_stub(name,
2848                                     &code,
2849                                     frame_complete,
2850                                     (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2851                                     oop_maps, false);
2852     return stub->entry_point();
2853   }
2854 
2855   // Initialization
2856   void generate_initial() {
2857     // Generates all stubs and initializes the entry points
2858 
2859     // This platform-specific stub is needed by generate_call_stub()
2860     StubRoutines::amd64::_mxcsr_std        = generate_fp_mask("mxcsr_std",        0x0000000000001F80);
2861 
2862     // entry points that exist in all platforms Note: This is code
2863     // that could be shared among different platforms - however the
2864     // benefit seems to be smaller than the disadvantage of having a
2865     // much more complicated generator structure. See also comment in
2866     // stubRoutines.hpp.
2867 
2868     StubRoutines::_forward_exception_entry = generate_forward_exception();
2869 
2870     StubRoutines::_call_stub_entry =
2871       generate_call_stub(StubRoutines::_call_stub_return_address);
2872 
2873     // is referenced by megamorphic call
2874     StubRoutines::_catch_exception_entry = generate_catch_exception();
2875 
2876     // atomic calls
2877     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
2878     StubRoutines::_atomic_xchg_ptr_entry     = generate_atomic_xchg_ptr();
2879     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
2880     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2881     StubRoutines::_atomic_add_entry          = generate_atomic_add();
2882     StubRoutines::_atomic_add_ptr_entry      = generate_atomic_add_ptr();
2883     StubRoutines::_fence_entry               = generate_orderaccess_fence();
2884 
2885     StubRoutines::_handler_for_unsafe_access_entry =
2886       generate_handler_for_unsafe_access();
2887 
2888     // platform dependent
2889     StubRoutines::amd64::_get_previous_fp_entry = generate_get_previous_fp();
2890 
2891     StubRoutines::amd64::_verify_mxcsr_entry    = generate_verify_mxcsr();
2892   }
2893 
2894   void generate_all() {
2895     // Generates all stubs and initializes the entry points
2896 
2897     // These entry points require SharedInfo::stack0 to be set up in
2898     // non-core builds and need to be relocatable, so they each
2899     // fabricate a RuntimeStub internally.
2900     StubRoutines::_throw_AbstractMethodError_entry =
2901       generate_throw_exception("AbstractMethodError throw_exception",
2902                                CAST_FROM_FN_PTR(address,
2903                                                 SharedRuntime::
2904                                                 throw_AbstractMethodError),
2905                                false);
2906 
2907     StubRoutines::_throw_IncompatibleClassChangeError_entry =
2908       generate_throw_exception("IncompatibleClassChangeError throw_exception",
2909                                CAST_FROM_FN_PTR(address,
2910                                                 SharedRuntime::
2911                                                 throw_IncompatibleClassChangeError),
2912                                false);
2913 
2914     StubRoutines::_throw_ArithmeticException_entry =
2915       generate_throw_exception("ArithmeticException throw_exception",
2916                                CAST_FROM_FN_PTR(address,
2917                                                 SharedRuntime::
2918                                                 throw_ArithmeticException),
2919                                true);
2920 
2921     StubRoutines::_throw_NullPointerException_entry =
2922       generate_throw_exception("NullPointerException throw_exception",
2923                                CAST_FROM_FN_PTR(address,
2924                                                 SharedRuntime::
2925                                                 throw_NullPointerException),
2926                                true);
2927 
2928     StubRoutines::_throw_NullPointerException_at_call_entry =
2929       generate_throw_exception("NullPointerException at call throw_exception",
2930                                CAST_FROM_FN_PTR(address,
2931                                                 SharedRuntime::
2932                                                 throw_NullPointerException_at_call),
2933                                false);
2934 
2935     StubRoutines::_throw_StackOverflowError_entry =
2936       generate_throw_exception("StackOverflowError throw_exception",
2937                                CAST_FROM_FN_PTR(address,
2938                                                 SharedRuntime::
2939                                                 throw_StackOverflowError),
2940                                false);
2941 
2942     // entry points that are platform specific
2943     StubRoutines::amd64::_f2i_fixup = generate_f2i_fixup();
2944     StubRoutines::amd64::_f2l_fixup = generate_f2l_fixup();
2945     StubRoutines::amd64::_d2i_fixup = generate_d2i_fixup();
2946     StubRoutines::amd64::_d2l_fixup = generate_d2l_fixup();
2947 
2948     StubRoutines::amd64::_float_sign_mask  = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
2949     StubRoutines::amd64::_float_sign_flip  = generate_fp_mask("float_sign_flip",  0x8000000080000000);
2950     StubRoutines::amd64::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
2951     StubRoutines::amd64::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
2952 
2953     // support for verify_oop (must happen after universe_init)
2954     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2955 
2956     // arraycopy stubs used by compilers
2957     generate_arraycopy_stubs();
2958   }
2959 
2960  public:
2961   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2962     if (all) {
2963       generate_all();
2964     } else {
2965       generate_initial();
2966     }
2967   }
2968 }; // end class declaration
2969 
2970 address StubGenerator::disjoint_byte_copy_entry  = NULL;
2971 address StubGenerator::disjoint_short_copy_entry = NULL;
2972 address StubGenerator::disjoint_int_copy_entry   = NULL;
2973 address StubGenerator::disjoint_long_copy_entry  = NULL;
2974 address StubGenerator::disjoint_oop_copy_entry   = NULL;
2975 
2976 address StubGenerator::byte_copy_entry  = NULL;
2977 address StubGenerator::short_copy_entry = NULL;
2978 address StubGenerator::int_copy_entry   = NULL;
2979 address StubGenerator::long_copy_entry  = NULL;
2980 address StubGenerator::oop_copy_entry   = NULL;
2981 
2982 address StubGenerator::checkcast_copy_entry = NULL;
2983 
2984 void StubGenerator_generate(CodeBuffer* code, bool all) {
2985   StubGenerator g(code, all);
2986 }