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   static address unsafe_copy_entry;
1048 
1049   //
1050   // Verify that a register contains clean 32-bits positive value
1051   // (high 32-bits are 0) so it could be used in 64-bits shifts.
1052   //
1053   //  Input:
1054   //    Rint  -  32-bits value
1055   //    Rtmp  -  scratch
1056   //
1057   void assert_clean_int(Register Rint, Register Rtmp) {
1058 #ifdef ASSERT
1059     Label L;
1060     assert_different_registers(Rtmp, Rint);
1061     __ movslq(Rtmp, Rint);
1062     __ cmpq(Rtmp, Rint);
1063     __ jcc(Assembler::equal, L);
1064     __ stop("high 32-bits of int value are not 0");
1065     __ bind(L);
1066 #endif
1067   }
1068 
1069   //  Generate overlap test for array copy stubs
1070   //
1071   //  Input:
1072   //     c_rarg0 - from
1073   //     c_rarg1 - to
1074   //     c_rarg2 - element count
1075   //
1076   //  Output:
1077   //     rax   - &from[element count - 1]
1078   //
1079   void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1080     assert(no_overlap_target != NULL, "must be generated");
1081     array_overlap_test(no_overlap_target, NULL, sf);
1082   }
1083   void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1084     array_overlap_test(NULL, &L_no_overlap, sf);
1085   }
1086   void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1087     const Register from     = c_rarg0;
1088     const Register to       = c_rarg1;
1089     const Register count    = c_rarg2;
1090     const Register end_from = rax;
1091 
1092     __ cmpq(to, from);
1093     __ leaq(end_from, Address(from, count, sf, 0));
1094     if (NOLp == NULL) {
1095       ExternalAddress no_overlap(no_overlap_target);
1096       __ jump_cc(Assembler::belowEqual, no_overlap);
1097       __ cmpq(to, end_from);
1098       __ jump_cc(Assembler::aboveEqual, no_overlap);
1099     } else {
1100       __ jcc(Assembler::belowEqual, (*NOLp));
1101       __ cmpq(to, end_from);
1102       __ jcc(Assembler::aboveEqual, (*NOLp));
1103     }
1104   }
1105 
1106   // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1107   //
1108   // Outputs:
1109   //    rdi - rcx
1110   //    rsi - rdx
1111   //    rdx - r8
1112   //    rcx - r9
1113   //
1114   // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1115   // are non-volatile.  r9 and r10 should not be used by the caller.
1116   //
1117   void setup_arg_regs(int nargs = 3) {
1118     const Register saved_rdi = r9;
1119     const Register saved_rsi = r10;
1120     assert(nargs == 3 || nargs == 4, "else fix");
1121 #ifdef _WIN64
1122     assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1123            "unexpected argument registers");
1124     if (nargs >= 4)
1125       __ movq(rax, r9);  // r9 is also saved_rdi
1126     __ movq(saved_rdi, rdi);
1127     __ movq(saved_rsi, rsi);
1128     __ movq(rdi, rcx); // c_rarg0
1129     __ movq(rsi, rdx); // c_rarg1
1130     __ movq(rdx, r8);  // c_rarg2
1131     if (nargs >= 4)
1132       __ movq(rcx, rax); // c_rarg3 (via rax)
1133 #else
1134     assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1135            "unexpected argument registers");
1136 #endif
1137   }
1138 
1139   void restore_arg_regs() {
1140     const Register saved_rdi = r9;
1141     const Register saved_rsi = r10;
1142 #ifdef _WIN64
1143     __ movq(rdi, saved_rdi);
1144     __ movq(rsi, saved_rsi);
1145 #endif
1146   }
1147 
1148   // Generate code for an array write pre barrier
1149   //
1150   //     addr    -  starting address
1151   //     count    -  element count
1152   //
1153   //     Destroy no registers!
1154   //
1155   void  gen_write_ref_array_pre_barrier(Register addr, Register count) {
1156 #if 0 // G1 - only
1157     assert_different_registers(addr, c_rarg1);
1158     assert_different_registers(count, c_rarg0);
1159     BarrierSet* bs = Universe::heap()->barrier_set();
1160     switch (bs->kind()) {
1161       case BarrierSet::G1SATBCT:
1162       case BarrierSet::G1SATBCTLogging:
1163         {
1164           __ pushaq();                      // push registers
1165           __ movq(c_rarg0, addr);
1166           __ movq(c_rarg1, count);
1167           __ call(RuntimeAddress(BarrierSet::static_write_ref_array_pre));
1168           __ popaq();
1169         }
1170         break;
1171       case BarrierSet::CardTableModRef:
1172       case BarrierSet::CardTableExtension:
1173       case BarrierSet::ModRef:
1174         break;
1175       default      :
1176         ShouldNotReachHere();
1177 
1178     }
1179 #endif // 0 G1 - only
1180   }
1181 
1182   //
1183   // Generate code for an array write post barrier
1184   //
1185   //  Input:
1186   //     start    - register containing starting address of destination array
1187   //     end      - register containing ending address of destination array
1188   //     scratch  - scratch register
1189   //
1190   //  The input registers are overwritten.
1191   //  The ending address is inclusive.
1192   void  gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1193     assert_different_registers(start, end, scratch);
1194     BarrierSet* bs = Universe::heap()->barrier_set();
1195     switch (bs->kind()) {
1196 #if 0 // G1 - only
1197       case BarrierSet::G1SATBCT:
1198       case BarrierSet::G1SATBCTLogging:
1199 
1200         {
1201           __ pushaq();                      // push registers (overkill)
1202           // must compute element count unless barrier set interface is changed (other platforms supply count)
1203           assert_different_registers(start, end, scratch);
1204           __ leaq(scratch, Address(end, wordSize));
1205           __ subq(scratch, start);
1206           __ shrq(scratch, LogBytesPerWord);
1207           __ movq(c_rarg0, start);
1208           __ movq(c_rarg1, scratch);
1209           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
1210           __ popaq();
1211         }
1212         break;
1213 #endif // 0 G1 - only
1214       case BarrierSet::CardTableModRef:
1215       case BarrierSet::CardTableExtension:
1216         {
1217           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1218           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1219 
1220           Label L_loop;
1221 
1222            __ shrq(start, CardTableModRefBS::card_shift);
1223            __ shrq(end, CardTableModRefBS::card_shift);
1224            __ subq(end, start); // number of bytes to copy
1225 
1226           const Register count = end; // 'end' register contains bytes count now
1227           __ lea(scratch, ExternalAddress((address)ct->byte_map_base));
1228           __ addq(start, scratch);
1229         __ BIND(L_loop);
1230           __ movb(Address(start, count, Address::times_1), 0);
1231           __ decrementq(count);
1232           __ jcc(Assembler::greaterEqual, L_loop);
1233         }
1234       }
1235    }
1236 
1237   // Copy big chunks forward
1238   //
1239   // Inputs:
1240   //   end_from     - source arrays end address
1241   //   end_to       - destination array end address
1242   //   qword_count  - 64-bits element count, negative
1243   //   to           - scratch
1244   //   L_copy_32_bytes - entry label
1245   //   L_copy_8_bytes  - exit  label
1246   //
1247   void copy_32_bytes_forward(Register end_from, Register end_to,
1248                              Register qword_count, Register to,
1249                              Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1250     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1251     Label L_loop;
1252     __ align(16);
1253   __ BIND(L_loop);
1254     __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1255     __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1256     __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1257     __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1258     __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1259     __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1260     __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1261     __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1262   __ BIND(L_copy_32_bytes);
1263     __ addq(qword_count, 4);
1264     __ jcc(Assembler::lessEqual, L_loop);
1265     __ subq(qword_count, 4);
1266     __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1267   }
1268 
1269 
1270   // Copy big chunks backward
1271   //
1272   // Inputs:
1273   //   from         - source arrays address
1274   //   dest         - destination array address
1275   //   qword_count  - 64-bits element count
1276   //   to           - scratch
1277   //   L_copy_32_bytes - entry label
1278   //   L_copy_8_bytes  - exit  label
1279   //
1280   void copy_32_bytes_backward(Register from, Register dest,
1281                               Register qword_count, Register to,
1282                               Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1283     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1284     Label L_loop;
1285     __ align(16);
1286   __ BIND(L_loop);
1287     __ movq(to, Address(from, qword_count, Address::times_8, 24));
1288     __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1289     __ movq(to, Address(from, qword_count, Address::times_8, 16));
1290     __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1291     __ movq(to, Address(from, qword_count, Address::times_8,  8));
1292     __ movq(Address(dest, qword_count, Address::times_8,  8), to);
1293     __ movq(to, Address(from, qword_count, Address::times_8,  0));
1294     __ movq(Address(dest, qword_count, Address::times_8,  0), to);
1295   __ BIND(L_copy_32_bytes);
1296     __ subq(qword_count, 4);
1297     __ jcc(Assembler::greaterEqual, L_loop);
1298     __ addq(qword_count, 4);
1299     __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1300   }
1301 
1302 
1303   // Arguments:
1304   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1305   //             ignored
1306   //   name    - stub name string
1307   //
1308   // Inputs:
1309   //   c_rarg0   - source array address
1310   //   c_rarg1   - destination array address
1311   //   c_rarg2   - element count, treated as ssize_t, can be zero
1312   //
1313   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1314   // we let the hardware handle it.  The one to eight bytes within words,
1315   // dwords or qwords that span cache line boundaries will still be loaded
1316   // and stored atomically.
1317   //
1318   // Side Effects:
1319   //   disjoint_byte_copy_entry is set to the no-overlap entry point
1320   //   used by generate_conjoint_byte_copy().
1321   //
1322   address generate_disjoint_byte_copy(bool aligned, const char *name) {
1323     __ align(CodeEntryAlignment);
1324     StubCodeMark mark(this, "StubRoutines", name);
1325     address start = __ pc();
1326 
1327     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1328     Label L_copy_byte, L_exit;
1329     const Register from        = rdi;  // source array address
1330     const Register to          = rsi;  // destination array address
1331     const Register count       = rdx;  // elements count
1332     const Register byte_count  = rcx;
1333     const Register qword_count = count;
1334     const Register end_from    = from; // source array end address
1335     const Register end_to      = to;   // destination array end address
1336     // End pointers are inclusive, and if count is not zero they point
1337     // to the last unit copied:  end_to[0] := end_from[0]
1338 
1339     __ enter(); // required for proper stackwalking of RuntimeStub frame
1340     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1341 
1342     disjoint_byte_copy_entry = __ pc();
1343     BLOCK_COMMENT("Entry:");
1344     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1345 
1346     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1347                       // r9 and r10 may be used to save non-volatile registers
1348 
1349     // 'from', 'to' and 'count' are now valid
1350     __ movq(byte_count, count);
1351     __ shrq(count, 3); // count => qword_count
1352 
1353     // Copy from low to high addresses.  Use 'to' as scratch.
1354     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1355     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1356     __ negq(qword_count); // make the count negative
1357     __ jmp(L_copy_32_bytes);
1358 
1359     // Copy trailing qwords
1360   __ BIND(L_copy_8_bytes);
1361     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1362     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1363     __ incrementq(qword_count);
1364     __ jcc(Assembler::notZero, L_copy_8_bytes);
1365 
1366     // Check for and copy trailing dword
1367   __ BIND(L_copy_4_bytes);
1368     __ testq(byte_count, 4);
1369     __ jccb(Assembler::zero, L_copy_2_bytes);
1370     __ movl(rax, Address(end_from, 8));
1371     __ movl(Address(end_to, 8), rax);
1372 
1373     __ addq(end_from, 4);
1374     __ addq(end_to, 4);
1375 
1376     // Check for and copy trailing word
1377   __ BIND(L_copy_2_bytes);
1378     __ testq(byte_count, 2);
1379     __ jccb(Assembler::zero, L_copy_byte);
1380     __ movw(rax, Address(end_from, 8));
1381     __ movw(Address(end_to, 8), rax);
1382 
1383     __ addq(end_from, 2);
1384     __ addq(end_to, 2);
1385 
1386     // Check for and copy trailing byte
1387   __ BIND(L_copy_byte);
1388     __ testq(byte_count, 1);
1389     __ jccb(Assembler::zero, L_exit);
1390     __ movb(rax, Address(end_from, 8));
1391     __ movb(Address(end_to, 8), rax);
1392 
1393   __ BIND(L_exit);
1394     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1395     restore_arg_regs();
1396     __ xorq(rax, rax); // return 0
1397     __ leave(); // required for proper stackwalking of RuntimeStub frame
1398     __ ret(0);
1399 
1400     // Copy in 32-bytes chunks
1401     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1402     __ jmp(L_copy_4_bytes);
1403 
1404     return start;
1405   }
1406 
1407   // Arguments:
1408   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1409   //             ignored
1410   //   name    - stub name string
1411   //
1412   // Inputs:
1413   //   c_rarg0   - source array address
1414   //   c_rarg1   - destination array address
1415   //   c_rarg2   - element count, treated as ssize_t, can be zero
1416   //
1417   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1418   // we let the hardware handle it.  The one to eight bytes within words,
1419   // dwords or qwords that span cache line boundaries will still be loaded
1420   // and stored atomically.
1421   //
1422   address generate_conjoint_byte_copy(bool aligned, const char *name) {
1423     __ align(CodeEntryAlignment);
1424     StubCodeMark mark(this, "StubRoutines", name);
1425     address start = __ pc();
1426 
1427     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1428     const Register from        = rdi;  // source array address
1429     const Register to          = rsi;  // destination array address
1430     const Register count       = rdx;  // elements count
1431     const Register byte_count  = rcx;
1432     const Register qword_count = count;
1433 
1434     __ enter(); // required for proper stackwalking of RuntimeStub frame
1435     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1436 
1437     byte_copy_entry = __ pc();
1438     BLOCK_COMMENT("Entry:");
1439     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1440 
1441     array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1442     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1443                       // r9 and r10 may be used to save non-volatile registers
1444 
1445     // 'from', 'to' and 'count' are now valid
1446     __ movq(byte_count, count);
1447     __ shrq(count, 3);   // count => qword_count
1448 
1449     // Copy from high to low addresses.
1450 
1451     // Check for and copy trailing byte
1452     __ testq(byte_count, 1);
1453     __ jcc(Assembler::zero, L_copy_2_bytes);
1454     __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1455     __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1456     __ decrementq(byte_count); // Adjust for possible trailing word
1457 
1458     // Check for and copy trailing word
1459   __ BIND(L_copy_2_bytes);
1460     __ testq(byte_count, 2);
1461     __ jcc(Assembler::zero, L_copy_4_bytes);
1462     __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1463     __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1464 
1465     // Check for and copy trailing dword
1466   __ BIND(L_copy_4_bytes);
1467     __ testq(byte_count, 4);
1468     __ jcc(Assembler::zero, L_copy_32_bytes);
1469     __ movl(rax, Address(from, qword_count, Address::times_8));
1470     __ movl(Address(to, qword_count, Address::times_8), rax);
1471     __ jmp(L_copy_32_bytes);
1472 
1473     // Copy trailing qwords
1474   __ BIND(L_copy_8_bytes);
1475     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1476     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1477     __ decrementq(qword_count);
1478     __ jcc(Assembler::notZero, L_copy_8_bytes);
1479 
1480     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1481     restore_arg_regs();
1482     __ xorq(rax, rax); // return 0
1483     __ leave(); // required for proper stackwalking of RuntimeStub frame
1484     __ ret(0);
1485 
1486     // Copy in 32-bytes chunks
1487     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1488 
1489     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1490     restore_arg_regs();
1491     __ xorq(rax, rax); // return 0
1492     __ leave(); // required for proper stackwalking of RuntimeStub frame
1493     __ ret(0);
1494 
1495     return start;
1496   }
1497 
1498   // Arguments:
1499   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1500   //             ignored
1501   //   name    - stub name string
1502   //
1503   // Inputs:
1504   //   c_rarg0   - source array address
1505   //   c_rarg1   - destination array address
1506   //   c_rarg2   - element count, treated as ssize_t, can be zero
1507   //
1508   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1509   // let the hardware handle it.  The two or four words within dwords
1510   // or qwords that span cache line boundaries will still be loaded
1511   // and stored atomically.
1512   //
1513   // Side Effects:
1514   //   disjoint_short_copy_entry is set to the no-overlap entry point
1515   //   used by generate_conjoint_short_copy().
1516   //
1517   address generate_disjoint_short_copy(bool aligned, const char *name) {
1518     __ align(CodeEntryAlignment);
1519     StubCodeMark mark(this, "StubRoutines", name);
1520     address start = __ pc();
1521 
1522     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1523     const Register from        = rdi;  // source array address
1524     const Register to          = rsi;  // destination array address
1525     const Register count       = rdx;  // elements count
1526     const Register word_count  = rcx;
1527     const Register qword_count = count;
1528     const Register end_from    = from; // source array end address
1529     const Register end_to      = to;   // destination array end address
1530     // End pointers are inclusive, and if count is not zero they point
1531     // to the last unit copied:  end_to[0] := end_from[0]
1532 
1533     __ enter(); // required for proper stackwalking of RuntimeStub frame
1534     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1535 
1536     disjoint_short_copy_entry = __ pc();
1537     BLOCK_COMMENT("Entry:");
1538     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1539 
1540     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1541                       // r9 and r10 may be used to save non-volatile registers
1542 
1543     // 'from', 'to' and 'count' are now valid
1544     __ movq(word_count, count);
1545     __ shrq(count, 2); // count => qword_count
1546 
1547     // Copy from low to high addresses.  Use 'to' as scratch.
1548     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1549     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1550     __ negq(qword_count);
1551     __ jmp(L_copy_32_bytes);
1552 
1553     // Copy trailing qwords
1554   __ BIND(L_copy_8_bytes);
1555     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1556     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1557     __ incrementq(qword_count);
1558     __ jcc(Assembler::notZero, L_copy_8_bytes);
1559 
1560     // Original 'dest' is trashed, so we can't use it as a
1561     // base register for a possible trailing word copy
1562 
1563     // Check for and copy trailing dword
1564   __ BIND(L_copy_4_bytes);
1565     __ testq(word_count, 2);
1566     __ jccb(Assembler::zero, L_copy_2_bytes);
1567     __ movl(rax, Address(end_from, 8));
1568     __ movl(Address(end_to, 8), rax);
1569 
1570     __ addq(end_from, 4);
1571     __ addq(end_to, 4);
1572 
1573     // Check for and copy trailing word
1574   __ BIND(L_copy_2_bytes);
1575     __ testq(word_count, 1);
1576     __ jccb(Assembler::zero, L_exit);
1577     __ movw(rax, Address(end_from, 8));
1578     __ movw(Address(end_to, 8), rax);
1579 
1580   __ BIND(L_exit);
1581     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1582     restore_arg_regs();
1583     __ xorq(rax, rax); // return 0
1584     __ leave(); // required for proper stackwalking of RuntimeStub frame
1585     __ ret(0);
1586 
1587     // Copy in 32-bytes chunks
1588     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1589     __ jmp(L_copy_4_bytes);
1590 
1591     return start;
1592   }
1593 
1594   // Arguments:
1595   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1596   //             ignored
1597   //   name    - stub name string
1598   //
1599   // Inputs:
1600   //   c_rarg0   - source array address
1601   //   c_rarg1   - destination array address
1602   //   c_rarg2   - element count, treated as ssize_t, can be zero
1603   //
1604   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1605   // let the hardware handle it.  The two or four words within dwords
1606   // or qwords that span cache line boundaries will still be loaded
1607   // and stored atomically.
1608   //
1609   address generate_conjoint_short_copy(bool aligned, const char *name) {
1610     __ align(CodeEntryAlignment);
1611     StubCodeMark mark(this, "StubRoutines", name);
1612     address start = __ pc();
1613 
1614     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1615     const Register from        = rdi;  // source array address
1616     const Register to          = rsi;  // destination array address
1617     const Register count       = rdx;  // elements count
1618     const Register word_count  = rcx;
1619     const Register qword_count = count;
1620 
1621     __ enter(); // required for proper stackwalking of RuntimeStub frame
1622     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1623 
1624     short_copy_entry = __ pc();
1625     BLOCK_COMMENT("Entry:");
1626     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1627 
1628     array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1629     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1630                       // r9 and r10 may be used to save non-volatile registers
1631 
1632     // 'from', 'to' and 'count' are now valid
1633     __ movq(word_count, count);
1634     __ shrq(count, 2); // count => qword_count
1635 
1636     // Copy from high to low addresses.  Use 'to' as scratch.
1637 
1638     // Check for and copy trailing word
1639     __ testq(word_count, 1);
1640     __ jccb(Assembler::zero, L_copy_4_bytes);
1641     __ movw(rax, Address(from, word_count, Address::times_2, -2));
1642     __ movw(Address(to, word_count, Address::times_2, -2), rax);
1643 
1644     // Check for and copy trailing dword
1645   __ BIND(L_copy_4_bytes);
1646     __ testq(word_count, 2);
1647     __ jcc(Assembler::zero, L_copy_32_bytes);
1648     __ movl(rax, Address(from, qword_count, Address::times_8));
1649     __ movl(Address(to, qword_count, Address::times_8), rax);
1650     __ jmp(L_copy_32_bytes);
1651 
1652     // Copy trailing qwords
1653   __ BIND(L_copy_8_bytes);
1654     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1655     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1656     __ decrementq(qword_count);
1657     __ jcc(Assembler::notZero, L_copy_8_bytes);
1658 
1659     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1660     restore_arg_regs();
1661     __ xorq(rax, rax); // return 0
1662     __ leave(); // required for proper stackwalking of RuntimeStub frame
1663     __ ret(0);
1664 
1665     // Copy in 32-bytes chunks
1666     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1667 
1668     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1669     restore_arg_regs();
1670     __ xorq(rax, rax); // return 0
1671     __ leave(); // required for proper stackwalking of RuntimeStub frame
1672     __ ret(0);
1673 
1674     return start;
1675   }
1676 
1677   // Arguments:
1678   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1679   //             ignored
1680   //   is_oop  - true => oop array, so generate store check code
1681   //   name    - stub name string
1682   //
1683   // Inputs:
1684   //   c_rarg0   - source array address
1685   //   c_rarg1   - destination array address
1686   //   c_rarg2   - element count, treated as ssize_t, can be zero
1687   //
1688   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1689   // the hardware handle it.  The two dwords within qwords that span
1690   // cache line boundaries will still be loaded and stored atomicly.
1691   //
1692   // Side Effects:
1693   //   disjoint_int_copy_entry is set to the no-overlap entry point
1694   //   used by generate_conjoint_int_oop_copy().
1695   //
1696   address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1697     __ align(CodeEntryAlignment);
1698     StubCodeMark mark(this, "StubRoutines", name);
1699     address start = __ pc();
1700 
1701     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1702     const Register from        = rdi;  // source array address
1703     const Register to          = rsi;  // destination array address
1704     const Register count       = rdx;  // elements count
1705     const Register dword_count = rcx;
1706     const Register qword_count = count;
1707     const Register end_from    = from; // source array end address
1708     const Register end_to      = to;   // destination array end address
1709     const Register saved_to    = r11;  // saved destination array address
1710     // End pointers are inclusive, and if count is not zero they point
1711     // to the last unit copied:  end_to[0] := end_from[0]
1712 
1713     __ enter(); // required for proper stackwalking of RuntimeStub frame
1714     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1715 
1716     (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1717 
1718     if (is_oop) {
1719       // no registers are destroyed by this call
1720       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1721     }
1722 
1723     BLOCK_COMMENT("Entry:");
1724     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1725 
1726     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1727                       // r9 and r10 may be used to save non-volatile registers
1728 
1729     if (is_oop) {
1730       __ movq(saved_to, to);
1731     }
1732 
1733     // 'from', 'to' and 'count' are now valid
1734     __ movq(dword_count, count);
1735     __ shrq(count, 1); // count => qword_count
1736 
1737     // Copy from low to high addresses.  Use 'to' as scratch.
1738     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1739     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1740     __ negq(qword_count);
1741     __ jmp(L_copy_32_bytes);
1742 
1743     // Copy trailing qwords
1744   __ BIND(L_copy_8_bytes);
1745     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1746     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1747     __ incrementq(qword_count);
1748     __ jcc(Assembler::notZero, L_copy_8_bytes);
1749 
1750     // Check for and copy trailing dword
1751   __ BIND(L_copy_4_bytes);
1752     __ testq(dword_count, 1); // Only byte test since the value is 0 or 1
1753     __ jccb(Assembler::zero, L_exit);
1754     __ movl(rax, Address(end_from, 8));
1755     __ movl(Address(end_to, 8), rax);
1756 
1757   __ BIND(L_exit);
1758     if (is_oop) {
1759       __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1760       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1761     }
1762     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1763     restore_arg_regs();
1764     __ xorq(rax, rax); // return 0
1765     __ leave(); // required for proper stackwalking of RuntimeStub frame
1766     __ ret(0);
1767 
1768     // Copy 32-bytes chunks
1769     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1770     __ jmp(L_copy_4_bytes);
1771 
1772     return start;
1773   }
1774 
1775   // Arguments:
1776   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1777   //             ignored
1778   //   is_oop  - true => oop array, so generate store check code
1779   //   name    - stub name string
1780   //
1781   // Inputs:
1782   //   c_rarg0   - source array address
1783   //   c_rarg1   - destination array address
1784   //   c_rarg2   - element count, treated as ssize_t, can be zero
1785   //
1786   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1787   // the hardware handle it.  The two dwords within qwords that span
1788   // cache line boundaries will still be loaded and stored atomicly.
1789   //
1790   address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1791     __ align(CodeEntryAlignment);
1792     StubCodeMark mark(this, "StubRoutines", name);
1793     address start = __ pc();
1794 
1795     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1796     const Register from        = rdi;  // source array address
1797     const Register to          = rsi;  // destination array address
1798     const Register count       = rdx;  // elements count
1799     const Register dword_count = rcx;
1800     const Register qword_count = count;
1801 
1802     __ enter(); // required for proper stackwalking of RuntimeStub frame
1803     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1804 
1805     if (is_oop) {
1806       // no registers are destroyed by this call
1807       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1808     }
1809 
1810     (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1811     BLOCK_COMMENT("Entry:");
1812     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1813 
1814     array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1815                        Address::times_4);
1816     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1817                       // r9 and r10 may be used to save non-volatile registers
1818 
1819     assert_clean_int(count, rax); // Make sure 'count' is clean int.
1820     // 'from', 'to' and 'count' are now valid
1821     __ movq(dword_count, count);
1822     __ shrq(count, 1); // count => qword_count
1823 
1824     // Copy from high to low addresses.  Use 'to' as scratch.
1825 
1826     // Check for and copy trailing dword
1827     __ testq(dword_count, 1);
1828     __ jcc(Assembler::zero, L_copy_32_bytes);
1829     __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1830     __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1831     __ jmp(L_copy_32_bytes);
1832 
1833     // Copy trailing qwords
1834   __ BIND(L_copy_8_bytes);
1835     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1836     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1837     __ decrementq(qword_count);
1838     __ jcc(Assembler::notZero, L_copy_8_bytes);
1839 
1840     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1841     if (is_oop) {
1842       __ jmp(L_exit);
1843     }
1844     restore_arg_regs();
1845     __ xorq(rax, rax); // return 0
1846     __ leave(); // required for proper stackwalking of RuntimeStub frame
1847     __ ret(0);
1848 
1849     // Copy in 32-bytes chunks
1850     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1851 
1852    inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1853    __ bind(L_exit);
1854      if (is_oop) {
1855        Register end_to = rdx;
1856        __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1857        gen_write_ref_array_post_barrier(to, end_to, rax);
1858      }
1859     restore_arg_regs();
1860     __ xorq(rax, rax); // return 0
1861     __ leave(); // required for proper stackwalking of RuntimeStub frame
1862     __ ret(0);
1863 
1864     return start;
1865   }
1866 
1867   // Arguments:
1868   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1869   //             ignored
1870   //   is_oop  - true => oop array, so generate store check code
1871   //   name    - stub name string
1872   //
1873   // Inputs:
1874   //   c_rarg0   - source array address
1875   //   c_rarg1   - destination array address
1876   //   c_rarg2   - element count, treated as ssize_t, can be zero
1877   //
1878  // Side Effects:
1879   //   disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1880   //   no-overlap entry point used by generate_conjoint_long_oop_copy().
1881   //
1882   address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1883     __ align(CodeEntryAlignment);
1884     StubCodeMark mark(this, "StubRoutines", name);
1885     address start = __ pc();
1886 
1887     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1888     const Register from        = rdi;  // source array address
1889     const Register to          = rsi;  // destination array address
1890     const Register qword_count = rdx;  // elements count
1891     const Register end_from    = from; // source array end address
1892     const Register end_to      = rcx;  // destination array end address
1893     const Register saved_to    = to;
1894     // End pointers are inclusive, and if count is not zero they point
1895     // to the last unit copied:  end_to[0] := end_from[0]
1896 
1897     __ enter(); // required for proper stackwalking of RuntimeStub frame
1898     // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1899     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1900 
1901     if (is_oop) {
1902       disjoint_oop_copy_entry  = __ pc();
1903       // no registers are destroyed by this call
1904       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1905     } else {
1906       disjoint_long_copy_entry = __ pc();
1907     }
1908     BLOCK_COMMENT("Entry:");
1909     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1910 
1911     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1912                       // r9 and r10 may be used to save non-volatile registers
1913 
1914     // 'from', 'to' and 'qword_count' are now valid
1915 
1916     // Copy from low to high addresses.  Use 'to' as scratch.
1917     __ leaq(end_from, Address(from, qword_count, Address::times_8, -8));
1918     __ leaq(end_to,   Address(to,   qword_count, Address::times_8, -8));
1919     __ negq(qword_count);
1920     __ jmp(L_copy_32_bytes);
1921 
1922     // Copy trailing qwords
1923   __ BIND(L_copy_8_bytes);
1924     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1925     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1926     __ incrementq(qword_count);
1927     __ jcc(Assembler::notZero, L_copy_8_bytes);
1928 
1929     if (is_oop) {
1930       __ jmp(L_exit);
1931     } else {
1932       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1933       restore_arg_regs();
1934       __ xorq(rax, rax); // return 0
1935       __ leave(); // required for proper stackwalking of RuntimeStub frame
1936       __ ret(0);
1937     }
1938 
1939     // Copy 64-byte chunks
1940     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1941 
1942     if (is_oop) {
1943     __ BIND(L_exit);
1944       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1945       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1946     } else {
1947       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1948     }
1949     restore_arg_regs();
1950     __ xorq(rax, rax); // return 0
1951     __ leave(); // required for proper stackwalking of RuntimeStub frame
1952     __ ret(0);
1953 
1954     return start;
1955   }
1956 
1957   // Arguments:
1958   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1959   //             ignored
1960   //   is_oop  - true => oop array, so generate store check code
1961   //   name    - stub name string
1962   //
1963   // Inputs:
1964   //   c_rarg0   - source array address
1965   //   c_rarg1   - destination array address
1966   //   c_rarg2   - element count, treated as ssize_t, can be zero
1967   //
1968   address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1969     __ align(CodeEntryAlignment);
1970     StubCodeMark mark(this, "StubRoutines", name);
1971     address start = __ pc();
1972 
1973     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1974     const Register from        = rdi;  // source array address
1975     const Register to          = rsi;  // destination array address
1976     const Register qword_count = rdx;  // elements count
1977     const Register saved_count = rcx;
1978 
1979     __ enter(); // required for proper stackwalking of RuntimeStub frame
1980     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1981 
1982     address disjoint_copy_entry = NULL;
1983     if (is_oop) {
1984       assert(!UseCompressedOops, "shouldn't be called for compressed oops");
1985       disjoint_copy_entry = disjoint_oop_copy_entry;
1986       oop_copy_entry  = __ pc();
1987       array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
1988     } else {
1989       disjoint_copy_entry = disjoint_long_copy_entry;
1990       long_copy_entry = __ pc();
1991       array_overlap_test(disjoint_long_copy_entry, Address::times_8);
1992     }
1993     BLOCK_COMMENT("Entry:");
1994     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1995 
1996     array_overlap_test(disjoint_copy_entry, Address::times_8);
1997     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1998                       // r9 and r10 may be used to save non-volatile registers
1999 
2000     // 'from', 'to' and 'qword_count' are now valid
2001 
2002     if (is_oop) {
2003       // Save to and count for store barrier
2004       __ movq(saved_count, qword_count);
2005       // No registers are destroyed by this call
2006       gen_write_ref_array_pre_barrier(to, saved_count);
2007     }
2008 
2009     __ jmp(L_copy_32_bytes);
2010 
2011     // Copy trailing qwords
2012   __ BIND(L_copy_8_bytes);
2013     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2014     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2015     __ decrementq(qword_count);
2016     __ jcc(Assembler::notZero, L_copy_8_bytes);
2017 
2018     if (is_oop) {
2019       __ jmp(L_exit);
2020     } else {
2021       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2022       restore_arg_regs();
2023       __ xorq(rax, rax); // return 0
2024       __ leave(); // required for proper stackwalking of RuntimeStub frame
2025       __ ret(0);
2026     }
2027 
2028     // Copy in 32-bytes chunks
2029     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2030 
2031     if (is_oop) {
2032     __ BIND(L_exit);
2033       __ leaq(rcx, Address(to, saved_count, Address::times_8, -8));
2034       gen_write_ref_array_post_barrier(to, rcx, rax);
2035       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2036     } else {
2037       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2038     }
2039     restore_arg_regs();
2040     __ xorq(rax, rax); // return 0
2041     __ leave(); // required for proper stackwalking of RuntimeStub frame
2042     __ ret(0);
2043 
2044     return start;
2045   }
2046 
2047 
2048   // Helper for generating a dynamic type check.
2049   // Smashes no registers.
2050   void generate_type_check(Register sub_klass,
2051                            Register super_check_offset,
2052                            Register super_klass,
2053                            Label& L_success) {
2054     assert_different_registers(sub_klass, super_check_offset, super_klass);
2055 
2056     BLOCK_COMMENT("type_check:");
2057 
2058     Label L_miss;
2059 
2060     // a couple of useful fields in sub_klass:
2061     int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
2062                      Klass::secondary_supers_offset_in_bytes());
2063     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
2064                      Klass::secondary_super_cache_offset_in_bytes());
2065     Address secondary_supers_addr(sub_klass, ss_offset);
2066     Address super_cache_addr(     sub_klass, sc_offset);
2067 
2068     // if the pointers are equal, we are done (e.g., String[] elements)
2069     __ cmpq(super_klass, sub_klass);
2070     __ jcc(Assembler::equal, L_success);
2071 
2072     // check the supertype display:
2073     Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
2074     __ cmpq(super_klass, super_check_addr); // test the super type
2075     __ jcc(Assembler::equal, L_success);
2076 
2077     // if it was a primary super, we can just fail immediately
2078     __ cmpl(super_check_offset, sc_offset);
2079     __ jcc(Assembler::notEqual, L_miss);
2080 
2081     // Now do a linear scan of the secondary super-klass chain.
2082     // The repne_scan instruction uses fixed registers, which we must spill.
2083     // (We need a couple more temps in any case.)
2084     // This code is rarely used, so simplicity is a virtue here.
2085     inc_counter_np(SharedRuntime::_partial_subtype_ctr);
2086     {
2087       __ pushq(rax);
2088       __ pushq(rcx);
2089       __ pushq(rdi);
2090       assert_different_registers(sub_klass, super_klass, rax, rcx, rdi);
2091 
2092       __ movq(rdi, secondary_supers_addr);
2093       // Load the array length.
2094       __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
2095       // Skip to start of data.
2096       __ addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
2097       // Scan rcx words at [rdi] for occurance of rax
2098       // Set NZ/Z based on last compare
2099       __ movq(rax, super_klass);
2100       if (UseCompressedOops) {
2101         // Compare against compressed form.  Don't need to uncompress because
2102         // looks like orig rax is restored in popq below.
2103         __ encode_heap_oop(rax);
2104         __ repne_scanl();
2105       } else {
2106          __ repne_scanq();
2107       }
2108 
2109       // Unspill the temp. registers:
2110       __ popq(rdi);
2111       __ popq(rcx);
2112       __ popq(rax);
2113 
2114       __ jcc(Assembler::notEqual, L_miss);
2115     }
2116 
2117     // Success.  Cache the super we found and proceed in triumph.
2118     __ movq(super_cache_addr, super_klass); // note: rax is dead
2119     __ jmp(L_success);
2120 
2121     // Fall through on failure!
2122     __ BIND(L_miss);
2123   }
2124 
2125   //
2126   //  Generate checkcasting array copy stub
2127   //
2128   //  Input:
2129   //    c_rarg0   - source array address
2130   //    c_rarg1   - destination array address
2131   //    c_rarg2   - element count, treated as ssize_t, can be zero
2132   //    c_rarg3   - size_t ckoff (super_check_offset)
2133   // not Win64
2134   //    c_rarg4   - oop ckval (super_klass)
2135   // Win64
2136   //    rsp+40    - oop ckval (super_klass)
2137   //
2138   //  Output:
2139   //    rax ==  0  -  success
2140   //    rax == -1^K - failure, where K is partial transfer count
2141   //
2142   address generate_checkcast_copy(const char *name) {
2143 
2144     Label L_load_element, L_store_element, L_do_card_marks, L_done;
2145 
2146     // Input registers (after setup_arg_regs)
2147     const Register from        = rdi;   // source array address
2148     const Register to          = rsi;   // destination array address
2149     const Register length      = rdx;   // elements count
2150     const Register ckoff       = rcx;   // super_check_offset
2151     const Register ckval       = r8;    // super_klass
2152 
2153     // Registers used as temps (r13, r14 are save-on-entry)
2154     const Register end_from    = from;  // source array end address
2155     const Register end_to      = r13;   // destination array end address
2156     const Register count       = rdx;   // -(count_remaining)
2157     const Register r14_length  = r14;   // saved copy of length
2158     // End pointers are inclusive, and if length is not zero they point
2159     // to the last unit copied:  end_to[0] := end_from[0]
2160 
2161     const Register rax_oop    = rax;    // actual oop copied
2162     const Register r11_klass  = r11;    // oop._klass
2163 
2164     //---------------------------------------------------------------
2165     // Assembler stub will be used for this call to arraycopy
2166     // if the two arrays are subtypes of Object[] but the
2167     // destination array type is not equal to or a supertype
2168     // of the source type.  Each element must be separately
2169     // checked.
2170 
2171     __ align(CodeEntryAlignment);
2172     StubCodeMark mark(this, "StubRoutines", name);
2173     address start = __ pc();
2174 
2175     __ enter(); // required for proper stackwalking of RuntimeStub frame
2176 
2177     checkcast_copy_entry  = __ pc();
2178     BLOCK_COMMENT("Entry:");
2179 
2180 #ifdef ASSERT
2181     // caller guarantees that the arrays really are different
2182     // otherwise, we would have to make conjoint checks
2183     { Label L;
2184       array_overlap_test(L, TIMES_OOP);
2185       __ stop("checkcast_copy within a single array");
2186       __ bind(L);
2187     }
2188 #endif //ASSERT
2189 
2190     // allocate spill slots for r13, r14
2191     enum {
2192       saved_r13_offset,
2193       saved_r14_offset,
2194       saved_rbp_offset,
2195       saved_rip_offset,
2196       saved_rarg0_offset
2197     };
2198     __ subq(rsp, saved_rbp_offset * wordSize);
2199     __ movq(Address(rsp, saved_r13_offset * wordSize), r13);
2200     __ movq(Address(rsp, saved_r14_offset * wordSize), r14);
2201     setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2202                        // ckoff => rcx, ckval => r8
2203                        // r9 and r10 may be used to save non-volatile registers
2204 #ifdef _WIN64
2205     // last argument (#4) is on stack on Win64
2206     const int ckval_offset = saved_rarg0_offset + 4;
2207     __ movq(ckval, Address(rsp, ckval_offset * wordSize));
2208 #endif
2209 
2210     // check that int operands are properly extended to size_t
2211     assert_clean_int(length, rax);
2212     assert_clean_int(ckoff, rax);
2213 
2214 #ifdef ASSERT
2215     BLOCK_COMMENT("assert consistent ckoff/ckval");
2216     // The ckoff and ckval must be mutually consistent,
2217     // even though caller generates both.
2218     { Label L;
2219       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2220                         Klass::super_check_offset_offset_in_bytes());
2221       __ cmpl(ckoff, Address(ckval, sco_offset));
2222       __ jcc(Assembler::equal, L);
2223       __ stop("super_check_offset inconsistent");
2224       __ bind(L);
2225     }
2226 #endif //ASSERT
2227 
2228     // Loop-invariant addresses.  They are exclusive end pointers.
2229     Address end_from_addr(from, length, TIMES_OOP, 0);
2230     Address   end_to_addr(to,   length, TIMES_OOP, 0);
2231     // Loop-variant addresses.  They assume post-incremented count < 0.
2232     Address from_element_addr(end_from, count, TIMES_OOP, 0);
2233     Address   to_element_addr(end_to,   count, TIMES_OOP, 0);
2234 
2235     gen_write_ref_array_pre_barrier(to, count);
2236 
2237     // Copy from low to high addresses, indexed from the end of each array.
2238     __ leaq(end_from, end_from_addr);
2239     __ leaq(end_to,   end_to_addr);
2240     __ movq(r14_length, length);        // save a copy of the length
2241     assert(length == count, "");        // else fix next line:
2242     __ negq(count);                     // negate and test the length
2243     __ jcc(Assembler::notZero, L_load_element);
2244 
2245     // Empty array:  Nothing to do.
2246     __ xorq(rax, rax);                  // return 0 on (trivial) success
2247     __ jmp(L_done);
2248 
2249     // ======== begin loop ========
2250     // (Loop is rotated; its entry is L_load_element.)
2251     // Loop control:
2252     //   for (count = -count; count != 0; count++)
2253     // Base pointers src, dst are biased by 8*(count-1),to last element.
2254     __ align(16);
2255 
2256     __ BIND(L_store_element);
2257     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop
2258     __ incrementq(count);               // increment the count toward zero
2259     __ jcc(Assembler::zero, L_do_card_marks);
2260 
2261     // ======== loop entry is here ========
2262     __ BIND(L_load_element);
2263     __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2264     __ testq(rax_oop, rax_oop);
2265     __ jcc(Assembler::zero, L_store_element);
2266 
2267     __ load_klass(r11_klass, rax_oop);// query the object klass
2268     generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2269     // ======== end loop ========
2270 
2271     // It was a real error; we must depend on the caller to finish the job.
2272     // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2273     // Emit GC store barriers for the oops we have copied (r14 + rdx),
2274     // and report their number to the caller.
2275     assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2276     __ leaq(end_to, to_element_addr);
2277     gen_write_ref_array_post_barrier(to, end_to, rcx);
2278     __ movq(rax, r14_length);           // original oops
2279     __ addq(rax, count);                // K = (original - remaining) oops
2280     __ notq(rax);                       // report (-1^K) to caller
2281     __ jmp(L_done);
2282 
2283     // Come here on success only.
2284     __ BIND(L_do_card_marks);
2285     __ addq(end_to, -wordSize);         // make an inclusive end pointer
2286     gen_write_ref_array_post_barrier(to, end_to, rcx);
2287     __ xorq(rax, rax);                  // return 0 on success
2288 
2289     // Common exit point (success or failure).
2290     __ BIND(L_done);
2291     __ movq(r13, Address(rsp, saved_r13_offset * wordSize));
2292     __ movq(r14, Address(rsp, saved_r14_offset * wordSize));
2293     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2294     restore_arg_regs();
2295     __ leave(); // required for proper stackwalking of RuntimeStub frame
2296     __ ret(0);
2297 
2298     return start;
2299   }
2300 
2301   //
2302   //  Generate 'unsafe' array copy stub
2303   //  Though just as safe as the other stubs, it takes an unscaled
2304   //  size_t argument instead of an element count.
2305   //
2306   //  Input:
2307   //    c_rarg0   - source array address
2308   //    c_rarg1   - destination array address
2309   //    c_rarg2   - byte count, treated as ssize_t, can be zero
2310   //
2311   // Examines the alignment of the operands and dispatches
2312   // to a long, int, short, or byte copy loop.
2313   //
2314   address generate_unsafe_copy(const char *name) {
2315 
2316     Label L_long_aligned, L_int_aligned, L_short_aligned;
2317 
2318     // Input registers (before setup_arg_regs)
2319     const Register from        = c_rarg0;  // source array address
2320     const Register to          = c_rarg1;  // destination array address
2321     const Register size        = c_rarg2;  // byte count (size_t)
2322 
2323     // Register used as a temp
2324     const Register bits        = rax;      // test copy of low bits
2325 
2326     __ align(CodeEntryAlignment);
2327     StubCodeMark mark(this, "StubRoutines", name);
2328     address start = __ pc();
2329 
2330     __ enter(); // required for proper stackwalking of RuntimeStub frame
2331 
2332     checkcast_copy_entry  = __ pc();
2333     BLOCK_COMMENT("Entry:");
2334 
2335     // bump this on entry, not on exit:
2336     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2337 
2338     __ movq(bits, from);
2339     __ orq(bits, to);
2340     __ orq(bits, size);
2341 
2342     __ testb(bits, BytesPerLong-1);
2343     __ jccb(Assembler::zero, L_long_aligned);
2344 
2345     __ testb(bits, BytesPerInt-1);
2346     __ jccb(Assembler::zero, L_int_aligned);
2347 
2348     __ testb(bits, BytesPerShort-1);
2349     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2350 
2351     __ BIND(L_short_aligned);
2352     __ shrq(size, LogBytesPerShort); // size => short_count
2353     __ jump(RuntimeAddress(short_copy_entry));
2354 
2355     __ BIND(L_int_aligned);
2356     __ shrq(size, LogBytesPerInt); // size => int_count
2357     __ jump(RuntimeAddress(int_copy_entry));
2358 
2359     __ BIND(L_long_aligned);
2360     __ shrq(size, LogBytesPerLong); // size => qword_count
2361     __ jump(RuntimeAddress(long_copy_entry));
2362 
2363     return start;
2364   }
2365 
2366   // Perform range checks on the proposed arraycopy.
2367   // Kills temp, but nothing else.
2368   // Also, clean the sign bits of src_pos and dst_pos.
2369   void arraycopy_range_checks(Register src,     // source array oop (c_rarg0)
2370                               Register src_pos, // source position (c_rarg1)
2371                               Register dst,     // destination array oo (c_rarg2)
2372                               Register dst_pos, // destination position (c_rarg3)
2373                               Register length,
2374                               Register temp,
2375                               Label& L_failed) {
2376     BLOCK_COMMENT("arraycopy_range_checks:");
2377 
2378     //  if (src_pos + length > arrayOop(src)->length())  FAIL;
2379     __ movl(temp, length);
2380     __ addl(temp, src_pos);             // src_pos + length
2381     __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2382     __ jcc(Assembler::above, L_failed);
2383 
2384     //  if (dst_pos + length > arrayOop(dst)->length())  FAIL;
2385     __ movl(temp, length);
2386     __ addl(temp, dst_pos);             // dst_pos + length
2387     __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2388     __ jcc(Assembler::above, L_failed);
2389 
2390     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2391     // Move with sign extension can be used since they are positive.
2392     __ movslq(src_pos, src_pos);
2393     __ movslq(dst_pos, dst_pos);
2394 
2395     BLOCK_COMMENT("arraycopy_range_checks done");
2396   }
2397 
2398   //
2399   //  Generate generic array copy stubs
2400   //
2401   //  Input:
2402   //    c_rarg0    -  src oop
2403   //    c_rarg1    -  src_pos (32-bits)
2404   //    c_rarg2    -  dst oop
2405   //    c_rarg3    -  dst_pos (32-bits)
2406   // not Win64
2407   //    c_rarg4    -  element count (32-bits)
2408   // Win64
2409   //    rsp+40     -  element count (32-bits)
2410   //
2411   //  Output:
2412   //    rax ==  0  -  success
2413   //    rax == -1^K - failure, where K is partial transfer count
2414   //
2415   address generate_generic_copy(const char *name) {
2416 
2417     Label L_failed, L_failed_0, L_objArray;
2418     Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2419 
2420     // Input registers
2421     const Register src        = c_rarg0;  // source array oop
2422     const Register src_pos    = c_rarg1;  // source position
2423     const Register dst        = c_rarg2;  // destination array oop
2424     const Register dst_pos    = c_rarg3;  // destination position
2425     // elements count is on stack on Win64
2426 #ifdef _WIN64
2427 #define C_RARG4 Address(rsp, 6 * wordSize)
2428 #else
2429 #define C_RARG4 c_rarg4
2430 #endif
2431 
2432     { int modulus = CodeEntryAlignment;
2433       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
2434       int advance = target - (__ offset() % modulus);
2435       if (advance < 0)  advance += modulus;
2436       if (advance > 0)  __ nop(advance);
2437     }
2438     StubCodeMark mark(this, "StubRoutines", name);
2439 
2440     // Short-hop target to L_failed.  Makes for denser prologue code.
2441     __ BIND(L_failed_0);
2442     __ jmp(L_failed);
2443     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2444 
2445     __ align(CodeEntryAlignment);
2446     address start = __ pc();
2447 
2448     __ enter(); // required for proper stackwalking of RuntimeStub frame
2449 
2450     // bump this on entry, not on exit:
2451     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2452 
2453     //-----------------------------------------------------------------------
2454     // Assembler stub will be used for this call to arraycopy
2455     // if the following conditions are met:
2456     //
2457     // (1) src and dst must not be null.
2458     // (2) src_pos must not be negative.
2459     // (3) dst_pos must not be negative.
2460     // (4) length  must not be negative.
2461     // (5) src klass and dst klass should be the same and not NULL.
2462     // (6) src and dst should be arrays.
2463     // (7) src_pos + length must not exceed length of src.
2464     // (8) dst_pos + length must not exceed length of dst.
2465     //
2466 
2467     //  if (src == NULL) return -1;
2468     __ testq(src, src);         // src oop
2469     size_t j1off = __ offset();
2470     __ jccb(Assembler::zero, L_failed_0);
2471 
2472     //  if (src_pos < 0) return -1;
2473     __ testl(src_pos, src_pos); // src_pos (32-bits)
2474     __ jccb(Assembler::negative, L_failed_0);
2475 
2476     //  if (dst == NULL) return -1;
2477     __ testq(dst, dst);         // dst oop
2478     __ jccb(Assembler::zero, L_failed_0);
2479 
2480     //  if (dst_pos < 0) return -1;
2481     __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2482     size_t j4off = __ offset();
2483     __ jccb(Assembler::negative, L_failed_0);
2484 
2485     // The first four tests are very dense code,
2486     // but not quite dense enough to put four
2487     // jumps in a 16-byte instruction fetch buffer.
2488     // That's good, because some branch predicters
2489     // do not like jumps so close together.
2490     // Make sure of this.
2491     guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2492 
2493     // registers used as temp
2494     const Register r11_length    = r11; // elements count to copy
2495     const Register r10_src_klass = r10; // array klass
2496     const Register r9_dst_klass  = r9;  // dest array klass
2497 
2498     //  if (length < 0) return -1;
2499     __ movl(r11_length, C_RARG4);       // length (elements count, 32-bits value)
2500     __ testl(r11_length, r11_length);
2501     __ jccb(Assembler::negative, L_failed_0);
2502 
2503     __ load_klass(r10_src_klass, src);
2504 #ifdef ASSERT
2505     //  assert(src->klass() != NULL);
2506     BLOCK_COMMENT("assert klasses not null");
2507     { Label L1, L2;
2508       __ testq(r10_src_klass, r10_src_klass);
2509       __ jcc(Assembler::notZero, L2);   // it is broken if klass is NULL
2510       __ bind(L1);
2511       __ stop("broken null klass");
2512       __ bind(L2);
2513       __ load_klass(r9_dst_klass, dst);
2514       __ cmpq(r9_dst_klass, 0);
2515       __ jcc(Assembler::equal, L1);     // this would be broken also
2516       BLOCK_COMMENT("assert done");
2517     }
2518 #endif
2519 
2520     // Load layout helper (32-bits)
2521     //
2522     //  | array_tag | element_type | log2_element_size | header_size |
2523     // 32          28             24                  16             0
2524     //
2525     //   array_tag: typeArray = 0xF, objArray = 0xD, normal instance = 0x0
2526     //
2527 
2528     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2529                     Klass::layout_helper_offset_in_bytes();
2530 
2531     const Register rax_lh = rax;  // layout helper
2532 
2533     __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2534 
2535     // Handle objArrays completely differently...
2536     // This is necessary because they have covariant element types.
2537     // The next type check (for src.klass == dst.klass), can get a
2538     // false negative in the case of object arrays.
2539     jint objArray_lh = LayoutHelper::for_array(T_OBJECT).as_int();
2540     __ cmpl(rax_lh, objArray_lh);
2541     __ jcc(Assembler::equal, L_objArray);
2542 
2543     //  if (src->klass() != dst->klass()) return -1;
2544     __ load_klass(r9_dst_klass, dst);
2545     __ cmpq(r10_src_klass, r9_dst_klass);
2546     __ jcc(Assembler::notEqual, L_failed);
2547 
2548     //  if (!src->has_length_field()) return -1;
2549     __ cmpl(rax_lh, LayoutHelper::_neutral_value);
2550     __ jcc(Assembler::greaterEqual, L_failed);
2551 
2552     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2553                            r10, L_failed);
2554 
2555     // typeArrayKlass
2556     //
2557     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2558     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2559     //
2560 
2561     const Register r10_offset = r10;    // array offset
2562     const Register rax_elsize = rax_lh; // element size
2563 
2564     __ movl(r10_offset, rax_lh);
2565     assert(LayoutHelper::_header_size_shift == 0, "");
2566     __ andl(r10_offset, LayoutHelper::_header_size_mask);   // array_offset
2567     __ addq(src, r10_offset);           // src array offset
2568     __ addq(dst, r10_offset);           // dst array offset
2569     BLOCK_COMMENT("choose copy loop based on element size");
2570     __ shrl(rax_lh, LayoutHelper::_element_size_shift);
2571     __ andl(rax_lh, LayoutHelper::_element_size_mask); // rax_lh -> rax_elsize
2572 
2573     // next registers should be set before the jump to corresponding stub
2574     const Register from     = c_rarg0;  // source array address
2575     const Register to       = c_rarg1;  // destination array address
2576     const Register count    = c_rarg2;  // elements count
2577 
2578     // 'from', 'to', 'count' registers should be set in such order
2579     // since they are the same as 'src', 'src_pos', 'dst'.
2580 
2581   __ BIND(L_copy_bytes);
2582     __ cmpl(rax_elsize, 0);
2583     __ jccb(Assembler::notEqual, L_copy_shorts);
2584     __ leaq(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2585     __ leaq(to,   Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2586     __ movslq(count, r11_length); // length
2587     __ jump(RuntimeAddress(byte_copy_entry));
2588 
2589   __ BIND(L_copy_shorts);
2590     __ cmpl(rax_elsize, LogBytesPerShort);
2591     __ jccb(Assembler::notEqual, L_copy_ints);
2592     __ leaq(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2593     __ leaq(to,   Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2594     __ movslq(count, r11_length); // length
2595     __ jump(RuntimeAddress(short_copy_entry));
2596 
2597   __ BIND(L_copy_ints);
2598     __ cmpl(rax_elsize, LogBytesPerInt);
2599     __ jccb(Assembler::notEqual, L_copy_longs);
2600     __ leaq(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2601     __ leaq(to,   Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2602     __ movslq(count, r11_length); // length
2603     __ jump(RuntimeAddress(int_copy_entry));
2604 
2605   __ BIND(L_copy_longs);
2606 #ifdef ASSERT
2607     { Label L;
2608       __ cmpl(rax_elsize, LogBytesPerLong);
2609       __ jcc(Assembler::equal, L);
2610       __ stop("must be long copy, but elsize is wrong");
2611       __ bind(L);
2612     }
2613 #endif
2614     __ leaq(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2615     __ leaq(to,   Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2616     __ movslq(count, r11_length); // length
2617     __ jump(RuntimeAddress(long_copy_entry));
2618 
2619     Label L_different_objArray, L_plain_copy, L_checkcast_copy;
2620 
2621     // objArrayKlass
2622   __ BIND(L_objArray);
2623     // live at this point:  r10_src_klass, src[_pos], dst[_pos]
2624 
2625     //  test array classes for subtyping
2626     __ load_klass(r9_dst_klass, dst);
2627     __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2628     __ jcc(Assembler::notEqual, L_different_objArray);
2629 
2630     // Identically typed arrays can be copied without element-wise checks.
2631     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2632                            r10, L_failed);
2633 
2634     __ leaq(from, Address(src, src_pos, TIMES_OOP,
2635                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2636     __ leaq(to,   Address(dst, dst_pos, TIMES_OOP,
2637                   arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2638     __ movslq(count, r11_length); // length
2639   __ BIND(L_plain_copy);
2640     __ jump(RuntimeAddress(oop_copy_entry));
2641 
2642   __ BIND(L_different_objArray);
2643     // live at this point:  r10_src_klass, !r11_length
2644     {
2645       // assert(r11_length == C_RARG4); // will reload from here
2646       Register r11_dst_klass = r11;
2647       __ load_klass(r11_dst_klass, dst);
2648 
2649       // Before looking at dst.length, make sure dst is also an objArray.
2650       __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2651       __ jcc(Assembler::notEqual, L_failed);
2652 
2653       // It is safe to examine both src.length and dst.length.
2654 #ifndef _WIN64
2655       arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2656                              rax, L_failed);
2657 #else
2658       __ movl(r11_length, C_RARG4);     // reload
2659       arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2660                              rax, L_failed);
2661       __ load_klass(r11_dst_klass, dst); // reload
2662 #endif
2663 
2664       // Marshal the base address arguments now, freeing registers.
2665       __ leaq(from, Address(src, src_pos, TIMES_OOP,
2666                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2667       __ leaq(to,   Address(dst, dst_pos, TIMES_OOP,
2668                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2669       __ movl(count, C_RARG4);          // length (reloaded)
2670       Register sco_temp = c_rarg3;      // this register is free now
2671       assert_different_registers(from, to, count, sco_temp,
2672                                  r11_dst_klass, r10_src_klass);
2673       assert_clean_int(count, sco_temp);
2674 
2675       // Generate the type check.
2676       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2677                         Klass::super_check_offset_offset_in_bytes());
2678       __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2679       assert_clean_int(sco_temp, rax);
2680       generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2681 
2682       // Fetch destination element klass from the objArrayKlass header.
2683       int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2684                        objArrayKlass::element_klass_offset_in_bytes());
2685       __ movq(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2686       __ movl(sco_temp,      Address(r11_dst_klass, sco_offset));
2687       assert_clean_int(sco_temp, rax);
2688 
2689       // the checkcast_copy loop needs two extra arguments:
2690       assert(c_rarg3 == sco_temp, "#3 already in place");
2691       __ movq(C_RARG4, r11_dst_klass);  // dst.klass.element_klass
2692       __ jump(RuntimeAddress(checkcast_copy_entry));
2693     }
2694 
2695   __ BIND(L_failed);
2696     __ xorq(rax, rax);
2697     __ notq(rax); // return -1
2698     __ leave();   // required for proper stackwalking of RuntimeStub frame
2699     __ ret(0);
2700 
2701     return start;
2702   }
2703 
2704 #undef length_arg
2705 
2706   void generate_arraycopy_stubs() {
2707     // Call the conjoint generation methods immediately after
2708     // the disjoint ones so that short branches from the former
2709     // to the latter can be generated.
2710     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2711     StubRoutines::_jbyte_arraycopy           = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2712 
2713     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2714     StubRoutines::_jshort_arraycopy          = generate_conjoint_short_copy(false, "jshort_arraycopy");
2715 
2716     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2717     StubRoutines::_jint_arraycopy            = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2718 
2719     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2720     StubRoutines::_jlong_arraycopy           = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2721 
2722 
2723     if (UseCompressedOops) {
2724       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2725       StubRoutines::_oop_arraycopy           = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2726     } else {
2727       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2728       StubRoutines::_oop_arraycopy           = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2729     }
2730 
2731     StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2732     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy");
2733     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy");
2734 
2735     // We don't generate specialized code for HeapWord-aligned source
2736     // arrays, so just use the code we've already generated
2737     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = StubRoutines::_jbyte_disjoint_arraycopy;
2738     StubRoutines::_arrayof_jbyte_arraycopy           = StubRoutines::_jbyte_arraycopy;
2739 
2740     StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2741     StubRoutines::_arrayof_jshort_arraycopy          = StubRoutines::_jshort_arraycopy;
2742 
2743     StubRoutines::_arrayof_jint_disjoint_arraycopy   = StubRoutines::_jint_disjoint_arraycopy;
2744     StubRoutines::_arrayof_jint_arraycopy            = StubRoutines::_jint_arraycopy;
2745 
2746     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = StubRoutines::_jlong_disjoint_arraycopy;
2747     StubRoutines::_arrayof_jlong_arraycopy           = StubRoutines::_jlong_arraycopy;
2748 
2749     StubRoutines::_arrayof_oop_disjoint_arraycopy    = StubRoutines::_oop_disjoint_arraycopy;
2750     StubRoutines::_arrayof_oop_arraycopy             = StubRoutines::_oop_arraycopy;
2751   }
2752 
2753 #undef __
2754 #define __ masm->
2755 
2756   // Continuation point for throwing of implicit exceptions that are
2757   // not handled in the current activation. Fabricates an exception
2758   // oop and initiates normal exception dispatching in this
2759   // frame. Since we need to preserve callee-saved values (currently
2760   // only for C2, but done for C1 as well) we need a callee-saved oop
2761   // map and therefore have to make these stubs into RuntimeStubs
2762   // rather than BufferBlobs.  If the compiler needs all registers to
2763   // be preserved between the fault point and the exception handler
2764   // then it must assume responsibility for that in
2765   // AbstractCompiler::continuation_for_implicit_null_exception or
2766   // continuation_for_implicit_division_by_zero_exception. All other
2767   // implicit exceptions (e.g., NullPointerException or
2768   // AbstractMethodError on entry) are either at call sites or
2769   // otherwise assume that stack unwinding will be initiated, so
2770   // caller saved registers were assumed volatile in the compiler.
2771   address generate_throw_exception(const char* name,
2772                                    address runtime_entry,
2773                                    bool restore_saved_exception_pc) {
2774     // Information about frame layout at time of blocking runtime call.
2775     // Note that we only have to preserve callee-saved registers since
2776     // the compilers are responsible for supplying a continuation point
2777     // if they expect all registers to be preserved.
2778     enum layout {
2779       rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2780       rbp_off2,
2781       return_off,
2782       return_off2,
2783       framesize // inclusive of return address
2784     };
2785 
2786     int insts_size = 512;
2787     int locs_size  = 64;
2788 
2789     CodeBuffer code(name, insts_size, locs_size);
2790     OopMapSet* oop_maps  = new OopMapSet();
2791     MacroAssembler* masm = new MacroAssembler(&code);
2792 
2793     address start = __ pc();
2794 
2795     // This is an inlined and slightly modified version of call_VM
2796     // which has the ability to fetch the return PC out of
2797     // thread-local storage and also sets up last_Java_sp slightly
2798     // differently than the real call_VM
2799     if (restore_saved_exception_pc) {
2800       __ movq(rax,
2801               Address(r15_thread,
2802                       in_bytes(JavaThread::saved_exception_pc_offset())));
2803       __ pushq(rax);
2804     }
2805 
2806     __ enter(); // required for proper stackwalking of RuntimeStub frame
2807 
2808     assert(is_even(framesize/2), "sp not 16-byte aligned");
2809 
2810     // return address and rbp are already in place
2811     __ subq(rsp, (framesize-4) << LogBytesPerInt); // prolog
2812 
2813     int frame_complete = __ pc() - start;
2814 
2815     // Set up last_Java_sp and last_Java_fp
2816     __ set_last_Java_frame(rsp, rbp, NULL);
2817 
2818     // Call runtime
2819     __ movq(c_rarg0, r15_thread);
2820     BLOCK_COMMENT("call runtime_entry");
2821     __ call(RuntimeAddress(runtime_entry));
2822 
2823     // Generate oop map
2824     OopMap* map = new OopMap(framesize, 0);
2825 
2826     oop_maps->add_gc_map(__ pc() - start, map);
2827 
2828     __ reset_last_Java_frame(true, false);
2829 
2830     __ leave(); // required for proper stackwalking of RuntimeStub frame
2831 
2832     // check for pending exceptions
2833 #ifdef ASSERT
2834     Label L;
2835     __ cmpq(Address(r15_thread, Thread::pending_exception_offset()),
2836             (int) NULL);
2837     __ jcc(Assembler::notEqual, L);
2838     __ should_not_reach_here();
2839     __ bind(L);
2840 #endif // ASSERT
2841     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2842 
2843 
2844     // codeBlob framesize is in words (not VMRegImpl::slot_size)
2845     RuntimeStub* stub =
2846       RuntimeStub::new_runtime_stub(name,
2847                                     &code,
2848                                     frame_complete,
2849                                     (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2850                                     oop_maps, false);
2851     return stub->entry_point();
2852   }
2853 
2854   // Initialization
2855   void generate_initial() {
2856     // Generates all stubs and initializes the entry points
2857 
2858     // This platform-specific stub is needed by generate_call_stub()
2859     StubRoutines::amd64::_mxcsr_std        = generate_fp_mask("mxcsr_std",        0x0000000000001F80);
2860 
2861     // entry points that exist in all platforms Note: This is code
2862     // that could be shared among different platforms - however the
2863     // benefit seems to be smaller than the disadvantage of having a
2864     // much more complicated generator structure. See also comment in
2865     // stubRoutines.hpp.
2866 
2867     StubRoutines::_forward_exception_entry = generate_forward_exception();
2868 
2869     StubRoutines::_call_stub_entry =
2870       generate_call_stub(StubRoutines::_call_stub_return_address);
2871 
2872     // is referenced by megamorphic call
2873     StubRoutines::_catch_exception_entry = generate_catch_exception();
2874 
2875     // atomic calls
2876     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
2877     StubRoutines::_atomic_xchg_ptr_entry     = generate_atomic_xchg_ptr();
2878     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
2879     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2880     StubRoutines::_atomic_add_entry          = generate_atomic_add();
2881     StubRoutines::_atomic_add_ptr_entry      = generate_atomic_add_ptr();
2882     StubRoutines::_fence_entry               = generate_orderaccess_fence();
2883 
2884     StubRoutines::_handler_for_unsafe_access_entry =
2885       generate_handler_for_unsafe_access();
2886 
2887     // platform dependent
2888     StubRoutines::amd64::_get_previous_fp_entry = generate_get_previous_fp();
2889 
2890     StubRoutines::amd64::_verify_mxcsr_entry    = generate_verify_mxcsr();
2891   }
2892 
2893   void generate_all() {
2894     // Generates all stubs and initializes the entry points
2895 
2896     // These entry points require SharedInfo::stack0 to be set up in
2897     // non-core builds and need to be relocatable, so they each
2898     // fabricate a RuntimeStub internally.
2899     StubRoutines::_throw_AbstractMethodError_entry =
2900       generate_throw_exception("AbstractMethodError throw_exception",
2901                                CAST_FROM_FN_PTR(address,
2902                                                 SharedRuntime::
2903                                                 throw_AbstractMethodError),
2904                                false);
2905 
2906     StubRoutines::_throw_IncompatibleClassChangeError_entry =
2907       generate_throw_exception("IncompatibleClassChangeError throw_exception",
2908                                CAST_FROM_FN_PTR(address,
2909                                                 SharedRuntime::
2910                                                 throw_IncompatibleClassChangeError),
2911                                false);
2912 
2913     StubRoutines::_throw_ArithmeticException_entry =
2914       generate_throw_exception("ArithmeticException throw_exception",
2915                                CAST_FROM_FN_PTR(address,
2916                                                 SharedRuntime::
2917                                                 throw_ArithmeticException),
2918                                true);
2919 
2920     StubRoutines::_throw_NullPointerException_entry =
2921       generate_throw_exception("NullPointerException throw_exception",
2922                                CAST_FROM_FN_PTR(address,
2923                                                 SharedRuntime::
2924                                                 throw_NullPointerException),
2925                                true);
2926 
2927     StubRoutines::_throw_NullPointerException_at_call_entry =
2928       generate_throw_exception("NullPointerException at call throw_exception",
2929                                CAST_FROM_FN_PTR(address,
2930                                                 SharedRuntime::
2931                                                 throw_NullPointerException_at_call),
2932                                false);
2933 
2934     StubRoutines::_throw_StackOverflowError_entry =
2935       generate_throw_exception("StackOverflowError throw_exception",
2936                                CAST_FROM_FN_PTR(address,
2937                                                 SharedRuntime::
2938                                                 throw_StackOverflowError),
2939                                false);
2940 
2941     // entry points that are platform specific
2942     StubRoutines::amd64::_f2i_fixup = generate_f2i_fixup();
2943     StubRoutines::amd64::_f2l_fixup = generate_f2l_fixup();
2944     StubRoutines::amd64::_d2i_fixup = generate_d2i_fixup();
2945     StubRoutines::amd64::_d2l_fixup = generate_d2l_fixup();
2946 
2947     StubRoutines::amd64::_float_sign_mask  = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
2948     StubRoutines::amd64::_float_sign_flip  = generate_fp_mask("float_sign_flip",  0x8000000080000000);
2949     StubRoutines::amd64::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
2950     StubRoutines::amd64::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
2951 
2952     // support for verify_oop (must happen after universe_init)
2953     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2954 
2955     // arraycopy stubs used by compilers
2956     generate_arraycopy_stubs();
2957   }
2958 
2959  public:
2960   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2961     if (all) {
2962       generate_all();
2963     } else {
2964       generate_initial();
2965     }
2966   }
2967 }; // end class declaration
2968 
2969 address StubGenerator::disjoint_byte_copy_entry  = NULL;
2970 address StubGenerator::disjoint_short_copy_entry = NULL;
2971 address StubGenerator::disjoint_int_copy_entry   = NULL;
2972 address StubGenerator::disjoint_long_copy_entry  = NULL;
2973 address StubGenerator::disjoint_oop_copy_entry   = NULL;
2974 
2975 address StubGenerator::byte_copy_entry  = NULL;
2976 address StubGenerator::short_copy_entry = NULL;
2977 address StubGenerator::int_copy_entry   = NULL;
2978 address StubGenerator::long_copy_entry  = NULL;
2979 address StubGenerator::oop_copy_entry   = NULL;
2980 
2981 address StubGenerator::checkcast_copy_entry = NULL;
2982 
2983 void StubGenerator_generate(CodeBuffer* code, bool all) {
2984   StubGenerator g(code, all);
2985 }