1 /*
   2  * Copyright 2003-2010 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 #define a__ ((Assembler*)_masm)->
  35 
  36 #ifdef PRODUCT
  37 #define BLOCK_COMMENT(str) /* nothing */
  38 #else
  39 #define BLOCK_COMMENT(str) __ block_comment(str)
  40 #endif
  41 
  42 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  43 const int MXCSR_MASK = 0xFFC0;  // Mask out any pending exceptions
  44 
  45 // Stub Code definitions
  46 
  47 static address handle_unsafe_access() {
  48   JavaThread* thread = JavaThread::current();
  49   address pc = thread->saved_exception_pc();
  50   // pc is the instruction which we must emulate
  51   // doing a no-op is fine:  return garbage from the load
  52   // therefore, compute npc
  53   address npc = Assembler::locate_next_instruction(pc);
  54 
  55   // request an async exception
  56   thread->set_pending_unsafe_access_error();
  57 
  58   // return address of next instruction to execute
  59   return npc;
  60 }
  61 
  62 class StubGenerator: public StubCodeGenerator {
  63  private:
  64 
  65 #ifdef PRODUCT
  66 #define inc_counter_np(counter) (0)
  67 #else
  68   void inc_counter_np_(int& counter) {
  69     __ incrementl(ExternalAddress((address)&counter));
  70   }
  71 #define inc_counter_np(counter) \
  72   BLOCK_COMMENT("inc_counter " #counter); \
  73   inc_counter_np_(counter);
  74 #endif
  75 
  76   // Call stubs are used to call Java from C
  77   //
  78   // Linux Arguments:
  79   //    c_rarg0:   call wrapper address                   address
  80   //    c_rarg1:   result                                 address
  81   //    c_rarg2:   result type                            BasicType
  82   //    c_rarg3:   method                                 methodOop
  83   //    c_rarg4:   (interpreter) entry point              address
  84   //    c_rarg5:   parameters                             intptr_t*
  85   //    16(rbp): parameter size (in words)              int
  86   //    24(rbp): thread                                 Thread*
  87   //
  88   //     [ return_from_Java     ] <--- rsp
  89   //     [ argument word n      ]
  90   //      ...
  91   // -12 [ argument word 1      ]
  92   // -11 [ saved r15            ] <--- rsp_after_call
  93   // -10 [ saved r14            ]
  94   //  -9 [ saved r13            ]
  95   //  -8 [ saved r12            ]
  96   //  -7 [ saved rbx            ]
  97   //  -6 [ call wrapper         ]
  98   //  -5 [ result               ]
  99   //  -4 [ result type          ]
 100   //  -3 [ method               ]
 101   //  -2 [ entry point          ]
 102   //  -1 [ parameters           ]
 103   //   0 [ saved rbp            ] <--- rbp
 104   //   1 [ return address       ]
 105   //   2 [ parameter size       ]
 106   //   3 [ thread               ]
 107   //
 108   // Windows Arguments:
 109   //    c_rarg0:   call wrapper address                   address
 110   //    c_rarg1:   result                                 address
 111   //    c_rarg2:   result type                            BasicType
 112   //    c_rarg3:   method                                 methodOop
 113   //    48(rbp): (interpreter) entry point              address
 114   //    56(rbp): parameters                             intptr_t*
 115   //    64(rbp): parameter size (in words)              int
 116   //    72(rbp): thread                                 Thread*
 117   //
 118   //     [ return_from_Java     ] <--- rsp
 119   //     [ argument word n      ]
 120   //      ...
 121   //  -8 [ argument word 1      ]
 122   //  -7 [ saved r15            ] <--- rsp_after_call
 123   //  -6 [ saved r14            ]
 124   //  -5 [ saved r13            ]
 125   //  -4 [ saved r12            ]
 126   //  -3 [ saved rdi            ]
 127   //  -2 [ saved rsi            ]
 128   //  -1 [ saved rbx            ]
 129   //   0 [ saved rbp            ] <--- rbp
 130   //   1 [ return address       ]
 131   //   2 [ call wrapper         ]
 132   //   3 [ result               ]
 133   //   4 [ result type          ]
 134   //   5 [ method               ]
 135   //   6 [ entry point          ]
 136   //   7 [ parameters           ]
 137   //   8 [ parameter size       ]
 138   //   9 [ thread               ]
 139   //
 140   //    Windows reserves the callers stack space for arguments 1-4.
 141   //    We spill c_rarg0-c_rarg3 to this space.
 142 
 143   // Call stub stack layout word offsets from rbp
 144   enum call_stub_layout {
 145 #ifdef _WIN64
 146     rsp_after_call_off = -7,
 147     r15_off            = rsp_after_call_off,
 148     r14_off            = -6,
 149     r13_off            = -5,
 150     r12_off            = -4,
 151     rdi_off            = -3,
 152     rsi_off            = -2,
 153     rbx_off            = -1,
 154     rbp_off            =  0,
 155     retaddr_off        =  1,
 156     call_wrapper_off   =  2,
 157     result_off         =  3,
 158     result_type_off    =  4,
 159     method_off         =  5,
 160     entry_point_off    =  6,
 161     parameters_off     =  7,
 162     parameter_size_off =  8,
 163     thread_off         =  9
 164 #else
 165     rsp_after_call_off = -12,
 166     mxcsr_off          = rsp_after_call_off,
 167     r15_off            = -11,
 168     r14_off            = -10,
 169     r13_off            = -9,
 170     r12_off            = -8,
 171     rbx_off            = -7,
 172     call_wrapper_off   = -6,
 173     result_off         = -5,
 174     result_type_off    = -4,
 175     method_off         = -3,
 176     entry_point_off    = -2,
 177     parameters_off     = -1,
 178     rbp_off            =  0,
 179     retaddr_off        =  1,
 180     parameter_size_off =  2,
 181     thread_off         =  3
 182 #endif
 183   };
 184 
 185   address generate_call_stub(address& return_address) {
 186     assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 187            (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 188            "adjust this code");
 189     StubCodeMark mark(this, "StubRoutines", "call_stub");
 190     address start = __ pc();
 191 
 192     // same as in generate_catch_exception()!
 193     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 194 
 195     const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 196     const Address result        (rbp, result_off         * wordSize);
 197     const Address result_type   (rbp, result_type_off    * wordSize);
 198     const Address method        (rbp, method_off         * wordSize);
 199     const Address entry_point   (rbp, entry_point_off    * wordSize);
 200     const Address parameters    (rbp, parameters_off     * wordSize);
 201     const Address parameter_size(rbp, parameter_size_off * wordSize);
 202 
 203     // same as in generate_catch_exception()!
 204     const Address thread        (rbp, thread_off         * wordSize);
 205 
 206     const Address r15_save(rbp, r15_off * wordSize);
 207     const Address r14_save(rbp, r14_off * wordSize);
 208     const Address r13_save(rbp, r13_off * wordSize);
 209     const Address r12_save(rbp, r12_off * wordSize);
 210     const Address rbx_save(rbp, rbx_off * wordSize);
 211 
 212     // stub code
 213     __ enter();
 214     __ subptr(rsp, -rsp_after_call_off * wordSize);
 215 
 216     // save register parameters
 217 #ifndef _WIN64
 218     __ movptr(parameters,   c_rarg5); // parameters
 219     __ movptr(entry_point,  c_rarg4); // entry_point
 220 #endif
 221 
 222     __ movptr(method,       c_rarg3); // method
 223     __ movl(result_type,  c_rarg2);   // result type
 224     __ movptr(result,       c_rarg1); // result
 225     __ movptr(call_wrapper, c_rarg0); // call wrapper
 226 
 227     // save regs belonging to calling function
 228     __ movptr(rbx_save, rbx);
 229     __ movptr(r12_save, r12);
 230     __ movptr(r13_save, r13);
 231     __ movptr(r14_save, r14);
 232     __ movptr(r15_save, r15);
 233 
 234 #ifdef _WIN64
 235     const Address rdi_save(rbp, rdi_off * wordSize);
 236     const Address rsi_save(rbp, rsi_off * wordSize);
 237 
 238     __ movptr(rsi_save, rsi);
 239     __ movptr(rdi_save, rdi);
 240 #else
 241     const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 242     {
 243       Label skip_ldmx;
 244       __ stmxcsr(mxcsr_save);
 245       __ movl(rax, mxcsr_save);
 246       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 247       ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
 248       __ cmp32(rax, mxcsr_std);
 249       __ jcc(Assembler::equal, skip_ldmx);
 250       __ ldmxcsr(mxcsr_std);
 251       __ bind(skip_ldmx);
 252     }
 253 #endif
 254 
 255     // Load up thread register
 256     __ movptr(r15_thread, thread);
 257     __ reinit_heapbase();
 258 
 259 #ifdef ASSERT
 260     // make sure we have no pending exceptions
 261     {
 262       Label L;
 263       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 264       __ jcc(Assembler::equal, L);
 265       __ stop("StubRoutines::call_stub: entered with pending exception");
 266       __ bind(L);
 267     }
 268 #endif
 269 
 270     // pass parameters if any
 271     BLOCK_COMMENT("pass parameters if any");
 272     Label parameters_done;
 273     __ movl(c_rarg3, parameter_size);
 274     __ testl(c_rarg3, c_rarg3);
 275     __ jcc(Assembler::zero, parameters_done);
 276 
 277     Label loop;
 278     __ movptr(c_rarg2, parameters);       // parameter pointer
 279     __ movl(c_rarg1, c_rarg3);            // parameter counter is in c_rarg1
 280     __ BIND(loop);
 281     if (TaggedStackInterpreter) {
 282       __ movl(rax, Address(c_rarg2, 0)); // get tag
 283       __ addptr(c_rarg2, wordSize);      // advance to next tag
 284       __ push(rax);                      // pass tag
 285     }
 286     __ movptr(rax, Address(c_rarg2, 0));// get parameter
 287     __ addptr(c_rarg2, wordSize);       // advance to next parameter
 288     __ decrementl(c_rarg1);             // decrement counter
 289     __ push(rax);                       // pass parameter
 290     __ jcc(Assembler::notZero, loop);
 291 
 292     // call Java function
 293     __ BIND(parameters_done);
 294     __ movptr(rbx, method);             // get methodOop
 295     __ movptr(c_rarg1, entry_point);    // get entry_point
 296     __ mov(r13, rsp);                   // set sender sp
 297     BLOCK_COMMENT("call Java function");
 298     __ call(c_rarg1);
 299 
 300     BLOCK_COMMENT("call_stub_return_address:");
 301     return_address = __ pc();
 302 
 303     // store result depending on type (everything that is not
 304     // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 305     __ movptr(c_rarg0, result);
 306     Label is_long, is_float, is_double, exit;
 307     __ movl(c_rarg1, result_type);
 308     __ cmpl(c_rarg1, T_OBJECT);
 309     __ jcc(Assembler::equal, is_long);
 310     __ cmpl(c_rarg1, T_LONG);
 311     __ jcc(Assembler::equal, is_long);
 312     __ cmpl(c_rarg1, T_FLOAT);
 313     __ jcc(Assembler::equal, is_float);
 314     __ cmpl(c_rarg1, T_DOUBLE);
 315     __ jcc(Assembler::equal, is_double);
 316 
 317     // handle T_INT case
 318     __ movl(Address(c_rarg0, 0), rax);
 319 
 320     __ BIND(exit);
 321 
 322     // pop parameters
 323     __ lea(rsp, rsp_after_call);
 324 
 325 #ifdef ASSERT
 326     // verify that threads correspond
 327     {
 328       Label L, S;
 329       __ cmpptr(r15_thread, thread);
 330       __ jcc(Assembler::notEqual, S);
 331       __ get_thread(rbx);
 332       __ cmpptr(r15_thread, rbx);
 333       __ jcc(Assembler::equal, L);
 334       __ bind(S);
 335       __ jcc(Assembler::equal, L);
 336       __ stop("StubRoutines::call_stub: threads must correspond");
 337       __ bind(L);
 338     }
 339 #endif
 340 
 341     // restore regs belonging to calling function
 342     __ movptr(r15, r15_save);
 343     __ movptr(r14, r14_save);
 344     __ movptr(r13, r13_save);
 345     __ movptr(r12, r12_save);
 346     __ movptr(rbx, rbx_save);
 347 
 348 #ifdef _WIN64
 349     __ movptr(rdi, rdi_save);
 350     __ movptr(rsi, rsi_save);
 351 #else
 352     __ ldmxcsr(mxcsr_save);
 353 #endif
 354 
 355     // restore rsp
 356     __ addptr(rsp, -rsp_after_call_off * wordSize);
 357 
 358     // return
 359     __ pop(rbp);
 360     __ ret(0);
 361 
 362     // handle return types different from T_INT
 363     __ BIND(is_long);
 364     __ movq(Address(c_rarg0, 0), rax);
 365     __ jmp(exit);
 366 
 367     __ BIND(is_float);
 368     __ movflt(Address(c_rarg0, 0), xmm0);
 369     __ jmp(exit);
 370 
 371     __ BIND(is_double);
 372     __ movdbl(Address(c_rarg0, 0), xmm0);
 373     __ jmp(exit);
 374 
 375     return start;
 376   }
 377 
 378   // Return point for a Java call if there's an exception thrown in
 379   // Java code.  The exception is caught and transformed into a
 380   // pending exception stored in JavaThread that can be tested from
 381   // within the VM.
 382   //
 383   // Note: Usually the parameters are removed by the callee. In case
 384   // of an exception crossing an activation frame boundary, that is
 385   // not the case if the callee is compiled code => need to setup the
 386   // rsp.
 387   //
 388   // rax: exception oop
 389 
 390   address generate_catch_exception() {
 391     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 392     address start = __ pc();
 393 
 394     // same as in generate_call_stub():
 395     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 396     const Address thread        (rbp, thread_off         * wordSize);
 397 
 398 #ifdef ASSERT
 399     // verify that threads correspond
 400     {
 401       Label L, S;
 402       __ cmpptr(r15_thread, thread);
 403       __ jcc(Assembler::notEqual, S);
 404       __ get_thread(rbx);
 405       __ cmpptr(r15_thread, rbx);
 406       __ jcc(Assembler::equal, L);
 407       __ bind(S);
 408       __ stop("StubRoutines::catch_exception: threads must correspond");
 409       __ bind(L);
 410     }
 411 #endif
 412 
 413     // set pending exception
 414     __ verify_oop(rax);
 415 
 416     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
 417     __ lea(rscratch1, ExternalAddress((address)__FILE__));
 418     __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 419     __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 420 
 421     // complete return to VM
 422     assert(StubRoutines::_call_stub_return_address != NULL,
 423            "_call_stub_return_address must have been generated before");
 424     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 425 
 426     return start;
 427   }
 428 
 429   // Continuation point for runtime calls returning with a pending
 430   // exception.  The pending exception check happened in the runtime
 431   // or native call stub.  The pending exception in Thread is
 432   // converted into a Java-level exception.
 433   //
 434   // Contract with Java-level exception handlers:
 435   // rax: exception
 436   // rdx: throwing pc
 437   //
 438   // NOTE: At entry of this stub, exception-pc must be on stack !!
 439 
 440   address generate_forward_exception() {
 441     StubCodeMark mark(this, "StubRoutines", "forward exception");
 442     address start = __ pc();
 443 
 444     // Upon entry, the sp points to the return address returning into
 445     // Java (interpreted or compiled) code; i.e., the return address
 446     // becomes the throwing pc.
 447     //
 448     // Arguments pushed before the runtime call are still on the stack
 449     // but the exception handler will reset the stack pointer ->
 450     // ignore them.  A potential result in registers can be ignored as
 451     // well.
 452 
 453 #ifdef ASSERT
 454     // make sure this code is only executed if there is a pending exception
 455     {
 456       Label L;
 457       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
 458       __ jcc(Assembler::notEqual, L);
 459       __ stop("StubRoutines::forward exception: no pending exception (1)");
 460       __ bind(L);
 461     }
 462 #endif
 463 
 464     // compute exception handler into rbx
 465     __ movptr(c_rarg0, Address(rsp, 0));
 466     BLOCK_COMMENT("call exception_handler_for_return_address");
 467     __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 468                          SharedRuntime::exception_handler_for_return_address),
 469                     c_rarg0);
 470     __ mov(rbx, rax);
 471 
 472     // setup rax & rdx, remove return address & clear pending exception
 473     __ pop(rdx);
 474     __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 475     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 476 
 477 #ifdef ASSERT
 478     // make sure exception is set
 479     {
 480       Label L;
 481       __ testptr(rax, rax);
 482       __ jcc(Assembler::notEqual, L);
 483       __ stop("StubRoutines::forward exception: no pending exception (2)");
 484       __ bind(L);
 485     }
 486 #endif
 487 
 488     // continue at exception handler (return address removed)
 489     // rax: exception
 490     // rbx: exception handler
 491     // rdx: throwing pc
 492     __ verify_oop(rax);
 493     __ jmp(rbx);
 494 
 495     return start;
 496   }
 497 
 498   // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
 499   //
 500   // Arguments :
 501   //    c_rarg0: exchange_value
 502   //    c_rarg0: dest
 503   //
 504   // Result:
 505   //    *dest <- ex, return (orig *dest)
 506   address generate_atomic_xchg() {
 507     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 508     address start = __ pc();
 509 
 510     __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
 511     __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
 512     __ ret(0);
 513 
 514     return start;
 515   }
 516 
 517   // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
 518   //
 519   // Arguments :
 520   //    c_rarg0: exchange_value
 521   //    c_rarg1: dest
 522   //
 523   // Result:
 524   //    *dest <- ex, return (orig *dest)
 525   address generate_atomic_xchg_ptr() {
 526     StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
 527     address start = __ pc();
 528 
 529     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 530     __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
 531     __ ret(0);
 532 
 533     return start;
 534   }
 535 
 536   // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
 537   //                                         jint compare_value)
 538   //
 539   // Arguments :
 540   //    c_rarg0: exchange_value
 541   //    c_rarg1: dest
 542   //    c_rarg2: compare_value
 543   //
 544   // Result:
 545   //    if ( compare_value == *dest ) {
 546   //       *dest = exchange_value
 547   //       return compare_value;
 548   //    else
 549   //       return *dest;
 550   address generate_atomic_cmpxchg() {
 551     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
 552     address start = __ pc();
 553 
 554     __ movl(rax, c_rarg2);
 555    if ( os::is_MP() ) __ lock();
 556     __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
 557     __ ret(0);
 558 
 559     return start;
 560   }
 561 
 562   // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
 563   //                                             volatile jlong* dest,
 564   //                                             jlong compare_value)
 565   // Arguments :
 566   //    c_rarg0: exchange_value
 567   //    c_rarg1: dest
 568   //    c_rarg2: compare_value
 569   //
 570   // Result:
 571   //    if ( compare_value == *dest ) {
 572   //       *dest = exchange_value
 573   //       return compare_value;
 574   //    else
 575   //       return *dest;
 576   address generate_atomic_cmpxchg_long() {
 577     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
 578     address start = __ pc();
 579 
 580     __ movq(rax, c_rarg2);
 581    if ( os::is_MP() ) __ lock();
 582     __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
 583     __ ret(0);
 584 
 585     return start;
 586   }
 587 
 588   // Support for jint atomic::add(jint add_value, volatile jint* dest)
 589   //
 590   // Arguments :
 591   //    c_rarg0: add_value
 592   //    c_rarg1: dest
 593   //
 594   // Result:
 595   //    *dest += add_value
 596   //    return *dest;
 597   address generate_atomic_add() {
 598     StubCodeMark mark(this, "StubRoutines", "atomic_add");
 599     address start = __ pc();
 600 
 601     __ movl(rax, c_rarg0);
 602    if ( os::is_MP() ) __ lock();
 603     __ xaddl(Address(c_rarg1, 0), c_rarg0);
 604     __ addl(rax, c_rarg0);
 605     __ ret(0);
 606 
 607     return start;
 608   }
 609 
 610   // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
 611   //
 612   // Arguments :
 613   //    c_rarg0: add_value
 614   //    c_rarg1: dest
 615   //
 616   // Result:
 617   //    *dest += add_value
 618   //    return *dest;
 619   address generate_atomic_add_ptr() {
 620     StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
 621     address start = __ pc();
 622 
 623     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 624    if ( os::is_MP() ) __ lock();
 625     __ xaddptr(Address(c_rarg1, 0), c_rarg0);
 626     __ addptr(rax, c_rarg0);
 627     __ ret(0);
 628 
 629     return start;
 630   }
 631 
 632   // Support for intptr_t OrderAccess::fence()
 633   //
 634   // Arguments :
 635   //
 636   // Result:
 637   address generate_orderaccess_fence() {
 638     StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 639     address start = __ pc();
 640     __ membar(Assembler::StoreLoad);
 641     __ ret(0);
 642 
 643     return start;
 644   }
 645 
 646   // Support for intptr_t get_previous_fp()
 647   //
 648   // This routine is used to find the previous frame pointer for the
 649   // caller (current_frame_guess). This is used as part of debugging
 650   // ps() is seemingly lost trying to find frames.
 651   // This code assumes that caller current_frame_guess) has a frame.
 652   address generate_get_previous_fp() {
 653     StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
 654     const Address old_fp(rbp, 0);
 655     const Address older_fp(rax, 0);
 656     address start = __ pc();
 657 
 658     __ enter();
 659     __ movptr(rax, old_fp); // callers fp
 660     __ movptr(rax, older_fp); // the frame for ps()
 661     __ pop(rbp);
 662     __ ret(0);
 663 
 664     return start;
 665   }
 666 
 667   //----------------------------------------------------------------------------------------------------
 668   // Support for void verify_mxcsr()
 669   //
 670   // This routine is used with -Xcheck:jni to verify that native
 671   // JNI code does not return to Java code without restoring the
 672   // MXCSR register to our expected state.
 673 
 674   address generate_verify_mxcsr() {
 675     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 676     address start = __ pc();
 677 
 678     const Address mxcsr_save(rsp, 0);
 679 
 680     if (CheckJNICalls) {
 681       Label ok_ret;
 682       __ push(rax);
 683       __ subptr(rsp, wordSize);      // allocate a temp location
 684       __ stmxcsr(mxcsr_save);
 685       __ movl(rax, mxcsr_save);
 686       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 687       __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
 688       __ jcc(Assembler::equal, ok_ret);
 689 
 690       __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 691 
 692       __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
 693 
 694       __ bind(ok_ret);
 695       __ addptr(rsp, wordSize);
 696       __ pop(rax);
 697     }
 698 
 699     __ ret(0);
 700 
 701     return start;
 702   }
 703 
 704   address generate_f2i_fixup() {
 705     StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 706     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 707 
 708     address start = __ pc();
 709 
 710     Label L;
 711 
 712     __ push(rax);
 713     __ push(c_rarg3);
 714     __ push(c_rarg2);
 715     __ push(c_rarg1);
 716 
 717     __ movl(rax, 0x7f800000);
 718     __ xorl(c_rarg3, c_rarg3);
 719     __ movl(c_rarg2, inout);
 720     __ movl(c_rarg1, c_rarg2);
 721     __ andl(c_rarg1, 0x7fffffff);
 722     __ cmpl(rax, c_rarg1); // NaN? -> 0
 723     __ jcc(Assembler::negative, L);
 724     __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 725     __ movl(c_rarg3, 0x80000000);
 726     __ movl(rax, 0x7fffffff);
 727     __ cmovl(Assembler::positive, c_rarg3, rax);
 728 
 729     __ bind(L);
 730     __ movptr(inout, c_rarg3);
 731 
 732     __ pop(c_rarg1);
 733     __ pop(c_rarg2);
 734     __ pop(c_rarg3);
 735     __ pop(rax);
 736 
 737     __ ret(0);
 738 
 739     return start;
 740   }
 741 
 742   address generate_f2l_fixup() {
 743     StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 744     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 745     address start = __ pc();
 746 
 747     Label L;
 748 
 749     __ push(rax);
 750     __ push(c_rarg3);
 751     __ push(c_rarg2);
 752     __ push(c_rarg1);
 753 
 754     __ movl(rax, 0x7f800000);
 755     __ xorl(c_rarg3, c_rarg3);
 756     __ movl(c_rarg2, inout);
 757     __ movl(c_rarg1, c_rarg2);
 758     __ andl(c_rarg1, 0x7fffffff);
 759     __ cmpl(rax, c_rarg1); // NaN? -> 0
 760     __ jcc(Assembler::negative, L);
 761     __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 762     __ mov64(c_rarg3, 0x8000000000000000);
 763     __ mov64(rax, 0x7fffffffffffffff);
 764     __ cmov(Assembler::positive, c_rarg3, rax);
 765 
 766     __ bind(L);
 767     __ movptr(inout, c_rarg3);
 768 
 769     __ pop(c_rarg1);
 770     __ pop(c_rarg2);
 771     __ pop(c_rarg3);
 772     __ pop(rax);
 773 
 774     __ ret(0);
 775 
 776     return start;
 777   }
 778 
 779   address generate_d2i_fixup() {
 780     StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 781     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 782 
 783     address start = __ pc();
 784 
 785     Label L;
 786 
 787     __ push(rax);
 788     __ push(c_rarg3);
 789     __ push(c_rarg2);
 790     __ push(c_rarg1);
 791     __ push(c_rarg0);
 792 
 793     __ movl(rax, 0x7ff00000);
 794     __ movq(c_rarg2, inout);
 795     __ movl(c_rarg3, c_rarg2);
 796     __ mov(c_rarg1, c_rarg2);
 797     __ mov(c_rarg0, c_rarg2);
 798     __ negl(c_rarg3);
 799     __ shrptr(c_rarg1, 0x20);
 800     __ orl(c_rarg3, c_rarg2);
 801     __ andl(c_rarg1, 0x7fffffff);
 802     __ xorl(c_rarg2, c_rarg2);
 803     __ shrl(c_rarg3, 0x1f);
 804     __ orl(c_rarg1, c_rarg3);
 805     __ cmpl(rax, c_rarg1);
 806     __ jcc(Assembler::negative, L); // NaN -> 0
 807     __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 808     __ movl(c_rarg2, 0x80000000);
 809     __ movl(rax, 0x7fffffff);
 810     __ cmov(Assembler::positive, c_rarg2, rax);
 811 
 812     __ bind(L);
 813     __ movptr(inout, c_rarg2);
 814 
 815     __ pop(c_rarg0);
 816     __ pop(c_rarg1);
 817     __ pop(c_rarg2);
 818     __ pop(c_rarg3);
 819     __ pop(rax);
 820 
 821     __ ret(0);
 822 
 823     return start;
 824   }
 825 
 826   address generate_d2l_fixup() {
 827     StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 828     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 829 
 830     address start = __ pc();
 831 
 832     Label L;
 833 
 834     __ push(rax);
 835     __ push(c_rarg3);
 836     __ push(c_rarg2);
 837     __ push(c_rarg1);
 838     __ push(c_rarg0);
 839 
 840     __ movl(rax, 0x7ff00000);
 841     __ movq(c_rarg2, inout);
 842     __ movl(c_rarg3, c_rarg2);
 843     __ mov(c_rarg1, c_rarg2);
 844     __ mov(c_rarg0, c_rarg2);
 845     __ negl(c_rarg3);
 846     __ shrptr(c_rarg1, 0x20);
 847     __ orl(c_rarg3, c_rarg2);
 848     __ andl(c_rarg1, 0x7fffffff);
 849     __ xorl(c_rarg2, c_rarg2);
 850     __ shrl(c_rarg3, 0x1f);
 851     __ orl(c_rarg1, c_rarg3);
 852     __ cmpl(rax, c_rarg1);
 853     __ jcc(Assembler::negative, L); // NaN -> 0
 854     __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 855     __ mov64(c_rarg2, 0x8000000000000000);
 856     __ mov64(rax, 0x7fffffffffffffff);
 857     __ cmovq(Assembler::positive, c_rarg2, rax);
 858 
 859     __ bind(L);
 860     __ movq(inout, c_rarg2);
 861 
 862     __ pop(c_rarg0);
 863     __ pop(c_rarg1);
 864     __ pop(c_rarg2);
 865     __ pop(c_rarg3);
 866     __ pop(rax);
 867 
 868     __ ret(0);
 869 
 870     return start;
 871   }
 872 
 873   address generate_fp_mask(const char *stub_name, int64_t mask) {
 874     StubCodeMark mark(this, "StubRoutines", stub_name);
 875 
 876     __ align(16);
 877     address start = __ pc();
 878 
 879     __ emit_data64( mask, relocInfo::none );
 880     __ emit_data64( mask, relocInfo::none );
 881 
 882     return start;
 883   }
 884 
 885   // The following routine generates a subroutine to throw an
 886   // asynchronous UnknownError when an unsafe access gets a fault that
 887   // could not be reasonably prevented by the programmer.  (Example:
 888   // SIGBUS/OBJERR.)
 889   address generate_handler_for_unsafe_access() {
 890     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 891     address start = __ pc();
 892 
 893     __ push(0);                       // hole for return address-to-be
 894     __ pusha();                       // push registers
 895     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 896 
 897     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 898     BLOCK_COMMENT("call handle_unsafe_access");
 899     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 900     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 901 
 902     __ movptr(next_pc, rax);          // stuff next address
 903     __ popa();
 904     __ ret(0);                        // jump to next address
 905 
 906     return start;
 907   }
 908 
 909   // Non-destructive plausibility checks for oops
 910   //
 911   // Arguments:
 912   //    all args on stack!
 913   //
 914   // Stack after saving c_rarg3:
 915   //    [tos + 0]: saved c_rarg3
 916   //    [tos + 1]: saved c_rarg2
 917   //    [tos + 2]: saved r12 (several TemplateTable methods use it)
 918   //    [tos + 3]: saved flags
 919   //    [tos + 4]: return address
 920   //  * [tos + 5]: error message (char*)
 921   //  * [tos + 6]: object to verify (oop)
 922   //  * [tos + 7]: saved rax - saved by caller and bashed
 923   //  * = popped on exit
 924   address generate_verify_oop() {
 925     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 926     address start = __ pc();
 927 
 928     Label exit, error;
 929 
 930     __ pushf();
 931     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 932 
 933     __ push(r12);
 934 
 935     // save c_rarg2 and c_rarg3
 936     __ push(c_rarg2);
 937     __ push(c_rarg3);
 938 
 939     enum {
 940            // After previous pushes.
 941            oop_to_verify = 6 * wordSize,
 942            saved_rax     = 7 * wordSize,
 943 
 944            // Before the call to MacroAssembler::debug(), see below.
 945            return_addr   = 16 * wordSize,
 946            error_msg     = 17 * wordSize
 947     };
 948 
 949     // get object
 950     __ movptr(rax, Address(rsp, oop_to_verify));
 951 
 952     // make sure object is 'reasonable'
 953     __ testptr(rax, rax);
 954     __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
 955     // Check if the oop is in the right area of memory
 956     __ movptr(c_rarg2, rax);
 957     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
 958     __ andptr(c_rarg2, c_rarg3);
 959     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
 960     __ cmpptr(c_rarg2, c_rarg3);
 961     __ jcc(Assembler::notZero, error);
 962 
 963     // set r12 to heapbase for load_klass()
 964     __ reinit_heapbase();
 965 
 966     // make sure klass is 'reasonable'
 967     __ load_klass(rax, rax);  // get klass
 968     __ testptr(rax, rax);
 969     __ jcc(Assembler::zero, error); // if klass is NULL it is broken
 970     // Check if the klass is in the right area of memory
 971     __ mov(c_rarg2, rax);
 972     __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
 973     __ andptr(c_rarg2, c_rarg3);
 974     __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
 975     __ cmpptr(c_rarg2, c_rarg3);
 976     __ jcc(Assembler::notZero, error);
 977 
 978     // make sure klass' klass is 'reasonable'
 979     __ load_klass(rax, rax);
 980     __ testptr(rax, rax);
 981     __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
 982     // Check if the klass' klass is in the right area of memory
 983     __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
 984     __ andptr(rax, c_rarg3);
 985     __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
 986     __ cmpptr(rax, c_rarg3);
 987     __ jcc(Assembler::notZero, error);
 988 
 989     // return if everything seems ok
 990     __ bind(exit);
 991     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
 992     __ pop(c_rarg3);                             // restore c_rarg3
 993     __ pop(c_rarg2);                             // restore c_rarg2
 994     __ pop(r12);                                 // restore r12
 995     __ popf();                                   // restore flags
 996     __ ret(3 * wordSize);                        // pop caller saved stuff
 997 
 998     // handle errors
 999     __ bind(error);
1000     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1001     __ pop(c_rarg3);                             // get saved c_rarg3 back
1002     __ pop(c_rarg2);                             // get saved c_rarg2 back
1003     __ pop(r12);                                 // get saved r12 back
1004     __ popf();                                   // get saved flags off stack --
1005                                                  // will be ignored
1006 
1007     __ pusha();                                  // push registers
1008                                                  // (rip is already
1009                                                  // already pushed)
1010     // debug(char* msg, int64_t pc, int64_t regs[])
1011     // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1012     // pushed all the registers, so now the stack looks like:
1013     //     [tos +  0] 16 saved registers
1014     //     [tos + 16] return address
1015     //   * [tos + 17] error message (char*)
1016     //   * [tos + 18] object to verify (oop)
1017     //   * [tos + 19] saved rax - saved by caller and bashed
1018     //   * = popped on exit
1019 
1020     __ movptr(c_rarg0, Address(rsp, error_msg));    // pass address of error message
1021     __ movptr(c_rarg1, Address(rsp, return_addr));  // pass return address
1022     __ movq(c_rarg2, rsp);                          // pass address of regs on stack
1023     __ mov(r12, rsp);                               // remember rsp
1024     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1025     __ andptr(rsp, -16);                            // align stack as required by ABI
1026     BLOCK_COMMENT("call MacroAssembler::debug");
1027     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1028     __ mov(rsp, r12);                               // restore rsp
1029     __ popa();                                      // pop registers (includes r12)
1030     __ ret(3 * wordSize);                           // pop caller saved stuff
1031 
1032     return start;
1033   }
1034 
1035   static address disjoint_byte_copy_entry;
1036   static address disjoint_short_copy_entry;
1037   static address disjoint_int_copy_entry;
1038   static address disjoint_long_copy_entry;
1039   static address disjoint_oop_copy_entry;
1040 
1041   static address byte_copy_entry;
1042   static address short_copy_entry;
1043   static address int_copy_entry;
1044   static address long_copy_entry;
1045   static address oop_copy_entry;
1046 
1047   static address checkcast_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     __ cmpptr(to, from);
1093     __ lea(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       __ cmpptr(to, end_from);
1098       __ jump_cc(Assembler::aboveEqual, no_overlap);
1099     } else {
1100       __ jcc(Assembler::belowEqual, (*NOLp));
1101       __ cmpptr(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       __ mov(rax, r9);  // r9 is also saved_rdi
1126     __ movptr(saved_rdi, rdi);
1127     __ movptr(saved_rsi, rsi);
1128     __ mov(rdi, rcx); // c_rarg0
1129     __ mov(rsi, rdx); // c_rarg1
1130     __ mov(rdx, r8);  // c_rarg2
1131     if (nargs >= 4)
1132       __ mov(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     __ movptr(rdi, saved_rdi);
1144     __ movptr(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     BarrierSet* bs = Universe::heap()->barrier_set();
1157     switch (bs->kind()) {
1158       case BarrierSet::G1SATBCT:
1159       case BarrierSet::G1SATBCTLogging:
1160         {
1161           __ pusha();                      // push registers
1162           if (count == c_rarg0) {
1163             if (addr == c_rarg1) {
1164               // exactly backwards!!
1165               __ xchgptr(c_rarg1, c_rarg0);
1166             } else {
1167               __ movptr(c_rarg1, count);
1168               __ movptr(c_rarg0, addr);
1169             }
1170 
1171           } else {
1172             __ movptr(c_rarg0, addr);
1173             __ movptr(c_rarg1, count);
1174           }
1175           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)));
1176           __ popa();
1177         }
1178         break;
1179       case BarrierSet::CardTableModRef:
1180       case BarrierSet::CardTableExtension:
1181       case BarrierSet::ModRef:
1182         break;
1183       default:
1184         ShouldNotReachHere();
1185 
1186     }
1187   }
1188 
1189   //
1190   // Generate code for an array write post barrier
1191   //
1192   //  Input:
1193   //     start    - register containing starting address of destination array
1194   //     end      - register containing ending address of destination array
1195   //     scratch  - scratch register
1196   //
1197   //  The input registers are overwritten.
1198   //  The ending address is inclusive.
1199   void  gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1200     assert_different_registers(start, end, scratch);
1201     BarrierSet* bs = Universe::heap()->barrier_set();
1202     switch (bs->kind()) {
1203       case BarrierSet::G1SATBCT:
1204       case BarrierSet::G1SATBCTLogging:
1205 
1206         {
1207           __ pusha();                      // push registers (overkill)
1208           // must compute element count unless barrier set interface is changed (other platforms supply count)
1209           assert_different_registers(start, end, scratch);
1210           __ lea(scratch, Address(end, BytesPerHeapOop));
1211           __ subptr(scratch, start);               // subtract start to get #bytes
1212           __ shrptr(scratch, LogBytesPerHeapOop);  // convert to element count
1213           __ mov(c_rarg0, start);
1214           __ mov(c_rarg1, scratch);
1215           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)));
1216           __ popa();
1217         }
1218         break;
1219       case BarrierSet::CardTableModRef:
1220       case BarrierSet::CardTableExtension:
1221         {
1222           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1223           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1224 
1225           Label L_loop;
1226 
1227            __ shrptr(start, CardTableModRefBS::card_shift);
1228            __ addptr(end, BytesPerHeapOop);
1229            __ shrptr(end, CardTableModRefBS::card_shift);
1230            __ subptr(end, start); // number of bytes to copy
1231 
1232           intptr_t disp = (intptr_t) ct->byte_map_base;
1233           if (__ is_simm32(disp)) {
1234             Address cardtable(noreg, noreg, Address::no_scale, disp);
1235             __ lea(scratch, cardtable);
1236           } else {
1237             ExternalAddress cardtable((address)disp);
1238             __ lea(scratch, cardtable);
1239           }
1240 
1241           const Register count = end; // 'end' register contains bytes count now
1242           __ addptr(start, scratch);
1243         __ BIND(L_loop);
1244           __ movb(Address(start, count, Address::times_1), 0);
1245           __ decrement(count);
1246           __ jcc(Assembler::greaterEqual, L_loop);
1247         }
1248         break;
1249       default:
1250         ShouldNotReachHere();
1251 
1252     }
1253   }
1254 
1255 
1256   // Copy big chunks forward
1257   //
1258   // Inputs:
1259   //   end_from     - source arrays end address
1260   //   end_to       - destination array end address
1261   //   qword_count  - 64-bits element count, negative
1262   //   to           - scratch
1263   //   L_copy_32_bytes - entry label
1264   //   L_copy_8_bytes  - exit  label
1265   //
1266   void copy_32_bytes_forward(Register end_from, Register end_to,
1267                              Register qword_count, Register to,
1268                              Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1269     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1270     Label L_loop;
1271     __ align(16);
1272   __ BIND(L_loop);
1273     if(UseUnalignedLoadStores) {
1274       __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1275       __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1276       __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1277       __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1278 
1279     } else {
1280       __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1281       __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1282       __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1283       __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1284       __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1285       __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1286       __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1287       __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1288     }
1289   __ BIND(L_copy_32_bytes);
1290     __ addptr(qword_count, 4);
1291     __ jcc(Assembler::lessEqual, L_loop);
1292     __ subptr(qword_count, 4);
1293     __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1294   }
1295 
1296 
1297   // Copy big chunks backward
1298   //
1299   // Inputs:
1300   //   from         - source arrays address
1301   //   dest         - destination array address
1302   //   qword_count  - 64-bits element count
1303   //   to           - scratch
1304   //   L_copy_32_bytes - entry label
1305   //   L_copy_8_bytes  - exit  label
1306   //
1307   void copy_32_bytes_backward(Register from, Register dest,
1308                               Register qword_count, Register to,
1309                               Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1310     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1311     Label L_loop;
1312     __ align(16);
1313   __ BIND(L_loop);
1314     if(UseUnalignedLoadStores) {
1315       __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1316       __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1317       __ movdqu(xmm1, Address(from, qword_count, Address::times_8,  0));
1318       __ movdqu(Address(dest, qword_count, Address::times_8,  0), xmm1);
1319 
1320     } else {
1321       __ movq(to, Address(from, qword_count, Address::times_8, 24));
1322       __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1323       __ movq(to, Address(from, qword_count, Address::times_8, 16));
1324       __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1325       __ movq(to, Address(from, qword_count, Address::times_8,  8));
1326       __ movq(Address(dest, qword_count, Address::times_8,  8), to);
1327       __ movq(to, Address(from, qword_count, Address::times_8,  0));
1328       __ movq(Address(dest, qword_count, Address::times_8,  0), to);
1329     }
1330   __ BIND(L_copy_32_bytes);
1331     __ subptr(qword_count, 4);
1332     __ jcc(Assembler::greaterEqual, L_loop);
1333     __ addptr(qword_count, 4);
1334     __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1335   }
1336 
1337 
1338   // Arguments:
1339   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1340   //             ignored
1341   //   name    - stub name string
1342   //
1343   // Inputs:
1344   //   c_rarg0   - source array address
1345   //   c_rarg1   - destination array address
1346   //   c_rarg2   - element count, treated as ssize_t, can be zero
1347   //
1348   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1349   // we let the hardware handle it.  The one to eight bytes within words,
1350   // dwords or qwords that span cache line boundaries will still be loaded
1351   // and stored atomically.
1352   //
1353   // Side Effects:
1354   //   disjoint_byte_copy_entry is set to the no-overlap entry point
1355   //   used by generate_conjoint_byte_copy().
1356   //
1357   address generate_disjoint_byte_copy(bool aligned, const char *name) {
1358     __ align(CodeEntryAlignment);
1359     StubCodeMark mark(this, "StubRoutines", name);
1360     address start = __ pc();
1361 
1362     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1363     Label L_copy_byte, L_exit;
1364     const Register from        = rdi;  // source array address
1365     const Register to          = rsi;  // destination array address
1366     const Register count       = rdx;  // elements count
1367     const Register byte_count  = rcx;
1368     const Register qword_count = count;
1369     const Register end_from    = from; // source array end address
1370     const Register end_to      = to;   // destination array end address
1371     // End pointers are inclusive, and if count is not zero they point
1372     // to the last unit copied:  end_to[0] := end_from[0]
1373 
1374     __ enter(); // required for proper stackwalking of RuntimeStub frame
1375     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1376 
1377     disjoint_byte_copy_entry = __ pc();
1378     BLOCK_COMMENT("Entry:");
1379     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1380 
1381     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1382                       // r9 and r10 may be used to save non-volatile registers
1383 
1384     // 'from', 'to' and 'count' are now valid
1385     __ movptr(byte_count, count);
1386     __ shrptr(count, 3); // count => qword_count
1387 
1388     // Copy from low to high addresses.  Use 'to' as scratch.
1389     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1390     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1391     __ negptr(qword_count); // make the count negative
1392     __ jmp(L_copy_32_bytes);
1393 
1394     // Copy trailing qwords
1395   __ BIND(L_copy_8_bytes);
1396     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1397     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1398     __ increment(qword_count);
1399     __ jcc(Assembler::notZero, L_copy_8_bytes);
1400 
1401     // Check for and copy trailing dword
1402   __ BIND(L_copy_4_bytes);
1403     __ testl(byte_count, 4);
1404     __ jccb(Assembler::zero, L_copy_2_bytes);
1405     __ movl(rax, Address(end_from, 8));
1406     __ movl(Address(end_to, 8), rax);
1407 
1408     __ addptr(end_from, 4);
1409     __ addptr(end_to, 4);
1410 
1411     // Check for and copy trailing word
1412   __ BIND(L_copy_2_bytes);
1413     __ testl(byte_count, 2);
1414     __ jccb(Assembler::zero, L_copy_byte);
1415     __ movw(rax, Address(end_from, 8));
1416     __ movw(Address(end_to, 8), rax);
1417 
1418     __ addptr(end_from, 2);
1419     __ addptr(end_to, 2);
1420 
1421     // Check for and copy trailing byte
1422   __ BIND(L_copy_byte);
1423     __ testl(byte_count, 1);
1424     __ jccb(Assembler::zero, L_exit);
1425     __ movb(rax, Address(end_from, 8));
1426     __ movb(Address(end_to, 8), rax);
1427 
1428   __ BIND(L_exit);
1429     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1430     restore_arg_regs();
1431     __ xorptr(rax, rax); // return 0
1432     __ leave(); // required for proper stackwalking of RuntimeStub frame
1433     __ ret(0);
1434 
1435     // Copy in 32-bytes chunks
1436     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1437     __ jmp(L_copy_4_bytes);
1438 
1439     return start;
1440   }
1441 
1442   // Arguments:
1443   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1444   //             ignored
1445   //   name    - stub name string
1446   //
1447   // Inputs:
1448   //   c_rarg0   - source array address
1449   //   c_rarg1   - destination array address
1450   //   c_rarg2   - element count, treated as ssize_t, can be zero
1451   //
1452   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1453   // we let the hardware handle it.  The one to eight bytes within words,
1454   // dwords or qwords that span cache line boundaries will still be loaded
1455   // and stored atomically.
1456   //
1457   address generate_conjoint_byte_copy(bool aligned, const char *name) {
1458     __ align(CodeEntryAlignment);
1459     StubCodeMark mark(this, "StubRoutines", name);
1460     address start = __ pc();
1461 
1462     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1463     const Register from        = rdi;  // source array address
1464     const Register to          = rsi;  // destination array address
1465     const Register count       = rdx;  // elements count
1466     const Register byte_count  = rcx;
1467     const Register qword_count = count;
1468 
1469     __ enter(); // required for proper stackwalking of RuntimeStub frame
1470     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1471 
1472     byte_copy_entry = __ pc();
1473     BLOCK_COMMENT("Entry:");
1474     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1475 
1476     array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1477     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1478                       // r9 and r10 may be used to save non-volatile registers
1479 
1480     // 'from', 'to' and 'count' are now valid
1481     __ movptr(byte_count, count);
1482     __ shrptr(count, 3);   // count => qword_count
1483 
1484     // Copy from high to low addresses.
1485 
1486     // Check for and copy trailing byte
1487     __ testl(byte_count, 1);
1488     __ jcc(Assembler::zero, L_copy_2_bytes);
1489     __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1490     __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1491     __ decrement(byte_count); // Adjust for possible trailing word
1492 
1493     // Check for and copy trailing word
1494   __ BIND(L_copy_2_bytes);
1495     __ testl(byte_count, 2);
1496     __ jcc(Assembler::zero, L_copy_4_bytes);
1497     __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1498     __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1499 
1500     // Check for and copy trailing dword
1501   __ BIND(L_copy_4_bytes);
1502     __ testl(byte_count, 4);
1503     __ jcc(Assembler::zero, L_copy_32_bytes);
1504     __ movl(rax, Address(from, qword_count, Address::times_8));
1505     __ movl(Address(to, qword_count, Address::times_8), rax);
1506     __ jmp(L_copy_32_bytes);
1507 
1508     // Copy trailing qwords
1509   __ BIND(L_copy_8_bytes);
1510     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1511     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1512     __ decrement(qword_count);
1513     __ jcc(Assembler::notZero, L_copy_8_bytes);
1514 
1515     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1516     restore_arg_regs();
1517     __ xorptr(rax, rax); // return 0
1518     __ leave(); // required for proper stackwalking of RuntimeStub frame
1519     __ ret(0);
1520 
1521     // Copy in 32-bytes chunks
1522     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1523 
1524     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1525     restore_arg_regs();
1526     __ xorptr(rax, rax); // return 0
1527     __ leave(); // required for proper stackwalking of RuntimeStub frame
1528     __ ret(0);
1529 
1530     return start;
1531   }
1532 
1533   // Arguments:
1534   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1535   //             ignored
1536   //   name    - stub name string
1537   //
1538   // Inputs:
1539   //   c_rarg0   - source array address
1540   //   c_rarg1   - destination array address
1541   //   c_rarg2   - element count, treated as ssize_t, can be zero
1542   //
1543   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1544   // let the hardware handle it.  The two or four words within dwords
1545   // or qwords that span cache line boundaries will still be loaded
1546   // and stored atomically.
1547   //
1548   // Side Effects:
1549   //   disjoint_short_copy_entry is set to the no-overlap entry point
1550   //   used by generate_conjoint_short_copy().
1551   //
1552   address generate_disjoint_short_copy(bool aligned, const char *name) {
1553     __ align(CodeEntryAlignment);
1554     StubCodeMark mark(this, "StubRoutines", name);
1555     address start = __ pc();
1556 
1557     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1558     const Register from        = rdi;  // source array address
1559     const Register to          = rsi;  // destination array address
1560     const Register count       = rdx;  // elements count
1561     const Register word_count  = rcx;
1562     const Register qword_count = count;
1563     const Register end_from    = from; // source array end address
1564     const Register end_to      = to;   // destination array end address
1565     // End pointers are inclusive, and if count is not zero they point
1566     // to the last unit copied:  end_to[0] := end_from[0]
1567 
1568     __ enter(); // required for proper stackwalking of RuntimeStub frame
1569     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1570 
1571     disjoint_short_copy_entry = __ pc();
1572     BLOCK_COMMENT("Entry:");
1573     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1574 
1575     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1576                       // r9 and r10 may be used to save non-volatile registers
1577 
1578     // 'from', 'to' and 'count' are now valid
1579     __ movptr(word_count, count);
1580     __ shrptr(count, 2); // count => qword_count
1581 
1582     // Copy from low to high addresses.  Use 'to' as scratch.
1583     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1584     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1585     __ negptr(qword_count);
1586     __ jmp(L_copy_32_bytes);
1587 
1588     // Copy trailing qwords
1589   __ BIND(L_copy_8_bytes);
1590     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1591     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1592     __ increment(qword_count);
1593     __ jcc(Assembler::notZero, L_copy_8_bytes);
1594 
1595     // Original 'dest' is trashed, so we can't use it as a
1596     // base register for a possible trailing word copy
1597 
1598     // Check for and copy trailing dword
1599   __ BIND(L_copy_4_bytes);
1600     __ testl(word_count, 2);
1601     __ jccb(Assembler::zero, L_copy_2_bytes);
1602     __ movl(rax, Address(end_from, 8));
1603     __ movl(Address(end_to, 8), rax);
1604 
1605     __ addptr(end_from, 4);
1606     __ addptr(end_to, 4);
1607 
1608     // Check for and copy trailing word
1609   __ BIND(L_copy_2_bytes);
1610     __ testl(word_count, 1);
1611     __ jccb(Assembler::zero, L_exit);
1612     __ movw(rax, Address(end_from, 8));
1613     __ movw(Address(end_to, 8), rax);
1614 
1615   __ BIND(L_exit);
1616     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1617     restore_arg_regs();
1618     __ xorptr(rax, rax); // return 0
1619     __ leave(); // required for proper stackwalking of RuntimeStub frame
1620     __ ret(0);
1621 
1622     // Copy in 32-bytes chunks
1623     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1624     __ jmp(L_copy_4_bytes);
1625 
1626     return start;
1627   }
1628 
1629   // Arguments:
1630   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1631   //             ignored
1632   //   name    - stub name string
1633   //
1634   // Inputs:
1635   //   c_rarg0   - source array address
1636   //   c_rarg1   - destination array address
1637   //   c_rarg2   - element count, treated as ssize_t, can be zero
1638   //
1639   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1640   // let the hardware handle it.  The two or four words within dwords
1641   // or qwords that span cache line boundaries will still be loaded
1642   // and stored atomically.
1643   //
1644   address generate_conjoint_short_copy(bool aligned, const char *name) {
1645     __ align(CodeEntryAlignment);
1646     StubCodeMark mark(this, "StubRoutines", name);
1647     address start = __ pc();
1648 
1649     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1650     const Register from        = rdi;  // source array address
1651     const Register to          = rsi;  // destination array address
1652     const Register count       = rdx;  // elements count
1653     const Register word_count  = rcx;
1654     const Register qword_count = count;
1655 
1656     __ enter(); // required for proper stackwalking of RuntimeStub frame
1657     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1658 
1659     short_copy_entry = __ pc();
1660     BLOCK_COMMENT("Entry:");
1661     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1662 
1663     array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1664     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1665                       // r9 and r10 may be used to save non-volatile registers
1666 
1667     // 'from', 'to' and 'count' are now valid
1668     __ movptr(word_count, count);
1669     __ shrptr(count, 2); // count => qword_count
1670 
1671     // Copy from high to low addresses.  Use 'to' as scratch.
1672 
1673     // Check for and copy trailing word
1674     __ testl(word_count, 1);
1675     __ jccb(Assembler::zero, L_copy_4_bytes);
1676     __ movw(rax, Address(from, word_count, Address::times_2, -2));
1677     __ movw(Address(to, word_count, Address::times_2, -2), rax);
1678 
1679     // Check for and copy trailing dword
1680   __ BIND(L_copy_4_bytes);
1681     __ testl(word_count, 2);
1682     __ jcc(Assembler::zero, L_copy_32_bytes);
1683     __ movl(rax, Address(from, qword_count, Address::times_8));
1684     __ movl(Address(to, qword_count, Address::times_8), rax);
1685     __ jmp(L_copy_32_bytes);
1686 
1687     // Copy trailing qwords
1688   __ BIND(L_copy_8_bytes);
1689     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1690     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1691     __ decrement(qword_count);
1692     __ jcc(Assembler::notZero, L_copy_8_bytes);
1693 
1694     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1695     restore_arg_regs();
1696     __ xorptr(rax, rax); // return 0
1697     __ leave(); // required for proper stackwalking of RuntimeStub frame
1698     __ ret(0);
1699 
1700     // Copy in 32-bytes chunks
1701     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1702 
1703     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1704     restore_arg_regs();
1705     __ xorptr(rax, rax); // return 0
1706     __ leave(); // required for proper stackwalking of RuntimeStub frame
1707     __ ret(0);
1708 
1709     return start;
1710   }
1711 
1712   // Arguments:
1713   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1714   //             ignored
1715   //   is_oop  - true => oop array, so generate store check code
1716   //   name    - stub name string
1717   //
1718   // Inputs:
1719   //   c_rarg0   - source array address
1720   //   c_rarg1   - destination array address
1721   //   c_rarg2   - element count, treated as ssize_t, can be zero
1722   //
1723   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1724   // the hardware handle it.  The two dwords within qwords that span
1725   // cache line boundaries will still be loaded and stored atomicly.
1726   //
1727   // Side Effects:
1728   //   disjoint_int_copy_entry is set to the no-overlap entry point
1729   //   used by generate_conjoint_int_oop_copy().
1730   //
1731   address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1732     __ align(CodeEntryAlignment);
1733     StubCodeMark mark(this, "StubRoutines", name);
1734     address start = __ pc();
1735 
1736     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1737     const Register from        = rdi;  // source array address
1738     const Register to          = rsi;  // destination array address
1739     const Register count       = rdx;  // elements count
1740     const Register dword_count = rcx;
1741     const Register qword_count = count;
1742     const Register end_from    = from; // source array end address
1743     const Register end_to      = to;   // destination array end address
1744     const Register saved_to    = r11;  // saved destination array address
1745     // End pointers are inclusive, and if count is not zero they point
1746     // to the last unit copied:  end_to[0] := end_from[0]
1747 
1748     __ enter(); // required for proper stackwalking of RuntimeStub frame
1749     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1750 
1751     (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1752 
1753     if (is_oop) {
1754       // no registers are destroyed by this call
1755       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1756     }
1757 
1758     BLOCK_COMMENT("Entry:");
1759     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1760 
1761     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1762                       // r9 and r10 may be used to save non-volatile registers
1763 
1764     if (is_oop) {
1765       __ movq(saved_to, to);
1766     }
1767 
1768     // 'from', 'to' and 'count' are now valid
1769     __ movptr(dword_count, count);
1770     __ shrptr(count, 1); // count => qword_count
1771 
1772     // Copy from low to high addresses.  Use 'to' as scratch.
1773     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1774     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1775     __ negptr(qword_count);
1776     __ jmp(L_copy_32_bytes);
1777 
1778     // Copy trailing qwords
1779   __ BIND(L_copy_8_bytes);
1780     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1781     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1782     __ increment(qword_count);
1783     __ jcc(Assembler::notZero, L_copy_8_bytes);
1784 
1785     // Check for and copy trailing dword
1786   __ BIND(L_copy_4_bytes);
1787     __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1788     __ jccb(Assembler::zero, L_exit);
1789     __ movl(rax, Address(end_from, 8));
1790     __ movl(Address(end_to, 8), rax);
1791 
1792   __ BIND(L_exit);
1793     if (is_oop) {
1794       __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1795       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1796     }
1797     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1798     restore_arg_regs();
1799     __ xorptr(rax, rax); // return 0
1800     __ leave(); // required for proper stackwalking of RuntimeStub frame
1801     __ ret(0);
1802 
1803     // Copy 32-bytes chunks
1804     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1805     __ jmp(L_copy_4_bytes);
1806 
1807     return start;
1808   }
1809 
1810   // Arguments:
1811   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1812   //             ignored
1813   //   is_oop  - true => oop array, so generate store check code
1814   //   name    - stub name string
1815   //
1816   // Inputs:
1817   //   c_rarg0   - source array address
1818   //   c_rarg1   - destination array address
1819   //   c_rarg2   - element count, treated as ssize_t, can be zero
1820   //
1821   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1822   // the hardware handle it.  The two dwords within qwords that span
1823   // cache line boundaries will still be loaded and stored atomicly.
1824   //
1825   address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1826     __ align(CodeEntryAlignment);
1827     StubCodeMark mark(this, "StubRoutines", name);
1828     address start = __ pc();
1829 
1830     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1831     const Register from        = rdi;  // source array address
1832     const Register to          = rsi;  // destination array address
1833     const Register count       = rdx;  // elements count
1834     const Register dword_count = rcx;
1835     const Register qword_count = count;
1836 
1837     __ enter(); // required for proper stackwalking of RuntimeStub frame
1838     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1839 
1840     if (is_oop) {
1841       // no registers are destroyed by this call
1842       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1843     }
1844 
1845     (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1846     BLOCK_COMMENT("Entry:");
1847     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1848 
1849     array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1850                        Address::times_4);
1851     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1852                       // r9 and r10 may be used to save non-volatile registers
1853 
1854     assert_clean_int(count, rax); // Make sure 'count' is clean int.
1855     // 'from', 'to' and 'count' are now valid
1856     __ movptr(dword_count, count);
1857     __ shrptr(count, 1); // count => qword_count
1858 
1859     // Copy from high to low addresses.  Use 'to' as scratch.
1860 
1861     // Check for and copy trailing dword
1862     __ testl(dword_count, 1);
1863     __ jcc(Assembler::zero, L_copy_32_bytes);
1864     __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1865     __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1866     __ jmp(L_copy_32_bytes);
1867 
1868     // Copy trailing qwords
1869   __ BIND(L_copy_8_bytes);
1870     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1871     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1872     __ decrement(qword_count);
1873     __ jcc(Assembler::notZero, L_copy_8_bytes);
1874 
1875     inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1876     if (is_oop) {
1877       __ jmp(L_exit);
1878     }
1879     restore_arg_regs();
1880     __ xorptr(rax, rax); // return 0
1881     __ leave(); // required for proper stackwalking of RuntimeStub frame
1882     __ ret(0);
1883 
1884     // Copy in 32-bytes chunks
1885     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1886 
1887    inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1888    __ bind(L_exit);
1889      if (is_oop) {
1890        Register end_to = rdx;
1891        __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1892        gen_write_ref_array_post_barrier(to, end_to, rax);
1893      }
1894     restore_arg_regs();
1895     __ xorptr(rax, rax); // return 0
1896     __ leave(); // required for proper stackwalking of RuntimeStub frame
1897     __ ret(0);
1898 
1899     return start;
1900   }
1901 
1902   // Arguments:
1903   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1904   //             ignored
1905   //   is_oop  - true => oop array, so generate store check code
1906   //   name    - stub name string
1907   //
1908   // Inputs:
1909   //   c_rarg0   - source array address
1910   //   c_rarg1   - destination array address
1911   //   c_rarg2   - element count, treated as ssize_t, can be zero
1912   //
1913  // Side Effects:
1914   //   disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1915   //   no-overlap entry point used by generate_conjoint_long_oop_copy().
1916   //
1917   address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1918     __ align(CodeEntryAlignment);
1919     StubCodeMark mark(this, "StubRoutines", name);
1920     address start = __ pc();
1921 
1922     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1923     const Register from        = rdi;  // source array address
1924     const Register to          = rsi;  // destination array address
1925     const Register qword_count = rdx;  // elements count
1926     const Register end_from    = from; // source array end address
1927     const Register end_to      = rcx;  // destination array end address
1928     const Register saved_to    = to;
1929     // End pointers are inclusive, and if count is not zero they point
1930     // to the last unit copied:  end_to[0] := end_from[0]
1931 
1932     __ enter(); // required for proper stackwalking of RuntimeStub frame
1933     // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1934     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1935 
1936     if (is_oop) {
1937       disjoint_oop_copy_entry  = __ pc();
1938       // no registers are destroyed by this call
1939       gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1940     } else {
1941       disjoint_long_copy_entry = __ pc();
1942     }
1943     BLOCK_COMMENT("Entry:");
1944     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1945 
1946     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1947                       // r9 and r10 may be used to save non-volatile registers
1948 
1949     // 'from', 'to' and 'qword_count' are now valid
1950 
1951     // Copy from low to high addresses.  Use 'to' as scratch.
1952     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1953     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1954     __ negptr(qword_count);
1955     __ jmp(L_copy_32_bytes);
1956 
1957     // Copy trailing qwords
1958   __ BIND(L_copy_8_bytes);
1959     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1960     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1961     __ increment(qword_count);
1962     __ jcc(Assembler::notZero, L_copy_8_bytes);
1963 
1964     if (is_oop) {
1965       __ jmp(L_exit);
1966     } else {
1967       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1968       restore_arg_regs();
1969       __ xorptr(rax, rax); // return 0
1970       __ leave(); // required for proper stackwalking of RuntimeStub frame
1971       __ ret(0);
1972     }
1973 
1974     // Copy 64-byte chunks
1975     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1976 
1977     if (is_oop) {
1978     __ BIND(L_exit);
1979       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1980       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1981     } else {
1982       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1983     }
1984     restore_arg_regs();
1985     __ xorptr(rax, rax); // return 0
1986     __ leave(); // required for proper stackwalking of RuntimeStub frame
1987     __ ret(0);
1988 
1989     return start;
1990   }
1991 
1992   // Arguments:
1993   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1994   //             ignored
1995   //   is_oop  - true => oop array, so generate store check code
1996   //   name    - stub name string
1997   //
1998   // Inputs:
1999   //   c_rarg0   - source array address
2000   //   c_rarg1   - destination array address
2001   //   c_rarg2   - element count, treated as ssize_t, can be zero
2002   //
2003   address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
2004     __ align(CodeEntryAlignment);
2005     StubCodeMark mark(this, "StubRoutines", name);
2006     address start = __ pc();
2007 
2008     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2009     const Register from        = rdi;  // source array address
2010     const Register to          = rsi;  // destination array address
2011     const Register qword_count = rdx;  // elements count
2012     const Register saved_count = rcx;
2013 
2014     __ enter(); // required for proper stackwalking of RuntimeStub frame
2015     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
2016 
2017     address disjoint_copy_entry = NULL;
2018     if (is_oop) {
2019       assert(!UseCompressedOops, "shouldn't be called for compressed oops");
2020       disjoint_copy_entry = disjoint_oop_copy_entry;
2021       oop_copy_entry  = __ pc();
2022       array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
2023     } else {
2024       disjoint_copy_entry = disjoint_long_copy_entry;
2025       long_copy_entry = __ pc();
2026       array_overlap_test(disjoint_long_copy_entry, Address::times_8);
2027     }
2028     BLOCK_COMMENT("Entry:");
2029     // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2030 
2031     array_overlap_test(disjoint_copy_entry, Address::times_8);
2032     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2033                       // r9 and r10 may be used to save non-volatile registers
2034 
2035     // 'from', 'to' and 'qword_count' are now valid
2036 
2037     if (is_oop) {
2038       // Save to and count for store barrier
2039       __ movptr(saved_count, qword_count);
2040       // No registers are destroyed by this call
2041       gen_write_ref_array_pre_barrier(to, saved_count);
2042     }
2043 
2044     __ jmp(L_copy_32_bytes);
2045 
2046     // Copy trailing qwords
2047   __ BIND(L_copy_8_bytes);
2048     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2049     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2050     __ decrement(qword_count);
2051     __ jcc(Assembler::notZero, L_copy_8_bytes);
2052 
2053     if (is_oop) {
2054       __ jmp(L_exit);
2055     } else {
2056       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2057       restore_arg_regs();
2058       __ xorptr(rax, rax); // return 0
2059       __ leave(); // required for proper stackwalking of RuntimeStub frame
2060       __ ret(0);
2061     }
2062 
2063     // Copy in 32-bytes chunks
2064     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2065 
2066     if (is_oop) {
2067     __ BIND(L_exit);
2068       __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2069       gen_write_ref_array_post_barrier(to, rcx, rax);
2070       inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2071     } else {
2072       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2073     }
2074     restore_arg_regs();
2075     __ xorptr(rax, rax); // return 0
2076     __ leave(); // required for proper stackwalking of RuntimeStub frame
2077     __ ret(0);
2078 
2079     return start;
2080   }
2081 
2082 
2083   // Helper for generating a dynamic type check.
2084   // Smashes no registers.
2085   void generate_type_check(Register sub_klass,
2086                            Register super_check_offset,
2087                            Register super_klass,
2088                            Label& L_success) {
2089     assert_different_registers(sub_klass, super_check_offset, super_klass);
2090 
2091     BLOCK_COMMENT("type_check:");
2092 
2093     Label L_miss;
2094 
2095     __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg,        &L_success, &L_miss, NULL,
2096                                      super_check_offset);
2097     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2098 
2099     // Fall through on failure!
2100     __ BIND(L_miss);
2101   }
2102 
2103   //
2104   //  Generate checkcasting array copy stub
2105   //
2106   //  Input:
2107   //    c_rarg0   - source array address
2108   //    c_rarg1   - destination array address
2109   //    c_rarg2   - element count, treated as ssize_t, can be zero
2110   //    c_rarg3   - size_t ckoff (super_check_offset)
2111   // not Win64
2112   //    c_rarg4   - oop ckval (super_klass)
2113   // Win64
2114   //    rsp+40    - oop ckval (super_klass)
2115   //
2116   //  Output:
2117   //    rax ==  0  -  success
2118   //    rax == -1^K - failure, where K is partial transfer count
2119   //
2120   address generate_checkcast_copy(const char *name) {
2121 
2122     Label L_load_element, L_store_element, L_do_card_marks, L_done;
2123 
2124     // Input registers (after setup_arg_regs)
2125     const Register from        = rdi;   // source array address
2126     const Register to          = rsi;   // destination array address
2127     const Register length      = rdx;   // elements count
2128     const Register ckoff       = rcx;   // super_check_offset
2129     const Register ckval       = r8;    // super_klass
2130 
2131     // Registers used as temps (r13, r14 are save-on-entry)
2132     const Register end_from    = from;  // source array end address
2133     const Register end_to      = r13;   // destination array end address
2134     const Register count       = rdx;   // -(count_remaining)
2135     const Register r14_length  = r14;   // saved copy of length
2136     // End pointers are inclusive, and if length is not zero they point
2137     // to the last unit copied:  end_to[0] := end_from[0]
2138 
2139     const Register rax_oop    = rax;    // actual oop copied
2140     const Register r11_klass  = r11;    // oop._klass
2141 
2142     //---------------------------------------------------------------
2143     // Assembler stub will be used for this call to arraycopy
2144     // if the two arrays are subtypes of Object[] but the
2145     // destination array type is not equal to or a supertype
2146     // of the source type.  Each element must be separately
2147     // checked.
2148 
2149     __ align(CodeEntryAlignment);
2150     StubCodeMark mark(this, "StubRoutines", name);
2151     address start = __ pc();
2152 
2153     __ enter(); // required for proper stackwalking of RuntimeStub frame
2154 
2155     checkcast_copy_entry  = __ pc();
2156     BLOCK_COMMENT("Entry:");
2157 
2158 #ifdef ASSERT
2159     // caller guarantees that the arrays really are different
2160     // otherwise, we would have to make conjoint checks
2161     { Label L;
2162       array_overlap_test(L, TIMES_OOP);
2163       __ stop("checkcast_copy within a single array");
2164       __ bind(L);
2165     }
2166 #endif //ASSERT
2167 
2168     // allocate spill slots for r13, r14
2169     enum {
2170       saved_r13_offset,
2171       saved_r14_offset,
2172       saved_rbp_offset,
2173       saved_rip_offset,
2174       saved_rarg0_offset
2175     };
2176     __ subptr(rsp, saved_rbp_offset * wordSize);
2177     __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2178     __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2179     setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2180                        // ckoff => rcx, ckval => r8
2181                        // r9 and r10 may be used to save non-volatile registers
2182 #ifdef _WIN64
2183     // last argument (#4) is on stack on Win64
2184     const int ckval_offset = saved_rarg0_offset + 4;
2185     __ movptr(ckval, Address(rsp, ckval_offset * wordSize));
2186 #endif
2187 
2188     // check that int operands are properly extended to size_t
2189     assert_clean_int(length, rax);
2190     assert_clean_int(ckoff, rax);
2191 
2192 #ifdef ASSERT
2193     BLOCK_COMMENT("assert consistent ckoff/ckval");
2194     // The ckoff and ckval must be mutually consistent,
2195     // even though caller generates both.
2196     { Label L;
2197       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2198                         Klass::super_check_offset_offset_in_bytes());
2199       __ cmpl(ckoff, Address(ckval, sco_offset));
2200       __ jcc(Assembler::equal, L);
2201       __ stop("super_check_offset inconsistent");
2202       __ bind(L);
2203     }
2204 #endif //ASSERT
2205 
2206     // Loop-invariant addresses.  They are exclusive end pointers.
2207     Address end_from_addr(from, length, TIMES_OOP, 0);
2208     Address   end_to_addr(to,   length, TIMES_OOP, 0);
2209     // Loop-variant addresses.  They assume post-incremented count < 0.
2210     Address from_element_addr(end_from, count, TIMES_OOP, 0);
2211     Address   to_element_addr(end_to,   count, TIMES_OOP, 0);
2212 
2213     gen_write_ref_array_pre_barrier(to, count);
2214 
2215     // Copy from low to high addresses, indexed from the end of each array.
2216     __ lea(end_from, end_from_addr);
2217     __ lea(end_to,   end_to_addr);
2218     __ movptr(r14_length, length);        // save a copy of the length
2219     assert(length == count, "");          // else fix next line:
2220     __ negptr(count);                     // negate and test the length
2221     __ jcc(Assembler::notZero, L_load_element);
2222 
2223     // Empty array:  Nothing to do.
2224     __ xorptr(rax, rax);                  // return 0 on (trivial) success
2225     __ jmp(L_done);
2226 
2227     // ======== begin loop ========
2228     // (Loop is rotated; its entry is L_load_element.)
2229     // Loop control:
2230     //   for (count = -count; count != 0; count++)
2231     // Base pointers src, dst are biased by 8*(count-1),to last element.
2232     __ align(16);
2233 
2234     __ BIND(L_store_element);
2235     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop
2236     __ increment(count);               // increment the count toward zero
2237     __ jcc(Assembler::zero, L_do_card_marks);
2238 
2239     // ======== loop entry is here ========
2240     __ BIND(L_load_element);
2241     __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2242     __ testptr(rax_oop, rax_oop);
2243     __ jcc(Assembler::zero, L_store_element);
2244 
2245     __ load_klass(r11_klass, rax_oop);// query the object klass
2246     generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2247     // ======== end loop ========
2248 
2249     // It was a real error; we must depend on the caller to finish the job.
2250     // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2251     // Emit GC store barriers for the oops we have copied (r14 + rdx),
2252     // and report their number to the caller.
2253     assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2254     __ lea(end_to, to_element_addr);
2255     __ addptr(end_to, -heapOopSize);      // make an inclusive end pointer
2256     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2257     __ movptr(rax, r14_length);           // original oops
2258     __ addptr(rax, count);                // K = (original - remaining) oops
2259     __ notptr(rax);                       // report (-1^K) to caller
2260     __ jmp(L_done);
2261 
2262     // Come here on success only.
2263     __ BIND(L_do_card_marks);
2264     __ addptr(end_to, -heapOopSize);         // make an inclusive end pointer
2265     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2266     __ xorptr(rax, rax);                  // return 0 on success
2267 
2268     // Common exit point (success or failure).
2269     __ BIND(L_done);
2270     __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2271     __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2272     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2273     restore_arg_regs();
2274     __ leave(); // required for proper stackwalking of RuntimeStub frame
2275     __ ret(0);
2276 
2277     return start;
2278   }
2279 
2280   //
2281   //  Generate 'unsafe' array copy stub
2282   //  Though just as safe as the other stubs, it takes an unscaled
2283   //  size_t argument instead of an element count.
2284   //
2285   //  Input:
2286   //    c_rarg0   - source array address
2287   //    c_rarg1   - destination array address
2288   //    c_rarg2   - byte count, treated as ssize_t, can be zero
2289   //
2290   // Examines the alignment of the operands and dispatches
2291   // to a long, int, short, or byte copy loop.
2292   //
2293   address generate_unsafe_copy(const char *name) {
2294 
2295     Label L_long_aligned, L_int_aligned, L_short_aligned;
2296 
2297     // Input registers (before setup_arg_regs)
2298     const Register from        = c_rarg0;  // source array address
2299     const Register to          = c_rarg1;  // destination array address
2300     const Register size        = c_rarg2;  // byte count (size_t)
2301 
2302     // Register used as a temp
2303     const Register bits        = rax;      // test copy of low bits
2304 
2305     __ align(CodeEntryAlignment);
2306     StubCodeMark mark(this, "StubRoutines", name);
2307     address start = __ pc();
2308 
2309     __ enter(); // required for proper stackwalking of RuntimeStub frame
2310 
2311     // bump this on entry, not on exit:
2312     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2313 
2314     __ mov(bits, from);
2315     __ orptr(bits, to);
2316     __ orptr(bits, size);
2317 
2318     __ testb(bits, BytesPerLong-1);
2319     __ jccb(Assembler::zero, L_long_aligned);
2320 
2321     __ testb(bits, BytesPerInt-1);
2322     __ jccb(Assembler::zero, L_int_aligned);
2323 
2324     __ testb(bits, BytesPerShort-1);
2325     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2326 
2327     __ BIND(L_short_aligned);
2328     __ shrptr(size, LogBytesPerShort); // size => short_count
2329     __ jump(RuntimeAddress(short_copy_entry));
2330 
2331     __ BIND(L_int_aligned);
2332     __ shrptr(size, LogBytesPerInt); // size => int_count
2333     __ jump(RuntimeAddress(int_copy_entry));
2334 
2335     __ BIND(L_long_aligned);
2336     __ shrptr(size, LogBytesPerLong); // size => qword_count
2337     __ jump(RuntimeAddress(long_copy_entry));
2338 
2339     return start;
2340   }
2341 
2342   // Perform range checks on the proposed arraycopy.
2343   // Kills temp, but nothing else.
2344   // Also, clean the sign bits of src_pos and dst_pos.
2345   void arraycopy_range_checks(Register src,     // source array oop (c_rarg0)
2346                               Register src_pos, // source position (c_rarg1)
2347                               Register dst,     // destination array oo (c_rarg2)
2348                               Register dst_pos, // destination position (c_rarg3)
2349                               Register length,
2350                               Register temp,
2351                               Label& L_failed) {
2352     BLOCK_COMMENT("arraycopy_range_checks:");
2353 
2354     //  if (src_pos + length > arrayOop(src)->length())  FAIL;
2355     __ movl(temp, length);
2356     __ addl(temp, src_pos);             // src_pos + length
2357     __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2358     __ jcc(Assembler::above, L_failed);
2359 
2360     //  if (dst_pos + length > arrayOop(dst)->length())  FAIL;
2361     __ movl(temp, length);
2362     __ addl(temp, dst_pos);             // dst_pos + length
2363     __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2364     __ jcc(Assembler::above, L_failed);
2365 
2366     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2367     // Move with sign extension can be used since they are positive.
2368     __ movslq(src_pos, src_pos);
2369     __ movslq(dst_pos, dst_pos);
2370 
2371     BLOCK_COMMENT("arraycopy_range_checks done");
2372   }
2373 
2374   //
2375   //  Generate generic array copy stubs
2376   //
2377   //  Input:
2378   //    c_rarg0    -  src oop
2379   //    c_rarg1    -  src_pos (32-bits)
2380   //    c_rarg2    -  dst oop
2381   //    c_rarg3    -  dst_pos (32-bits)
2382   // not Win64
2383   //    c_rarg4    -  element count (32-bits)
2384   // Win64
2385   //    rsp+40     -  element count (32-bits)
2386   //
2387   //  Output:
2388   //    rax ==  0  -  success
2389   //    rax == -1^K - failure, where K is partial transfer count
2390   //
2391   address generate_generic_copy(const char *name) {
2392 
2393     Label L_failed, L_failed_0, L_objArray;
2394     Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2395 
2396     // Input registers
2397     const Register src        = c_rarg0;  // source array oop
2398     const Register src_pos    = c_rarg1;  // source position
2399     const Register dst        = c_rarg2;  // destination array oop
2400     const Register dst_pos    = c_rarg3;  // destination position
2401     // elements count is on stack on Win64
2402 #ifdef _WIN64
2403 #define C_RARG4 Address(rsp, 6 * wordSize)
2404 #else
2405 #define C_RARG4 c_rarg4
2406 #endif
2407 
2408     { int modulus = CodeEntryAlignment;
2409       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
2410       int advance = target - (__ offset() % modulus);
2411       if (advance < 0)  advance += modulus;
2412       if (advance > 0)  __ nop(advance);
2413     }
2414     StubCodeMark mark(this, "StubRoutines", name);
2415 
2416     // Short-hop target to L_failed.  Makes for denser prologue code.
2417     __ BIND(L_failed_0);
2418     __ jmp(L_failed);
2419     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2420 
2421     __ align(CodeEntryAlignment);
2422     address start = __ pc();
2423 
2424     __ enter(); // required for proper stackwalking of RuntimeStub frame
2425 
2426     // bump this on entry, not on exit:
2427     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2428 
2429     //-----------------------------------------------------------------------
2430     // Assembler stub will be used for this call to arraycopy
2431     // if the following conditions are met:
2432     //
2433     // (1) src and dst must not be null.
2434     // (2) src_pos must not be negative.
2435     // (3) dst_pos must not be negative.
2436     // (4) length  must not be negative.
2437     // (5) src klass and dst klass should be the same and not NULL.
2438     // (6) src and dst should be arrays.
2439     // (7) src_pos + length must not exceed length of src.
2440     // (8) dst_pos + length must not exceed length of dst.
2441     //
2442 
2443     //  if (src == NULL) return -1;
2444     __ testptr(src, src);         // src oop
2445     size_t j1off = __ offset();
2446     __ jccb(Assembler::zero, L_failed_0);
2447 
2448     //  if (src_pos < 0) return -1;
2449     __ testl(src_pos, src_pos); // src_pos (32-bits)
2450     __ jccb(Assembler::negative, L_failed_0);
2451 
2452     //  if (dst == NULL) return -1;
2453     __ testptr(dst, dst);         // dst oop
2454     __ jccb(Assembler::zero, L_failed_0);
2455 
2456     //  if (dst_pos < 0) return -1;
2457     __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2458     size_t j4off = __ offset();
2459     __ jccb(Assembler::negative, L_failed_0);
2460 
2461     // The first four tests are very dense code,
2462     // but not quite dense enough to put four
2463     // jumps in a 16-byte instruction fetch buffer.
2464     // That's good, because some branch predicters
2465     // do not like jumps so close together.
2466     // Make sure of this.
2467     guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2468 
2469     // registers used as temp
2470     const Register r11_length    = r11; // elements count to copy
2471     const Register r10_src_klass = r10; // array klass
2472     const Register r9_dst_klass  = r9;  // dest array klass
2473 
2474     //  if (length < 0) return -1;
2475     __ movl(r11_length, C_RARG4);       // length (elements count, 32-bits value)
2476     __ testl(r11_length, r11_length);
2477     __ jccb(Assembler::negative, L_failed_0);
2478 
2479     __ load_klass(r10_src_klass, src);
2480 #ifdef ASSERT
2481     //  assert(src->klass() != NULL);
2482     BLOCK_COMMENT("assert klasses not null");
2483     { Label L1, L2;
2484       __ testptr(r10_src_klass, r10_src_klass);
2485       __ jcc(Assembler::notZero, L2);   // it is broken if klass is NULL
2486       __ bind(L1);
2487       __ stop("broken null klass");
2488       __ bind(L2);
2489       __ load_klass(r9_dst_klass, dst);
2490       __ cmpq(r9_dst_klass, 0);
2491       __ jcc(Assembler::equal, L1);     // this would be broken also
2492       BLOCK_COMMENT("assert done");
2493     }
2494 #endif
2495 
2496     // Load layout helper (32-bits)
2497     //
2498     //  |array_tag|     | header_size | element_type |     |log2_element_size|
2499     // 32        30    24            16              8     2                 0
2500     //
2501     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2502     //
2503 
2504     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2505                     Klass::layout_helper_offset_in_bytes();
2506 
2507     const Register rax_lh = rax;  // layout helper
2508 
2509     __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2510 
2511     // Handle objArrays completely differently...
2512     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2513     __ cmpl(rax_lh, objArray_lh);
2514     __ jcc(Assembler::equal, L_objArray);
2515 
2516     //  if (src->klass() != dst->klass()) return -1;
2517     __ load_klass(r9_dst_klass, dst);
2518     __ cmpq(r10_src_klass, r9_dst_klass);
2519     __ jcc(Assembler::notEqual, L_failed);
2520 
2521     //  if (!src->is_Array()) return -1;
2522     __ cmpl(rax_lh, Klass::_lh_neutral_value);
2523     __ jcc(Assembler::greaterEqual, L_failed);
2524 
2525     // At this point, it is known to be a typeArray (array_tag 0x3).
2526 #ifdef ASSERT
2527     { Label L;
2528       __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2529       __ jcc(Assembler::greaterEqual, L);
2530       __ stop("must be a primitive array");
2531       __ bind(L);
2532     }
2533 #endif
2534 
2535     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2536                            r10, L_failed);
2537 
2538     // typeArrayKlass
2539     //
2540     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2541     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2542     //
2543 
2544     const Register r10_offset = r10;    // array offset
2545     const Register rax_elsize = rax_lh; // element size
2546 
2547     __ movl(r10_offset, rax_lh);
2548     __ shrl(r10_offset, Klass::_lh_header_size_shift);
2549     __ andptr(r10_offset, Klass::_lh_header_size_mask);   // array_offset
2550     __ addptr(src, r10_offset);           // src array offset
2551     __ addptr(dst, r10_offset);           // dst array offset
2552     BLOCK_COMMENT("choose copy loop based on element size");
2553     __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2554 
2555     // next registers should be set before the jump to corresponding stub
2556     const Register from     = c_rarg0;  // source array address
2557     const Register to       = c_rarg1;  // destination array address
2558     const Register count    = c_rarg2;  // elements count
2559 
2560     // 'from', 'to', 'count' registers should be set in such order
2561     // since they are the same as 'src', 'src_pos', 'dst'.
2562 
2563   __ BIND(L_copy_bytes);
2564     __ cmpl(rax_elsize, 0);
2565     __ jccb(Assembler::notEqual, L_copy_shorts);
2566     __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2567     __ lea(to,   Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2568     __ movl2ptr(count, r11_length); // length
2569     __ jump(RuntimeAddress(byte_copy_entry));
2570 
2571   __ BIND(L_copy_shorts);
2572     __ cmpl(rax_elsize, LogBytesPerShort);
2573     __ jccb(Assembler::notEqual, L_copy_ints);
2574     __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2575     __ lea(to,   Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2576     __ movl2ptr(count, r11_length); // length
2577     __ jump(RuntimeAddress(short_copy_entry));
2578 
2579   __ BIND(L_copy_ints);
2580     __ cmpl(rax_elsize, LogBytesPerInt);
2581     __ jccb(Assembler::notEqual, L_copy_longs);
2582     __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2583     __ lea(to,   Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2584     __ movl2ptr(count, r11_length); // length
2585     __ jump(RuntimeAddress(int_copy_entry));
2586 
2587   __ BIND(L_copy_longs);
2588 #ifdef ASSERT
2589     { Label L;
2590       __ cmpl(rax_elsize, LogBytesPerLong);
2591       __ jcc(Assembler::equal, L);
2592       __ stop("must be long copy, but elsize is wrong");
2593       __ bind(L);
2594     }
2595 #endif
2596     __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2597     __ lea(to,   Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2598     __ movl2ptr(count, r11_length); // length
2599     __ jump(RuntimeAddress(long_copy_entry));
2600 
2601     // objArrayKlass
2602   __ BIND(L_objArray);
2603     // live at this point:  r10_src_klass, src[_pos], dst[_pos]
2604 
2605     Label L_plain_copy, L_checkcast_copy;
2606     //  test array classes for subtyping
2607     __ load_klass(r9_dst_klass, dst);
2608     __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2609     __ jcc(Assembler::notEqual, L_checkcast_copy);
2610 
2611     // Identically typed arrays can be copied without element-wise checks.
2612     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2613                            r10, L_failed);
2614 
2615     __ lea(from, Address(src, src_pos, TIMES_OOP,
2616                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2617     __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2618                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2619     __ movl2ptr(count, r11_length); // length
2620   __ BIND(L_plain_copy);
2621     __ jump(RuntimeAddress(oop_copy_entry));
2622 
2623   __ BIND(L_checkcast_copy);
2624     // live at this point:  r10_src_klass, !r11_length
2625     {
2626       // assert(r11_length == C_RARG4); // will reload from here
2627       Register r11_dst_klass = r11;
2628       __ load_klass(r11_dst_klass, dst);
2629 
2630       // Before looking at dst.length, make sure dst is also an objArray.
2631       __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2632       __ jcc(Assembler::notEqual, L_failed);
2633 
2634       // It is safe to examine both src.length and dst.length.
2635 #ifndef _WIN64
2636       arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2637                              rax, L_failed);
2638 #else
2639       __ movl(r11_length, C_RARG4);     // reload
2640       arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2641                              rax, L_failed);
2642       __ load_klass(r11_dst_klass, dst); // reload
2643 #endif
2644 
2645       // Marshal the base address arguments now, freeing registers.
2646       __ lea(from, Address(src, src_pos, TIMES_OOP,
2647                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2648       __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2649                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2650       __ movl(count, C_RARG4);          // length (reloaded)
2651       Register sco_temp = c_rarg3;      // this register is free now
2652       assert_different_registers(from, to, count, sco_temp,
2653                                  r11_dst_klass, r10_src_klass);
2654       assert_clean_int(count, sco_temp);
2655 
2656       // Generate the type check.
2657       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2658                         Klass::super_check_offset_offset_in_bytes());
2659       __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2660       assert_clean_int(sco_temp, rax);
2661       generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2662 
2663       // Fetch destination element klass from the objArrayKlass header.
2664       int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2665                        objArrayKlass::element_klass_offset_in_bytes());
2666       __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2667       __ movl(sco_temp,      Address(r11_dst_klass, sco_offset));
2668       assert_clean_int(sco_temp, rax);
2669 
2670       // the checkcast_copy loop needs two extra arguments:
2671       assert(c_rarg3 == sco_temp, "#3 already in place");
2672       __ movptr(C_RARG4, r11_dst_klass);  // dst.klass.element_klass
2673       __ jump(RuntimeAddress(checkcast_copy_entry));
2674     }
2675 
2676   __ BIND(L_failed);
2677     __ xorptr(rax, rax);
2678     __ notptr(rax); // return -1
2679     __ leave();   // required for proper stackwalking of RuntimeStub frame
2680     __ ret(0);
2681 
2682     return start;
2683   }
2684 
2685 #undef length_arg
2686 
2687   void generate_arraycopy_stubs() {
2688     // Call the conjoint generation methods immediately after
2689     // the disjoint ones so that short branches from the former
2690     // to the latter can be generated.
2691     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2692     StubRoutines::_jbyte_arraycopy           = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2693 
2694     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2695     StubRoutines::_jshort_arraycopy          = generate_conjoint_short_copy(false, "jshort_arraycopy");
2696 
2697     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2698     StubRoutines::_jint_arraycopy            = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2699 
2700     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2701     StubRoutines::_jlong_arraycopy           = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2702 
2703 
2704     if (UseCompressedOops) {
2705       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2706       StubRoutines::_oop_arraycopy           = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2707     } else {
2708       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2709       StubRoutines::_oop_arraycopy           = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2710     }
2711 
2712     StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2713     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy");
2714     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy");
2715 
2716     // We don't generate specialized code for HeapWord-aligned source
2717     // arrays, so just use the code we've already generated
2718     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = StubRoutines::_jbyte_disjoint_arraycopy;
2719     StubRoutines::_arrayof_jbyte_arraycopy           = StubRoutines::_jbyte_arraycopy;
2720 
2721     StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2722     StubRoutines::_arrayof_jshort_arraycopy          = StubRoutines::_jshort_arraycopy;
2723 
2724     StubRoutines::_arrayof_jint_disjoint_arraycopy   = StubRoutines::_jint_disjoint_arraycopy;
2725     StubRoutines::_arrayof_jint_arraycopy            = StubRoutines::_jint_arraycopy;
2726 
2727     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = StubRoutines::_jlong_disjoint_arraycopy;
2728     StubRoutines::_arrayof_jlong_arraycopy           = StubRoutines::_jlong_arraycopy;
2729 
2730     StubRoutines::_arrayof_oop_disjoint_arraycopy    = StubRoutines::_oop_disjoint_arraycopy;
2731     StubRoutines::_arrayof_oop_arraycopy             = StubRoutines::_oop_arraycopy;
2732   }
2733 
2734   void generate_math_stubs() {
2735     {
2736       StubCodeMark mark(this, "StubRoutines", "log");
2737       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2738 
2739       __ subq(rsp, 8);
2740       __ movdbl(Address(rsp, 0), xmm0);
2741       __ fld_d(Address(rsp, 0));
2742       __ flog();
2743       __ fstp_d(Address(rsp, 0));
2744       __ movdbl(xmm0, Address(rsp, 0));
2745       __ addq(rsp, 8);
2746       __ ret(0);
2747     }
2748     {
2749       StubCodeMark mark(this, "StubRoutines", "log10");
2750       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2751 
2752       __ subq(rsp, 8);
2753       __ movdbl(Address(rsp, 0), xmm0);
2754       __ fld_d(Address(rsp, 0));
2755       __ flog10();
2756       __ fstp_d(Address(rsp, 0));
2757       __ movdbl(xmm0, Address(rsp, 0));
2758       __ addq(rsp, 8);
2759       __ ret(0);
2760     }
2761     {
2762       StubCodeMark mark(this, "StubRoutines", "sin");
2763       StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2764 
2765       __ subq(rsp, 8);
2766       __ movdbl(Address(rsp, 0), xmm0);
2767       __ fld_d(Address(rsp, 0));
2768       __ trigfunc('s');
2769       __ fstp_d(Address(rsp, 0));
2770       __ movdbl(xmm0, Address(rsp, 0));
2771       __ addq(rsp, 8);
2772       __ ret(0);
2773     }
2774     {
2775       StubCodeMark mark(this, "StubRoutines", "cos");
2776       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2777 
2778       __ subq(rsp, 8);
2779       __ movdbl(Address(rsp, 0), xmm0);
2780       __ fld_d(Address(rsp, 0));
2781       __ trigfunc('c');
2782       __ fstp_d(Address(rsp, 0));
2783       __ movdbl(xmm0, Address(rsp, 0));
2784       __ addq(rsp, 8);
2785       __ ret(0);
2786     }
2787     {
2788       StubCodeMark mark(this, "StubRoutines", "tan");
2789       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2790 
2791       __ subq(rsp, 8);
2792       __ movdbl(Address(rsp, 0), xmm0);
2793       __ fld_d(Address(rsp, 0));
2794       __ trigfunc('t');
2795       __ fstp_d(Address(rsp, 0));
2796       __ movdbl(xmm0, Address(rsp, 0));
2797       __ addq(rsp, 8);
2798       __ ret(0);
2799     }
2800 
2801     // The intrinsic version of these seem to return the same value as
2802     // the strict version.
2803     StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2804     StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2805   }
2806 
2807 #undef __
2808 #define __ masm->
2809 
2810   // Continuation point for throwing of implicit exceptions that are
2811   // not handled in the current activation. Fabricates an exception
2812   // oop and initiates normal exception dispatching in this
2813   // frame. Since we need to preserve callee-saved values (currently
2814   // only for C2, but done for C1 as well) we need a callee-saved oop
2815   // map and therefore have to make these stubs into RuntimeStubs
2816   // rather than BufferBlobs.  If the compiler needs all registers to
2817   // be preserved between the fault point and the exception handler
2818   // then it must assume responsibility for that in
2819   // AbstractCompiler::continuation_for_implicit_null_exception or
2820   // continuation_for_implicit_division_by_zero_exception. All other
2821   // implicit exceptions (e.g., NullPointerException or
2822   // AbstractMethodError on entry) are either at call sites or
2823   // otherwise assume that stack unwinding will be initiated, so
2824   // caller saved registers were assumed volatile in the compiler.
2825   address generate_throw_exception(const char* name,
2826                                    address runtime_entry,
2827                                    bool restore_saved_exception_pc) {
2828     // Information about frame layout at time of blocking runtime call.
2829     // Note that we only have to preserve callee-saved registers since
2830     // the compilers are responsible for supplying a continuation point
2831     // if they expect all registers to be preserved.
2832     enum layout {
2833       rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2834       rbp_off2,
2835       return_off,
2836       return_off2,
2837       framesize // inclusive of return address
2838     };
2839 
2840     int insts_size = 512;
2841     int locs_size  = 64;
2842 
2843     CodeBuffer code(name, insts_size, locs_size);
2844     OopMapSet* oop_maps  = new OopMapSet();
2845     MacroAssembler* masm = new MacroAssembler(&code);
2846 
2847     address start = __ pc();
2848 
2849     // This is an inlined and slightly modified version of call_VM
2850     // which has the ability to fetch the return PC out of
2851     // thread-local storage and also sets up last_Java_sp slightly
2852     // differently than the real call_VM
2853     if (restore_saved_exception_pc) {
2854       __ movptr(rax,
2855                 Address(r15_thread,
2856                         in_bytes(JavaThread::saved_exception_pc_offset())));
2857       __ push(rax);
2858     }
2859 
2860     __ enter(); // required for proper stackwalking of RuntimeStub frame
2861 
2862     assert(is_even(framesize/2), "sp not 16-byte aligned");
2863 
2864     // return address and rbp are already in place
2865     __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2866 
2867     int frame_complete = __ pc() - start;
2868 
2869     // Set up last_Java_sp and last_Java_fp
2870     __ set_last_Java_frame(rsp, rbp, NULL);
2871 
2872     // Call runtime
2873     __ movptr(c_rarg0, r15_thread);
2874     BLOCK_COMMENT("call runtime_entry");
2875     __ call(RuntimeAddress(runtime_entry));
2876 
2877     // Generate oop map
2878     OopMap* map = new OopMap(framesize, 0);
2879 
2880     oop_maps->add_gc_map(__ pc() - start, map);
2881 
2882     __ reset_last_Java_frame(true, false);
2883 
2884     __ leave(); // required for proper stackwalking of RuntimeStub frame
2885 
2886     // check for pending exceptions
2887 #ifdef ASSERT
2888     Label L;
2889     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
2890             (int32_t) NULL_WORD);
2891     __ jcc(Assembler::notEqual, L);
2892     __ should_not_reach_here();
2893     __ bind(L);
2894 #endif // ASSERT
2895     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2896 
2897 
2898     // codeBlob framesize is in words (not VMRegImpl::slot_size)
2899     RuntimeStub* stub =
2900       RuntimeStub::new_runtime_stub(name,
2901                                     &code,
2902                                     frame_complete,
2903                                     (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2904                                     oop_maps, false);
2905     return stub->entry_point();
2906   }
2907 
2908   // Initialization
2909   void generate_initial() {
2910     // Generates all stubs and initializes the entry points
2911 
2912     // This platform-specific stub is needed by generate_call_stub()
2913     StubRoutines::x86::_mxcsr_std        = generate_fp_mask("mxcsr_std",        0x0000000000001F80);
2914 
2915     // entry points that exist in all platforms Note: This is code
2916     // that could be shared among different platforms - however the
2917     // benefit seems to be smaller than the disadvantage of having a
2918     // much more complicated generator structure. See also comment in
2919     // stubRoutines.hpp.
2920 
2921     StubRoutines::_forward_exception_entry = generate_forward_exception();
2922 
2923     StubRoutines::_call_stub_entry =
2924       generate_call_stub(StubRoutines::_call_stub_return_address);
2925 
2926     // is referenced by megamorphic call
2927     StubRoutines::_catch_exception_entry = generate_catch_exception();
2928 
2929     // atomic calls
2930     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
2931     StubRoutines::_atomic_xchg_ptr_entry     = generate_atomic_xchg_ptr();
2932     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
2933     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2934     StubRoutines::_atomic_add_entry          = generate_atomic_add();
2935     StubRoutines::_atomic_add_ptr_entry      = generate_atomic_add_ptr();
2936     StubRoutines::_fence_entry               = generate_orderaccess_fence();
2937 
2938     StubRoutines::_handler_for_unsafe_access_entry =
2939       generate_handler_for_unsafe_access();
2940 
2941     // platform dependent
2942     StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
2943 
2944     StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
2945   }
2946 
2947   void generate_all() {
2948     // Generates all stubs and initializes the entry points
2949 
2950     // These entry points require SharedInfo::stack0 to be set up in
2951     // non-core builds and need to be relocatable, so they each
2952     // fabricate a RuntimeStub internally.
2953     StubRoutines::_throw_AbstractMethodError_entry =
2954       generate_throw_exception("AbstractMethodError throw_exception",
2955                                CAST_FROM_FN_PTR(address,
2956                                                 SharedRuntime::
2957                                                 throw_AbstractMethodError),
2958                                false);
2959 
2960     StubRoutines::_throw_IncompatibleClassChangeError_entry =
2961       generate_throw_exception("IncompatibleClassChangeError throw_exception",
2962                                CAST_FROM_FN_PTR(address,
2963                                                 SharedRuntime::
2964                                                 throw_IncompatibleClassChangeError),
2965                                false);
2966 
2967     StubRoutines::_throw_ArithmeticException_entry =
2968       generate_throw_exception("ArithmeticException throw_exception",
2969                                CAST_FROM_FN_PTR(address,
2970                                                 SharedRuntime::
2971                                                 throw_ArithmeticException),
2972                                true);
2973 
2974     StubRoutines::_throw_NullPointerException_entry =
2975       generate_throw_exception("NullPointerException throw_exception",
2976                                CAST_FROM_FN_PTR(address,
2977                                                 SharedRuntime::
2978                                                 throw_NullPointerException),
2979                                true);
2980 
2981     StubRoutines::_throw_NullPointerException_at_call_entry =
2982       generate_throw_exception("NullPointerException at call throw_exception",
2983                                CAST_FROM_FN_PTR(address,
2984                                                 SharedRuntime::
2985                                                 throw_NullPointerException_at_call),
2986                                false);
2987 
2988     StubRoutines::_throw_StackOverflowError_entry =
2989       generate_throw_exception("StackOverflowError throw_exception",
2990                                CAST_FROM_FN_PTR(address,
2991                                                 SharedRuntime::
2992                                                 throw_StackOverflowError),
2993                                false);
2994 
2995     // entry points that are platform specific
2996     StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
2997     StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
2998     StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
2999     StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3000 
3001     StubRoutines::x86::_float_sign_mask  = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
3002     StubRoutines::x86::_float_sign_flip  = generate_fp_mask("float_sign_flip",  0x8000000080000000);
3003     StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3004     StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3005 
3006     // support for verify_oop (must happen after universe_init)
3007     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3008 
3009     // arraycopy stubs used by compilers
3010     generate_arraycopy_stubs();
3011 
3012     // generic method handle stubs
3013     if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) {
3014       for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
3015            ek < MethodHandles::_EK_LIMIT;
3016            ek = MethodHandles::EntryKind(1 + (int)ek)) {
3017         StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
3018         MethodHandles::generate_method_handle_stub(_masm, ek);
3019       }
3020     }
3021 
3022     generate_math_stubs();
3023   }
3024 
3025  public:
3026   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3027     if (all) {
3028       generate_all();
3029     } else {
3030       generate_initial();
3031     }
3032   }
3033 }; // end class declaration
3034 
3035 address StubGenerator::disjoint_byte_copy_entry  = NULL;
3036 address StubGenerator::disjoint_short_copy_entry = NULL;
3037 address StubGenerator::disjoint_int_copy_entry   = NULL;
3038 address StubGenerator::disjoint_long_copy_entry  = NULL;
3039 address StubGenerator::disjoint_oop_copy_entry   = NULL;
3040 
3041 address StubGenerator::byte_copy_entry  = NULL;
3042 address StubGenerator::short_copy_entry = NULL;
3043 address StubGenerator::int_copy_entry   = NULL;
3044 address StubGenerator::long_copy_entry  = NULL;
3045 address StubGenerator::oop_copy_entry   = NULL;
3046 
3047 address StubGenerator::checkcast_copy_entry = NULL;
3048 
3049 void StubGenerator_generate(CodeBuffer* code, bool all) {
3050   StubGenerator g(code, all);
3051 }