1 /*
   2  * Copyright (c) 2003, 2012, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "nativeInst_x86.hpp"
  30 #include "oops/instanceOop.hpp"
  31 #include "oops/method.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "oops/oop.inline.hpp"
  34 #include "prims/methodHandles.hpp"
  35 #include "runtime/frame.inline.hpp"
  36 #include "runtime/handles.inline.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/stubCodeGenerator.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "runtime/thread.inline.hpp"
  41 #include "utilities/top.hpp"
  42 #ifdef COMPILER2
  43 #include "opto/runtime.hpp"
  44 #endif
  45 
  46 // Declaration and definition of StubGenerator (no .hpp file).
  47 // For a more detailed description of the stub routine structure
  48 // see the comment in stubRoutines.hpp
  49 
  50 #define __ _masm->
  51 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
  52 #define a__ ((Assembler*)_masm)->
  53 
  54 #ifdef PRODUCT
  55 #define BLOCK_COMMENT(str) /* nothing */
  56 #else
  57 #define BLOCK_COMMENT(str) __ block_comment(str)
  58 #endif
  59 
  60 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  61 const int MXCSR_MASK = 0xFFC0;  // Mask out any pending exceptions
  62 
  63 // Stub Code definitions
  64 
  65 static address handle_unsafe_access() {
  66   JavaThread* thread = JavaThread::current();
  67   address pc = thread->saved_exception_pc();
  68   // pc is the instruction which we must emulate
  69   // doing a no-op is fine:  return garbage from the load
  70   // therefore, compute npc
  71   address npc = Assembler::locate_next_instruction(pc);
  72 
  73   // request an async exception
  74   thread->set_pending_unsafe_access_error();
  75 
  76   // return address of next instruction to execute
  77   return npc;
  78 }
  79 
  80 class StubGenerator: public StubCodeGenerator {
  81  private:
  82 
  83 #ifdef PRODUCT
  84 #define inc_counter_np(counter) (0)
  85 #else
  86   void inc_counter_np_(int& counter) {
  87     // This can destroy rscratch1 if counter is far from the code cache
  88     __ incrementl(ExternalAddress((address)&counter));
  89   }
  90 #define inc_counter_np(counter) \
  91   BLOCK_COMMENT("inc_counter " #counter); \
  92   inc_counter_np_(counter);
  93 #endif
  94 
  95   // Call stubs are used to call Java from C
  96   //
  97   // Linux Arguments:
  98   //    c_rarg0:   call wrapper address                   address
  99   //    c_rarg1:   result                                 address
 100   //    c_rarg2:   result type                            BasicType
 101   //    c_rarg3:   method                                 Method*
 102   //    c_rarg4:   (interpreter) entry point              address
 103   //    c_rarg5:   parameters                             intptr_t*
 104   //    16(rbp): parameter size (in words)              int
 105   //    24(rbp): thread                                 Thread*
 106   //
 107   //     [ return_from_Java     ] <--- rsp
 108   //     [ argument word n      ]
 109   //      ...
 110   // -12 [ argument word 1      ]
 111   // -11 [ saved r15            ] <--- rsp_after_call
 112   // -10 [ saved r14            ]
 113   //  -9 [ saved r13            ]
 114   //  -8 [ saved r12            ]
 115   //  -7 [ saved rbx            ]
 116   //  -6 [ call wrapper         ]
 117   //  -5 [ result               ]
 118   //  -4 [ result type          ]
 119   //  -3 [ method               ]
 120   //  -2 [ entry point          ]
 121   //  -1 [ parameters           ]
 122   //   0 [ saved rbp            ] <--- rbp
 123   //   1 [ return address       ]
 124   //   2 [ parameter size       ]
 125   //   3 [ thread               ]
 126   //
 127   // Windows Arguments:
 128   //    c_rarg0:   call wrapper address                   address
 129   //    c_rarg1:   result                                 address
 130   //    c_rarg2:   result type                            BasicType
 131   //    c_rarg3:   method                                 Method*
 132   //    48(rbp): (interpreter) entry point              address
 133   //    56(rbp): parameters                             intptr_t*
 134   //    64(rbp): parameter size (in words)              int
 135   //    72(rbp): thread                                 Thread*
 136   //
 137   //     [ return_from_Java     ] <--- rsp
 138   //     [ argument word n      ]
 139   //      ...
 140   // -28 [ argument word 1      ]
 141   // -27 [ saved xmm15          ] <--- rsp_after_call
 142   //     [ saved xmm7-xmm14     ]
 143   //  -9 [ saved xmm6           ] (each xmm register takes 2 slots)
 144   //  -7 [ saved r15            ]
 145   //  -6 [ saved r14            ]
 146   //  -5 [ saved r13            ]
 147   //  -4 [ saved r12            ]
 148   //  -3 [ saved rdi            ]
 149   //  -2 [ saved rsi            ]
 150   //  -1 [ saved rbx            ]
 151   //   0 [ saved rbp            ] <--- rbp
 152   //   1 [ return address       ]
 153   //   2 [ call wrapper         ]
 154   //   3 [ result               ]
 155   //   4 [ result type          ]
 156   //   5 [ method               ]
 157   //   6 [ entry point          ]
 158   //   7 [ parameters           ]
 159   //   8 [ parameter size       ]
 160   //   9 [ thread               ]
 161   //
 162   //    Windows reserves the callers stack space for arguments 1-4.
 163   //    We spill c_rarg0-c_rarg3 to this space.
 164 
 165   // Call stub stack layout word offsets from rbp
 166   enum call_stub_layout {
 167 #ifdef _WIN64
 168     xmm_save_first     = 6,  // save from xmm6
 169     xmm_save_last      = 15, // to xmm15
 170     xmm_save_base      = -9,
 171     rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
 172     r15_off            = -7,
 173     r14_off            = -6,
 174     r13_off            = -5,
 175     r12_off            = -4,
 176     rdi_off            = -3,
 177     rsi_off            = -2,
 178     rbx_off            = -1,
 179     rbp_off            =  0,
 180     retaddr_off        =  1,
 181     call_wrapper_off   =  2,
 182     result_off         =  3,
 183     result_type_off    =  4,
 184     method_off         =  5,
 185     entry_point_off    =  6,
 186     parameters_off     =  7,
 187     parameter_size_off =  8,
 188     thread_off         =  9
 189 #else
 190     rsp_after_call_off = -12,
 191     mxcsr_off          = rsp_after_call_off,
 192     r15_off            = -11,
 193     r14_off            = -10,
 194     r13_off            = -9,
 195     r12_off            = -8,
 196     rbx_off            = -7,
 197     call_wrapper_off   = -6,
 198     result_off         = -5,
 199     result_type_off    = -4,
 200     method_off         = -3,
 201     entry_point_off    = -2,
 202     parameters_off     = -1,
 203     rbp_off            =  0,
 204     retaddr_off        =  1,
 205     parameter_size_off =  2,
 206     thread_off         =  3
 207 #endif
 208   };
 209 
 210 #ifdef _WIN64
 211   Address xmm_save(int reg) {
 212     assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
 213     return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
 214   }
 215 #endif
 216 
 217   address generate_call_stub(address& return_address) {
 218     assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 219            (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 220            "adjust this code");
 221     StubCodeMark mark(this, "StubRoutines", "call_stub");
 222     address start = __ pc();
 223 
 224     // same as in generate_catch_exception()!
 225     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 226 
 227     const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 228     const Address result        (rbp, result_off         * wordSize);
 229     const Address result_type   (rbp, result_type_off    * wordSize);
 230     const Address method        (rbp, method_off         * wordSize);
 231     const Address entry_point   (rbp, entry_point_off    * wordSize);
 232     const Address parameters    (rbp, parameters_off     * wordSize);
 233     const Address parameter_size(rbp, parameter_size_off * wordSize);
 234 
 235     // same as in generate_catch_exception()!
 236     const Address thread        (rbp, thread_off         * wordSize);
 237 
 238     const Address r15_save(rbp, r15_off * wordSize);
 239     const Address r14_save(rbp, r14_off * wordSize);
 240     const Address r13_save(rbp, r13_off * wordSize);
 241     const Address r12_save(rbp, r12_off * wordSize);
 242     const Address rbx_save(rbp, rbx_off * wordSize);
 243 
 244     // stub code
 245     __ enter();
 246     __ subptr(rsp, -rsp_after_call_off * wordSize);
 247 
 248     // save register parameters
 249 #ifndef _WIN64
 250     __ movptr(parameters,   c_rarg5); // parameters
 251     __ movptr(entry_point,  c_rarg4); // entry_point
 252 #endif
 253 
 254     __ movptr(method,       c_rarg3); // method
 255     __ movl(result_type,  c_rarg2);   // result type
 256     __ movptr(result,       c_rarg1); // result
 257     __ movptr(call_wrapper, c_rarg0); // call wrapper
 258 
 259     // save regs belonging to calling function
 260     __ movptr(rbx_save, rbx);
 261     __ movptr(r12_save, r12);
 262     __ movptr(r13_save, r13);
 263     __ movptr(r14_save, r14);
 264     __ movptr(r15_save, r15);
 265 #ifdef _WIN64
 266     for (int i = 6; i <= 15; i++) {
 267       __ movdqu(xmm_save(i), as_XMMRegister(i));
 268     }
 269 
 270     const Address rdi_save(rbp, rdi_off * wordSize);
 271     const Address rsi_save(rbp, rsi_off * wordSize);
 272 
 273     __ movptr(rsi_save, rsi);
 274     __ movptr(rdi_save, rdi);
 275 #else
 276     const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 277     {
 278       Label skip_ldmx;
 279       __ stmxcsr(mxcsr_save);
 280       __ movl(rax, mxcsr_save);
 281       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 282       ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
 283       __ cmp32(rax, mxcsr_std);
 284       __ jcc(Assembler::equal, skip_ldmx);
 285       __ ldmxcsr(mxcsr_std);
 286       __ bind(skip_ldmx);
 287     }
 288 #endif
 289 
 290     // Load up thread register
 291     __ movptr(r15_thread, thread);
 292     __ reinit_heapbase();
 293 
 294 #ifdef ASSERT
 295     // make sure we have no pending exceptions
 296     {
 297       Label L;
 298       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 299       __ jcc(Assembler::equal, L);
 300       __ stop("StubRoutines::call_stub: entered with pending exception");
 301       __ bind(L);
 302     }
 303 #endif
 304 
 305     // pass parameters if any
 306     BLOCK_COMMENT("pass parameters if any");
 307     Label parameters_done;
 308     __ movl(c_rarg3, parameter_size);
 309     __ testl(c_rarg3, c_rarg3);
 310     __ jcc(Assembler::zero, parameters_done);
 311 
 312     Label loop;
 313     __ movptr(c_rarg2, parameters);       // parameter pointer
 314     __ movl(c_rarg1, c_rarg3);            // parameter counter is in c_rarg1
 315     __ BIND(loop);
 316     __ movptr(rax, Address(c_rarg2, 0));// get parameter
 317     __ addptr(c_rarg2, wordSize);       // advance to next parameter
 318     __ decrementl(c_rarg1);             // decrement counter
 319     __ push(rax);                       // pass parameter
 320     __ jcc(Assembler::notZero, loop);
 321 
 322     // call Java function
 323     __ BIND(parameters_done);
 324     __ movptr(rbx, method);             // get Method*
 325     __ movptr(c_rarg1, entry_point);    // get entry_point
 326     __ mov(r13, rsp);                   // set sender sp
 327     BLOCK_COMMENT("call Java function");
 328     __ call(c_rarg1);
 329 
 330     BLOCK_COMMENT("call_stub_return_address:");
 331     return_address = __ pc();
 332 
 333     // store result depending on type (everything that is not
 334     // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 335     __ movptr(c_rarg0, result);
 336     Label is_long, is_float, is_double, exit;
 337     __ movl(c_rarg1, result_type);
 338     __ cmpl(c_rarg1, T_OBJECT);
 339     __ jcc(Assembler::equal, is_long);
 340     __ cmpl(c_rarg1, T_LONG);
 341     __ jcc(Assembler::equal, is_long);
 342     __ cmpl(c_rarg1, T_FLOAT);
 343     __ jcc(Assembler::equal, is_float);
 344     __ cmpl(c_rarg1, T_DOUBLE);
 345     __ jcc(Assembler::equal, is_double);
 346 
 347     // handle T_INT case
 348     __ movl(Address(c_rarg0, 0), rax);
 349 
 350     __ BIND(exit);
 351 
 352     // pop parameters
 353     __ lea(rsp, rsp_after_call);
 354 
 355 #ifdef ASSERT
 356     // verify that threads correspond
 357     {
 358       Label L, S;
 359       __ cmpptr(r15_thread, thread);
 360       __ jcc(Assembler::notEqual, S);
 361       __ get_thread(rbx);
 362       __ cmpptr(r15_thread, rbx);
 363       __ jcc(Assembler::equal, L);
 364       __ bind(S);
 365       __ jcc(Assembler::equal, L);
 366       __ stop("StubRoutines::call_stub: threads must correspond");
 367       __ bind(L);
 368     }
 369 #endif
 370 
 371     // restore regs belonging to calling function
 372 #ifdef _WIN64
 373     for (int i = 15; i >= 6; i--) {
 374       __ movdqu(as_XMMRegister(i), xmm_save(i));
 375     }
 376 #endif
 377     __ movptr(r15, r15_save);
 378     __ movptr(r14, r14_save);
 379     __ movptr(r13, r13_save);
 380     __ movptr(r12, r12_save);
 381     __ movptr(rbx, rbx_save);
 382 
 383 #ifdef _WIN64
 384     __ movptr(rdi, rdi_save);
 385     __ movptr(rsi, rsi_save);
 386 #else
 387     __ ldmxcsr(mxcsr_save);
 388 #endif
 389 
 390     // restore rsp
 391     __ addptr(rsp, -rsp_after_call_off * wordSize);
 392 
 393     // return
 394     __ pop(rbp);
 395     __ ret(0);
 396 
 397     // handle return types different from T_INT
 398     __ BIND(is_long);
 399     __ movq(Address(c_rarg0, 0), rax);
 400     __ jmp(exit);
 401 
 402     __ BIND(is_float);
 403     __ movflt(Address(c_rarg0, 0), xmm0);
 404     __ jmp(exit);
 405 
 406     __ BIND(is_double);
 407     __ movdbl(Address(c_rarg0, 0), xmm0);
 408     __ jmp(exit);
 409 
 410     return start;
 411   }
 412 
 413   // Return point for a Java call if there's an exception thrown in
 414   // Java code.  The exception is caught and transformed into a
 415   // pending exception stored in JavaThread that can be tested from
 416   // within the VM.
 417   //
 418   // Note: Usually the parameters are removed by the callee. In case
 419   // of an exception crossing an activation frame boundary, that is
 420   // not the case if the callee is compiled code => need to setup the
 421   // rsp.
 422   //
 423   // rax: exception oop
 424 
 425   address generate_catch_exception() {
 426     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 427     address start = __ pc();
 428 
 429     // same as in generate_call_stub():
 430     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 431     const Address thread        (rbp, thread_off         * wordSize);
 432 
 433 #ifdef ASSERT
 434     // verify that threads correspond
 435     {
 436       Label L, S;
 437       __ cmpptr(r15_thread, thread);
 438       __ jcc(Assembler::notEqual, S);
 439       __ get_thread(rbx);
 440       __ cmpptr(r15_thread, rbx);
 441       __ jcc(Assembler::equal, L);
 442       __ bind(S);
 443       __ stop("StubRoutines::catch_exception: threads must correspond");
 444       __ bind(L);
 445     }
 446 #endif
 447 
 448     // set pending exception
 449     __ verify_oop(rax);
 450 
 451     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
 452     __ lea(rscratch1, ExternalAddress((address)__FILE__));
 453     __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 454     __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 455 
 456     // complete return to VM
 457     assert(StubRoutines::_call_stub_return_address != NULL,
 458            "_call_stub_return_address must have been generated before");
 459     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 460 
 461     return start;
 462   }
 463 
 464   // Continuation point for runtime calls returning with a pending
 465   // exception.  The pending exception check happened in the runtime
 466   // or native call stub.  The pending exception in Thread is
 467   // converted into a Java-level exception.
 468   //
 469   // Contract with Java-level exception handlers:
 470   // rax: exception
 471   // rdx: throwing pc
 472   //
 473   // NOTE: At entry of this stub, exception-pc must be on stack !!
 474 
 475   address generate_forward_exception() {
 476     StubCodeMark mark(this, "StubRoutines", "forward exception");
 477     address start = __ pc();
 478 
 479     // Upon entry, the sp points to the return address returning into
 480     // Java (interpreted or compiled) code; i.e., the return address
 481     // becomes the throwing pc.
 482     //
 483     // Arguments pushed before the runtime call are still on the stack
 484     // but the exception handler will reset the stack pointer ->
 485     // ignore them.  A potential result in registers can be ignored as
 486     // well.
 487 
 488 #ifdef ASSERT
 489     // make sure this code is only executed if there is a pending exception
 490     {
 491       Label L;
 492       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
 493       __ jcc(Assembler::notEqual, L);
 494       __ stop("StubRoutines::forward exception: no pending exception (1)");
 495       __ bind(L);
 496     }
 497 #endif
 498 
 499     // compute exception handler into rbx
 500     __ movptr(c_rarg0, Address(rsp, 0));
 501     BLOCK_COMMENT("call exception_handler_for_return_address");
 502     __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 503                          SharedRuntime::exception_handler_for_return_address),
 504                     r15_thread, c_rarg0);
 505     __ mov(rbx, rax);
 506 
 507     // setup rax & rdx, remove return address & clear pending exception
 508     __ pop(rdx);
 509     __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 510     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 511 
 512 #ifdef ASSERT
 513     // make sure exception is set
 514     {
 515       Label L;
 516       __ testptr(rax, rax);
 517       __ jcc(Assembler::notEqual, L);
 518       __ stop("StubRoutines::forward exception: no pending exception (2)");
 519       __ bind(L);
 520     }
 521 #endif
 522 
 523     // continue at exception handler (return address removed)
 524     // rax: exception
 525     // rbx: exception handler
 526     // rdx: throwing pc
 527     __ verify_oop(rax);
 528     __ jmp(rbx);
 529 
 530     return start;
 531   }
 532 
 533   // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
 534   //
 535   // Arguments :
 536   //    c_rarg0: exchange_value
 537   //    c_rarg0: dest
 538   //
 539   // Result:
 540   //    *dest <- ex, return (orig *dest)
 541   address generate_atomic_xchg() {
 542     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 543     address start = __ pc();
 544 
 545     __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
 546     __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
 547     __ ret(0);
 548 
 549     return start;
 550   }
 551 
 552   // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
 553   //
 554   // Arguments :
 555   //    c_rarg0: exchange_value
 556   //    c_rarg1: dest
 557   //
 558   // Result:
 559   //    *dest <- ex, return (orig *dest)
 560   address generate_atomic_xchg_ptr() {
 561     StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
 562     address start = __ pc();
 563 
 564     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 565     __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
 566     __ ret(0);
 567 
 568     return start;
 569   }
 570 
 571   // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
 572   //                                         jint compare_value)
 573   //
 574   // Arguments :
 575   //    c_rarg0: exchange_value
 576   //    c_rarg1: dest
 577   //    c_rarg2: compare_value
 578   //
 579   // Result:
 580   //    if ( compare_value == *dest ) {
 581   //       *dest = exchange_value
 582   //       return compare_value;
 583   //    else
 584   //       return *dest;
 585   address generate_atomic_cmpxchg() {
 586     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
 587     address start = __ pc();
 588 
 589     __ movl(rax, c_rarg2);
 590    if ( os::is_MP() ) __ lock();
 591     __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
 592     __ ret(0);
 593 
 594     return start;
 595   }
 596 
 597   // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
 598   //                                             volatile jlong* dest,
 599   //                                             jlong compare_value)
 600   // Arguments :
 601   //    c_rarg0: exchange_value
 602   //    c_rarg1: dest
 603   //    c_rarg2: compare_value
 604   //
 605   // Result:
 606   //    if ( compare_value == *dest ) {
 607   //       *dest = exchange_value
 608   //       return compare_value;
 609   //    else
 610   //       return *dest;
 611   address generate_atomic_cmpxchg_long() {
 612     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
 613     address start = __ pc();
 614 
 615     __ movq(rax, c_rarg2);
 616    if ( os::is_MP() ) __ lock();
 617     __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
 618     __ ret(0);
 619 
 620     return start;
 621   }
 622 
 623   // Support for jint atomic::add(jint add_value, volatile jint* dest)
 624   //
 625   // Arguments :
 626   //    c_rarg0: add_value
 627   //    c_rarg1: dest
 628   //
 629   // Result:
 630   //    *dest += add_value
 631   //    return *dest;
 632   address generate_atomic_add() {
 633     StubCodeMark mark(this, "StubRoutines", "atomic_add");
 634     address start = __ pc();
 635 
 636     __ movl(rax, c_rarg0);
 637    if ( os::is_MP() ) __ lock();
 638     __ xaddl(Address(c_rarg1, 0), c_rarg0);
 639     __ addl(rax, c_rarg0);
 640     __ ret(0);
 641 
 642     return start;
 643   }
 644 
 645   // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
 646   //
 647   // Arguments :
 648   //    c_rarg0: add_value
 649   //    c_rarg1: dest
 650   //
 651   // Result:
 652   //    *dest += add_value
 653   //    return *dest;
 654   address generate_atomic_add_ptr() {
 655     StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
 656     address start = __ pc();
 657 
 658     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 659    if ( os::is_MP() ) __ lock();
 660     __ xaddptr(Address(c_rarg1, 0), c_rarg0);
 661     __ addptr(rax, c_rarg0);
 662     __ ret(0);
 663 
 664     return start;
 665   }
 666 
 667   // Support for intptr_t OrderAccess::fence()
 668   //
 669   // Arguments :
 670   //
 671   // Result:
 672   address generate_orderaccess_fence() {
 673     StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 674     address start = __ pc();
 675     __ membar(Assembler::StoreLoad);
 676     __ ret(0);
 677 
 678     return start;
 679   }
 680 
 681   // Support for intptr_t get_previous_fp()
 682   //
 683   // This routine is used to find the previous frame pointer for the
 684   // caller (current_frame_guess). This is used as part of debugging
 685   // ps() is seemingly lost trying to find frames.
 686   // This code assumes that caller current_frame_guess) has a frame.
 687   address generate_get_previous_fp() {
 688     StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
 689     const Address old_fp(rbp, 0);
 690     const Address older_fp(rax, 0);
 691     address start = __ pc();
 692 
 693     __ enter();
 694     __ movptr(rax, old_fp); // callers fp
 695     __ movptr(rax, older_fp); // the frame for ps()
 696     __ pop(rbp);
 697     __ ret(0);
 698 
 699     return start;
 700   }
 701 
 702   // Support for intptr_t get_previous_sp()
 703   //
 704   // This routine is used to find the previous stack pointer for the
 705   // caller.
 706   address generate_get_previous_sp() {
 707     StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
 708     address start = __ pc();
 709 
 710     __ movptr(rax, rsp);
 711     __ addptr(rax, 8); // return address is at the top of the stack.
 712     __ ret(0);
 713 
 714     return start;
 715   }
 716 
 717   //----------------------------------------------------------------------------------------------------
 718   // Support for void verify_mxcsr()
 719   //
 720   // This routine is used with -Xcheck:jni to verify that native
 721   // JNI code does not return to Java code without restoring the
 722   // MXCSR register to our expected state.
 723 
 724   address generate_verify_mxcsr() {
 725     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 726     address start = __ pc();
 727 
 728     const Address mxcsr_save(rsp, 0);
 729 
 730     if (CheckJNICalls) {
 731       Label ok_ret;
 732       __ push(rax);
 733       __ subptr(rsp, wordSize);      // allocate a temp location
 734       __ stmxcsr(mxcsr_save);
 735       __ movl(rax, mxcsr_save);
 736       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 737       __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
 738       __ jcc(Assembler::equal, ok_ret);
 739 
 740       __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 741 
 742       __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
 743 
 744       __ bind(ok_ret);
 745       __ addptr(rsp, wordSize);
 746       __ pop(rax);
 747     }
 748 
 749     __ ret(0);
 750 
 751     return start;
 752   }
 753 
 754   address generate_f2i_fixup() {
 755     StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 756     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 757 
 758     address start = __ pc();
 759 
 760     Label L;
 761 
 762     __ push(rax);
 763     __ push(c_rarg3);
 764     __ push(c_rarg2);
 765     __ push(c_rarg1);
 766 
 767     __ movl(rax, 0x7f800000);
 768     __ xorl(c_rarg3, c_rarg3);
 769     __ movl(c_rarg2, inout);
 770     __ movl(c_rarg1, c_rarg2);
 771     __ andl(c_rarg1, 0x7fffffff);
 772     __ cmpl(rax, c_rarg1); // NaN? -> 0
 773     __ jcc(Assembler::negative, L);
 774     __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 775     __ movl(c_rarg3, 0x80000000);
 776     __ movl(rax, 0x7fffffff);
 777     __ cmovl(Assembler::positive, c_rarg3, rax);
 778 
 779     __ bind(L);
 780     __ movptr(inout, c_rarg3);
 781 
 782     __ pop(c_rarg1);
 783     __ pop(c_rarg2);
 784     __ pop(c_rarg3);
 785     __ pop(rax);
 786 
 787     __ ret(0);
 788 
 789     return start;
 790   }
 791 
 792   address generate_f2l_fixup() {
 793     StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 794     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 795     address start = __ pc();
 796 
 797     Label L;
 798 
 799     __ push(rax);
 800     __ push(c_rarg3);
 801     __ push(c_rarg2);
 802     __ push(c_rarg1);
 803 
 804     __ movl(rax, 0x7f800000);
 805     __ xorl(c_rarg3, c_rarg3);
 806     __ movl(c_rarg2, inout);
 807     __ movl(c_rarg1, c_rarg2);
 808     __ andl(c_rarg1, 0x7fffffff);
 809     __ cmpl(rax, c_rarg1); // NaN? -> 0
 810     __ jcc(Assembler::negative, L);
 811     __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 812     __ mov64(c_rarg3, 0x8000000000000000);
 813     __ mov64(rax, 0x7fffffffffffffff);
 814     __ cmov(Assembler::positive, c_rarg3, rax);
 815 
 816     __ bind(L);
 817     __ movptr(inout, c_rarg3);
 818 
 819     __ pop(c_rarg1);
 820     __ pop(c_rarg2);
 821     __ pop(c_rarg3);
 822     __ pop(rax);
 823 
 824     __ ret(0);
 825 
 826     return start;
 827   }
 828 
 829   address generate_d2i_fixup() {
 830     StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 831     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 832 
 833     address start = __ pc();
 834 
 835     Label L;
 836 
 837     __ push(rax);
 838     __ push(c_rarg3);
 839     __ push(c_rarg2);
 840     __ push(c_rarg1);
 841     __ push(c_rarg0);
 842 
 843     __ movl(rax, 0x7ff00000);
 844     __ movq(c_rarg2, inout);
 845     __ movl(c_rarg3, c_rarg2);
 846     __ mov(c_rarg1, c_rarg2);
 847     __ mov(c_rarg0, c_rarg2);
 848     __ negl(c_rarg3);
 849     __ shrptr(c_rarg1, 0x20);
 850     __ orl(c_rarg3, c_rarg2);
 851     __ andl(c_rarg1, 0x7fffffff);
 852     __ xorl(c_rarg2, c_rarg2);
 853     __ shrl(c_rarg3, 0x1f);
 854     __ orl(c_rarg1, c_rarg3);
 855     __ cmpl(rax, c_rarg1);
 856     __ jcc(Assembler::negative, L); // NaN -> 0
 857     __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 858     __ movl(c_rarg2, 0x80000000);
 859     __ movl(rax, 0x7fffffff);
 860     __ cmov(Assembler::positive, c_rarg2, rax);
 861 
 862     __ bind(L);
 863     __ movptr(inout, c_rarg2);
 864 
 865     __ pop(c_rarg0);
 866     __ pop(c_rarg1);
 867     __ pop(c_rarg2);
 868     __ pop(c_rarg3);
 869     __ pop(rax);
 870 
 871     __ ret(0);
 872 
 873     return start;
 874   }
 875 
 876   address generate_d2l_fixup() {
 877     StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 878     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 879 
 880     address start = __ pc();
 881 
 882     Label L;
 883 
 884     __ push(rax);
 885     __ push(c_rarg3);
 886     __ push(c_rarg2);
 887     __ push(c_rarg1);
 888     __ push(c_rarg0);
 889 
 890     __ movl(rax, 0x7ff00000);
 891     __ movq(c_rarg2, inout);
 892     __ movl(c_rarg3, c_rarg2);
 893     __ mov(c_rarg1, c_rarg2);
 894     __ mov(c_rarg0, c_rarg2);
 895     __ negl(c_rarg3);
 896     __ shrptr(c_rarg1, 0x20);
 897     __ orl(c_rarg3, c_rarg2);
 898     __ andl(c_rarg1, 0x7fffffff);
 899     __ xorl(c_rarg2, c_rarg2);
 900     __ shrl(c_rarg3, 0x1f);
 901     __ orl(c_rarg1, c_rarg3);
 902     __ cmpl(rax, c_rarg1);
 903     __ jcc(Assembler::negative, L); // NaN -> 0
 904     __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 905     __ mov64(c_rarg2, 0x8000000000000000);
 906     __ mov64(rax, 0x7fffffffffffffff);
 907     __ cmovq(Assembler::positive, c_rarg2, rax);
 908 
 909     __ bind(L);
 910     __ movq(inout, c_rarg2);
 911 
 912     __ pop(c_rarg0);
 913     __ pop(c_rarg1);
 914     __ pop(c_rarg2);
 915     __ pop(c_rarg3);
 916     __ pop(rax);
 917 
 918     __ ret(0);
 919 
 920     return start;
 921   }
 922 
 923   address generate_fp_mask(const char *stub_name, int64_t mask) {
 924     __ align(CodeEntryAlignment);
 925     StubCodeMark mark(this, "StubRoutines", stub_name);
 926     address start = __ pc();
 927 
 928     __ emit_data64( mask, relocInfo::none );
 929     __ emit_data64( mask, relocInfo::none );
 930 
 931     return start;
 932   }
 933 
 934   // The following routine generates a subroutine to throw an
 935   // asynchronous UnknownError when an unsafe access gets a fault that
 936   // could not be reasonably prevented by the programmer.  (Example:
 937   // SIGBUS/OBJERR.)
 938   address generate_handler_for_unsafe_access() {
 939     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 940     address start = __ pc();
 941 
 942     __ push(0);                       // hole for return address-to-be
 943     __ pusha();                       // push registers
 944     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 945 
 946     // FIXME: this probably needs alignment logic
 947 
 948     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 949     BLOCK_COMMENT("call handle_unsafe_access");
 950     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 951     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 952 
 953     __ movptr(next_pc, rax);          // stuff next address
 954     __ popa();
 955     __ ret(0);                        // jump to next address
 956 
 957     return start;
 958   }
 959 
 960   // Non-destructive plausibility checks for oops
 961   //
 962   // Arguments:
 963   //    all args on stack!
 964   //
 965   // Stack after saving c_rarg3:
 966   //    [tos + 0]: saved c_rarg3
 967   //    [tos + 1]: saved c_rarg2
 968   //    [tos + 2]: saved r12 (several TemplateTable methods use it)
 969   //    [tos + 3]: saved flags
 970   //    [tos + 4]: return address
 971   //  * [tos + 5]: error message (char*)
 972   //  * [tos + 6]: object to verify (oop)
 973   //  * [tos + 7]: saved rax - saved by caller and bashed
 974   //  * [tos + 8]: saved r10 (rscratch1) - saved by caller
 975   //  * = popped on exit
 976   address generate_verify_oop() {
 977     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 978     address start = __ pc();
 979 
 980     Label exit, error;
 981 
 982     __ pushf();
 983     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 984 
 985     __ push(r12);
 986 
 987     // save c_rarg2 and c_rarg3
 988     __ push(c_rarg2);
 989     __ push(c_rarg3);
 990 
 991     enum {
 992            // After previous pushes.
 993            oop_to_verify = 6 * wordSize,
 994            saved_rax     = 7 * wordSize,
 995            saved_r10     = 8 * wordSize,
 996 
 997            // Before the call to MacroAssembler::debug(), see below.
 998            return_addr   = 16 * wordSize,
 999            error_msg     = 17 * wordSize
1000     };
1001 
1002     // get object
1003     __ movptr(rax, Address(rsp, oop_to_verify));
1004 
1005     // make sure object is 'reasonable'
1006     __ testptr(rax, rax);
1007     __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1008     // Check if the oop is in the right area of memory
1009     __ movptr(c_rarg2, rax);
1010     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1011     __ andptr(c_rarg2, c_rarg3);
1012     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1013     __ cmpptr(c_rarg2, c_rarg3);
1014     __ jcc(Assembler::notZero, error);
1015 
1016     // set r12 to heapbase for load_klass()
1017     __ reinit_heapbase();
1018 
1019     // make sure klass is 'reasonable', which is not zero.
1020     __ load_klass(rax, rax);  // get klass
1021     __ testptr(rax, rax);
1022     __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1023     // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers
1024 
1025     // return if everything seems ok
1026     __ bind(exit);
1027     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1028     __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1029     __ pop(c_rarg3);                             // restore c_rarg3
1030     __ pop(c_rarg2);                             // restore c_rarg2
1031     __ pop(r12);                                 // restore r12
1032     __ popf();                                   // restore flags
1033     __ ret(4 * wordSize);                        // pop caller saved stuff
1034 
1035     // handle errors
1036     __ bind(error);
1037     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1038     __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1039     __ pop(c_rarg3);                             // get saved c_rarg3 back
1040     __ pop(c_rarg2);                             // get saved c_rarg2 back
1041     __ pop(r12);                                 // get saved r12 back
1042     __ popf();                                   // get saved flags off stack --
1043                                                  // will be ignored
1044 
1045     __ pusha();                                  // push registers
1046                                                  // (rip is already
1047                                                  // already pushed)
1048     // debug(char* msg, int64_t pc, int64_t regs[])
1049     // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1050     // pushed all the registers, so now the stack looks like:
1051     //     [tos +  0] 16 saved registers
1052     //     [tos + 16] return address
1053     //   * [tos + 17] error message (char*)
1054     //   * [tos + 18] object to verify (oop)
1055     //   * [tos + 19] saved rax - saved by caller and bashed
1056     //   * [tos + 20] saved r10 (rscratch1) - saved by caller
1057     //   * = popped on exit
1058 
1059     __ movptr(c_rarg0, Address(rsp, error_msg));    // pass address of error message
1060     __ movptr(c_rarg1, Address(rsp, return_addr));  // pass return address
1061     __ movq(c_rarg2, rsp);                          // pass address of regs on stack
1062     __ mov(r12, rsp);                               // remember rsp
1063     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1064     __ andptr(rsp, -16);                            // align stack as required by ABI
1065     BLOCK_COMMENT("call MacroAssembler::debug");
1066     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1067     __ mov(rsp, r12);                               // restore rsp
1068     __ popa();                                      // pop registers (includes r12)
1069     __ ret(4 * wordSize);                           // pop caller saved stuff
1070 
1071     return start;
1072   }
1073 
1074   //
1075   // Verify that a register contains clean 32-bits positive value
1076   // (high 32-bits are 0) so it could be used in 64-bits shifts.
1077   //
1078   //  Input:
1079   //    Rint  -  32-bits value
1080   //    Rtmp  -  scratch
1081   //
1082   void assert_clean_int(Register Rint, Register Rtmp) {
1083 #ifdef ASSERT
1084     Label L;
1085     assert_different_registers(Rtmp, Rint);
1086     __ movslq(Rtmp, Rint);
1087     __ cmpq(Rtmp, Rint);
1088     __ jcc(Assembler::equal, L);
1089     __ stop("high 32-bits of int value are not 0");
1090     __ bind(L);
1091 #endif
1092   }
1093 
1094   //  Generate overlap test for array copy stubs
1095   //
1096   //  Input:
1097   //     c_rarg0 - from
1098   //     c_rarg1 - to
1099   //     c_rarg2 - element count
1100   //
1101   //  Output:
1102   //     rax   - &from[element count - 1]
1103   //
1104   void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1105     assert(no_overlap_target != NULL, "must be generated");
1106     array_overlap_test(no_overlap_target, NULL, sf);
1107   }
1108   void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1109     array_overlap_test(NULL, &L_no_overlap, sf);
1110   }
1111   void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1112     const Register from     = c_rarg0;
1113     const Register to       = c_rarg1;
1114     const Register count    = c_rarg2;
1115     const Register end_from = rax;
1116 
1117     __ cmpptr(to, from);
1118     __ lea(end_from, Address(from, count, sf, 0));
1119     if (NOLp == NULL) {
1120       ExternalAddress no_overlap(no_overlap_target);
1121       __ jump_cc(Assembler::belowEqual, no_overlap);
1122       __ cmpptr(to, end_from);
1123       __ jump_cc(Assembler::aboveEqual, no_overlap);
1124     } else {
1125       __ jcc(Assembler::belowEqual, (*NOLp));
1126       __ cmpptr(to, end_from);
1127       __ jcc(Assembler::aboveEqual, (*NOLp));
1128     }
1129   }
1130 
1131   // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1132   //
1133   // Outputs:
1134   //    rdi - rcx
1135   //    rsi - rdx
1136   //    rdx - r8
1137   //    rcx - r9
1138   //
1139   // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1140   // are non-volatile.  r9 and r10 should not be used by the caller.
1141   //
1142   void setup_arg_regs(int nargs = 3) {
1143     const Register saved_rdi = r9;
1144     const Register saved_rsi = r10;
1145     assert(nargs == 3 || nargs == 4, "else fix");
1146 #ifdef _WIN64
1147     assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1148            "unexpected argument registers");
1149     if (nargs >= 4)
1150       __ mov(rax, r9);  // r9 is also saved_rdi
1151     __ movptr(saved_rdi, rdi);
1152     __ movptr(saved_rsi, rsi);
1153     __ mov(rdi, rcx); // c_rarg0
1154     __ mov(rsi, rdx); // c_rarg1
1155     __ mov(rdx, r8);  // c_rarg2
1156     if (nargs >= 4)
1157       __ mov(rcx, rax); // c_rarg3 (via rax)
1158 #else
1159     assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1160            "unexpected argument registers");
1161 #endif
1162   }
1163 
1164   void restore_arg_regs() {
1165     const Register saved_rdi = r9;
1166     const Register saved_rsi = r10;
1167 #ifdef _WIN64
1168     __ movptr(rdi, saved_rdi);
1169     __ movptr(rsi, saved_rsi);
1170 #endif
1171   }
1172 
1173   // Generate code for an array write pre barrier
1174   //
1175   //     addr    -  starting address
1176   //     count   -  element count
1177   //     tmp     - scratch register
1178   //
1179   //     Destroy no registers!
1180   //
1181   void  gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1182     BarrierSet* bs = Universe::heap()->barrier_set();
1183     switch (bs->kind()) {
1184       case BarrierSet::G1SATBCT:
1185       case BarrierSet::G1SATBCTLogging:
1186         // With G1, don't generate the call if we statically know that the target in uninitialized
1187         if (!dest_uninitialized) {
1188            __ pusha();                      // push registers
1189            if (count == c_rarg0) {
1190              if (addr == c_rarg1) {
1191                // exactly backwards!!
1192                __ xchgptr(c_rarg1, c_rarg0);
1193              } else {
1194                __ movptr(c_rarg1, count);
1195                __ movptr(c_rarg0, addr);
1196              }
1197            } else {
1198              __ movptr(c_rarg0, addr);
1199              __ movptr(c_rarg1, count);
1200            }
1201            __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1202            __ popa();
1203         }
1204          break;
1205       case BarrierSet::CardTableModRef:
1206       case BarrierSet::CardTableExtension:
1207       case BarrierSet::ModRef:
1208         break;
1209       default:
1210         ShouldNotReachHere();
1211 
1212     }
1213   }
1214 
1215   //
1216   // Generate code for an array write post barrier
1217   //
1218   //  Input:
1219   //     start    - register containing starting address of destination array
1220   //     end      - register containing ending address of destination array
1221   //     scratch  - scratch register
1222   //
1223   //  The input registers are overwritten.
1224   //  The ending address is inclusive.
1225   void  gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1226     assert_different_registers(start, end, scratch);
1227     BarrierSet* bs = Universe::heap()->barrier_set();
1228     switch (bs->kind()) {
1229       case BarrierSet::G1SATBCT:
1230       case BarrierSet::G1SATBCTLogging:
1231 
1232         {
1233           __ pusha();                      // push registers (overkill)
1234           // must compute element count unless barrier set interface is changed (other platforms supply count)
1235           assert_different_registers(start, end, scratch);
1236           __ lea(scratch, Address(end, BytesPerHeapOop));
1237           __ subptr(scratch, start);               // subtract start to get #bytes
1238           __ shrptr(scratch, LogBytesPerHeapOop);  // convert to element count
1239           __ mov(c_rarg0, start);
1240           __ mov(c_rarg1, scratch);
1241           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1242           __ popa();
1243         }
1244         break;
1245       case BarrierSet::CardTableModRef:
1246       case BarrierSet::CardTableExtension:
1247         {
1248           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1249           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1250 
1251           Label L_loop;
1252 
1253            __ shrptr(start, CardTableModRefBS::card_shift);
1254            __ addptr(end, BytesPerHeapOop);
1255            __ shrptr(end, CardTableModRefBS::card_shift);
1256            __ subptr(end, start); // number of bytes to copy
1257 
1258           intptr_t disp = (intptr_t) ct->byte_map_base;
1259           if (Assembler::is_simm32(disp)) {
1260             Address cardtable(noreg, noreg, Address::no_scale, disp);
1261             __ lea(scratch, cardtable);
1262           } else {
1263             ExternalAddress cardtable((address)disp);
1264             __ lea(scratch, cardtable);
1265           }
1266 
1267           const Register count = end; // 'end' register contains bytes count now
1268           __ addptr(start, scratch);
1269         __ BIND(L_loop);
1270           __ movb(Address(start, count, Address::times_1), 0);
1271           __ decrement(count);
1272           __ jcc(Assembler::greaterEqual, L_loop);
1273         }
1274         break;
1275       default:
1276         ShouldNotReachHere();
1277 
1278     }
1279   }
1280 
1281 
1282   // Copy big chunks forward
1283   //
1284   // Inputs:
1285   //   end_from     - source arrays end address
1286   //   end_to       - destination array end address
1287   //   qword_count  - 64-bits element count, negative
1288   //   to           - scratch
1289   //   L_copy_32_bytes - entry label
1290   //   L_copy_8_bytes  - exit  label
1291   //
1292   void copy_32_bytes_forward(Register end_from, Register end_to,
1293                              Register qword_count, Register to,
1294                              Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1295     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1296     Label L_loop;
1297     __ align(OptoLoopAlignment);
1298   __ BIND(L_loop);
1299     if(UseUnalignedLoadStores) {
1300       __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1301       __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1302       __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1303       __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1304 
1305     } else {
1306       __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1307       __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1308       __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1309       __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1310       __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1311       __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1312       __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1313       __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1314     }
1315   __ BIND(L_copy_32_bytes);
1316     __ addptr(qword_count, 4);
1317     __ jcc(Assembler::lessEqual, L_loop);
1318     __ subptr(qword_count, 4);
1319     __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1320   }
1321 
1322 
1323   // Copy big chunks backward
1324   //
1325   // Inputs:
1326   //   from         - source arrays address
1327   //   dest         - destination array address
1328   //   qword_count  - 64-bits element count
1329   //   to           - scratch
1330   //   L_copy_32_bytes - entry label
1331   //   L_copy_8_bytes  - exit  label
1332   //
1333   void copy_32_bytes_backward(Register from, Register dest,
1334                               Register qword_count, Register to,
1335                               Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1336     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1337     Label L_loop;
1338     __ align(OptoLoopAlignment);
1339   __ BIND(L_loop);
1340     if(UseUnalignedLoadStores) {
1341       __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1342       __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1343       __ movdqu(xmm1, Address(from, qword_count, Address::times_8,  0));
1344       __ movdqu(Address(dest, qword_count, Address::times_8,  0), xmm1);
1345 
1346     } else {
1347       __ movq(to, Address(from, qword_count, Address::times_8, 24));
1348       __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1349       __ movq(to, Address(from, qword_count, Address::times_8, 16));
1350       __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1351       __ movq(to, Address(from, qword_count, Address::times_8,  8));
1352       __ movq(Address(dest, qword_count, Address::times_8,  8), to);
1353       __ movq(to, Address(from, qword_count, Address::times_8,  0));
1354       __ movq(Address(dest, qword_count, Address::times_8,  0), to);
1355     }
1356   __ BIND(L_copy_32_bytes);
1357     __ subptr(qword_count, 4);
1358     __ jcc(Assembler::greaterEqual, L_loop);
1359     __ addptr(qword_count, 4);
1360     __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1361   }
1362 
1363 
1364   // Arguments:
1365   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1366   //             ignored
1367   //   name    - stub name string
1368   //
1369   // Inputs:
1370   //   c_rarg0   - source array address
1371   //   c_rarg1   - destination array address
1372   //   c_rarg2   - element count, treated as ssize_t, can be zero
1373   //
1374   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1375   // we let the hardware handle it.  The one to eight bytes within words,
1376   // dwords or qwords that span cache line boundaries will still be loaded
1377   // and stored atomically.
1378   //
1379   // Side Effects:
1380   //   disjoint_byte_copy_entry is set to the no-overlap entry point
1381   //   used by generate_conjoint_byte_copy().
1382   //
1383   address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1384     __ align(CodeEntryAlignment);
1385     StubCodeMark mark(this, "StubRoutines", name);
1386     address start = __ pc();
1387 
1388     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1389     Label L_copy_byte, L_exit;
1390     const Register from        = rdi;  // source array address
1391     const Register to          = rsi;  // destination array address
1392     const Register count       = rdx;  // elements count
1393     const Register byte_count  = rcx;
1394     const Register qword_count = count;
1395     const Register end_from    = from; // source array end address
1396     const Register end_to      = to;   // destination array end address
1397     // End pointers are inclusive, and if count is not zero they point
1398     // to the last unit copied:  end_to[0] := end_from[0]
1399 
1400     __ enter(); // required for proper stackwalking of RuntimeStub frame
1401     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1402 
1403     if (entry != NULL) {
1404       *entry = __ pc();
1405        // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1406       BLOCK_COMMENT("Entry:");
1407     }
1408 
1409     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1410                       // r9 and r10 may be used to save non-volatile registers
1411 
1412     // 'from', 'to' and 'count' are now valid
1413     __ movptr(byte_count, count);
1414     __ shrptr(count, 3); // count => qword_count
1415 
1416     // Copy from low to high addresses.  Use 'to' as scratch.
1417     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1418     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1419     __ negptr(qword_count); // make the count negative
1420     __ jmp(L_copy_32_bytes);
1421 
1422     // Copy trailing qwords
1423   __ BIND(L_copy_8_bytes);
1424     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1425     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1426     __ increment(qword_count);
1427     __ jcc(Assembler::notZero, L_copy_8_bytes);
1428 
1429     // Check for and copy trailing dword
1430   __ BIND(L_copy_4_bytes);
1431     __ testl(byte_count, 4);
1432     __ jccb(Assembler::zero, L_copy_2_bytes);
1433     __ movl(rax, Address(end_from, 8));
1434     __ movl(Address(end_to, 8), rax);
1435 
1436     __ addptr(end_from, 4);
1437     __ addptr(end_to, 4);
1438 
1439     // Check for and copy trailing word
1440   __ BIND(L_copy_2_bytes);
1441     __ testl(byte_count, 2);
1442     __ jccb(Assembler::zero, L_copy_byte);
1443     __ movw(rax, Address(end_from, 8));
1444     __ movw(Address(end_to, 8), rax);
1445 
1446     __ addptr(end_from, 2);
1447     __ addptr(end_to, 2);
1448 
1449     // Check for and copy trailing byte
1450   __ BIND(L_copy_byte);
1451     __ testl(byte_count, 1);
1452     __ jccb(Assembler::zero, L_exit);
1453     __ movb(rax, Address(end_from, 8));
1454     __ movb(Address(end_to, 8), rax);
1455 
1456   __ BIND(L_exit);
1457     restore_arg_regs();
1458     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1459     __ xorptr(rax, rax); // return 0
1460     __ leave(); // required for proper stackwalking of RuntimeStub frame
1461     __ ret(0);
1462 
1463     // Copy in 32-bytes chunks
1464     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1465     __ jmp(L_copy_4_bytes);
1466 
1467     return start;
1468   }
1469 
1470   // Arguments:
1471   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1472   //             ignored
1473   //   name    - stub name string
1474   //
1475   // Inputs:
1476   //   c_rarg0   - source array address
1477   //   c_rarg1   - destination array address
1478   //   c_rarg2   - element count, treated as ssize_t, can be zero
1479   //
1480   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1481   // we let the hardware handle it.  The one to eight bytes within words,
1482   // dwords or qwords that span cache line boundaries will still be loaded
1483   // and stored atomically.
1484   //
1485   address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1486                                       address* entry, const char *name) {
1487     __ align(CodeEntryAlignment);
1488     StubCodeMark mark(this, "StubRoutines", name);
1489     address start = __ pc();
1490 
1491     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1492     const Register from        = rdi;  // source array address
1493     const Register to          = rsi;  // destination array address
1494     const Register count       = rdx;  // elements count
1495     const Register byte_count  = rcx;
1496     const Register qword_count = count;
1497 
1498     __ enter(); // required for proper stackwalking of RuntimeStub frame
1499     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1500 
1501     if (entry != NULL) {
1502       *entry = __ pc();
1503       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1504       BLOCK_COMMENT("Entry:");
1505     }
1506 
1507     array_overlap_test(nooverlap_target, Address::times_1);
1508     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1509                       // r9 and r10 may be used to save non-volatile registers
1510 
1511     // 'from', 'to' and 'count' are now valid
1512     __ movptr(byte_count, count);
1513     __ shrptr(count, 3);   // count => qword_count
1514 
1515     // Copy from high to low addresses.
1516 
1517     // Check for and copy trailing byte
1518     __ testl(byte_count, 1);
1519     __ jcc(Assembler::zero, L_copy_2_bytes);
1520     __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1521     __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1522     __ decrement(byte_count); // Adjust for possible trailing word
1523 
1524     // Check for and copy trailing word
1525   __ BIND(L_copy_2_bytes);
1526     __ testl(byte_count, 2);
1527     __ jcc(Assembler::zero, L_copy_4_bytes);
1528     __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1529     __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1530 
1531     // Check for and copy trailing dword
1532   __ BIND(L_copy_4_bytes);
1533     __ testl(byte_count, 4);
1534     __ jcc(Assembler::zero, L_copy_32_bytes);
1535     __ movl(rax, Address(from, qword_count, Address::times_8));
1536     __ movl(Address(to, qword_count, Address::times_8), rax);
1537     __ jmp(L_copy_32_bytes);
1538 
1539     // Copy trailing qwords
1540   __ BIND(L_copy_8_bytes);
1541     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1542     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1543     __ decrement(qword_count);
1544     __ jcc(Assembler::notZero, L_copy_8_bytes);
1545 
1546     restore_arg_regs();
1547     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1548     __ xorptr(rax, rax); // return 0
1549     __ leave(); // required for proper stackwalking of RuntimeStub frame
1550     __ ret(0);
1551 
1552     // Copy in 32-bytes chunks
1553     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1554 
1555     restore_arg_regs();
1556     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1557     __ xorptr(rax, rax); // return 0
1558     __ leave(); // required for proper stackwalking of RuntimeStub frame
1559     __ ret(0);
1560 
1561     return start;
1562   }
1563 
1564   // Arguments:
1565   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1566   //             ignored
1567   //   name    - stub name string
1568   //
1569   // Inputs:
1570   //   c_rarg0   - source array address
1571   //   c_rarg1   - destination array address
1572   //   c_rarg2   - element count, treated as ssize_t, can be zero
1573   //
1574   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1575   // let the hardware handle it.  The two or four words within dwords
1576   // or qwords that span cache line boundaries will still be loaded
1577   // and stored atomically.
1578   //
1579   // Side Effects:
1580   //   disjoint_short_copy_entry is set to the no-overlap entry point
1581   //   used by generate_conjoint_short_copy().
1582   //
1583   address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1584     __ align(CodeEntryAlignment);
1585     StubCodeMark mark(this, "StubRoutines", name);
1586     address start = __ pc();
1587 
1588     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1589     const Register from        = rdi;  // source array address
1590     const Register to          = rsi;  // destination array address
1591     const Register count       = rdx;  // elements count
1592     const Register word_count  = rcx;
1593     const Register qword_count = count;
1594     const Register end_from    = from; // source array end address
1595     const Register end_to      = to;   // destination array end address
1596     // End pointers are inclusive, and if count is not zero they point
1597     // to the last unit copied:  end_to[0] := end_from[0]
1598 
1599     __ enter(); // required for proper stackwalking of RuntimeStub frame
1600     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1601 
1602     if (entry != NULL) {
1603       *entry = __ pc();
1604       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1605       BLOCK_COMMENT("Entry:");
1606     }
1607 
1608     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1609                       // r9 and r10 may be used to save non-volatile registers
1610 
1611     // 'from', 'to' and 'count' are now valid
1612     __ movptr(word_count, count);
1613     __ shrptr(count, 2); // count => qword_count
1614 
1615     // Copy from low to high addresses.  Use 'to' as scratch.
1616     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1617     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1618     __ negptr(qword_count);
1619     __ jmp(L_copy_32_bytes);
1620 
1621     // Copy trailing qwords
1622   __ BIND(L_copy_8_bytes);
1623     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1624     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1625     __ increment(qword_count);
1626     __ jcc(Assembler::notZero, L_copy_8_bytes);
1627 
1628     // Original 'dest' is trashed, so we can't use it as a
1629     // base register for a possible trailing word copy
1630 
1631     // Check for and copy trailing dword
1632   __ BIND(L_copy_4_bytes);
1633     __ testl(word_count, 2);
1634     __ jccb(Assembler::zero, L_copy_2_bytes);
1635     __ movl(rax, Address(end_from, 8));
1636     __ movl(Address(end_to, 8), rax);
1637 
1638     __ addptr(end_from, 4);
1639     __ addptr(end_to, 4);
1640 
1641     // Check for and copy trailing word
1642   __ BIND(L_copy_2_bytes);
1643     __ testl(word_count, 1);
1644     __ jccb(Assembler::zero, L_exit);
1645     __ movw(rax, Address(end_from, 8));
1646     __ movw(Address(end_to, 8), rax);
1647 
1648   __ BIND(L_exit);
1649     restore_arg_regs();
1650     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1651     __ xorptr(rax, rax); // return 0
1652     __ leave(); // required for proper stackwalking of RuntimeStub frame
1653     __ ret(0);
1654 
1655     // Copy in 32-bytes chunks
1656     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1657     __ jmp(L_copy_4_bytes);
1658 
1659     return start;
1660   }
1661 
1662   address generate_fill(BasicType t, bool aligned, const char *name) {
1663     __ align(CodeEntryAlignment);
1664     StubCodeMark mark(this, "StubRoutines", name);
1665     address start = __ pc();
1666 
1667     BLOCK_COMMENT("Entry:");
1668 
1669     const Register to       = c_rarg0;  // source array address
1670     const Register value    = c_rarg1;  // value
1671     const Register count    = c_rarg2;  // elements count
1672 
1673     __ enter(); // required for proper stackwalking of RuntimeStub frame
1674 
1675     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1676 
1677     __ leave(); // required for proper stackwalking of RuntimeStub frame
1678     __ ret(0);
1679     return start;
1680   }
1681 
1682   // Arguments:
1683   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1684   //             ignored
1685   //   name    - stub name string
1686   //
1687   // Inputs:
1688   //   c_rarg0   - source array address
1689   //   c_rarg1   - destination array address
1690   //   c_rarg2   - element count, treated as ssize_t, can be zero
1691   //
1692   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1693   // let the hardware handle it.  The two or four words within dwords
1694   // or qwords that span cache line boundaries will still be loaded
1695   // and stored atomically.
1696   //
1697   address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1698                                        address *entry, const char *name) {
1699     __ align(CodeEntryAlignment);
1700     StubCodeMark mark(this, "StubRoutines", name);
1701     address start = __ pc();
1702 
1703     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1704     const Register from        = rdi;  // source array address
1705     const Register to          = rsi;  // destination array address
1706     const Register count       = rdx;  // elements count
1707     const Register word_count  = rcx;
1708     const Register qword_count = count;
1709 
1710     __ enter(); // required for proper stackwalking of RuntimeStub frame
1711     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1712 
1713     if (entry != NULL) {
1714       *entry = __ pc();
1715       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1716       BLOCK_COMMENT("Entry:");
1717     }
1718 
1719     array_overlap_test(nooverlap_target, Address::times_2);
1720     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1721                       // r9 and r10 may be used to save non-volatile registers
1722 
1723     // 'from', 'to' and 'count' are now valid
1724     __ movptr(word_count, count);
1725     __ shrptr(count, 2); // count => qword_count
1726 
1727     // Copy from high to low addresses.  Use 'to' as scratch.
1728 
1729     // Check for and copy trailing word
1730     __ testl(word_count, 1);
1731     __ jccb(Assembler::zero, L_copy_4_bytes);
1732     __ movw(rax, Address(from, word_count, Address::times_2, -2));
1733     __ movw(Address(to, word_count, Address::times_2, -2), rax);
1734 
1735     // Check for and copy trailing dword
1736   __ BIND(L_copy_4_bytes);
1737     __ testl(word_count, 2);
1738     __ jcc(Assembler::zero, L_copy_32_bytes);
1739     __ movl(rax, Address(from, qword_count, Address::times_8));
1740     __ movl(Address(to, qword_count, Address::times_8), rax);
1741     __ jmp(L_copy_32_bytes);
1742 
1743     // Copy trailing qwords
1744   __ BIND(L_copy_8_bytes);
1745     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1746     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1747     __ decrement(qword_count);
1748     __ jcc(Assembler::notZero, L_copy_8_bytes);
1749 
1750     restore_arg_regs();
1751     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1752     __ xorptr(rax, rax); // return 0
1753     __ leave(); // required for proper stackwalking of RuntimeStub frame
1754     __ ret(0);
1755 
1756     // Copy in 32-bytes chunks
1757     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1758 
1759     restore_arg_regs();
1760     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1761     __ xorptr(rax, rax); // return 0
1762     __ leave(); // required for proper stackwalking of RuntimeStub frame
1763     __ ret(0);
1764 
1765     return start;
1766   }
1767 
1768   // Arguments:
1769   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1770   //             ignored
1771   //   is_oop  - true => oop array, so generate store check code
1772   //   name    - stub name string
1773   //
1774   // Inputs:
1775   //   c_rarg0   - source array address
1776   //   c_rarg1   - destination array address
1777   //   c_rarg2   - element count, treated as ssize_t, can be zero
1778   //
1779   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1780   // the hardware handle it.  The two dwords within qwords that span
1781   // cache line boundaries will still be loaded and stored atomicly.
1782   //
1783   // Side Effects:
1784   //   disjoint_int_copy_entry is set to the no-overlap entry point
1785   //   used by generate_conjoint_int_oop_copy().
1786   //
1787   address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1788                                          const char *name, bool dest_uninitialized = false) {
1789     __ align(CodeEntryAlignment);
1790     StubCodeMark mark(this, "StubRoutines", name);
1791     address start = __ pc();
1792 
1793     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1794     const Register from        = rdi;  // source array address
1795     const Register to          = rsi;  // destination array address
1796     const Register count       = rdx;  // elements count
1797     const Register dword_count = rcx;
1798     const Register qword_count = count;
1799     const Register end_from    = from; // source array end address
1800     const Register end_to      = to;   // destination array end address
1801     const Register saved_to    = r11;  // saved destination array address
1802     // End pointers are inclusive, and if count is not zero they point
1803     // to the last unit copied:  end_to[0] := end_from[0]
1804 
1805     __ enter(); // required for proper stackwalking of RuntimeStub frame
1806     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1807 
1808     if (entry != NULL) {
1809       *entry = __ pc();
1810       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1811       BLOCK_COMMENT("Entry:");
1812     }
1813 
1814     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1815                       // r9 and r10 may be used to save non-volatile registers
1816     if (is_oop) {
1817       __ movq(saved_to, to);
1818       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1819     }
1820 
1821     // 'from', 'to' and 'count' are now valid
1822     __ movptr(dword_count, count);
1823     __ shrptr(count, 1); // count => qword_count
1824 
1825     // Copy from low to high addresses.  Use 'to' as scratch.
1826     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1827     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1828     __ negptr(qword_count);
1829     __ jmp(L_copy_32_bytes);
1830 
1831     // Copy trailing qwords
1832   __ BIND(L_copy_8_bytes);
1833     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1834     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1835     __ increment(qword_count);
1836     __ jcc(Assembler::notZero, L_copy_8_bytes);
1837 
1838     // Check for and copy trailing dword
1839   __ BIND(L_copy_4_bytes);
1840     __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1841     __ jccb(Assembler::zero, L_exit);
1842     __ movl(rax, Address(end_from, 8));
1843     __ movl(Address(end_to, 8), rax);
1844 
1845   __ BIND(L_exit);
1846     if (is_oop) {
1847       __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1848       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1849     }
1850     restore_arg_regs();
1851     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1852     __ xorptr(rax, rax); // return 0
1853     __ leave(); // required for proper stackwalking of RuntimeStub frame
1854     __ ret(0);
1855 
1856     // Copy 32-bytes chunks
1857     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1858     __ jmp(L_copy_4_bytes);
1859 
1860     return start;
1861   }
1862 
1863   // Arguments:
1864   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1865   //             ignored
1866   //   is_oop  - true => oop array, so generate store check code
1867   //   name    - stub name string
1868   //
1869   // Inputs:
1870   //   c_rarg0   - source array address
1871   //   c_rarg1   - destination array address
1872   //   c_rarg2   - element count, treated as ssize_t, can be zero
1873   //
1874   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1875   // the hardware handle it.  The two dwords within qwords that span
1876   // cache line boundaries will still be loaded and stored atomicly.
1877   //
1878   address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1879                                          address *entry, const char *name,
1880                                          bool dest_uninitialized = false) {
1881     __ align(CodeEntryAlignment);
1882     StubCodeMark mark(this, "StubRoutines", name);
1883     address start = __ pc();
1884 
1885     Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1886     const Register from        = rdi;  // source array address
1887     const Register to          = rsi;  // destination array address
1888     const Register count       = rdx;  // elements count
1889     const Register dword_count = rcx;
1890     const Register qword_count = count;
1891 
1892     __ enter(); // required for proper stackwalking of RuntimeStub frame
1893     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1894 
1895     if (entry != NULL) {
1896       *entry = __ pc();
1897        // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1898       BLOCK_COMMENT("Entry:");
1899     }
1900 
1901     array_overlap_test(nooverlap_target, Address::times_4);
1902     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1903                       // r9 and r10 may be used to save non-volatile registers
1904 
1905     if (is_oop) {
1906       // no registers are destroyed by this call
1907       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1908     }
1909 
1910     assert_clean_int(count, rax); // Make sure 'count' is clean int.
1911     // 'from', 'to' and 'count' are now valid
1912     __ movptr(dword_count, count);
1913     __ shrptr(count, 1); // count => qword_count
1914 
1915     // Copy from high to low addresses.  Use 'to' as scratch.
1916 
1917     // Check for and copy trailing dword
1918     __ testl(dword_count, 1);
1919     __ jcc(Assembler::zero, L_copy_32_bytes);
1920     __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1921     __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1922     __ jmp(L_copy_32_bytes);
1923 
1924     // Copy trailing qwords
1925   __ BIND(L_copy_8_bytes);
1926     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1927     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1928     __ decrement(qword_count);
1929     __ jcc(Assembler::notZero, L_copy_8_bytes);
1930 
1931     if (is_oop) {
1932       __ jmp(L_exit);
1933     }
1934     restore_arg_regs();
1935     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1936     __ xorptr(rax, rax); // return 0
1937     __ leave(); // required for proper stackwalking of RuntimeStub frame
1938     __ ret(0);
1939 
1940     // Copy in 32-bytes chunks
1941     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1942 
1943    __ bind(L_exit);
1944      if (is_oop) {
1945        Register end_to = rdx;
1946        __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1947        gen_write_ref_array_post_barrier(to, end_to, rax);
1948      }
1949     restore_arg_regs();
1950     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1951     __ xorptr(rax, rax); // return 0
1952     __ leave(); // required for proper stackwalking of RuntimeStub frame
1953     __ ret(0);
1954 
1955     return start;
1956   }
1957 
1958   // Arguments:
1959   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1960   //             ignored
1961   //   is_oop  - true => oop array, so generate store check code
1962   //   name    - stub name string
1963   //
1964   // Inputs:
1965   //   c_rarg0   - source array address
1966   //   c_rarg1   - destination array address
1967   //   c_rarg2   - element count, treated as ssize_t, can be zero
1968   //
1969  // Side Effects:
1970   //   disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1971   //   no-overlap entry point used by generate_conjoint_long_oop_copy().
1972   //
1973   address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
1974                                           const char *name, bool dest_uninitialized = false) {
1975     __ align(CodeEntryAlignment);
1976     StubCodeMark mark(this, "StubRoutines", name);
1977     address start = __ pc();
1978 
1979     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1980     const Register from        = rdi;  // source array address
1981     const Register to          = rsi;  // destination array address
1982     const Register qword_count = rdx;  // elements count
1983     const Register end_from    = from; // source array end address
1984     const Register end_to      = rcx;  // destination array end address
1985     const Register saved_to    = to;
1986     // End pointers are inclusive, and if count is not zero they point
1987     // to the last unit copied:  end_to[0] := end_from[0]
1988 
1989     __ enter(); // required for proper stackwalking of RuntimeStub frame
1990     // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1991     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1992 
1993     if (entry != NULL) {
1994       *entry = __ pc();
1995       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1996       BLOCK_COMMENT("Entry:");
1997     }
1998 
1999     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2000                       // r9 and r10 may be used to save non-volatile registers
2001     // 'from', 'to' and 'qword_count' are now valid
2002     if (is_oop) {
2003       // no registers are destroyed by this call
2004       gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2005     }
2006 
2007     // Copy from low to high addresses.  Use 'to' as scratch.
2008     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2009     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
2010     __ negptr(qword_count);
2011     __ jmp(L_copy_32_bytes);
2012 
2013     // Copy trailing qwords
2014   __ BIND(L_copy_8_bytes);
2015     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2016     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2017     __ increment(qword_count);
2018     __ jcc(Assembler::notZero, L_copy_8_bytes);
2019 
2020     if (is_oop) {
2021       __ jmp(L_exit);
2022     } else {
2023       restore_arg_regs();
2024       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2025       __ xorptr(rax, rax); // return 0
2026       __ leave(); // required for proper stackwalking of RuntimeStub frame
2027       __ ret(0);
2028     }
2029 
2030     // Copy 64-byte chunks
2031     copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2032 
2033     if (is_oop) {
2034     __ BIND(L_exit);
2035       gen_write_ref_array_post_barrier(saved_to, end_to, rax);
2036     }
2037     restore_arg_regs();
2038     if (is_oop) {
2039       inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2040     } else {
2041       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2042     }
2043     __ xorptr(rax, rax); // return 0
2044     __ leave(); // required for proper stackwalking of RuntimeStub frame
2045     __ ret(0);
2046 
2047     return start;
2048   }
2049 
2050   // Arguments:
2051   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2052   //             ignored
2053   //   is_oop  - true => oop array, so generate store check code
2054   //   name    - stub name string
2055   //
2056   // Inputs:
2057   //   c_rarg0   - source array address
2058   //   c_rarg1   - destination array address
2059   //   c_rarg2   - element count, treated as ssize_t, can be zero
2060   //
2061   address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2062                                           address nooverlap_target, address *entry,
2063                                           const char *name, bool dest_uninitialized = false) {
2064     __ align(CodeEntryAlignment);
2065     StubCodeMark mark(this, "StubRoutines", name);
2066     address start = __ pc();
2067 
2068     Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2069     const Register from        = rdi;  // source array address
2070     const Register to          = rsi;  // destination array address
2071     const Register qword_count = rdx;  // elements count
2072     const Register saved_count = rcx;
2073 
2074     __ enter(); // required for proper stackwalking of RuntimeStub frame
2075     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
2076 
2077     if (entry != NULL) {
2078       *entry = __ pc();
2079       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2080       BLOCK_COMMENT("Entry:");
2081     }
2082 
2083     array_overlap_test(nooverlap_target, Address::times_8);
2084     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2085                       // r9 and r10 may be used to save non-volatile registers
2086     // 'from', 'to' and 'qword_count' are now valid
2087     if (is_oop) {
2088       // Save to and count for store barrier
2089       __ movptr(saved_count, qword_count);
2090       // No registers are destroyed by this call
2091       gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2092     }
2093 
2094     __ jmp(L_copy_32_bytes);
2095 
2096     // Copy trailing qwords
2097   __ BIND(L_copy_8_bytes);
2098     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2099     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2100     __ decrement(qword_count);
2101     __ jcc(Assembler::notZero, L_copy_8_bytes);
2102 
2103     if (is_oop) {
2104       __ jmp(L_exit);
2105     } else {
2106       restore_arg_regs();
2107       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2108       __ xorptr(rax, rax); // return 0
2109       __ leave(); // required for proper stackwalking of RuntimeStub frame
2110       __ ret(0);
2111     }
2112 
2113     // Copy in 32-bytes chunks
2114     copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2115 
2116     if (is_oop) {
2117     __ BIND(L_exit);
2118       __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2119       gen_write_ref_array_post_barrier(to, rcx, rax);
2120     }
2121     restore_arg_regs();
2122     if (is_oop) {
2123       inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2124     } else {
2125       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2126     }
2127     __ xorptr(rax, rax); // return 0
2128     __ leave(); // required for proper stackwalking of RuntimeStub frame
2129     __ ret(0);
2130 
2131     return start;
2132   }
2133 
2134 
2135   // Helper for generating a dynamic type check.
2136   // Smashes no registers.
2137   void generate_type_check(Register sub_klass,
2138                            Register super_check_offset,
2139                            Register super_klass,
2140                            Label& L_success) {
2141     assert_different_registers(sub_klass, super_check_offset, super_klass);
2142 
2143     BLOCK_COMMENT("type_check:");
2144 
2145     Label L_miss;
2146 
2147     __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg,        &L_success, &L_miss, NULL,
2148                                      super_check_offset);
2149     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2150 
2151     // Fall through on failure!
2152     __ BIND(L_miss);
2153   }
2154 
2155   //
2156   //  Generate checkcasting array copy stub
2157   //
2158   //  Input:
2159   //    c_rarg0   - source array address
2160   //    c_rarg1   - destination array address
2161   //    c_rarg2   - element count, treated as ssize_t, can be zero
2162   //    c_rarg3   - size_t ckoff (super_check_offset)
2163   // not Win64
2164   //    c_rarg4   - oop ckval (super_klass)
2165   // Win64
2166   //    rsp+40    - oop ckval (super_klass)
2167   //
2168   //  Output:
2169   //    rax ==  0  -  success
2170   //    rax == -1^K - failure, where K is partial transfer count
2171   //
2172   address generate_checkcast_copy(const char *name, address *entry,
2173                                   bool dest_uninitialized = false) {
2174 
2175     Label L_load_element, L_store_element, L_do_card_marks, L_done;
2176 
2177     // Input registers (after setup_arg_regs)
2178     const Register from        = rdi;   // source array address
2179     const Register to          = rsi;   // destination array address
2180     const Register length      = rdx;   // elements count
2181     const Register ckoff       = rcx;   // super_check_offset
2182     const Register ckval       = r8;    // super_klass
2183 
2184     // Registers used as temps (r13, r14 are save-on-entry)
2185     const Register end_from    = from;  // source array end address
2186     const Register end_to      = r13;   // destination array end address
2187     const Register count       = rdx;   // -(count_remaining)
2188     const Register r14_length  = r14;   // saved copy of length
2189     // End pointers are inclusive, and if length is not zero they point
2190     // to the last unit copied:  end_to[0] := end_from[0]
2191 
2192     const Register rax_oop    = rax;    // actual oop copied
2193     const Register r11_klass  = r11;    // oop._klass
2194 
2195     //---------------------------------------------------------------
2196     // Assembler stub will be used for this call to arraycopy
2197     // if the two arrays are subtypes of Object[] but the
2198     // destination array type is not equal to or a supertype
2199     // of the source type.  Each element must be separately
2200     // checked.
2201 
2202     __ align(CodeEntryAlignment);
2203     StubCodeMark mark(this, "StubRoutines", name);
2204     address start = __ pc();
2205 
2206     __ enter(); // required for proper stackwalking of RuntimeStub frame
2207 
2208 #ifdef ASSERT
2209     // caller guarantees that the arrays really are different
2210     // otherwise, we would have to make conjoint checks
2211     { Label L;
2212       array_overlap_test(L, TIMES_OOP);
2213       __ stop("checkcast_copy within a single array");
2214       __ bind(L);
2215     }
2216 #endif //ASSERT
2217 
2218     setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2219                        // ckoff => rcx, ckval => r8
2220                        // r9 and r10 may be used to save non-volatile registers
2221 #ifdef _WIN64
2222     // last argument (#4) is on stack on Win64
2223     __ movptr(ckval, Address(rsp, 6 * wordSize));
2224 #endif
2225 
2226     // Caller of this entry point must set up the argument registers.
2227     if (entry != NULL) {
2228       *entry = __ pc();
2229       BLOCK_COMMENT("Entry:");
2230     }
2231 
2232     // allocate spill slots for r13, r14
2233     enum {
2234       saved_r13_offset,
2235       saved_r14_offset,
2236       saved_rbp_offset
2237     };
2238     __ subptr(rsp, saved_rbp_offset * wordSize);
2239     __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2240     __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2241 
2242     // check that int operands are properly extended to size_t
2243     assert_clean_int(length, rax);
2244     assert_clean_int(ckoff, rax);
2245 
2246 #ifdef ASSERT
2247     BLOCK_COMMENT("assert consistent ckoff/ckval");
2248     // The ckoff and ckval must be mutually consistent,
2249     // even though caller generates both.
2250     { Label L;
2251       int sco_offset = in_bytes(Klass::super_check_offset_offset());
2252       __ cmpl(ckoff, Address(ckval, sco_offset));
2253       __ jcc(Assembler::equal, L);
2254       __ stop("super_check_offset inconsistent");
2255       __ bind(L);
2256     }
2257 #endif //ASSERT
2258 
2259     // Loop-invariant addresses.  They are exclusive end pointers.
2260     Address end_from_addr(from, length, TIMES_OOP, 0);
2261     Address   end_to_addr(to,   length, TIMES_OOP, 0);
2262     // Loop-variant addresses.  They assume post-incremented count < 0.
2263     Address from_element_addr(end_from, count, TIMES_OOP, 0);
2264     Address   to_element_addr(end_to,   count, TIMES_OOP, 0);
2265 
2266     gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2267 
2268     // Copy from low to high addresses, indexed from the end of each array.
2269     __ lea(end_from, end_from_addr);
2270     __ lea(end_to,   end_to_addr);
2271     __ movptr(r14_length, length);        // save a copy of the length
2272     assert(length == count, "");          // else fix next line:
2273     __ negptr(count);                     // negate and test the length
2274     __ jcc(Assembler::notZero, L_load_element);
2275 
2276     // Empty array:  Nothing to do.
2277     __ xorptr(rax, rax);                  // return 0 on (trivial) success
2278     __ jmp(L_done);
2279 
2280     // ======== begin loop ========
2281     // (Loop is rotated; its entry is L_load_element.)
2282     // Loop control:
2283     //   for (count = -count; count != 0; count++)
2284     // Base pointers src, dst are biased by 8*(count-1),to last element.
2285     __ align(OptoLoopAlignment);
2286 
2287     __ BIND(L_store_element);
2288     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop
2289     __ increment(count);               // increment the count toward zero
2290     __ jcc(Assembler::zero, L_do_card_marks);
2291 
2292     // ======== loop entry is here ========
2293     __ BIND(L_load_element);
2294     __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2295     __ testptr(rax_oop, rax_oop);
2296     __ jcc(Assembler::zero, L_store_element);
2297 
2298     __ load_klass(r11_klass, rax_oop);// query the object klass
2299     generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2300     // ======== end loop ========
2301 
2302     // It was a real error; we must depend on the caller to finish the job.
2303     // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2304     // Emit GC store barriers for the oops we have copied (r14 + rdx),
2305     // and report their number to the caller.
2306     assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2307     __ lea(end_to, to_element_addr);
2308     __ addptr(end_to, -heapOopSize);      // make an inclusive end pointer
2309     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2310     __ movptr(rax, r14_length);           // original oops
2311     __ addptr(rax, count);                // K = (original - remaining) oops
2312     __ notptr(rax);                       // report (-1^K) to caller
2313     __ jmp(L_done);
2314 
2315     // Come here on success only.
2316     __ BIND(L_do_card_marks);
2317     __ addptr(end_to, -heapOopSize);         // make an inclusive end pointer
2318     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2319     __ xorptr(rax, rax);                  // return 0 on success
2320 
2321     // Common exit point (success or failure).
2322     __ BIND(L_done);
2323     __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2324     __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2325     restore_arg_regs();
2326     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2327     __ leave(); // required for proper stackwalking of RuntimeStub frame
2328     __ ret(0);
2329 
2330     return start;
2331   }
2332 
2333   //
2334   //  Generate 'unsafe' array copy stub
2335   //  Though just as safe as the other stubs, it takes an unscaled
2336   //  size_t argument instead of an element count.
2337   //
2338   //  Input:
2339   //    c_rarg0   - source array address
2340   //    c_rarg1   - destination array address
2341   //    c_rarg2   - byte count, treated as ssize_t, can be zero
2342   //
2343   // Examines the alignment of the operands and dispatches
2344   // to a long, int, short, or byte copy loop.
2345   //
2346   address generate_unsafe_copy(const char *name,
2347                                address byte_copy_entry, address short_copy_entry,
2348                                address int_copy_entry, address long_copy_entry) {
2349 
2350     Label L_long_aligned, L_int_aligned, L_short_aligned;
2351 
2352     // Input registers (before setup_arg_regs)
2353     const Register from        = c_rarg0;  // source array address
2354     const Register to          = c_rarg1;  // destination array address
2355     const Register size        = c_rarg2;  // byte count (size_t)
2356 
2357     // Register used as a temp
2358     const Register bits        = rax;      // test copy of low bits
2359 
2360     __ align(CodeEntryAlignment);
2361     StubCodeMark mark(this, "StubRoutines", name);
2362     address start = __ pc();
2363 
2364     __ enter(); // required for proper stackwalking of RuntimeStub frame
2365 
2366     // bump this on entry, not on exit:
2367     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2368 
2369     __ mov(bits, from);
2370     __ orptr(bits, to);
2371     __ orptr(bits, size);
2372 
2373     __ testb(bits, BytesPerLong-1);
2374     __ jccb(Assembler::zero, L_long_aligned);
2375 
2376     __ testb(bits, BytesPerInt-1);
2377     __ jccb(Assembler::zero, L_int_aligned);
2378 
2379     __ testb(bits, BytesPerShort-1);
2380     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2381 
2382     __ BIND(L_short_aligned);
2383     __ shrptr(size, LogBytesPerShort); // size => short_count
2384     __ jump(RuntimeAddress(short_copy_entry));
2385 
2386     __ BIND(L_int_aligned);
2387     __ shrptr(size, LogBytesPerInt); // size => int_count
2388     __ jump(RuntimeAddress(int_copy_entry));
2389 
2390     __ BIND(L_long_aligned);
2391     __ shrptr(size, LogBytesPerLong); // size => qword_count
2392     __ jump(RuntimeAddress(long_copy_entry));
2393 
2394     return start;
2395   }
2396 
2397   // Perform range checks on the proposed arraycopy.
2398   // Kills temp, but nothing else.
2399   // Also, clean the sign bits of src_pos and dst_pos.
2400   void arraycopy_range_checks(Register src,     // source array oop (c_rarg0)
2401                               Register src_pos, // source position (c_rarg1)
2402                               Register dst,     // destination array oo (c_rarg2)
2403                               Register dst_pos, // destination position (c_rarg3)
2404                               Register length,
2405                               Register temp,
2406                               Label& L_failed) {
2407     BLOCK_COMMENT("arraycopy_range_checks:");
2408 
2409     //  if (src_pos + length > arrayOop(src)->length())  FAIL;
2410     __ movl(temp, length);
2411     __ addl(temp, src_pos);             // src_pos + length
2412     __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2413     __ jcc(Assembler::above, L_failed);
2414 
2415     //  if (dst_pos + length > arrayOop(dst)->length())  FAIL;
2416     __ movl(temp, length);
2417     __ addl(temp, dst_pos);             // dst_pos + length
2418     __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2419     __ jcc(Assembler::above, L_failed);
2420 
2421     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2422     // Move with sign extension can be used since they are positive.
2423     __ movslq(src_pos, src_pos);
2424     __ movslq(dst_pos, dst_pos);
2425 
2426     BLOCK_COMMENT("arraycopy_range_checks done");
2427   }
2428 
2429   //
2430   //  Generate generic array copy stubs
2431   //
2432   //  Input:
2433   //    c_rarg0    -  src oop
2434   //    c_rarg1    -  src_pos (32-bits)
2435   //    c_rarg2    -  dst oop
2436   //    c_rarg3    -  dst_pos (32-bits)
2437   // not Win64
2438   //    c_rarg4    -  element count (32-bits)
2439   // Win64
2440   //    rsp+40     -  element count (32-bits)
2441   //
2442   //  Output:
2443   //    rax ==  0  -  success
2444   //    rax == -1^K - failure, where K is partial transfer count
2445   //
2446   address generate_generic_copy(const char *name,
2447                                 address byte_copy_entry, address short_copy_entry,
2448                                 address int_copy_entry, address oop_copy_entry,
2449                                 address long_copy_entry, address checkcast_copy_entry) {
2450 
2451     Label L_failed, L_failed_0, L_objArray;
2452     Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2453 
2454     // Input registers
2455     const Register src        = c_rarg0;  // source array oop
2456     const Register src_pos    = c_rarg1;  // source position
2457     const Register dst        = c_rarg2;  // destination array oop
2458     const Register dst_pos    = c_rarg3;  // destination position
2459 #ifndef _WIN64
2460     const Register length     = c_rarg4;
2461 #else
2462     const Address  length(rsp, 6 * wordSize);  // elements count is on stack on Win64
2463 #endif
2464 
2465     { int modulus = CodeEntryAlignment;
2466       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
2467       int advance = target - (__ offset() % modulus);
2468       if (advance < 0)  advance += modulus;
2469       if (advance > 0)  __ nop(advance);
2470     }
2471     StubCodeMark mark(this, "StubRoutines", name);
2472 
2473     // Short-hop target to L_failed.  Makes for denser prologue code.
2474     __ BIND(L_failed_0);
2475     __ jmp(L_failed);
2476     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2477 
2478     __ align(CodeEntryAlignment);
2479     address start = __ pc();
2480 
2481     __ enter(); // required for proper stackwalking of RuntimeStub frame
2482 
2483     // bump this on entry, not on exit:
2484     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2485 
2486     //-----------------------------------------------------------------------
2487     // Assembler stub will be used for this call to arraycopy
2488     // if the following conditions are met:
2489     //
2490     // (1) src and dst must not be null.
2491     // (2) src_pos must not be negative.
2492     // (3) dst_pos must not be negative.
2493     // (4) length  must not be negative.
2494     // (5) src klass and dst klass should be the same and not NULL.
2495     // (6) src and dst should be arrays.
2496     // (7) src_pos + length must not exceed length of src.
2497     // (8) dst_pos + length must not exceed length of dst.
2498     //
2499 
2500     //  if (src == NULL) return -1;
2501     __ testptr(src, src);         // src oop
2502     size_t j1off = __ offset();
2503     __ jccb(Assembler::zero, L_failed_0);
2504 
2505     //  if (src_pos < 0) return -1;
2506     __ testl(src_pos, src_pos); // src_pos (32-bits)
2507     __ jccb(Assembler::negative, L_failed_0);
2508 
2509     //  if (dst == NULL) return -1;
2510     __ testptr(dst, dst);         // dst oop
2511     __ jccb(Assembler::zero, L_failed_0);
2512 
2513     //  if (dst_pos < 0) return -1;
2514     __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2515     size_t j4off = __ offset();
2516     __ jccb(Assembler::negative, L_failed_0);
2517 
2518     // The first four tests are very dense code,
2519     // but not quite dense enough to put four
2520     // jumps in a 16-byte instruction fetch buffer.
2521     // That's good, because some branch predicters
2522     // do not like jumps so close together.
2523     // Make sure of this.
2524     guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2525 
2526     // registers used as temp
2527     const Register r11_length    = r11; // elements count to copy
2528     const Register r10_src_klass = r10; // array klass
2529 
2530     //  if (length < 0) return -1;
2531     __ movl(r11_length, length);        // length (elements count, 32-bits value)
2532     __ testl(r11_length, r11_length);
2533     __ jccb(Assembler::negative, L_failed_0);
2534 
2535     __ load_klass(r10_src_klass, src);
2536 #ifdef ASSERT
2537     //  assert(src->klass() != NULL);
2538     {
2539       BLOCK_COMMENT("assert klasses not null {");
2540       Label L1, L2;
2541       __ testptr(r10_src_klass, r10_src_klass);
2542       __ jcc(Assembler::notZero, L2);   // it is broken if klass is NULL
2543       __ bind(L1);
2544       __ stop("broken null klass");
2545       __ bind(L2);
2546       __ load_klass(rax, dst);
2547       __ cmpq(rax, 0);
2548       __ jcc(Assembler::equal, L1);     // this would be broken also
2549       BLOCK_COMMENT("} assert klasses not null done");
2550     }
2551 #endif
2552 
2553     // Load layout helper (32-bits)
2554     //
2555     //  |array_tag|     | header_size | element_type |     |log2_element_size|
2556     // 32        30    24            16              8     2                 0
2557     //
2558     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2559     //
2560 
2561     const int lh_offset = in_bytes(Klass::layout_helper_offset());
2562 
2563     // Handle objArrays completely differently...
2564     const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2565     __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2566     __ jcc(Assembler::equal, L_objArray);
2567 
2568     //  if (src->klass() != dst->klass()) return -1;
2569     __ load_klass(rax, dst);
2570     __ cmpq(r10_src_klass, rax);
2571     __ jcc(Assembler::notEqual, L_failed);
2572 
2573     const Register rax_lh = rax;  // layout helper
2574     __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2575 
2576     //  if (!src->is_Array()) return -1;
2577     __ cmpl(rax_lh, Klass::_lh_neutral_value);
2578     __ jcc(Assembler::greaterEqual, L_failed);
2579 
2580     // At this point, it is known to be a typeArray (array_tag 0x3).
2581 #ifdef ASSERT
2582     {
2583       BLOCK_COMMENT("assert primitive array {");
2584       Label L;
2585       __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2586       __ jcc(Assembler::greaterEqual, L);
2587       __ stop("must be a primitive array");
2588       __ bind(L);
2589       BLOCK_COMMENT("} assert primitive array done");
2590     }
2591 #endif
2592 
2593     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2594                            r10, L_failed);
2595 
2596     // TypeArrayKlass
2597     //
2598     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2599     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2600     //
2601 
2602     const Register r10_offset = r10;    // array offset
2603     const Register rax_elsize = rax_lh; // element size
2604 
2605     __ movl(r10_offset, rax_lh);
2606     __ shrl(r10_offset, Klass::_lh_header_size_shift);
2607     __ andptr(r10_offset, Klass::_lh_header_size_mask);   // array_offset
2608     __ addptr(src, r10_offset);           // src array offset
2609     __ addptr(dst, r10_offset);           // dst array offset
2610     BLOCK_COMMENT("choose copy loop based on element size");
2611     __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2612 
2613     // next registers should be set before the jump to corresponding stub
2614     const Register from     = c_rarg0;  // source array address
2615     const Register to       = c_rarg1;  // destination array address
2616     const Register count    = c_rarg2;  // elements count
2617 
2618     // 'from', 'to', 'count' registers should be set in such order
2619     // since they are the same as 'src', 'src_pos', 'dst'.
2620 
2621   __ BIND(L_copy_bytes);
2622     __ cmpl(rax_elsize, 0);
2623     __ jccb(Assembler::notEqual, L_copy_shorts);
2624     __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2625     __ lea(to,   Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2626     __ movl2ptr(count, r11_length); // length
2627     __ jump(RuntimeAddress(byte_copy_entry));
2628 
2629   __ BIND(L_copy_shorts);
2630     __ cmpl(rax_elsize, LogBytesPerShort);
2631     __ jccb(Assembler::notEqual, L_copy_ints);
2632     __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2633     __ lea(to,   Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2634     __ movl2ptr(count, r11_length); // length
2635     __ jump(RuntimeAddress(short_copy_entry));
2636 
2637   __ BIND(L_copy_ints);
2638     __ cmpl(rax_elsize, LogBytesPerInt);
2639     __ jccb(Assembler::notEqual, L_copy_longs);
2640     __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2641     __ lea(to,   Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2642     __ movl2ptr(count, r11_length); // length
2643     __ jump(RuntimeAddress(int_copy_entry));
2644 
2645   __ BIND(L_copy_longs);
2646 #ifdef ASSERT
2647     {
2648       BLOCK_COMMENT("assert long copy {");
2649       Label L;
2650       __ cmpl(rax_elsize, LogBytesPerLong);
2651       __ jcc(Assembler::equal, L);
2652       __ stop("must be long copy, but elsize is wrong");
2653       __ bind(L);
2654       BLOCK_COMMENT("} assert long copy done");
2655     }
2656 #endif
2657     __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2658     __ lea(to,   Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2659     __ movl2ptr(count, r11_length); // length
2660     __ jump(RuntimeAddress(long_copy_entry));
2661 
2662     // ObjArrayKlass
2663   __ BIND(L_objArray);
2664     // live at this point:  r10_src_klass, r11_length, src[_pos], dst[_pos]
2665 
2666     Label L_plain_copy, L_checkcast_copy;
2667     //  test array classes for subtyping
2668     __ load_klass(rax, dst);
2669     __ cmpq(r10_src_klass, rax); // usual case is exact equality
2670     __ jcc(Assembler::notEqual, L_checkcast_copy);
2671 
2672     // Identically typed arrays can be copied without element-wise checks.
2673     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2674                            r10, L_failed);
2675 
2676     __ lea(from, Address(src, src_pos, TIMES_OOP,
2677                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2678     __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2679                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2680     __ movl2ptr(count, r11_length); // length
2681   __ BIND(L_plain_copy);
2682     __ jump(RuntimeAddress(oop_copy_entry));
2683 
2684   __ BIND(L_checkcast_copy);
2685     // live at this point:  r10_src_klass, r11_length, rax (dst_klass)
2686     {
2687       // Before looking at dst.length, make sure dst is also an objArray.
2688       __ cmpl(Address(rax, lh_offset), objArray_lh);
2689       __ jcc(Assembler::notEqual, L_failed);
2690 
2691       // It is safe to examine both src.length and dst.length.
2692       arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2693                              rax, L_failed);
2694 
2695       const Register r11_dst_klass = r11;
2696       __ load_klass(r11_dst_klass, dst); // reload
2697 
2698       // Marshal the base address arguments now, freeing registers.
2699       __ lea(from, Address(src, src_pos, TIMES_OOP,
2700                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2701       __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2702                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2703       __ movl(count, length);           // length (reloaded)
2704       Register sco_temp = c_rarg3;      // this register is free now
2705       assert_different_registers(from, to, count, sco_temp,
2706                                  r11_dst_klass, r10_src_klass);
2707       assert_clean_int(count, sco_temp);
2708 
2709       // Generate the type check.
2710       const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2711       __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2712       assert_clean_int(sco_temp, rax);
2713       generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2714 
2715       // Fetch destination element klass from the ObjArrayKlass header.
2716       int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2717       __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2718       __ movl(  sco_temp,      Address(r11_dst_klass, sco_offset));
2719       assert_clean_int(sco_temp, rax);
2720 
2721       // the checkcast_copy loop needs two extra arguments:
2722       assert(c_rarg3 == sco_temp, "#3 already in place");
2723       // Set up arguments for checkcast_copy_entry.
2724       setup_arg_regs(4);
2725       __ movptr(r8, r11_dst_klass);  // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2726       __ jump(RuntimeAddress(checkcast_copy_entry));
2727     }
2728 
2729   __ BIND(L_failed);
2730     __ xorptr(rax, rax);
2731     __ notptr(rax); // return -1
2732     __ leave();   // required for proper stackwalking of RuntimeStub frame
2733     __ ret(0);
2734 
2735     return start;
2736   }
2737 
2738   void generate_arraycopy_stubs() {
2739     address entry;
2740     address entry_jbyte_arraycopy;
2741     address entry_jshort_arraycopy;
2742     address entry_jint_arraycopy;
2743     address entry_oop_arraycopy;
2744     address entry_jlong_arraycopy;
2745     address entry_checkcast_arraycopy;
2746 
2747     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, &entry,
2748                                                                            "jbyte_disjoint_arraycopy");
2749     StubRoutines::_jbyte_arraycopy           = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2750                                                                            "jbyte_arraycopy");
2751 
2752     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2753                                                                             "jshort_disjoint_arraycopy");
2754     StubRoutines::_jshort_arraycopy          = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2755                                                                             "jshort_arraycopy");
2756 
2757     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_oop_copy(false, false, &entry,
2758                                                                               "jint_disjoint_arraycopy");
2759     StubRoutines::_jint_arraycopy            = generate_conjoint_int_oop_copy(false, false, entry,
2760                                                                               &entry_jint_arraycopy, "jint_arraycopy");
2761 
2762     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, false, &entry,
2763                                                                                "jlong_disjoint_arraycopy");
2764     StubRoutines::_jlong_arraycopy           = generate_conjoint_long_oop_copy(false, false, entry,
2765                                                                                &entry_jlong_arraycopy, "jlong_arraycopy");
2766 
2767 
2768     if (UseCompressedOops) {
2769       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_int_oop_copy(false, true, &entry,
2770                                                                               "oop_disjoint_arraycopy");
2771       StubRoutines::_oop_arraycopy           = generate_conjoint_int_oop_copy(false, true, entry,
2772                                                                               &entry_oop_arraycopy, "oop_arraycopy");
2773       StubRoutines::_oop_disjoint_arraycopy_uninit  = generate_disjoint_int_oop_copy(false, true, &entry,
2774                                                                                      "oop_disjoint_arraycopy_uninit",
2775                                                                                      /*dest_uninitialized*/true);
2776       StubRoutines::_oop_arraycopy_uninit           = generate_conjoint_int_oop_copy(false, true, entry,
2777                                                                                      NULL, "oop_arraycopy_uninit",
2778                                                                                      /*dest_uninitialized*/true);
2779     } else {
2780       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, true, &entry,
2781                                                                                "oop_disjoint_arraycopy");
2782       StubRoutines::_oop_arraycopy           = generate_conjoint_long_oop_copy(false, true, entry,
2783                                                                                &entry_oop_arraycopy, "oop_arraycopy");
2784       StubRoutines::_oop_disjoint_arraycopy_uninit  = generate_disjoint_long_oop_copy(false, true, &entry,
2785                                                                                       "oop_disjoint_arraycopy_uninit",
2786                                                                                       /*dest_uninitialized*/true);
2787       StubRoutines::_oop_arraycopy_uninit           = generate_conjoint_long_oop_copy(false, true, entry,
2788                                                                                       NULL, "oop_arraycopy_uninit",
2789                                                                                       /*dest_uninitialized*/true);
2790     }
2791 
2792     StubRoutines::_checkcast_arraycopy        = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2793     StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2794                                                                         /*dest_uninitialized*/true);
2795 
2796     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy",
2797                                                               entry_jbyte_arraycopy,
2798                                                               entry_jshort_arraycopy,
2799                                                               entry_jint_arraycopy,
2800                                                               entry_jlong_arraycopy);
2801     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy",
2802                                                                entry_jbyte_arraycopy,
2803                                                                entry_jshort_arraycopy,
2804                                                                entry_jint_arraycopy,
2805                                                                entry_oop_arraycopy,
2806                                                                entry_jlong_arraycopy,
2807                                                                entry_checkcast_arraycopy);
2808 
2809     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2810     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2811     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2812     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2813     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2814     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2815 
2816     // We don't generate specialized code for HeapWord-aligned source
2817     // arrays, so just use the code we've already generated
2818     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = StubRoutines::_jbyte_disjoint_arraycopy;
2819     StubRoutines::_arrayof_jbyte_arraycopy           = StubRoutines::_jbyte_arraycopy;
2820 
2821     StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2822     StubRoutines::_arrayof_jshort_arraycopy          = StubRoutines::_jshort_arraycopy;
2823 
2824     StubRoutines::_arrayof_jint_disjoint_arraycopy   = StubRoutines::_jint_disjoint_arraycopy;
2825     StubRoutines::_arrayof_jint_arraycopy            = StubRoutines::_jint_arraycopy;
2826 
2827     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = StubRoutines::_jlong_disjoint_arraycopy;
2828     StubRoutines::_arrayof_jlong_arraycopy           = StubRoutines::_jlong_arraycopy;
2829 
2830     StubRoutines::_arrayof_oop_disjoint_arraycopy    = StubRoutines::_oop_disjoint_arraycopy;
2831     StubRoutines::_arrayof_oop_arraycopy             = StubRoutines::_oop_arraycopy;
2832 
2833     StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit    = StubRoutines::_oop_disjoint_arraycopy_uninit;
2834     StubRoutines::_arrayof_oop_arraycopy_uninit             = StubRoutines::_oop_arraycopy_uninit;
2835   }
2836 
2837   void generate_math_stubs() {
2838     {
2839       StubCodeMark mark(this, "StubRoutines", "log");
2840       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2841 
2842       __ subq(rsp, 8);
2843       __ movdbl(Address(rsp, 0), xmm0);
2844       __ fld_d(Address(rsp, 0));
2845       __ flog();
2846       __ fstp_d(Address(rsp, 0));
2847       __ movdbl(xmm0, Address(rsp, 0));
2848       __ addq(rsp, 8);
2849       __ ret(0);
2850     }
2851     {
2852       StubCodeMark mark(this, "StubRoutines", "log10");
2853       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2854 
2855       __ subq(rsp, 8);
2856       __ movdbl(Address(rsp, 0), xmm0);
2857       __ fld_d(Address(rsp, 0));
2858       __ flog10();
2859       __ fstp_d(Address(rsp, 0));
2860       __ movdbl(xmm0, Address(rsp, 0));
2861       __ addq(rsp, 8);
2862       __ ret(0);
2863     }
2864     {
2865       StubCodeMark mark(this, "StubRoutines", "sin");
2866       StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2867 
2868       __ subq(rsp, 8);
2869       __ movdbl(Address(rsp, 0), xmm0);
2870       __ fld_d(Address(rsp, 0));
2871       __ trigfunc('s');
2872       __ fstp_d(Address(rsp, 0));
2873       __ movdbl(xmm0, Address(rsp, 0));
2874       __ addq(rsp, 8);
2875       __ ret(0);
2876     }
2877     {
2878       StubCodeMark mark(this, "StubRoutines", "cos");
2879       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2880 
2881       __ subq(rsp, 8);
2882       __ movdbl(Address(rsp, 0), xmm0);
2883       __ fld_d(Address(rsp, 0));
2884       __ trigfunc('c');
2885       __ fstp_d(Address(rsp, 0));
2886       __ movdbl(xmm0, Address(rsp, 0));
2887       __ addq(rsp, 8);
2888       __ ret(0);
2889     }
2890     {
2891       StubCodeMark mark(this, "StubRoutines", "tan");
2892       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2893 
2894       __ subq(rsp, 8);
2895       __ movdbl(Address(rsp, 0), xmm0);
2896       __ fld_d(Address(rsp, 0));
2897       __ trigfunc('t');
2898       __ fstp_d(Address(rsp, 0));
2899       __ movdbl(xmm0, Address(rsp, 0));
2900       __ addq(rsp, 8);
2901       __ ret(0);
2902     }
2903     {
2904       StubCodeMark mark(this, "StubRoutines", "exp");
2905       StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2906 
2907       __ subq(rsp, 8);
2908       __ movdbl(Address(rsp, 0), xmm0);
2909       __ fld_d(Address(rsp, 0));
2910       __ exp_with_fallback(0);
2911       __ fstp_d(Address(rsp, 0));
2912       __ movdbl(xmm0, Address(rsp, 0));
2913       __ addq(rsp, 8);
2914       __ ret(0);
2915     }
2916     {
2917       StubCodeMark mark(this, "StubRoutines", "pow");
2918       StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2919 
2920       __ subq(rsp, 8);
2921       __ movdbl(Address(rsp, 0), xmm1);
2922       __ fld_d(Address(rsp, 0));
2923       __ movdbl(Address(rsp, 0), xmm0);
2924       __ fld_d(Address(rsp, 0));
2925       __ pow_with_fallback(0);
2926       __ fstp_d(Address(rsp, 0));
2927       __ movdbl(xmm0, Address(rsp, 0));
2928       __ addq(rsp, 8);
2929       __ ret(0);
2930     }
2931   }
2932 
2933   // AES intrinsic stubs
2934   enum {AESBlockSize = 16};
2935 
2936   address generate_key_shuffle_mask() {
2937     __ align(16);
2938     StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2939     address start = __ pc();
2940     __ emit_data64( 0x0405060700010203, relocInfo::none );
2941     __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
2942     return start;
2943   }
2944 
2945   // Utility routine for loading a 128-bit key word in little endian format
2946   // can optionally specify that the shuffle mask is already in an xmmregister
2947   void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2948     __ movdqu(xmmdst, Address(key, offset));
2949     if (xmm_shuf_mask != NULL) {
2950       __ pshufb(xmmdst, xmm_shuf_mask);
2951     } else {
2952       __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2953     }
2954   }
2955 
2956   // Arguments:
2957   //
2958   // Inputs:
2959   //   c_rarg0   - source byte array address
2960   //   c_rarg1   - destination byte array address
2961   //   c_rarg2   - K (key) in little endian int array
2962   //
2963   address generate_aescrypt_encryptBlock() {
2964     assert(UseAES, "need AES instructions and misaligned SSE support");
2965     __ align(CodeEntryAlignment);
2966     StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2967     Label L_doLast;
2968     address start = __ pc();
2969 
2970     const Register from        = c_rarg0;  // source array address
2971     const Register to          = c_rarg1;  // destination array address
2972     const Register key         = c_rarg2;  // key array address
2973     const Register keylen      = rax;
2974 
2975     const XMMRegister xmm_result = xmm0;
2976     const XMMRegister xmm_key_shuf_mask = xmm1;
2977     // On win64 xmm6-xmm15 must be preserved so don't use them.
2978     const XMMRegister xmm_temp1  = xmm2;
2979     const XMMRegister xmm_temp2  = xmm3;
2980     const XMMRegister xmm_temp3  = xmm4;
2981     const XMMRegister xmm_temp4  = xmm5;
2982 
2983     __ enter(); // required for proper stackwalking of RuntimeStub frame
2984 
2985     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2986     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2987 
2988     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2989     __ movdqu(xmm_result, Address(from, 0));  // get 16 bytes of input
2990 
2991     // For encryption, the java expanded key ordering is just what we need
2992     // we don't know if the key is aligned, hence not using load-execute form
2993 
2994     load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2995     __ pxor(xmm_result, xmm_temp1);
2996     
2997     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2998     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2999     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3000     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3001 
3002     __ aesenc(xmm_result, xmm_temp1);
3003     __ aesenc(xmm_result, xmm_temp2);
3004     __ aesenc(xmm_result, xmm_temp3);
3005     __ aesenc(xmm_result, xmm_temp4);
3006 
3007     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3008     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3009     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3010     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3011 
3012     __ aesenc(xmm_result, xmm_temp1);
3013     __ aesenc(xmm_result, xmm_temp2);
3014     __ aesenc(xmm_result, xmm_temp3);
3015     __ aesenc(xmm_result, xmm_temp4);
3016 
3017     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3018     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3019     
3020     __ cmpl(keylen, 44);
3021     __ jccb(Assembler::equal, L_doLast);
3022 
3023     __ aesenc(xmm_result, xmm_temp1);
3024     __ aesenc(xmm_result, xmm_temp2);
3025 
3026     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3027     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3028 
3029     __ cmpl(keylen, 52);
3030     __ jccb(Assembler::equal, L_doLast);
3031 
3032     __ aesenc(xmm_result, xmm_temp1);
3033     __ aesenc(xmm_result, xmm_temp2);
3034     
3035     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3036     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3037 
3038     __ BIND(L_doLast);
3039     __ aesenc(xmm_result, xmm_temp1);
3040     __ aesenclast(xmm_result, xmm_temp2);
3041     __ movdqu(Address(to, 0), xmm_result);        // store the result
3042     __ xorptr(rax, rax); // return 0
3043     __ leave(); // required for proper stackwalking of RuntimeStub frame
3044     __ ret(0);
3045 
3046     return start;
3047   }
3048 
3049 
3050   // Arguments:
3051   //
3052   // Inputs:
3053   //   c_rarg0   - source byte array address
3054   //   c_rarg1   - destination byte array address
3055   //   c_rarg2   - K (key) in little endian int array
3056   //
3057   address generate_aescrypt_decryptBlock() {
3058     assert(UseAES, "need AES instructions and misaligned SSE support");
3059     __ align(CodeEntryAlignment);
3060     StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3061     Label L_doLast;
3062     address start = __ pc();
3063 
3064     const Register from        = c_rarg0;  // source array address
3065     const Register to          = c_rarg1;  // destination array address
3066     const Register key         = c_rarg2;  // key array address
3067     const Register keylen      = rax;
3068 
3069     const XMMRegister xmm_result = xmm0;
3070     const XMMRegister xmm_key_shuf_mask = xmm1;
3071     // On win64 xmm6-xmm15 must be preserved so don't use them.
3072     const XMMRegister xmm_temp1  = xmm2;
3073     const XMMRegister xmm_temp2  = xmm3;
3074     const XMMRegister xmm_temp3  = xmm4;
3075     const XMMRegister xmm_temp4  = xmm5;
3076 
3077     __ enter(); // required for proper stackwalking of RuntimeStub frame
3078 
3079     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3080     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3081 
3082     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3083     __ movdqu(xmm_result, Address(from, 0));
3084 
3085     // for decryption java expanded key ordering is rotated one position from what we want
3086     // so we start from 0x10 here and hit 0x00 last
3087     // we don't know if the key is aligned, hence not using load-execute form
3088     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3089     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3090     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3091     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3092 
3093     __ pxor  (xmm_result, xmm_temp1);
3094     __ aesdec(xmm_result, xmm_temp2);
3095     __ aesdec(xmm_result, xmm_temp3);
3096     __ aesdec(xmm_result, xmm_temp4);
3097 
3098     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3099     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3100     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3101     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3102 
3103     __ aesdec(xmm_result, xmm_temp1);
3104     __ aesdec(xmm_result, xmm_temp2);
3105     __ aesdec(xmm_result, xmm_temp3);
3106     __ aesdec(xmm_result, xmm_temp4);
3107 
3108     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3109     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3110     load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3111     
3112     __ cmpl(keylen, 44);
3113     __ jccb(Assembler::equal, L_doLast);
3114 
3115     __ aesdec(xmm_result, xmm_temp1);
3116     __ aesdec(xmm_result, xmm_temp2);
3117 
3118     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3119     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3120 
3121     __ cmpl(keylen, 52);
3122     __ jccb(Assembler::equal, L_doLast);
3123 
3124     __ aesdec(xmm_result, xmm_temp1);
3125     __ aesdec(xmm_result, xmm_temp2);
3126     
3127     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3128     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3129 
3130     __ BIND(L_doLast);
3131     __ aesdec(xmm_result, xmm_temp1);
3132     __ aesdec(xmm_result, xmm_temp2);
3133 
3134     // for decryption the aesdeclast operation is always on key+0x00
3135     __ aesdeclast(xmm_result, xmm_temp3);
3136     __ movdqu(Address(to, 0), xmm_result);  // store the result
3137     __ xorptr(rax, rax); // return 0
3138     __ leave(); // required for proper stackwalking of RuntimeStub frame
3139     __ ret(0);
3140 
3141     return start;
3142   }
3143 
3144 
3145   // Arguments:
3146   //
3147   // Inputs:
3148   //   c_rarg0   - source byte array address
3149   //   c_rarg1   - destination byte array address
3150   //   c_rarg2   - K (key) in little endian int array
3151   //   c_rarg3   - r vector byte array address
3152   //   c_rarg4   - input length
3153   //
3154   address generate_cipherBlockChaining_encryptAESCrypt() {
3155     assert(UseAES, "need AES instructions and misaligned SSE support");
3156     __ align(CodeEntryAlignment);
3157     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3158     address start = __ pc();
3159 
3160     Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3161     const Register from        = c_rarg0;  // source array address
3162     const Register to          = c_rarg1;  // destination array address
3163     const Register key         = c_rarg2;  // key array address
3164     const Register rvec        = c_rarg3;  // r byte array initialized from initvector array address
3165                                            // and left with the results of the last encryption block
3166 #ifndef _WIN64
3167     const Register len_reg     = c_rarg4;  // src len (must be multiple of blocksize 16)
3168 #else
3169     const Address  len_mem(rsp, 6 * wordSize);  // length is on stack on Win64
3170     const Register len_reg     = r10;      // pick the first volatile windows register
3171 #endif
3172     const Register pos         = rax;
3173 
3174     // xmm register assignments for the loops below
3175     const XMMRegister xmm_result = xmm0;
3176     const XMMRegister xmm_temp   = xmm1;
3177     // keys 0-10 preloaded into xmm2-xmm12
3178     const int XMM_REG_NUM_KEY_FIRST = 2;
3179     const int XMM_REG_NUM_KEY_LAST  = 15;
3180     const XMMRegister xmm_key0   = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3181     const XMMRegister xmm_key10  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3182     const XMMRegister xmm_key11  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3183     const XMMRegister xmm_key12  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3184     const XMMRegister xmm_key13  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3185 
3186     __ enter(); // required for proper stackwalking of RuntimeStub frame
3187 
3188 #ifdef _WIN64
3189     // on win64, fill len_reg from stack position
3190     __ movl(len_reg, len_mem);
3191     // save the xmm registers which must be preserved 6-15
3192     __ subptr(rsp, -rsp_after_call_off * wordSize);
3193     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3194       __ movdqu(xmm_save(i), as_XMMRegister(i));
3195     }
3196 #endif
3197 
3198     const XMMRegister xmm_key_shuf_mask = xmm_temp;  // used temporarily to swap key bytes up front
3199     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3200     // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3201     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3202       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3203       offset += 0x10;
3204     }
3205     __ movdqu(xmm_result, Address(rvec, 0x00));   // initialize xmm_result with r vec
3206 
3207     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3208     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3209     __ cmpl(rax, 44);
3210     __ jcc(Assembler::notEqual, L_key_192_256);
3211 
3212     // 128 bit code follows here
3213     __ movptr(pos, 0);
3214     __ align(OptoLoopAlignment);
3215 
3216     __ BIND(L_loopTop_128);
3217     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3218     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3219     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3220     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3221       __ aesenc(xmm_result, as_XMMRegister(rnum));
3222     }
3223     __ aesenclast(xmm_result, xmm_key10);
3224     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3225     // no need to store r to memory until we exit
3226     __ addptr(pos, AESBlockSize);
3227     __ subptr(len_reg, AESBlockSize);
3228     __ jcc(Assembler::notEqual, L_loopTop_128);
3229 
3230     __ BIND(L_exit);
3231     __ movdqu(Address(rvec, 0), xmm_result);     // final value of r stored in rvec of CipherBlockChaining object
3232 
3233 #ifdef _WIN64
3234     // restore xmm regs belonging to calling function
3235     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3236       __ movdqu(as_XMMRegister(i), xmm_save(i));
3237     }
3238 #endif
3239     __ movl(rax, 0); // return 0 (why?)
3240     __ leave(); // required for proper stackwalking of RuntimeStub frame
3241     __ ret(0);
3242 
3243     __ BIND(L_key_192_256);
3244     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3245     load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3246     load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3247     __ cmpl(rax, 52);
3248     __ jcc(Assembler::notEqual, L_key_256);
3249 
3250     // 192-bit code follows here (could be changed to use more xmm registers)
3251     __ movptr(pos, 0);
3252     __ align(OptoLoopAlignment);
3253     
3254     __ BIND(L_loopTop_192);
3255     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3256     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3257     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3258     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3259       __ aesenc(xmm_result, as_XMMRegister(rnum));
3260     }
3261     __ aesenclast(xmm_result, xmm_key12);
3262     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3263     // no need to store r to memory until we exit
3264     __ addptr(pos, AESBlockSize);
3265     __ subptr(len_reg, AESBlockSize);
3266     __ jcc(Assembler::notEqual, L_loopTop_192);
3267     __ jmp(L_exit);
3268 
3269     __ BIND(L_key_256);
3270     // 256-bit code follows here (could be changed to use more xmm registers)
3271     load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3272     __ movptr(pos, 0);
3273     __ align(OptoLoopAlignment);
3274     
3275     __ BIND(L_loopTop_256);
3276     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3277     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3278     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3279     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3280       __ aesenc(xmm_result, as_XMMRegister(rnum));
3281     }
3282     load_key(xmm_temp, key, 0xe0);
3283     __ aesenclast(xmm_result, xmm_temp);
3284     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3285     // no need to store r to memory until we exit
3286     __ addptr(pos, AESBlockSize);
3287     __ subptr(len_reg, AESBlockSize);
3288     __ jcc(Assembler::notEqual, L_loopTop_256);
3289     __ jmp(L_exit);
3290 
3291     return start;
3292   }
3293 
3294 
3295 
3296   // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3297   // to hide instruction latency
3298   //
3299   // Arguments:
3300   //
3301   // Inputs:
3302   //   c_rarg0   - source byte array address
3303   //   c_rarg1   - destination byte array address
3304   //   c_rarg2   - K (key) in little endian int array
3305   //   c_rarg3   - r vector byte array address
3306   //   c_rarg4   - input length
3307   //
3308 
3309   address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3310     assert(UseAES, "need AES instructions and misaligned SSE support");
3311     __ align(CodeEntryAlignment);
3312     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3313     address start = __ pc();
3314 
3315     Label L_exit, L_key_192_256, L_key_256;
3316     Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128;
3317     Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
3318     const Register from        = c_rarg0;  // source array address
3319     const Register to          = c_rarg1;  // destination array address
3320     const Register key         = c_rarg2;  // key array address
3321     const Register rvec        = c_rarg3;  // r byte array initialized from initvector array address
3322                                            // and left with the results of the last encryption block
3323 #ifndef _WIN64
3324     const Register len_reg     = c_rarg4;  // src len (must be multiple of blocksize 16)
3325 #else
3326     const Address  len_mem(rsp, 6 * wordSize);  // length is on stack on Win64
3327     const Register len_reg     = r10;      // pick the first volatile windows register
3328 #endif
3329     const Register pos         = rax;
3330 
3331     // keys 0-10 preloaded into xmm2-xmm12
3332     const int XMM_REG_NUM_KEY_FIRST = 5;
3333     const int XMM_REG_NUM_KEY_LAST  = 15;
3334     const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3335     const XMMRegister xmm_key_last  = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3336 
3337     __ enter(); // required for proper stackwalking of RuntimeStub frame
3338 
3339 #ifdef _WIN64
3340     // on win64, fill len_reg from stack position
3341     __ movl(len_reg, len_mem);
3342     // save the xmm registers which must be preserved 6-15
3343     __ subptr(rsp, -rsp_after_call_off * wordSize);
3344     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3345       __ movdqu(xmm_save(i), as_XMMRegister(i));
3346     }
3347 #endif
3348     // the java expanded key ordering is rotated one position from what we want
3349     // so we start from 0x10 here and hit 0x00 last
3350     const XMMRegister xmm_key_shuf_mask = xmm1;  // used temporarily to swap key bytes up front
3351     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3352     // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3353     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3354       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3355       offset += 0x10;
3356     }
3357     load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3358 
3359     const XMMRegister xmm_prev_block_cipher = xmm1;  // holds cipher of previous block
3360 
3361     // registers holding the four results in the parallelized loop
3362     const XMMRegister xmm_result0 = xmm0;
3363     const XMMRegister xmm_result1 = xmm2;
3364     const XMMRegister xmm_result2 = xmm3;
3365     const XMMRegister xmm_result3 = xmm4;
3366 
3367     __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00));   // initialize with initial rvec
3368 
3369     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3370     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3371     __ cmpl(rax, 44);
3372     __ jcc(Assembler::notEqual, L_key_192_256);
3373 
3374 
3375     // 128-bit code follows here, parallelized
3376     __ movptr(pos, 0);
3377     __ align(OptoLoopAlignment);
3378     __ BIND(L_multiBlock_loopTop_128);
3379     __ cmpptr(len_reg, 4*AESBlockSize);           // see if at least 4 blocks left
3380     __ jcc(Assembler::less, L_singleBlock_loopTop_128);
3381 
3382     __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize));   // get next 4 blocks into xmmresult registers
3383     __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize));
3384     __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize));
3385     __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize));
3386 
3387 #define DoFour(opc, src_reg)                    \
3388     __ opc(xmm_result0, src_reg);               \
3389     __ opc(xmm_result1, src_reg);               \
3390     __ opc(xmm_result2, src_reg);               \
3391     __ opc(xmm_result3, src_reg);
3392 
3393     DoFour(pxor, xmm_key_first);
3394     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3395       DoFour(aesdec, as_XMMRegister(rnum));
3396     }
3397     DoFour(aesdeclast, xmm_key_last);
3398     // for each result, xor with the r vector of previous cipher block
3399     __ pxor(xmm_result0, xmm_prev_block_cipher);
3400     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize));
3401     __ pxor(xmm_result1, xmm_prev_block_cipher);
3402     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize));
3403     __ pxor(xmm_result2, xmm_prev_block_cipher);
3404     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize));
3405     __ pxor(xmm_result3, xmm_prev_block_cipher);
3406     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize));   // this will carry over to next set of blocks
3407 
3408     __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0);     // store 4 results into the next 64 bytes of output
3409     __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1);
3410     __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2);
3411     __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3);
3412 
3413     __ addptr(pos, 4*AESBlockSize);
3414     __ subptr(len_reg, 4*AESBlockSize);
3415     __ jmp(L_multiBlock_loopTop_128);
3416 
3417     // registers used in the non-parallelized loops
3418     // xmm register assignments for the loops below
3419     const XMMRegister xmm_result = xmm0;
3420     const XMMRegister xmm_prev_block_cipher_save = xmm2;
3421     const XMMRegister xmm_key11 = xmm3;
3422     const XMMRegister xmm_key12 = xmm4;
3423     const XMMRegister xmm_temp  = xmm4;
3424 
3425     __ align(OptoLoopAlignment);
3426     __ BIND(L_singleBlock_loopTop_128);
3427     __ cmpptr(len_reg, 0);           // any blocks left??
3428     __ jcc(Assembler::equal, L_exit);
3429     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
3430     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3431     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3432     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3433       __ aesdec(xmm_result, as_XMMRegister(rnum));
3434     }
3435     __ aesdeclast(xmm_result, xmm_key_last);
3436     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3437     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3438     // no need to store r to memory until we exit
3439     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);              // set up next r vector with cipher input from this block
3440 
3441     __ addptr(pos, AESBlockSize);
3442     __ subptr(len_reg, AESBlockSize);
3443     __ jmp(L_singleBlock_loopTop_128);
3444 
3445 
3446     __ BIND(L_exit);
3447     __ movdqu(Address(rvec, 0), xmm_prev_block_cipher);     // final value of r stored in rvec of CipherBlockChaining object
3448 #ifdef _WIN64
3449     // restore regs belonging to calling function
3450     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3451       __ movdqu(as_XMMRegister(i), xmm_save(i));
3452     }
3453 #endif
3454     __ movl(rax, 0); // return 0 (why?)
3455     __ leave(); // required for proper stackwalking of RuntimeStub frame
3456     __ ret(0);
3457 
3458 
3459     __ BIND(L_key_192_256);
3460     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3461     load_key(xmm_key11, key, 0xb0);
3462     __ cmpl(rax, 52);
3463     __ jcc(Assembler::notEqual, L_key_256);
3464 
3465     // 192-bit code follows here (could be optimized to use parallelism)
3466     load_key(xmm_key12, key, 0xc0);     // 192-bit key goes up to c0
3467     __ movptr(pos, 0);
3468     __ align(OptoLoopAlignment);
3469     
3470     __ BIND(L_singleBlock_loopTop_192);
3471     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
3472     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3473     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3474     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3475       __ aesdec(xmm_result, as_XMMRegister(rnum));
3476     }
3477     __ aesdec(xmm_result, xmm_key11);
3478     __ aesdec(xmm_result, xmm_key12);
3479     __ aesdeclast(xmm_result, xmm_key_last);                    // xmm15 always came from key+0
3480     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3481     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);  // store into the next 16 bytes of output
3482     // no need to store r to memory until we exit
3483     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);  // set up next r vector with cipher input from this block
3484     __ addptr(pos, AESBlockSize);
3485     __ subptr(len_reg, AESBlockSize);
3486     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
3487     __ jmp(L_exit);
3488 
3489     __ BIND(L_key_256);
3490     // 256-bit code follows here (could be optimized to use parallelism)
3491     __ movptr(pos, 0);
3492     __ align(OptoLoopAlignment);
3493     
3494     __ BIND(L_singleBlock_loopTop_256);
3495     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3496     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3497     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3498     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3499       __ aesdec(xmm_result, as_XMMRegister(rnum));
3500     }
3501     __ aesdec(xmm_result, xmm_key11);
3502     load_key(xmm_temp, key, 0xc0);
3503     __ aesdec(xmm_result, xmm_temp);
3504     load_key(xmm_temp, key, 0xd0);
3505     __ aesdec(xmm_result, xmm_temp);
3506     load_key(xmm_temp, key, 0xe0);     // 256-bit key goes up to e0
3507     __ aesdec(xmm_result, xmm_temp);
3508     __ aesdeclast(xmm_result, xmm_key_last);          // xmm15 came from key+0
3509     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3510     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);  // store into the next 16 bytes of output
3511     // no need to store r to memory until we exit
3512     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);  // set up next r vector with cipher input from this block
3513     __ addptr(pos, AESBlockSize);
3514     __ subptr(len_reg, AESBlockSize);
3515     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
3516     __ jmp(L_exit);
3517 
3518     return start;
3519   }
3520 
3521 
3522 
3523 #undef __
3524 #define __ masm->
3525 
3526   // Continuation point for throwing of implicit exceptions that are
3527   // not handled in the current activation. Fabricates an exception
3528   // oop and initiates normal exception dispatching in this
3529   // frame. Since we need to preserve callee-saved values (currently
3530   // only for C2, but done for C1 as well) we need a callee-saved oop
3531   // map and therefore have to make these stubs into RuntimeStubs
3532   // rather than BufferBlobs.  If the compiler needs all registers to
3533   // be preserved between the fault point and the exception handler
3534   // then it must assume responsibility for that in
3535   // AbstractCompiler::continuation_for_implicit_null_exception or
3536   // continuation_for_implicit_division_by_zero_exception. All other
3537   // implicit exceptions (e.g., NullPointerException or
3538   // AbstractMethodError on entry) are either at call sites or
3539   // otherwise assume that stack unwinding will be initiated, so
3540   // caller saved registers were assumed volatile in the compiler.
3541   address generate_throw_exception(const char* name,
3542                                    address runtime_entry,
3543                                    Register arg1 = noreg,
3544                                    Register arg2 = noreg) {
3545     // Information about frame layout at time of blocking runtime call.
3546     // Note that we only have to preserve callee-saved registers since
3547     // the compilers are responsible for supplying a continuation point
3548     // if they expect all registers to be preserved.
3549     enum layout {
3550       rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3551       rbp_off2,
3552       return_off,
3553       return_off2,
3554       framesize // inclusive of return address
3555     };
3556 
3557     int insts_size = 512;
3558     int locs_size  = 64;
3559 
3560     CodeBuffer code(name, insts_size, locs_size);
3561     OopMapSet* oop_maps  = new OopMapSet();
3562     MacroAssembler* masm = new MacroAssembler(&code);
3563 
3564     address start = __ pc();
3565 
3566     // This is an inlined and slightly modified version of call_VM
3567     // which has the ability to fetch the return PC out of
3568     // thread-local storage and also sets up last_Java_sp slightly
3569     // differently than the real call_VM
3570 
3571     __ enter(); // required for proper stackwalking of RuntimeStub frame
3572 
3573     assert(is_even(framesize/2), "sp not 16-byte aligned");
3574 
3575     // return address and rbp are already in place
3576     __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3577 
3578     int frame_complete = __ pc() - start;
3579 
3580     // Set up last_Java_sp and last_Java_fp
3581     address the_pc = __ pc();
3582     __ set_last_Java_frame(rsp, rbp, the_pc);
3583     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
3584 
3585     // Call runtime
3586     if (arg1 != noreg) {
3587       assert(arg2 != c_rarg1, "clobbered");
3588       __ movptr(c_rarg1, arg1);
3589     }
3590     if (arg2 != noreg) {
3591       __ movptr(c_rarg2, arg2);
3592     }
3593     __ movptr(c_rarg0, r15_thread);
3594     BLOCK_COMMENT("call runtime_entry");
3595     __ call(RuntimeAddress(runtime_entry));
3596 
3597     // Generate oop map
3598     OopMap* map = new OopMap(framesize, 0);
3599 
3600     oop_maps->add_gc_map(the_pc - start, map);
3601 
3602     __ reset_last_Java_frame(true, true);
3603 
3604     __ leave(); // required for proper stackwalking of RuntimeStub frame
3605 
3606     // check for pending exceptions
3607 #ifdef ASSERT
3608     Label L;
3609     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3610             (int32_t) NULL_WORD);
3611     __ jcc(Assembler::notEqual, L);
3612     __ should_not_reach_here();
3613     __ bind(L);
3614 #endif // ASSERT
3615     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3616 
3617 
3618     // codeBlob framesize is in words (not VMRegImpl::slot_size)
3619     RuntimeStub* stub =
3620       RuntimeStub::new_runtime_stub(name,
3621                                     &code,
3622                                     frame_complete,
3623                                     (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3624                                     oop_maps, false);
3625     return stub->entry_point();
3626   }
3627 
3628   // Initialization
3629   void generate_initial() {
3630     // Generates all stubs and initializes the entry points
3631 
3632     // This platform-specific stub is needed by generate_call_stub()
3633     StubRoutines::x86::_mxcsr_std        = generate_fp_mask("mxcsr_std",        0x0000000000001F80);
3634 
3635     // entry points that exist in all platforms Note: This is code
3636     // that could be shared among different platforms - however the
3637     // benefit seems to be smaller than the disadvantage of having a
3638     // much more complicated generator structure. See also comment in
3639     // stubRoutines.hpp.
3640 
3641     StubRoutines::_forward_exception_entry = generate_forward_exception();
3642 
3643     StubRoutines::_call_stub_entry =
3644       generate_call_stub(StubRoutines::_call_stub_return_address);
3645 
3646     // is referenced by megamorphic call
3647     StubRoutines::_catch_exception_entry = generate_catch_exception();
3648 
3649     // atomic calls
3650     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
3651     StubRoutines::_atomic_xchg_ptr_entry     = generate_atomic_xchg_ptr();
3652     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
3653     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3654     StubRoutines::_atomic_add_entry          = generate_atomic_add();
3655     StubRoutines::_atomic_add_ptr_entry      = generate_atomic_add_ptr();
3656     StubRoutines::_fence_entry               = generate_orderaccess_fence();
3657 
3658     StubRoutines::_handler_for_unsafe_access_entry =
3659       generate_handler_for_unsafe_access();
3660 
3661     // platform dependent
3662     StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3663     StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
3664 
3665     StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
3666 
3667     // Build this early so it's available for the interpreter.
3668     StubRoutines::_throw_StackOverflowError_entry =
3669       generate_throw_exception("StackOverflowError throw_exception",
3670                                CAST_FROM_FN_PTR(address,
3671                                                 SharedRuntime::
3672                                                 throw_StackOverflowError));
3673   }
3674 
3675   void generate_all() {
3676     // Generates all stubs and initializes the entry points
3677 
3678     // These entry points require SharedInfo::stack0 to be set up in
3679     // non-core builds and need to be relocatable, so they each
3680     // fabricate a RuntimeStub internally.
3681     StubRoutines::_throw_AbstractMethodError_entry =
3682       generate_throw_exception("AbstractMethodError throw_exception",
3683                                CAST_FROM_FN_PTR(address,
3684                                                 SharedRuntime::
3685                                                 throw_AbstractMethodError));
3686 
3687     StubRoutines::_throw_IncompatibleClassChangeError_entry =
3688       generate_throw_exception("IncompatibleClassChangeError throw_exception",
3689                                CAST_FROM_FN_PTR(address,
3690                                                 SharedRuntime::
3691                                                 throw_IncompatibleClassChangeError));
3692 
3693     StubRoutines::_throw_NullPointerException_at_call_entry =
3694       generate_throw_exception("NullPointerException at call throw_exception",
3695                                CAST_FROM_FN_PTR(address,
3696                                                 SharedRuntime::
3697                                                 throw_NullPointerException_at_call));
3698 
3699     // entry points that are platform specific
3700     StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3701     StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3702     StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3703     StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3704 
3705     StubRoutines::x86::_float_sign_mask  = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
3706     StubRoutines::x86::_float_sign_flip  = generate_fp_mask("float_sign_flip",  0x8000000080000000);
3707     StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3708     StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3709 
3710     // support for verify_oop (must happen after universe_init)
3711     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3712 
3713     // arraycopy stubs used by compilers
3714     generate_arraycopy_stubs();
3715 
3716     generate_math_stubs();
3717 
3718     // don't bother generating these AES intrinsic stubs unless global flag is set
3719     if (UseAESIntrinsics) {
3720       StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask();  // needed by the others
3721 
3722       StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3723       StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3724       StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3725       StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3726     }
3727   }
3728 
3729  public:
3730   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3731     if (all) {
3732       generate_all();
3733     } else {
3734       generate_initial();
3735     }
3736   }
3737 }; // end class declaration
3738 
3739 void StubGenerator_generate(CodeBuffer* code, bool all) {
3740   StubGenerator g(code, all);
3741 }