1 /*
   2  * Copyright (c) 2003, 2015, 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) ((void)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   // -60 [ argument word 1      ]
 141   // -59 [ saved xmm31          ] <--- rsp after_call
 142   //     [ saved xmm16-xmm30    ] (EVEX enabled, else the space is blank)
 143   // -27 [ saved xmm15          ]
 144   //     [ saved xmm7-xmm14     ]
 145   //  -9 [ saved xmm6           ] (each xmm register takes 2 slots)
 146   //  -7 [ saved r15            ]
 147   //  -6 [ saved r14            ]
 148   //  -5 [ saved r13            ]
 149   //  -4 [ saved r12            ]
 150   //  -3 [ saved rdi            ]
 151   //  -2 [ saved rsi            ]
 152   //  -1 [ saved rbx            ]
 153   //   0 [ saved rbp            ] <--- rbp
 154   //   1 [ return address       ]
 155   //   2 [ call wrapper         ]
 156   //   3 [ result               ]
 157   //   4 [ result type          ]
 158   //   5 [ method               ]
 159   //   6 [ entry point          ]
 160   //   7 [ parameters           ]
 161   //   8 [ parameter size       ]
 162   //   9 [ thread               ]
 163   //
 164   //    Windows reserves the callers stack space for arguments 1-4.
 165   //    We spill c_rarg0-c_rarg3 to this space.
 166 
 167   // Call stub stack layout word offsets from rbp
 168   enum call_stub_layout {
 169 #ifdef _WIN64
 170     xmm_save_first     = 6,  // save from xmm6
 171     xmm_save_last      = 31, // to xmm31
 172     xmm_save_base      = -9,
 173     rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
 174     r15_off            = -7,
 175     r14_off            = -6,
 176     r13_off            = -5,
 177     r12_off            = -4,
 178     rdi_off            = -3,
 179     rsi_off            = -2,
 180     rbx_off            = -1,
 181     rbp_off            =  0,
 182     retaddr_off        =  1,
 183     call_wrapper_off   =  2,
 184     result_off         =  3,
 185     result_type_off    =  4,
 186     method_off         =  5,
 187     entry_point_off    =  6,
 188     parameters_off     =  7,
 189     parameter_size_off =  8,
 190     thread_off         =  9
 191 #else
 192     rsp_after_call_off = -12,
 193     mxcsr_off          = rsp_after_call_off,
 194     r15_off            = -11,
 195     r14_off            = -10,
 196     r13_off            = -9,
 197     r12_off            = -8,
 198     rbx_off            = -7,
 199     call_wrapper_off   = -6,
 200     result_off         = -5,
 201     result_type_off    = -4,
 202     method_off         = -3,
 203     entry_point_off    = -2,
 204     parameters_off     = -1,
 205     rbp_off            =  0,
 206     retaddr_off        =  1,
 207     parameter_size_off =  2,
 208     thread_off         =  3
 209 #endif
 210   };
 211 
 212 #ifdef _WIN64
 213   Address xmm_save(int reg) {
 214     assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
 215     return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
 216   }
 217 #endif
 218 
 219   address generate_call_stub(address& return_address) {
 220     assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 221            (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 222            "adjust this code");
 223     StubCodeMark mark(this, "StubRoutines", "call_stub");
 224     address start = __ pc();
 225 
 226     // same as in generate_catch_exception()!
 227     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 228 
 229     const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 230     const Address result        (rbp, result_off         * wordSize);
 231     const Address result_type   (rbp, result_type_off    * wordSize);
 232     const Address method        (rbp, method_off         * wordSize);
 233     const Address entry_point   (rbp, entry_point_off    * wordSize);
 234     const Address parameters    (rbp, parameters_off     * wordSize);
 235     const Address parameter_size(rbp, parameter_size_off * wordSize);
 236 
 237     // same as in generate_catch_exception()!
 238     const Address thread        (rbp, thread_off         * wordSize);
 239 
 240     const Address r15_save(rbp, r15_off * wordSize);
 241     const Address r14_save(rbp, r14_off * wordSize);
 242     const Address r13_save(rbp, r13_off * wordSize);
 243     const Address r12_save(rbp, r12_off * wordSize);
 244     const Address rbx_save(rbp, rbx_off * wordSize);
 245 
 246     // stub code
 247     __ enter();
 248     __ subptr(rsp, -rsp_after_call_off * wordSize);
 249 
 250     // save register parameters
 251 #ifndef _WIN64
 252     __ movptr(parameters,   c_rarg5); // parameters
 253     __ movptr(entry_point,  c_rarg4); // entry_point
 254 #endif
 255 
 256     __ movptr(method,       c_rarg3); // method
 257     __ movl(result_type,  c_rarg2);   // result type
 258     __ movptr(result,       c_rarg1); // result
 259     __ movptr(call_wrapper, c_rarg0); // call wrapper
 260 
 261     // save regs belonging to calling function
 262     __ movptr(rbx_save, rbx);
 263     __ movptr(r12_save, r12);
 264     __ movptr(r13_save, r13);
 265     __ movptr(r14_save, r14);
 266     __ movptr(r15_save, r15);
 267     if (UseAVX > 2) {
 268       __ movl(rbx, 0xffff);
 269       __ kmovql(k1, rbx);
 270     }
 271 #ifdef _WIN64
 272     if (UseAVX > 2) {
 273       for (int i = 6; i <= 31; i++) {
 274         __ movdqu(xmm_save(i), as_XMMRegister(i));
 275       }
 276     } else {
 277       for (int i = 6; i <= 15; i++) {
 278         __ movdqu(xmm_save(i), as_XMMRegister(i));
 279       }
 280     }
 281 
 282     const Address rdi_save(rbp, rdi_off * wordSize);
 283     const Address rsi_save(rbp, rsi_off * wordSize);
 284 
 285     __ movptr(rsi_save, rsi);
 286     __ movptr(rdi_save, rdi);
 287 #else
 288     const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 289     {
 290       Label skip_ldmx;
 291       __ stmxcsr(mxcsr_save);
 292       __ movl(rax, mxcsr_save);
 293       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 294       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 295       __ cmp32(rax, mxcsr_std);
 296       __ jcc(Assembler::equal, skip_ldmx);
 297       __ ldmxcsr(mxcsr_std);
 298       __ bind(skip_ldmx);
 299     }
 300 #endif
 301 
 302     // Load up thread register
 303     __ movptr(r15_thread, thread);
 304     __ reinit_heapbase();
 305 
 306 #ifdef ASSERT
 307     // make sure we have no pending exceptions
 308     {
 309       Label L;
 310       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 311       __ jcc(Assembler::equal, L);
 312       __ stop("StubRoutines::call_stub: entered with pending exception");
 313       __ bind(L);
 314     }
 315 #endif
 316 
 317     // pass parameters if any
 318     BLOCK_COMMENT("pass parameters if any");
 319     Label parameters_done;
 320     __ movl(c_rarg3, parameter_size);
 321     __ testl(c_rarg3, c_rarg3);
 322     __ jcc(Assembler::zero, parameters_done);
 323 
 324     Label loop;
 325     __ movptr(c_rarg2, parameters);       // parameter pointer
 326     __ movl(c_rarg1, c_rarg3);            // parameter counter is in c_rarg1
 327     __ BIND(loop);
 328     __ movptr(rax, Address(c_rarg2, 0));// get parameter
 329     __ addptr(c_rarg2, wordSize);       // advance to next parameter
 330     __ decrementl(c_rarg1);             // decrement counter
 331     __ push(rax);                       // pass parameter
 332     __ jcc(Assembler::notZero, loop);
 333 
 334     // call Java function
 335     __ BIND(parameters_done);
 336     __ movptr(rbx, method);             // get Method*
 337     __ movptr(c_rarg1, entry_point);    // get entry_point
 338     __ mov(r13, rsp);                   // set sender sp
 339     BLOCK_COMMENT("call Java function");
 340     __ call(c_rarg1);
 341 
 342     BLOCK_COMMENT("call_stub_return_address:");
 343     return_address = __ pc();
 344 
 345     // store result depending on type (everything that is not
 346     // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 347     __ movptr(c_rarg0, result);
 348     Label is_long, is_float, is_double, exit;
 349     __ movl(c_rarg1, result_type);
 350     __ cmpl(c_rarg1, T_OBJECT);
 351     __ jcc(Assembler::equal, is_long);
 352     __ cmpl(c_rarg1, T_LONG);
 353     __ jcc(Assembler::equal, is_long);
 354     __ cmpl(c_rarg1, T_FLOAT);
 355     __ jcc(Assembler::equal, is_float);
 356     __ cmpl(c_rarg1, T_DOUBLE);
 357     __ jcc(Assembler::equal, is_double);
 358 
 359     // handle T_INT case
 360     __ movl(Address(c_rarg0, 0), rax);
 361 
 362     __ BIND(exit);
 363 
 364     // pop parameters
 365     __ lea(rsp, rsp_after_call);
 366 
 367 #ifdef ASSERT
 368     // verify that threads correspond
 369     {
 370       Label L, S;
 371       __ cmpptr(r15_thread, thread);
 372       __ jcc(Assembler::notEqual, S);
 373       __ get_thread(rbx);
 374       __ cmpptr(r15_thread, rbx);
 375       __ jcc(Assembler::equal, L);
 376       __ bind(S);
 377       __ jcc(Assembler::equal, L);
 378       __ stop("StubRoutines::call_stub: threads must correspond");
 379       __ bind(L);
 380     }
 381 #endif
 382 
 383     // restore regs belonging to calling function
 384 #ifdef _WIN64
 385     int xmm_ub = 15;
 386     if (UseAVX > 2) {
 387       xmm_ub = 31;
 388     }
 389     // emit the restores for xmm regs
 390     for (int i = 6; i <= xmm_ub; i++) {
 391       __ movdqu(as_XMMRegister(i), xmm_save(i));
 392     }
 393 #endif
 394     __ movptr(r15, r15_save);
 395     __ movptr(r14, r14_save);
 396     __ movptr(r13, r13_save);
 397     __ movptr(r12, r12_save);
 398     __ movptr(rbx, rbx_save);
 399 
 400 #ifdef _WIN64
 401     __ movptr(rdi, rdi_save);
 402     __ movptr(rsi, rsi_save);
 403 #else
 404     __ ldmxcsr(mxcsr_save);
 405 #endif
 406 
 407     // restore rsp
 408     __ addptr(rsp, -rsp_after_call_off * wordSize);
 409 
 410     // return
 411     __ pop(rbp);
 412     __ ret(0);
 413 
 414     // handle return types different from T_INT
 415     __ BIND(is_long);
 416     __ movq(Address(c_rarg0, 0), rax);
 417     __ jmp(exit);
 418 
 419     __ BIND(is_float);
 420     __ movflt(Address(c_rarg0, 0), xmm0);
 421     __ jmp(exit);
 422 
 423     __ BIND(is_double);
 424     __ movdbl(Address(c_rarg0, 0), xmm0);
 425     __ jmp(exit);
 426 
 427     return start;
 428   }
 429 
 430   // Return point for a Java call if there's an exception thrown in
 431   // Java code.  The exception is caught and transformed into a
 432   // pending exception stored in JavaThread that can be tested from
 433   // within the VM.
 434   //
 435   // Note: Usually the parameters are removed by the callee. In case
 436   // of an exception crossing an activation frame boundary, that is
 437   // not the case if the callee is compiled code => need to setup the
 438   // rsp.
 439   //
 440   // rax: exception oop
 441 
 442   address generate_catch_exception() {
 443     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 444     address start = __ pc();
 445 
 446     // same as in generate_call_stub():
 447     const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 448     const Address thread        (rbp, thread_off         * wordSize);
 449 
 450 #ifdef ASSERT
 451     // verify that threads correspond
 452     {
 453       Label L, S;
 454       __ cmpptr(r15_thread, thread);
 455       __ jcc(Assembler::notEqual, S);
 456       __ get_thread(rbx);
 457       __ cmpptr(r15_thread, rbx);
 458       __ jcc(Assembler::equal, L);
 459       __ bind(S);
 460       __ stop("StubRoutines::catch_exception: threads must correspond");
 461       __ bind(L);
 462     }
 463 #endif
 464 
 465     // set pending exception
 466     __ verify_oop(rax);
 467 
 468     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
 469     __ lea(rscratch1, ExternalAddress((address)__FILE__));
 470     __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 471     __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 472 
 473     // complete return to VM
 474     assert(StubRoutines::_call_stub_return_address != NULL,
 475            "_call_stub_return_address must have been generated before");
 476     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 477 
 478     return start;
 479   }
 480 
 481   // Continuation point for runtime calls returning with a pending
 482   // exception.  The pending exception check happened in the runtime
 483   // or native call stub.  The pending exception in Thread is
 484   // converted into a Java-level exception.
 485   //
 486   // Contract with Java-level exception handlers:
 487   // rax: exception
 488   // rdx: throwing pc
 489   //
 490   // NOTE: At entry of this stub, exception-pc must be on stack !!
 491 
 492   address generate_forward_exception() {
 493     StubCodeMark mark(this, "StubRoutines", "forward exception");
 494     address start = __ pc();
 495 
 496     // Upon entry, the sp points to the return address returning into
 497     // Java (interpreted or compiled) code; i.e., the return address
 498     // becomes the throwing pc.
 499     //
 500     // Arguments pushed before the runtime call are still on the stack
 501     // but the exception handler will reset the stack pointer ->
 502     // ignore them.  A potential result in registers can be ignored as
 503     // well.
 504 
 505 #ifdef ASSERT
 506     // make sure this code is only executed if there is a pending exception
 507     {
 508       Label L;
 509       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
 510       __ jcc(Assembler::notEqual, L);
 511       __ stop("StubRoutines::forward exception: no pending exception (1)");
 512       __ bind(L);
 513     }
 514 #endif
 515 
 516     // compute exception handler into rbx
 517     __ movptr(c_rarg0, Address(rsp, 0));
 518     BLOCK_COMMENT("call exception_handler_for_return_address");
 519     __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 520                          SharedRuntime::exception_handler_for_return_address),
 521                     r15_thread, c_rarg0);
 522     __ mov(rbx, rax);
 523 
 524     // setup rax & rdx, remove return address & clear pending exception
 525     __ pop(rdx);
 526     __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 527     __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 528 
 529 #ifdef ASSERT
 530     // make sure exception is set
 531     {
 532       Label L;
 533       __ testptr(rax, rax);
 534       __ jcc(Assembler::notEqual, L);
 535       __ stop("StubRoutines::forward exception: no pending exception (2)");
 536       __ bind(L);
 537     }
 538 #endif
 539 
 540     // continue at exception handler (return address removed)
 541     // rax: exception
 542     // rbx: exception handler
 543     // rdx: throwing pc
 544     __ verify_oop(rax);
 545     __ jmp(rbx);
 546 
 547     return start;
 548   }
 549 
 550   // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
 551   //
 552   // Arguments :
 553   //    c_rarg0: exchange_value
 554   //    c_rarg0: dest
 555   //
 556   // Result:
 557   //    *dest <- ex, return (orig *dest)
 558   address generate_atomic_xchg() {
 559     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 560     address start = __ pc();
 561 
 562     __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
 563     __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
 564     __ ret(0);
 565 
 566     return start;
 567   }
 568 
 569   // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
 570   //
 571   // Arguments :
 572   //    c_rarg0: exchange_value
 573   //    c_rarg1: dest
 574   //
 575   // Result:
 576   //    *dest <- ex, return (orig *dest)
 577   address generate_atomic_xchg_ptr() {
 578     StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
 579     address start = __ pc();
 580 
 581     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 582     __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
 583     __ ret(0);
 584 
 585     return start;
 586   }
 587 
 588   // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
 589   //                                         jint compare_value)
 590   //
 591   // Arguments :
 592   //    c_rarg0: exchange_value
 593   //    c_rarg1: dest
 594   //    c_rarg2: compare_value
 595   //
 596   // Result:
 597   //    if ( compare_value == *dest ) {
 598   //       *dest = exchange_value
 599   //       return compare_value;
 600   //    else
 601   //       return *dest;
 602   address generate_atomic_cmpxchg() {
 603     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
 604     address start = __ pc();
 605 
 606     __ movl(rax, c_rarg2);
 607    if ( os::is_MP() ) __ lock();
 608     __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
 609     __ ret(0);
 610 
 611     return start;
 612   }
 613 
 614   // Support for jbyte atomic::atomic_cmpxchg(jbyte exchange_value, volatile jbyte* dest,
 615   //                                          jbyte compare_value)
 616   //
 617   // Arguments :
 618   //    c_rarg0: exchange_value
 619   //    c_rarg1: dest
 620   //    c_rarg2: compare_value
 621   //
 622   // Result:
 623   //    if ( compare_value == *dest ) {
 624   //       *dest = exchange_value
 625   //       return compare_value;
 626   //    else
 627   //       return *dest;
 628   address generate_atomic_cmpxchg_byte() {
 629     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
 630     address start = __ pc();
 631 
 632     __ movsbq(rax, c_rarg2);
 633    if ( os::is_MP() ) __ lock();
 634     __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
 635     __ ret(0);
 636 
 637     return start;
 638   }
 639 
 640   // Support for jlong atomic::atomic_cmpxchg(jlong exchange_value,
 641   //                                          volatile jlong* dest,
 642   //                                          jlong compare_value)
 643   // Arguments :
 644   //    c_rarg0: exchange_value
 645   //    c_rarg1: dest
 646   //    c_rarg2: compare_value
 647   //
 648   // Result:
 649   //    if ( compare_value == *dest ) {
 650   //       *dest = exchange_value
 651   //       return compare_value;
 652   //    else
 653   //       return *dest;
 654   address generate_atomic_cmpxchg_long() {
 655     StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
 656     address start = __ pc();
 657 
 658     __ movq(rax, c_rarg2);
 659    if ( os::is_MP() ) __ lock();
 660     __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
 661     __ ret(0);
 662 
 663     return start;
 664   }
 665 
 666   // Support for jint atomic::add(jint add_value, volatile jint* dest)
 667   //
 668   // Arguments :
 669   //    c_rarg0: add_value
 670   //    c_rarg1: dest
 671   //
 672   // Result:
 673   //    *dest += add_value
 674   //    return *dest;
 675   address generate_atomic_add() {
 676     StubCodeMark mark(this, "StubRoutines", "atomic_add");
 677     address start = __ pc();
 678 
 679     __ movl(rax, c_rarg0);
 680    if ( os::is_MP() ) __ lock();
 681     __ xaddl(Address(c_rarg1, 0), c_rarg0);
 682     __ addl(rax, c_rarg0);
 683     __ ret(0);
 684 
 685     return start;
 686   }
 687 
 688   // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
 689   //
 690   // Arguments :
 691   //    c_rarg0: add_value
 692   //    c_rarg1: dest
 693   //
 694   // Result:
 695   //    *dest += add_value
 696   //    return *dest;
 697   address generate_atomic_add_ptr() {
 698     StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
 699     address start = __ pc();
 700 
 701     __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
 702    if ( os::is_MP() ) __ lock();
 703     __ xaddptr(Address(c_rarg1, 0), c_rarg0);
 704     __ addptr(rax, c_rarg0);
 705     __ ret(0);
 706 
 707     return start;
 708   }
 709 
 710   // Support for intptr_t OrderAccess::fence()
 711   //
 712   // Arguments :
 713   //
 714   // Result:
 715   address generate_orderaccess_fence() {
 716     StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 717     address start = __ pc();
 718     __ membar(Assembler::StoreLoad);
 719     __ ret(0);
 720 
 721     return start;
 722   }
 723 
 724   // Support for intptr_t get_previous_fp()
 725   //
 726   // This routine is used to find the previous frame pointer for the
 727   // caller (current_frame_guess). This is used as part of debugging
 728   // ps() is seemingly lost trying to find frames.
 729   // This code assumes that caller current_frame_guess) has a frame.
 730   address generate_get_previous_fp() {
 731     StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
 732     const Address old_fp(rbp, 0);
 733     const Address older_fp(rax, 0);
 734     address start = __ pc();
 735 
 736     __ enter();
 737     __ movptr(rax, old_fp); // callers fp
 738     __ movptr(rax, older_fp); // the frame for ps()
 739     __ pop(rbp);
 740     __ ret(0);
 741 
 742     return start;
 743   }
 744 
 745   // Support for intptr_t get_previous_sp()
 746   //
 747   // This routine is used to find the previous stack pointer for the
 748   // caller.
 749   address generate_get_previous_sp() {
 750     StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
 751     address start = __ pc();
 752 
 753     __ movptr(rax, rsp);
 754     __ addptr(rax, 8); // return address is at the top of the stack.
 755     __ ret(0);
 756 
 757     return start;
 758   }
 759 
 760   //----------------------------------------------------------------------------------------------------
 761   // Support for void verify_mxcsr()
 762   //
 763   // This routine is used with -Xcheck:jni to verify that native
 764   // JNI code does not return to Java code without restoring the
 765   // MXCSR register to our expected state.
 766 
 767   address generate_verify_mxcsr() {
 768     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 769     address start = __ pc();
 770 
 771     const Address mxcsr_save(rsp, 0);
 772 
 773     if (CheckJNICalls) {
 774       Label ok_ret;
 775       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 776       __ push(rax);
 777       __ subptr(rsp, wordSize);      // allocate a temp location
 778       __ stmxcsr(mxcsr_save);
 779       __ movl(rax, mxcsr_save);
 780       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 781       __ cmp32(rax, mxcsr_std);
 782       __ jcc(Assembler::equal, ok_ret);
 783 
 784       __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 785 
 786       __ ldmxcsr(mxcsr_std);
 787 
 788       __ bind(ok_ret);
 789       __ addptr(rsp, wordSize);
 790       __ pop(rax);
 791     }
 792 
 793     __ ret(0);
 794 
 795     return start;
 796   }
 797 
 798   address generate_f2i_fixup() {
 799     StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 800     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 801 
 802     address start = __ pc();
 803 
 804     Label L;
 805 
 806     __ push(rax);
 807     __ push(c_rarg3);
 808     __ push(c_rarg2);
 809     __ push(c_rarg1);
 810 
 811     __ movl(rax, 0x7f800000);
 812     __ xorl(c_rarg3, c_rarg3);
 813     __ movl(c_rarg2, inout);
 814     __ movl(c_rarg1, c_rarg2);
 815     __ andl(c_rarg1, 0x7fffffff);
 816     __ cmpl(rax, c_rarg1); // NaN? -> 0
 817     __ jcc(Assembler::negative, L);
 818     __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 819     __ movl(c_rarg3, 0x80000000);
 820     __ movl(rax, 0x7fffffff);
 821     __ cmovl(Assembler::positive, c_rarg3, rax);
 822 
 823     __ bind(L);
 824     __ movptr(inout, c_rarg3);
 825 
 826     __ pop(c_rarg1);
 827     __ pop(c_rarg2);
 828     __ pop(c_rarg3);
 829     __ pop(rax);
 830 
 831     __ ret(0);
 832 
 833     return start;
 834   }
 835 
 836   address generate_f2l_fixup() {
 837     StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 838     Address inout(rsp, 5 * wordSize); // return address + 4 saves
 839     address start = __ pc();
 840 
 841     Label L;
 842 
 843     __ push(rax);
 844     __ push(c_rarg3);
 845     __ push(c_rarg2);
 846     __ push(c_rarg1);
 847 
 848     __ movl(rax, 0x7f800000);
 849     __ xorl(c_rarg3, c_rarg3);
 850     __ movl(c_rarg2, inout);
 851     __ movl(c_rarg1, c_rarg2);
 852     __ andl(c_rarg1, 0x7fffffff);
 853     __ cmpl(rax, c_rarg1); // NaN? -> 0
 854     __ jcc(Assembler::negative, L);
 855     __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 856     __ mov64(c_rarg3, 0x8000000000000000);
 857     __ mov64(rax, 0x7fffffffffffffff);
 858     __ cmov(Assembler::positive, c_rarg3, rax);
 859 
 860     __ bind(L);
 861     __ movptr(inout, c_rarg3);
 862 
 863     __ pop(c_rarg1);
 864     __ pop(c_rarg2);
 865     __ pop(c_rarg3);
 866     __ pop(rax);
 867 
 868     __ ret(0);
 869 
 870     return start;
 871   }
 872 
 873   address generate_d2i_fixup() {
 874     StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 875     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 876 
 877     address start = __ pc();
 878 
 879     Label L;
 880 
 881     __ push(rax);
 882     __ push(c_rarg3);
 883     __ push(c_rarg2);
 884     __ push(c_rarg1);
 885     __ push(c_rarg0);
 886 
 887     __ movl(rax, 0x7ff00000);
 888     __ movq(c_rarg2, inout);
 889     __ movl(c_rarg3, c_rarg2);
 890     __ mov(c_rarg1, c_rarg2);
 891     __ mov(c_rarg0, c_rarg2);
 892     __ negl(c_rarg3);
 893     __ shrptr(c_rarg1, 0x20);
 894     __ orl(c_rarg3, c_rarg2);
 895     __ andl(c_rarg1, 0x7fffffff);
 896     __ xorl(c_rarg2, c_rarg2);
 897     __ shrl(c_rarg3, 0x1f);
 898     __ orl(c_rarg1, c_rarg3);
 899     __ cmpl(rax, c_rarg1);
 900     __ jcc(Assembler::negative, L); // NaN -> 0
 901     __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 902     __ movl(c_rarg2, 0x80000000);
 903     __ movl(rax, 0x7fffffff);
 904     __ cmov(Assembler::positive, c_rarg2, rax);
 905 
 906     __ bind(L);
 907     __ movptr(inout, c_rarg2);
 908 
 909     __ pop(c_rarg0);
 910     __ pop(c_rarg1);
 911     __ pop(c_rarg2);
 912     __ pop(c_rarg3);
 913     __ pop(rax);
 914 
 915     __ ret(0);
 916 
 917     return start;
 918   }
 919 
 920   address generate_d2l_fixup() {
 921     StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 922     Address inout(rsp, 6 * wordSize); // return address + 5 saves
 923 
 924     address start = __ pc();
 925 
 926     Label L;
 927 
 928     __ push(rax);
 929     __ push(c_rarg3);
 930     __ push(c_rarg2);
 931     __ push(c_rarg1);
 932     __ push(c_rarg0);
 933 
 934     __ movl(rax, 0x7ff00000);
 935     __ movq(c_rarg2, inout);
 936     __ movl(c_rarg3, c_rarg2);
 937     __ mov(c_rarg1, c_rarg2);
 938     __ mov(c_rarg0, c_rarg2);
 939     __ negl(c_rarg3);
 940     __ shrptr(c_rarg1, 0x20);
 941     __ orl(c_rarg3, c_rarg2);
 942     __ andl(c_rarg1, 0x7fffffff);
 943     __ xorl(c_rarg2, c_rarg2);
 944     __ shrl(c_rarg3, 0x1f);
 945     __ orl(c_rarg1, c_rarg3);
 946     __ cmpl(rax, c_rarg1);
 947     __ jcc(Assembler::negative, L); // NaN -> 0
 948     __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 949     __ mov64(c_rarg2, 0x8000000000000000);
 950     __ mov64(rax, 0x7fffffffffffffff);
 951     __ cmovq(Assembler::positive, c_rarg2, rax);
 952 
 953     __ bind(L);
 954     __ movq(inout, c_rarg2);
 955 
 956     __ pop(c_rarg0);
 957     __ pop(c_rarg1);
 958     __ pop(c_rarg2);
 959     __ pop(c_rarg3);
 960     __ pop(rax);
 961 
 962     __ ret(0);
 963 
 964     return start;
 965   }
 966 
 967   address generate_fp_mask(const char *stub_name, int64_t mask) {
 968     __ align(CodeEntryAlignment);
 969     StubCodeMark mark(this, "StubRoutines", stub_name);
 970     address start = __ pc();
 971 
 972     __ emit_data64( mask, relocInfo::none );
 973     __ emit_data64( mask, relocInfo::none );
 974 
 975     return start;
 976   }
 977 
 978   // The following routine generates a subroutine to throw an
 979   // asynchronous UnknownError when an unsafe access gets a fault that
 980   // could not be reasonably prevented by the programmer.  (Example:
 981   // SIGBUS/OBJERR.)
 982   address generate_handler_for_unsafe_access() {
 983     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 984     address start = __ pc();
 985 
 986     __ push(0);                       // hole for return address-to-be
 987     __ pusha();                       // push registers
 988     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 989 
 990     // FIXME: this probably needs alignment logic
 991 
 992     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 993     BLOCK_COMMENT("call handle_unsafe_access");
 994     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 995     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 996 
 997     __ movptr(next_pc, rax);          // stuff next address
 998     __ popa();
 999     __ ret(0);                        // jump to next address
1000 
1001     return start;
1002   }
1003 
1004   // Non-destructive plausibility checks for oops
1005   //
1006   // Arguments:
1007   //    all args on stack!
1008   //
1009   // Stack after saving c_rarg3:
1010   //    [tos + 0]: saved c_rarg3
1011   //    [tos + 1]: saved c_rarg2
1012   //    [tos + 2]: saved r12 (several TemplateTable methods use it)
1013   //    [tos + 3]: saved flags
1014   //    [tos + 4]: return address
1015   //  * [tos + 5]: error message (char*)
1016   //  * [tos + 6]: object to verify (oop)
1017   //  * [tos + 7]: saved rax - saved by caller and bashed
1018   //  * [tos + 8]: saved r10 (rscratch1) - saved by caller
1019   //  * = popped on exit
1020   address generate_verify_oop() {
1021     StubCodeMark mark(this, "StubRoutines", "verify_oop");
1022     address start = __ pc();
1023 
1024     Label exit, error;
1025 
1026     __ pushf();
1027     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1028 
1029     __ push(r12);
1030 
1031     // save c_rarg2 and c_rarg3
1032     __ push(c_rarg2);
1033     __ push(c_rarg3);
1034 
1035     enum {
1036            // After previous pushes.
1037            oop_to_verify = 6 * wordSize,
1038            saved_rax     = 7 * wordSize,
1039            saved_r10     = 8 * wordSize,
1040 
1041            // Before the call to MacroAssembler::debug(), see below.
1042            return_addr   = 16 * wordSize,
1043            error_msg     = 17 * wordSize
1044     };
1045 
1046     // get object
1047     __ movptr(rax, Address(rsp, oop_to_verify));
1048 
1049     // make sure object is 'reasonable'
1050     __ testptr(rax, rax);
1051     __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1052     // Check if the oop is in the right area of memory
1053     __ movptr(c_rarg2, rax);
1054     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1055     __ andptr(c_rarg2, c_rarg3);
1056     __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1057     __ cmpptr(c_rarg2, c_rarg3);
1058     __ jcc(Assembler::notZero, error);
1059 
1060     // set r12 to heapbase for load_klass()
1061     __ reinit_heapbase();
1062 
1063     // make sure klass is 'reasonable', which is not zero.
1064     __ load_klass(rax, rax);  // get klass
1065     __ testptr(rax, rax);
1066     __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1067 
1068     // return if everything seems ok
1069     __ bind(exit);
1070     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1071     __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1072     __ pop(c_rarg3);                             // restore c_rarg3
1073     __ pop(c_rarg2);                             // restore c_rarg2
1074     __ pop(r12);                                 // restore r12
1075     __ popf();                                   // restore flags
1076     __ ret(4 * wordSize);                        // pop caller saved stuff
1077 
1078     // handle errors
1079     __ bind(error);
1080     __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1081     __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1082     __ pop(c_rarg3);                             // get saved c_rarg3 back
1083     __ pop(c_rarg2);                             // get saved c_rarg2 back
1084     __ pop(r12);                                 // get saved r12 back
1085     __ popf();                                   // get saved flags off stack --
1086                                                  // will be ignored
1087 
1088     __ pusha();                                  // push registers
1089                                                  // (rip is already
1090                                                  // already pushed)
1091     // debug(char* msg, int64_t pc, int64_t regs[])
1092     // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1093     // pushed all the registers, so now the stack looks like:
1094     //     [tos +  0] 16 saved registers
1095     //     [tos + 16] return address
1096     //   * [tos + 17] error message (char*)
1097     //   * [tos + 18] object to verify (oop)
1098     //   * [tos + 19] saved rax - saved by caller and bashed
1099     //   * [tos + 20] saved r10 (rscratch1) - saved by caller
1100     //   * = popped on exit
1101 
1102     __ movptr(c_rarg0, Address(rsp, error_msg));    // pass address of error message
1103     __ movptr(c_rarg1, Address(rsp, return_addr));  // pass return address
1104     __ movq(c_rarg2, rsp);                          // pass address of regs on stack
1105     __ mov(r12, rsp);                               // remember rsp
1106     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1107     __ andptr(rsp, -16);                            // align stack as required by ABI
1108     BLOCK_COMMENT("call MacroAssembler::debug");
1109     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1110     __ mov(rsp, r12);                               // restore rsp
1111     __ popa();                                      // pop registers (includes r12)
1112     __ ret(4 * wordSize);                           // pop caller saved stuff
1113 
1114     return start;
1115   }
1116 
1117   //
1118   // Verify that a register contains clean 32-bits positive value
1119   // (high 32-bits are 0) so it could be used in 64-bits shifts.
1120   //
1121   //  Input:
1122   //    Rint  -  32-bits value
1123   //    Rtmp  -  scratch
1124   //
1125   void assert_clean_int(Register Rint, Register Rtmp) {
1126 #ifdef ASSERT
1127     Label L;
1128     assert_different_registers(Rtmp, Rint);
1129     __ movslq(Rtmp, Rint);
1130     __ cmpq(Rtmp, Rint);
1131     __ jcc(Assembler::equal, L);
1132     __ stop("high 32-bits of int value are not 0");
1133     __ bind(L);
1134 #endif
1135   }
1136 
1137   //  Generate overlap test for array copy stubs
1138   //
1139   //  Input:
1140   //     c_rarg0 - from
1141   //     c_rarg1 - to
1142   //     c_rarg2 - element count
1143   //
1144   //  Output:
1145   //     rax   - &from[element count - 1]
1146   //
1147   void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1148     assert(no_overlap_target != NULL, "must be generated");
1149     array_overlap_test(no_overlap_target, NULL, sf);
1150   }
1151   void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1152     array_overlap_test(NULL, &L_no_overlap, sf);
1153   }
1154   void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1155     const Register from     = c_rarg0;
1156     const Register to       = c_rarg1;
1157     const Register count    = c_rarg2;
1158     const Register end_from = rax;
1159 
1160     __ cmpptr(to, from);
1161     __ lea(end_from, Address(from, count, sf, 0));
1162     if (NOLp == NULL) {
1163       ExternalAddress no_overlap(no_overlap_target);
1164       __ jump_cc(Assembler::belowEqual, no_overlap);
1165       __ cmpptr(to, end_from);
1166       __ jump_cc(Assembler::aboveEqual, no_overlap);
1167     } else {
1168       __ jcc(Assembler::belowEqual, (*NOLp));
1169       __ cmpptr(to, end_from);
1170       __ jcc(Assembler::aboveEqual, (*NOLp));
1171     }
1172   }
1173 
1174   // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1175   //
1176   // Outputs:
1177   //    rdi - rcx
1178   //    rsi - rdx
1179   //    rdx - r8
1180   //    rcx - r9
1181   //
1182   // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1183   // are non-volatile.  r9 and r10 should not be used by the caller.
1184   //
1185   void setup_arg_regs(int nargs = 3) {
1186     const Register saved_rdi = r9;
1187     const Register saved_rsi = r10;
1188     assert(nargs == 3 || nargs == 4, "else fix");
1189 #ifdef _WIN64
1190     assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1191            "unexpected argument registers");
1192     if (nargs >= 4)
1193       __ mov(rax, r9);  // r9 is also saved_rdi
1194     __ movptr(saved_rdi, rdi);
1195     __ movptr(saved_rsi, rsi);
1196     __ mov(rdi, rcx); // c_rarg0
1197     __ mov(rsi, rdx); // c_rarg1
1198     __ mov(rdx, r8);  // c_rarg2
1199     if (nargs >= 4)
1200       __ mov(rcx, rax); // c_rarg3 (via rax)
1201 #else
1202     assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1203            "unexpected argument registers");
1204 #endif
1205   }
1206 
1207   void restore_arg_regs() {
1208     const Register saved_rdi = r9;
1209     const Register saved_rsi = r10;
1210 #ifdef _WIN64
1211     __ movptr(rdi, saved_rdi);
1212     __ movptr(rsi, saved_rsi);
1213 #endif
1214   }
1215 
1216   // Generate code for an array write pre barrier
1217   //
1218   //     addr    -  starting address
1219   //     count   -  element count
1220   //     tmp     - scratch register
1221   //
1222   //     Destroy no registers!
1223   //
1224   void  gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1225     BarrierSet* bs = Universe::heap()->barrier_set();
1226     switch (bs->kind()) {
1227       case BarrierSet::G1SATBCTLogging:
1228         // With G1, don't generate the call if we statically know that the target in uninitialized
1229         if (!dest_uninitialized) {
1230            __ pusha();                      // push registers
1231            if (count == c_rarg0) {
1232              if (addr == c_rarg1) {
1233                // exactly backwards!!
1234                __ xchgptr(c_rarg1, c_rarg0);
1235              } else {
1236                __ movptr(c_rarg1, count);
1237                __ movptr(c_rarg0, addr);
1238              }
1239            } else {
1240              __ movptr(c_rarg0, addr);
1241              __ movptr(c_rarg1, count);
1242            }
1243            __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1244            __ popa();
1245         }
1246          break;
1247       case BarrierSet::CardTableModRef:
1248       case BarrierSet::CardTableExtension:
1249       case BarrierSet::ModRef:
1250         break;
1251       default:
1252         ShouldNotReachHere();
1253 
1254     }
1255   }
1256 
1257   //
1258   // Generate code for an array write post barrier
1259   //
1260   //  Input:
1261   //     start    - register containing starting address of destination array
1262   //     count    - elements count
1263   //     scratch  - scratch register
1264   //
1265   //  The input registers are overwritten.
1266   //
1267   void  gen_write_ref_array_post_barrier(Register start, Register count, Register scratch) {
1268     assert_different_registers(start, count, scratch);
1269     BarrierSet* bs = Universe::heap()->barrier_set();
1270     switch (bs->kind()) {
1271       case BarrierSet::G1SATBCTLogging:
1272         {
1273           __ pusha();             // push registers (overkill)
1274           if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
1275             assert_different_registers(c_rarg1, start);
1276             __ mov(c_rarg1, count);
1277             __ mov(c_rarg0, start);
1278           } else {
1279             assert_different_registers(c_rarg0, count);
1280             __ mov(c_rarg0, start);
1281             __ mov(c_rarg1, count);
1282           }
1283           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1284           __ popa();
1285         }
1286         break;
1287       case BarrierSet::CardTableModRef:
1288       case BarrierSet::CardTableExtension:
1289         {
1290           CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
1291           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1292 
1293           Label L_loop;
1294           const Register end = count;
1295 
1296           __ leaq(end, Address(start, count, TIMES_OOP, 0));  // end == start+count*oop_size
1297           __ subptr(end, BytesPerHeapOop); // end - 1 to make inclusive
1298           __ shrptr(start, CardTableModRefBS::card_shift);
1299           __ shrptr(end,   CardTableModRefBS::card_shift);
1300           __ subptr(end, start); // end --> cards count
1301 
1302           int64_t disp = (int64_t) ct->byte_map_base;
1303           __ mov64(scratch, disp);
1304           __ addptr(start, scratch);
1305         __ BIND(L_loop);
1306           __ movb(Address(start, count, Address::times_1), 0);
1307           __ decrement(count);
1308           __ jcc(Assembler::greaterEqual, L_loop);
1309         }
1310         break;
1311       default:
1312         ShouldNotReachHere();
1313 
1314     }
1315   }
1316 
1317 
1318   // Copy big chunks forward
1319   //
1320   // Inputs:
1321   //   end_from     - source arrays end address
1322   //   end_to       - destination array end address
1323   //   qword_count  - 64-bits element count, negative
1324   //   to           - scratch
1325   //   L_copy_bytes - entry label
1326   //   L_copy_8_bytes  - exit  label
1327   //
1328   void copy_bytes_forward(Register end_from, Register end_to,
1329                              Register qword_count, Register to,
1330                              Label& L_copy_bytes, Label& L_copy_8_bytes) {
1331     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1332     Label L_loop;
1333     __ align(OptoLoopAlignment);
1334     if (UseUnalignedLoadStores) {
1335       Label L_end;
1336       // Copy 64-bytes per iteration
1337       __ BIND(L_loop);
1338       if (UseAVX > 2) {
1339         __ evmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1340         __ evmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1341       } else if (UseAVX == 2) {
1342         __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1343         __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1344         __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1345         __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1346       } else {
1347         __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1348         __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1349         __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1350         __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1351         __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1352         __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1353         __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1354         __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1355       }
1356       __ BIND(L_copy_bytes);
1357       __ addptr(qword_count, 8);
1358       __ jcc(Assembler::lessEqual, L_loop);
1359       __ subptr(qword_count, 4);  // sub(8) and add(4)
1360       __ jccb(Assembler::greater, L_end);
1361       // Copy trailing 32 bytes
1362       if (UseAVX >= 2) {
1363         __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1364         __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1365       } else {
1366         __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1367         __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1368         __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1369         __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1370       }
1371       __ addptr(qword_count, 4);
1372       __ BIND(L_end);
1373       if (UseAVX >= 2) {
1374         // clean upper bits of YMM registers
1375         __ vpxor(xmm0, xmm0);
1376         __ vpxor(xmm1, xmm1);
1377       }
1378     } else {
1379       // Copy 32-bytes per iteration
1380       __ BIND(L_loop);
1381       __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1382       __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1383       __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1384       __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1385       __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1386       __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1387       __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1388       __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1389 
1390       __ BIND(L_copy_bytes);
1391       __ addptr(qword_count, 4);
1392       __ jcc(Assembler::lessEqual, L_loop);
1393     }
1394     __ subptr(qword_count, 4);
1395     __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1396   }
1397 
1398   // Copy big chunks backward
1399   //
1400   // Inputs:
1401   //   from         - source arrays address
1402   //   dest         - destination array address
1403   //   qword_count  - 64-bits element count
1404   //   to           - scratch
1405   //   L_copy_bytes - entry label
1406   //   L_copy_8_bytes  - exit  label
1407   //
1408   void copy_bytes_backward(Register from, Register dest,
1409                               Register qword_count, Register to,
1410                               Label& L_copy_bytes, Label& L_copy_8_bytes) {
1411     DEBUG_ONLY(__ stop("enter at entry label, not here"));
1412     Label L_loop;
1413     __ align(OptoLoopAlignment);
1414     if (UseUnalignedLoadStores) {
1415       Label L_end;
1416       // Copy 64-bytes per iteration
1417       __ BIND(L_loop);
1418       if (UseAVX > 2) {
1419         __ evmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32), Assembler::AVX_512bit);
1420         __ evmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0, Assembler::AVX_512bit);
1421       } else if (UseAVX == 2) {
1422         __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1423         __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1424         __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8,  0));
1425         __ vmovdqu(Address(dest, qword_count, Address::times_8,  0), xmm1);
1426       } else {
1427         __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1428         __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1429         __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1430         __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1431         __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1432         __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1433         __ movdqu(xmm3, Address(from, qword_count, Address::times_8,  0));
1434         __ movdqu(Address(dest, qword_count, Address::times_8,  0), xmm3);
1435       }
1436       __ BIND(L_copy_bytes);
1437       __ subptr(qword_count, 8);
1438       __ jcc(Assembler::greaterEqual, L_loop);
1439 
1440       __ addptr(qword_count, 4);  // add(8) and sub(4)
1441       __ jccb(Assembler::less, L_end);
1442       // Copy trailing 32 bytes
1443       if (UseAVX >= 2) {
1444         __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1445         __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1446       } else {
1447         __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1448         __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1449         __ movdqu(xmm1, Address(from, qword_count, Address::times_8,  0));
1450         __ movdqu(Address(dest, qword_count, Address::times_8,  0), xmm1);
1451       }
1452       __ subptr(qword_count, 4);
1453       __ BIND(L_end);
1454       if (UseAVX >= 2) {
1455         // clean upper bits of YMM registers
1456         __ vpxor(xmm0, xmm0);
1457         __ vpxor(xmm1, xmm1);
1458       }
1459     } else {
1460       // Copy 32-bytes per iteration
1461       __ BIND(L_loop);
1462       __ movq(to, Address(from, qword_count, Address::times_8, 24));
1463       __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1464       __ movq(to, Address(from, qword_count, Address::times_8, 16));
1465       __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1466       __ movq(to, Address(from, qword_count, Address::times_8,  8));
1467       __ movq(Address(dest, qword_count, Address::times_8,  8), to);
1468       __ movq(to, Address(from, qword_count, Address::times_8,  0));
1469       __ movq(Address(dest, qword_count, Address::times_8,  0), to);
1470 
1471       __ BIND(L_copy_bytes);
1472       __ subptr(qword_count, 4);
1473       __ jcc(Assembler::greaterEqual, L_loop);
1474     }
1475     __ addptr(qword_count, 4);
1476     __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1477   }
1478 
1479 
1480   // Arguments:
1481   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1482   //             ignored
1483   //   name    - stub name string
1484   //
1485   // Inputs:
1486   //   c_rarg0   - source array address
1487   //   c_rarg1   - destination array address
1488   //   c_rarg2   - element count, treated as ssize_t, can be zero
1489   //
1490   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1491   // we let the hardware handle it.  The one to eight bytes within words,
1492   // dwords or qwords that span cache line boundaries will still be loaded
1493   // and stored atomically.
1494   //
1495   // Side Effects:
1496   //   disjoint_byte_copy_entry is set to the no-overlap entry point
1497   //   used by generate_conjoint_byte_copy().
1498   //
1499   address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1500     __ align(CodeEntryAlignment);
1501     StubCodeMark mark(this, "StubRoutines", name);
1502     address start = __ pc();
1503 
1504     Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1505     Label L_copy_byte, L_exit;
1506     const Register from        = rdi;  // source array address
1507     const Register to          = rsi;  // destination array address
1508     const Register count       = rdx;  // elements count
1509     const Register byte_count  = rcx;
1510     const Register qword_count = count;
1511     const Register end_from    = from; // source array end address
1512     const Register end_to      = to;   // destination array end address
1513     // End pointers are inclusive, and if count is not zero they point
1514     // to the last unit copied:  end_to[0] := end_from[0]
1515 
1516     __ enter(); // required for proper stackwalking of RuntimeStub frame
1517     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1518 
1519     if (entry != NULL) {
1520       *entry = __ pc();
1521        // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1522       BLOCK_COMMENT("Entry:");
1523     }
1524 
1525     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1526                       // r9 and r10 may be used to save non-volatile registers
1527 
1528     // 'from', 'to' and 'count' are now valid
1529     __ movptr(byte_count, count);
1530     __ shrptr(count, 3); // count => qword_count
1531 
1532     // Copy from low to high addresses.  Use 'to' as scratch.
1533     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1534     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1535     __ negptr(qword_count); // make the count negative
1536     __ jmp(L_copy_bytes);
1537 
1538     // Copy trailing qwords
1539   __ BIND(L_copy_8_bytes);
1540     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1541     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1542     __ increment(qword_count);
1543     __ jcc(Assembler::notZero, L_copy_8_bytes);
1544 
1545     // Check for and copy trailing dword
1546   __ BIND(L_copy_4_bytes);
1547     __ testl(byte_count, 4);
1548     __ jccb(Assembler::zero, L_copy_2_bytes);
1549     __ movl(rax, Address(end_from, 8));
1550     __ movl(Address(end_to, 8), rax);
1551 
1552     __ addptr(end_from, 4);
1553     __ addptr(end_to, 4);
1554 
1555     // Check for and copy trailing word
1556   __ BIND(L_copy_2_bytes);
1557     __ testl(byte_count, 2);
1558     __ jccb(Assembler::zero, L_copy_byte);
1559     __ movw(rax, Address(end_from, 8));
1560     __ movw(Address(end_to, 8), rax);
1561 
1562     __ addptr(end_from, 2);
1563     __ addptr(end_to, 2);
1564 
1565     // Check for and copy trailing byte
1566   __ BIND(L_copy_byte);
1567     __ testl(byte_count, 1);
1568     __ jccb(Assembler::zero, L_exit);
1569     __ movb(rax, Address(end_from, 8));
1570     __ movb(Address(end_to, 8), rax);
1571 
1572   __ BIND(L_exit);
1573     restore_arg_regs();
1574     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1575     __ xorptr(rax, rax); // return 0
1576     __ leave(); // required for proper stackwalking of RuntimeStub frame
1577     __ ret(0);
1578 
1579     // Copy in multi-bytes chunks
1580     copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1581     __ jmp(L_copy_4_bytes);
1582 
1583     return start;
1584   }
1585 
1586   // Arguments:
1587   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1588   //             ignored
1589   //   name    - stub name string
1590   //
1591   // Inputs:
1592   //   c_rarg0   - source array address
1593   //   c_rarg1   - destination array address
1594   //   c_rarg2   - element count, treated as ssize_t, can be zero
1595   //
1596   // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1597   // we let the hardware handle it.  The one to eight bytes within words,
1598   // dwords or qwords that span cache line boundaries will still be loaded
1599   // and stored atomically.
1600   //
1601   address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1602                                       address* entry, const char *name) {
1603     __ align(CodeEntryAlignment);
1604     StubCodeMark mark(this, "StubRoutines", name);
1605     address start = __ pc();
1606 
1607     Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1608     const Register from        = rdi;  // source array address
1609     const Register to          = rsi;  // destination array address
1610     const Register count       = rdx;  // elements count
1611     const Register byte_count  = rcx;
1612     const Register qword_count = count;
1613 
1614     __ enter(); // required for proper stackwalking of RuntimeStub frame
1615     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1616 
1617     if (entry != NULL) {
1618       *entry = __ pc();
1619       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1620       BLOCK_COMMENT("Entry:");
1621     }
1622 
1623     array_overlap_test(nooverlap_target, Address::times_1);
1624     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1625                       // r9 and r10 may be used to save non-volatile registers
1626 
1627     // 'from', 'to' and 'count' are now valid
1628     __ movptr(byte_count, count);
1629     __ shrptr(count, 3);   // count => qword_count
1630 
1631     // Copy from high to low addresses.
1632 
1633     // Check for and copy trailing byte
1634     __ testl(byte_count, 1);
1635     __ jcc(Assembler::zero, L_copy_2_bytes);
1636     __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1637     __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1638     __ decrement(byte_count); // Adjust for possible trailing word
1639 
1640     // Check for and copy trailing word
1641   __ BIND(L_copy_2_bytes);
1642     __ testl(byte_count, 2);
1643     __ jcc(Assembler::zero, L_copy_4_bytes);
1644     __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1645     __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1646 
1647     // Check for and copy trailing dword
1648   __ BIND(L_copy_4_bytes);
1649     __ testl(byte_count, 4);
1650     __ jcc(Assembler::zero, L_copy_bytes);
1651     __ movl(rax, Address(from, qword_count, Address::times_8));
1652     __ movl(Address(to, qword_count, Address::times_8), rax);
1653     __ jmp(L_copy_bytes);
1654 
1655     // Copy trailing qwords
1656   __ BIND(L_copy_8_bytes);
1657     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1658     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1659     __ decrement(qword_count);
1660     __ jcc(Assembler::notZero, L_copy_8_bytes);
1661 
1662     restore_arg_regs();
1663     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1664     __ xorptr(rax, rax); // return 0
1665     __ leave(); // required for proper stackwalking of RuntimeStub frame
1666     __ ret(0);
1667 
1668     // Copy in multi-bytes chunks
1669     copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1670 
1671     restore_arg_regs();
1672     inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1673     __ xorptr(rax, rax); // return 0
1674     __ leave(); // required for proper stackwalking of RuntimeStub frame
1675     __ ret(0);
1676 
1677     return start;
1678   }
1679 
1680   // Arguments:
1681   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1682   //             ignored
1683   //   name    - stub name string
1684   //
1685   // Inputs:
1686   //   c_rarg0   - source array address
1687   //   c_rarg1   - destination array address
1688   //   c_rarg2   - element count, treated as ssize_t, can be zero
1689   //
1690   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1691   // let the hardware handle it.  The two or four words within dwords
1692   // or qwords that span cache line boundaries will still be loaded
1693   // and stored atomically.
1694   //
1695   // Side Effects:
1696   //   disjoint_short_copy_entry is set to the no-overlap entry point
1697   //   used by generate_conjoint_short_copy().
1698   //
1699   address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1700     __ align(CodeEntryAlignment);
1701     StubCodeMark mark(this, "StubRoutines", name);
1702     address start = __ pc();
1703 
1704     Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1705     const Register from        = rdi;  // source array address
1706     const Register to          = rsi;  // destination array address
1707     const Register count       = rdx;  // elements count
1708     const Register word_count  = rcx;
1709     const Register qword_count = count;
1710     const Register end_from    = from; // source array end address
1711     const Register end_to      = to;   // destination array end address
1712     // End pointers are inclusive, and if count is not zero they point
1713     // to the last unit copied:  end_to[0] := end_from[0]
1714 
1715     __ enter(); // required for proper stackwalking of RuntimeStub frame
1716     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1717 
1718     if (entry != NULL) {
1719       *entry = __ pc();
1720       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1721       BLOCK_COMMENT("Entry:");
1722     }
1723 
1724     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1725                       // r9 and r10 may be used to save non-volatile registers
1726 
1727     // 'from', 'to' and 'count' are now valid
1728     __ movptr(word_count, count);
1729     __ shrptr(count, 2); // count => qword_count
1730 
1731     // Copy from low to high addresses.  Use 'to' as scratch.
1732     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1733     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1734     __ negptr(qword_count);
1735     __ jmp(L_copy_bytes);
1736 
1737     // Copy trailing qwords
1738   __ BIND(L_copy_8_bytes);
1739     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1740     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1741     __ increment(qword_count);
1742     __ jcc(Assembler::notZero, L_copy_8_bytes);
1743 
1744     // Original 'dest' is trashed, so we can't use it as a
1745     // base register for a possible trailing word copy
1746 
1747     // Check for and copy trailing dword
1748   __ BIND(L_copy_4_bytes);
1749     __ testl(word_count, 2);
1750     __ jccb(Assembler::zero, L_copy_2_bytes);
1751     __ movl(rax, Address(end_from, 8));
1752     __ movl(Address(end_to, 8), rax);
1753 
1754     __ addptr(end_from, 4);
1755     __ addptr(end_to, 4);
1756 
1757     // Check for and copy trailing word
1758   __ BIND(L_copy_2_bytes);
1759     __ testl(word_count, 1);
1760     __ jccb(Assembler::zero, L_exit);
1761     __ movw(rax, Address(end_from, 8));
1762     __ movw(Address(end_to, 8), rax);
1763 
1764   __ BIND(L_exit);
1765     restore_arg_regs();
1766     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1767     __ xorptr(rax, rax); // return 0
1768     __ leave(); // required for proper stackwalking of RuntimeStub frame
1769     __ ret(0);
1770 
1771     // Copy in multi-bytes chunks
1772     copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1773     __ jmp(L_copy_4_bytes);
1774 
1775     return start;
1776   }
1777 
1778   address generate_fill(BasicType t, bool aligned, const char *name) {
1779     __ align(CodeEntryAlignment);
1780     StubCodeMark mark(this, "StubRoutines", name);
1781     address start = __ pc();
1782 
1783     BLOCK_COMMENT("Entry:");
1784 
1785     const Register to       = c_rarg0;  // source array address
1786     const Register value    = c_rarg1;  // value
1787     const Register count    = c_rarg2;  // elements count
1788 
1789     __ enter(); // required for proper stackwalking of RuntimeStub frame
1790 
1791     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1792 
1793     __ leave(); // required for proper stackwalking of RuntimeStub frame
1794     __ ret(0);
1795     return start;
1796   }
1797 
1798   // Arguments:
1799   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1800   //             ignored
1801   //   name    - stub name string
1802   //
1803   // Inputs:
1804   //   c_rarg0   - source array address
1805   //   c_rarg1   - destination array address
1806   //   c_rarg2   - element count, treated as ssize_t, can be zero
1807   //
1808   // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1809   // let the hardware handle it.  The two or four words within dwords
1810   // or qwords that span cache line boundaries will still be loaded
1811   // and stored atomically.
1812   //
1813   address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1814                                        address *entry, const char *name) {
1815     __ align(CodeEntryAlignment);
1816     StubCodeMark mark(this, "StubRoutines", name);
1817     address start = __ pc();
1818 
1819     Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1820     const Register from        = rdi;  // source array address
1821     const Register to          = rsi;  // destination array address
1822     const Register count       = rdx;  // elements count
1823     const Register word_count  = rcx;
1824     const Register qword_count = count;
1825 
1826     __ enter(); // required for proper stackwalking of RuntimeStub frame
1827     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1828 
1829     if (entry != NULL) {
1830       *entry = __ pc();
1831       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1832       BLOCK_COMMENT("Entry:");
1833     }
1834 
1835     array_overlap_test(nooverlap_target, Address::times_2);
1836     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1837                       // r9 and r10 may be used to save non-volatile registers
1838 
1839     // 'from', 'to' and 'count' are now valid
1840     __ movptr(word_count, count);
1841     __ shrptr(count, 2); // count => qword_count
1842 
1843     // Copy from high to low addresses.  Use 'to' as scratch.
1844 
1845     // Check for and copy trailing word
1846     __ testl(word_count, 1);
1847     __ jccb(Assembler::zero, L_copy_4_bytes);
1848     __ movw(rax, Address(from, word_count, Address::times_2, -2));
1849     __ movw(Address(to, word_count, Address::times_2, -2), rax);
1850 
1851     // Check for and copy trailing dword
1852   __ BIND(L_copy_4_bytes);
1853     __ testl(word_count, 2);
1854     __ jcc(Assembler::zero, L_copy_bytes);
1855     __ movl(rax, Address(from, qword_count, Address::times_8));
1856     __ movl(Address(to, qword_count, Address::times_8), rax);
1857     __ jmp(L_copy_bytes);
1858 
1859     // Copy trailing qwords
1860   __ BIND(L_copy_8_bytes);
1861     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1862     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1863     __ decrement(qword_count);
1864     __ jcc(Assembler::notZero, L_copy_8_bytes);
1865 
1866     restore_arg_regs();
1867     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1868     __ xorptr(rax, rax); // return 0
1869     __ leave(); // required for proper stackwalking of RuntimeStub frame
1870     __ ret(0);
1871 
1872     // Copy in multi-bytes chunks
1873     copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1874 
1875     restore_arg_regs();
1876     inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1877     __ xorptr(rax, rax); // return 0
1878     __ leave(); // required for proper stackwalking of RuntimeStub frame
1879     __ ret(0);
1880 
1881     return start;
1882   }
1883 
1884   // Arguments:
1885   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1886   //             ignored
1887   //   is_oop  - true => oop array, so generate store check code
1888   //   name    - stub name string
1889   //
1890   // Inputs:
1891   //   c_rarg0   - source array address
1892   //   c_rarg1   - destination array address
1893   //   c_rarg2   - element count, treated as ssize_t, can be zero
1894   //
1895   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1896   // the hardware handle it.  The two dwords within qwords that span
1897   // cache line boundaries will still be loaded and stored atomicly.
1898   //
1899   // Side Effects:
1900   //   disjoint_int_copy_entry is set to the no-overlap entry point
1901   //   used by generate_conjoint_int_oop_copy().
1902   //
1903   address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1904                                          const char *name, bool dest_uninitialized = false) {
1905     __ align(CodeEntryAlignment);
1906     StubCodeMark mark(this, "StubRoutines", name);
1907     address start = __ pc();
1908 
1909     Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1910     const Register from        = rdi;  // source array address
1911     const Register to          = rsi;  // destination array address
1912     const Register count       = rdx;  // elements count
1913     const Register dword_count = rcx;
1914     const Register qword_count = count;
1915     const Register end_from    = from; // source array end address
1916     const Register end_to      = to;   // destination array end address
1917     const Register saved_to    = r11;  // saved destination array address
1918     // End pointers are inclusive, and if count is not zero they point
1919     // to the last unit copied:  end_to[0] := end_from[0]
1920 
1921     __ enter(); // required for proper stackwalking of RuntimeStub frame
1922     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
1923 
1924     if (entry != NULL) {
1925       *entry = __ pc();
1926       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1927       BLOCK_COMMENT("Entry:");
1928     }
1929 
1930     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1931                       // r9 and r10 may be used to save non-volatile registers
1932     if (is_oop) {
1933       __ movq(saved_to, to);
1934       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1935     }
1936 
1937     // 'from', 'to' and 'count' are now valid
1938     __ movptr(dword_count, count);
1939     __ shrptr(count, 1); // count => qword_count
1940 
1941     // Copy from low to high addresses.  Use 'to' as scratch.
1942     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1943     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
1944     __ negptr(qword_count);
1945     __ jmp(L_copy_bytes);
1946 
1947     // Copy trailing qwords
1948   __ BIND(L_copy_8_bytes);
1949     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1950     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1951     __ increment(qword_count);
1952     __ jcc(Assembler::notZero, L_copy_8_bytes);
1953 
1954     // Check for and copy trailing dword
1955   __ BIND(L_copy_4_bytes);
1956     __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1957     __ jccb(Assembler::zero, L_exit);
1958     __ movl(rax, Address(end_from, 8));
1959     __ movl(Address(end_to, 8), rax);
1960 
1961   __ BIND(L_exit);
1962     if (is_oop) {
1963       gen_write_ref_array_post_barrier(saved_to, dword_count, rax);
1964     }
1965     restore_arg_regs();
1966     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1967     __ xorptr(rax, rax); // return 0
1968     __ leave(); // required for proper stackwalking of RuntimeStub frame
1969     __ ret(0);
1970 
1971     // Copy in multi-bytes chunks
1972     copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1973     __ jmp(L_copy_4_bytes);
1974 
1975     return start;
1976   }
1977 
1978   // Arguments:
1979   //   aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1980   //             ignored
1981   //   is_oop  - true => oop array, so generate store check code
1982   //   name    - stub name string
1983   //
1984   // Inputs:
1985   //   c_rarg0   - source array address
1986   //   c_rarg1   - destination array address
1987   //   c_rarg2   - element count, treated as ssize_t, can be zero
1988   //
1989   // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1990   // the hardware handle it.  The two dwords within qwords that span
1991   // cache line boundaries will still be loaded and stored atomicly.
1992   //
1993   address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1994                                          address *entry, const char *name,
1995                                          bool dest_uninitialized = false) {
1996     __ align(CodeEntryAlignment);
1997     StubCodeMark mark(this, "StubRoutines", name);
1998     address start = __ pc();
1999 
2000     Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
2001     const Register from        = rdi;  // source array address
2002     const Register to          = rsi;  // destination array address
2003     const Register count       = rdx;  // elements count
2004     const Register dword_count = rcx;
2005     const Register qword_count = count;
2006 
2007     __ enter(); // required for proper stackwalking of RuntimeStub frame
2008     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
2009 
2010     if (entry != NULL) {
2011       *entry = __ pc();
2012        // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2013       BLOCK_COMMENT("Entry:");
2014     }
2015 
2016     array_overlap_test(nooverlap_target, Address::times_4);
2017     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2018                       // r9 and r10 may be used to save non-volatile registers
2019 
2020     if (is_oop) {
2021       // no registers are destroyed by this call
2022       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2023     }
2024 
2025     assert_clean_int(count, rax); // Make sure 'count' is clean int.
2026     // 'from', 'to' and 'count' are now valid
2027     __ movptr(dword_count, count);
2028     __ shrptr(count, 1); // count => qword_count
2029 
2030     // Copy from high to low addresses.  Use 'to' as scratch.
2031 
2032     // Check for and copy trailing dword
2033     __ testl(dword_count, 1);
2034     __ jcc(Assembler::zero, L_copy_bytes);
2035     __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2036     __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2037     __ jmp(L_copy_bytes);
2038 
2039     // Copy trailing qwords
2040   __ BIND(L_copy_8_bytes);
2041     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2042     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2043     __ decrement(qword_count);
2044     __ jcc(Assembler::notZero, L_copy_8_bytes);
2045 
2046     if (is_oop) {
2047       __ jmp(L_exit);
2048     }
2049     restore_arg_regs();
2050     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2051     __ xorptr(rax, rax); // return 0
2052     __ leave(); // required for proper stackwalking of RuntimeStub frame
2053     __ ret(0);
2054 
2055     // Copy in multi-bytes chunks
2056     copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2057 
2058   __ BIND(L_exit);
2059     if (is_oop) {
2060       gen_write_ref_array_post_barrier(to, dword_count, rax);
2061     }
2062     restore_arg_regs();
2063     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2064     __ xorptr(rax, rax); // return 0
2065     __ leave(); // required for proper stackwalking of RuntimeStub frame
2066     __ ret(0);
2067 
2068     return start;
2069   }
2070 
2071   // Arguments:
2072   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2073   //             ignored
2074   //   is_oop  - true => oop array, so generate store check code
2075   //   name    - stub name string
2076   //
2077   // Inputs:
2078   //   c_rarg0   - source array address
2079   //   c_rarg1   - destination array address
2080   //   c_rarg2   - element count, treated as ssize_t, can be zero
2081   //
2082  // Side Effects:
2083   //   disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2084   //   no-overlap entry point used by generate_conjoint_long_oop_copy().
2085   //
2086   address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2087                                           const char *name, bool dest_uninitialized = false) {
2088     __ align(CodeEntryAlignment);
2089     StubCodeMark mark(this, "StubRoutines", name);
2090     address start = __ pc();
2091 
2092     Label L_copy_bytes, L_copy_8_bytes, L_exit;
2093     const Register from        = rdi;  // source array address
2094     const Register to          = rsi;  // destination array address
2095     const Register qword_count = rdx;  // elements count
2096     const Register end_from    = from; // source array end address
2097     const Register end_to      = rcx;  // destination array end address
2098     const Register saved_to    = to;
2099     const Register saved_count = r11;
2100     // End pointers are inclusive, and if count is not zero they point
2101     // to the last unit copied:  end_to[0] := end_from[0]
2102 
2103     __ enter(); // required for proper stackwalking of RuntimeStub frame
2104     // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2105     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
2106 
2107     if (entry != NULL) {
2108       *entry = __ pc();
2109       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2110       BLOCK_COMMENT("Entry:");
2111     }
2112 
2113     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2114                       // r9 and r10 may be used to save non-volatile registers
2115     // 'from', 'to' and 'qword_count' are now valid
2116     if (is_oop) {
2117       // Save to and count for store barrier
2118       __ movptr(saved_count, qword_count);
2119       // no registers are destroyed by this call
2120       gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2121     }
2122 
2123     // Copy from low to high addresses.  Use 'to' as scratch.
2124     __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2125     __ lea(end_to,   Address(to,   qword_count, Address::times_8, -8));
2126     __ negptr(qword_count);
2127     __ jmp(L_copy_bytes);
2128 
2129     // Copy trailing qwords
2130   __ BIND(L_copy_8_bytes);
2131     __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2132     __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2133     __ increment(qword_count);
2134     __ jcc(Assembler::notZero, L_copy_8_bytes);
2135 
2136     if (is_oop) {
2137       __ jmp(L_exit);
2138     } else {
2139       restore_arg_regs();
2140       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2141       __ xorptr(rax, rax); // return 0
2142       __ leave(); // required for proper stackwalking of RuntimeStub frame
2143       __ ret(0);
2144     }
2145 
2146     // Copy in multi-bytes chunks
2147     copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2148 
2149     if (is_oop) {
2150     __ BIND(L_exit);
2151       gen_write_ref_array_post_barrier(saved_to, saved_count, rax);
2152     }
2153     restore_arg_regs();
2154     if (is_oop) {
2155       inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2156     } else {
2157       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2158     }
2159     __ xorptr(rax, rax); // return 0
2160     __ leave(); // required for proper stackwalking of RuntimeStub frame
2161     __ ret(0);
2162 
2163     return start;
2164   }
2165 
2166   // Arguments:
2167   //   aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2168   //             ignored
2169   //   is_oop  - true => oop array, so generate store check code
2170   //   name    - stub name string
2171   //
2172   // Inputs:
2173   //   c_rarg0   - source array address
2174   //   c_rarg1   - destination array address
2175   //   c_rarg2   - element count, treated as ssize_t, can be zero
2176   //
2177   address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2178                                           address nooverlap_target, address *entry,
2179                                           const char *name, bool dest_uninitialized = false) {
2180     __ align(CodeEntryAlignment);
2181     StubCodeMark mark(this, "StubRoutines", name);
2182     address start = __ pc();
2183 
2184     Label L_copy_bytes, L_copy_8_bytes, L_exit;
2185     const Register from        = rdi;  // source array address
2186     const Register to          = rsi;  // destination array address
2187     const Register qword_count = rdx;  // elements count
2188     const Register saved_count = rcx;
2189 
2190     __ enter(); // required for proper stackwalking of RuntimeStub frame
2191     assert_clean_int(c_rarg2, rax);    // Make sure 'count' is clean int.
2192 
2193     if (entry != NULL) {
2194       *entry = __ pc();
2195       // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2196       BLOCK_COMMENT("Entry:");
2197     }
2198 
2199     array_overlap_test(nooverlap_target, Address::times_8);
2200     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2201                       // r9 and r10 may be used to save non-volatile registers
2202     // 'from', 'to' and 'qword_count' are now valid
2203     if (is_oop) {
2204       // Save to and count for store barrier
2205       __ movptr(saved_count, qword_count);
2206       // No registers are destroyed by this call
2207       gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2208     }
2209 
2210     __ jmp(L_copy_bytes);
2211 
2212     // Copy trailing qwords
2213   __ BIND(L_copy_8_bytes);
2214     __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2215     __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2216     __ decrement(qword_count);
2217     __ jcc(Assembler::notZero, L_copy_8_bytes);
2218 
2219     if (is_oop) {
2220       __ jmp(L_exit);
2221     } else {
2222       restore_arg_regs();
2223       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2224       __ xorptr(rax, rax); // return 0
2225       __ leave(); // required for proper stackwalking of RuntimeStub frame
2226       __ ret(0);
2227     }
2228 
2229     // Copy in multi-bytes chunks
2230     copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2231 
2232     if (is_oop) {
2233     __ BIND(L_exit);
2234       gen_write_ref_array_post_barrier(to, saved_count, rax);
2235     }
2236     restore_arg_regs();
2237     if (is_oop) {
2238       inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2239     } else {
2240       inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2241     }
2242     __ xorptr(rax, rax); // return 0
2243     __ leave(); // required for proper stackwalking of RuntimeStub frame
2244     __ ret(0);
2245 
2246     return start;
2247   }
2248 
2249 
2250   // Helper for generating a dynamic type check.
2251   // Smashes no registers.
2252   void generate_type_check(Register sub_klass,
2253                            Register super_check_offset,
2254                            Register super_klass,
2255                            Label& L_success) {
2256     assert_different_registers(sub_klass, super_check_offset, super_klass);
2257 
2258     BLOCK_COMMENT("type_check:");
2259 
2260     Label L_miss;
2261 
2262     __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg,        &L_success, &L_miss, NULL,
2263                                      super_check_offset);
2264     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2265 
2266     // Fall through on failure!
2267     __ BIND(L_miss);
2268   }
2269 
2270   //
2271   //  Generate checkcasting array copy stub
2272   //
2273   //  Input:
2274   //    c_rarg0   - source array address
2275   //    c_rarg1   - destination array address
2276   //    c_rarg2   - element count, treated as ssize_t, can be zero
2277   //    c_rarg3   - size_t ckoff (super_check_offset)
2278   // not Win64
2279   //    c_rarg4   - oop ckval (super_klass)
2280   // Win64
2281   //    rsp+40    - oop ckval (super_klass)
2282   //
2283   //  Output:
2284   //    rax ==  0  -  success
2285   //    rax == -1^K - failure, where K is partial transfer count
2286   //
2287   address generate_checkcast_copy(const char *name, address *entry,
2288                                   bool dest_uninitialized = false) {
2289 
2290     Label L_load_element, L_store_element, L_do_card_marks, L_done;
2291 
2292     // Input registers (after setup_arg_regs)
2293     const Register from        = rdi;   // source array address
2294     const Register to          = rsi;   // destination array address
2295     const Register length      = rdx;   // elements count
2296     const Register ckoff       = rcx;   // super_check_offset
2297     const Register ckval       = r8;    // super_klass
2298 
2299     // Registers used as temps (r13, r14 are save-on-entry)
2300     const Register end_from    = from;  // source array end address
2301     const Register end_to      = r13;   // destination array end address
2302     const Register count       = rdx;   // -(count_remaining)
2303     const Register r14_length  = r14;   // saved copy of length
2304     // End pointers are inclusive, and if length is not zero they point
2305     // to the last unit copied:  end_to[0] := end_from[0]
2306 
2307     const Register rax_oop    = rax;    // actual oop copied
2308     const Register r11_klass  = r11;    // oop._klass
2309 
2310     //---------------------------------------------------------------
2311     // Assembler stub will be used for this call to arraycopy
2312     // if the two arrays are subtypes of Object[] but the
2313     // destination array type is not equal to or a supertype
2314     // of the source type.  Each element must be separately
2315     // checked.
2316 
2317     __ align(CodeEntryAlignment);
2318     StubCodeMark mark(this, "StubRoutines", name);
2319     address start = __ pc();
2320 
2321     __ enter(); // required for proper stackwalking of RuntimeStub frame
2322 
2323 #ifdef ASSERT
2324     // caller guarantees that the arrays really are different
2325     // otherwise, we would have to make conjoint checks
2326     { Label L;
2327       array_overlap_test(L, TIMES_OOP);
2328       __ stop("checkcast_copy within a single array");
2329       __ bind(L);
2330     }
2331 #endif //ASSERT
2332 
2333     setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2334                        // ckoff => rcx, ckval => r8
2335                        // r9 and r10 may be used to save non-volatile registers
2336 #ifdef _WIN64
2337     // last argument (#4) is on stack on Win64
2338     __ movptr(ckval, Address(rsp, 6 * wordSize));
2339 #endif
2340 
2341     // Caller of this entry point must set up the argument registers.
2342     if (entry != NULL) {
2343       *entry = __ pc();
2344       BLOCK_COMMENT("Entry:");
2345     }
2346 
2347     // allocate spill slots for r13, r14
2348     enum {
2349       saved_r13_offset,
2350       saved_r14_offset,
2351       saved_rbp_offset
2352     };
2353     __ subptr(rsp, saved_rbp_offset * wordSize);
2354     __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2355     __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2356 
2357     // check that int operands are properly extended to size_t
2358     assert_clean_int(length, rax);
2359     assert_clean_int(ckoff, rax);
2360 
2361 #ifdef ASSERT
2362     BLOCK_COMMENT("assert consistent ckoff/ckval");
2363     // The ckoff and ckval must be mutually consistent,
2364     // even though caller generates both.
2365     { Label L;
2366       int sco_offset = in_bytes(Klass::super_check_offset_offset());
2367       __ cmpl(ckoff, Address(ckval, sco_offset));
2368       __ jcc(Assembler::equal, L);
2369       __ stop("super_check_offset inconsistent");
2370       __ bind(L);
2371     }
2372 #endif //ASSERT
2373 
2374     // Loop-invariant addresses.  They are exclusive end pointers.
2375     Address end_from_addr(from, length, TIMES_OOP, 0);
2376     Address   end_to_addr(to,   length, TIMES_OOP, 0);
2377     // Loop-variant addresses.  They assume post-incremented count < 0.
2378     Address from_element_addr(end_from, count, TIMES_OOP, 0);
2379     Address   to_element_addr(end_to,   count, TIMES_OOP, 0);
2380 
2381     gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2382 
2383     // Copy from low to high addresses, indexed from the end of each array.
2384     __ lea(end_from, end_from_addr);
2385     __ lea(end_to,   end_to_addr);
2386     __ movptr(r14_length, length);        // save a copy of the length
2387     assert(length == count, "");          // else fix next line:
2388     __ negptr(count);                     // negate and test the length
2389     __ jcc(Assembler::notZero, L_load_element);
2390 
2391     // Empty array:  Nothing to do.
2392     __ xorptr(rax, rax);                  // return 0 on (trivial) success
2393     __ jmp(L_done);
2394 
2395     // ======== begin loop ========
2396     // (Loop is rotated; its entry is L_load_element.)
2397     // Loop control:
2398     //   for (count = -count; count != 0; count++)
2399     // Base pointers src, dst are biased by 8*(count-1),to last element.
2400     __ align(OptoLoopAlignment);
2401 
2402     __ BIND(L_store_element);
2403     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop
2404     __ increment(count);               // increment the count toward zero
2405     __ jcc(Assembler::zero, L_do_card_marks);
2406 
2407     // ======== loop entry is here ========
2408     __ BIND(L_load_element);
2409     __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2410     __ testptr(rax_oop, rax_oop);
2411     __ jcc(Assembler::zero, L_store_element);
2412 
2413     __ load_klass(r11_klass, rax_oop);// query the object klass
2414     generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2415     // ======== end loop ========
2416 
2417     // It was a real error; we must depend on the caller to finish the job.
2418     // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2419     // Emit GC store barriers for the oops we have copied (r14 + rdx),
2420     // and report their number to the caller.
2421     assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2422     Label L_post_barrier;
2423     __ addptr(r14_length, count);     // K = (original - remaining) oops
2424     __ movptr(rax, r14_length);       // save the value
2425     __ notptr(rax);                   // report (-1^K) to caller (does not affect flags)
2426     __ jccb(Assembler::notZero, L_post_barrier);
2427     __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2428 
2429     // Come here on success only.
2430     __ BIND(L_do_card_marks);
2431     __ xorptr(rax, rax);              // return 0 on success
2432 
2433     __ BIND(L_post_barrier);
2434     gen_write_ref_array_post_barrier(to, r14_length, rscratch1);
2435 
2436     // Common exit point (success or failure).
2437     __ BIND(L_done);
2438     __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2439     __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2440     restore_arg_regs();
2441     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2442     __ leave(); // required for proper stackwalking of RuntimeStub frame
2443     __ ret(0);
2444 
2445     return start;
2446   }
2447 
2448   //
2449   //  Generate 'unsafe' array copy stub
2450   //  Though just as safe as the other stubs, it takes an unscaled
2451   //  size_t argument instead of an element count.
2452   //
2453   //  Input:
2454   //    c_rarg0   - source array address
2455   //    c_rarg1   - destination array address
2456   //    c_rarg2   - byte count, treated as ssize_t, can be zero
2457   //
2458   // Examines the alignment of the operands and dispatches
2459   // to a long, int, short, or byte copy loop.
2460   //
2461   address generate_unsafe_copy(const char *name,
2462                                address byte_copy_entry, address short_copy_entry,
2463                                address int_copy_entry, address long_copy_entry) {
2464 
2465     Label L_long_aligned, L_int_aligned, L_short_aligned;
2466 
2467     // Input registers (before setup_arg_regs)
2468     const Register from        = c_rarg0;  // source array address
2469     const Register to          = c_rarg1;  // destination array address
2470     const Register size        = c_rarg2;  // byte count (size_t)
2471 
2472     // Register used as a temp
2473     const Register bits        = rax;      // test copy of low bits
2474 
2475     __ align(CodeEntryAlignment);
2476     StubCodeMark mark(this, "StubRoutines", name);
2477     address start = __ pc();
2478 
2479     __ enter(); // required for proper stackwalking of RuntimeStub frame
2480 
2481     // bump this on entry, not on exit:
2482     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2483 
2484     __ mov(bits, from);
2485     __ orptr(bits, to);
2486     __ orptr(bits, size);
2487 
2488     __ testb(bits, BytesPerLong-1);
2489     __ jccb(Assembler::zero, L_long_aligned);
2490 
2491     __ testb(bits, BytesPerInt-1);
2492     __ jccb(Assembler::zero, L_int_aligned);
2493 
2494     __ testb(bits, BytesPerShort-1);
2495     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2496 
2497     __ BIND(L_short_aligned);
2498     __ shrptr(size, LogBytesPerShort); // size => short_count
2499     __ jump(RuntimeAddress(short_copy_entry));
2500 
2501     __ BIND(L_int_aligned);
2502     __ shrptr(size, LogBytesPerInt); // size => int_count
2503     __ jump(RuntimeAddress(int_copy_entry));
2504 
2505     __ BIND(L_long_aligned);
2506     __ shrptr(size, LogBytesPerLong); // size => qword_count
2507     __ jump(RuntimeAddress(long_copy_entry));
2508 
2509     return start;
2510   }
2511 
2512   // Perform range checks on the proposed arraycopy.
2513   // Kills temp, but nothing else.
2514   // Also, clean the sign bits of src_pos and dst_pos.
2515   void arraycopy_range_checks(Register src,     // source array oop (c_rarg0)
2516                               Register src_pos, // source position (c_rarg1)
2517                               Register dst,     // destination array oo (c_rarg2)
2518                               Register dst_pos, // destination position (c_rarg3)
2519                               Register length,
2520                               Register temp,
2521                               Label& L_failed) {
2522     BLOCK_COMMENT("arraycopy_range_checks:");
2523 
2524     //  if (src_pos + length > arrayOop(src)->length())  FAIL;
2525     __ movl(temp, length);
2526     __ addl(temp, src_pos);             // src_pos + length
2527     __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2528     __ jcc(Assembler::above, L_failed);
2529 
2530     //  if (dst_pos + length > arrayOop(dst)->length())  FAIL;
2531     __ movl(temp, length);
2532     __ addl(temp, dst_pos);             // dst_pos + length
2533     __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2534     __ jcc(Assembler::above, L_failed);
2535 
2536     // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2537     // Move with sign extension can be used since they are positive.
2538     __ movslq(src_pos, src_pos);
2539     __ movslq(dst_pos, dst_pos);
2540 
2541     BLOCK_COMMENT("arraycopy_range_checks done");
2542   }
2543 
2544   //
2545   //  Generate generic array copy stubs
2546   //
2547   //  Input:
2548   //    c_rarg0    -  src oop
2549   //    c_rarg1    -  src_pos (32-bits)
2550   //    c_rarg2    -  dst oop
2551   //    c_rarg3    -  dst_pos (32-bits)
2552   // not Win64
2553   //    c_rarg4    -  element count (32-bits)
2554   // Win64
2555   //    rsp+40     -  element count (32-bits)
2556   //
2557   //  Output:
2558   //    rax ==  0  -  success
2559   //    rax == -1^K - failure, where K is partial transfer count
2560   //
2561   address generate_generic_copy(const char *name,
2562                                 address byte_copy_entry, address short_copy_entry,
2563                                 address int_copy_entry, address oop_copy_entry,
2564                                 address long_copy_entry, address checkcast_copy_entry) {
2565 
2566     Label L_failed, L_failed_0, L_objArray;
2567     Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2568 
2569     // Input registers
2570     const Register src        = c_rarg0;  // source array oop
2571     const Register src_pos    = c_rarg1;  // source position
2572     const Register dst        = c_rarg2;  // destination array oop
2573     const Register dst_pos    = c_rarg3;  // destination position
2574 #ifndef _WIN64
2575     const Register length     = c_rarg4;
2576 #else
2577     const Address  length(rsp, 6 * wordSize);  // elements count is on stack on Win64
2578 #endif
2579 
2580     { int modulus = CodeEntryAlignment;
2581       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
2582       int advance = target - (__ offset() % modulus);
2583       if (advance < 0)  advance += modulus;
2584       if (advance > 0)  __ nop(advance);
2585     }
2586     StubCodeMark mark(this, "StubRoutines", name);
2587 
2588     // Short-hop target to L_failed.  Makes for denser prologue code.
2589     __ BIND(L_failed_0);
2590     __ jmp(L_failed);
2591     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2592 
2593     __ align(CodeEntryAlignment);
2594     address start = __ pc();
2595 
2596     __ enter(); // required for proper stackwalking of RuntimeStub frame
2597 
2598     // bump this on entry, not on exit:
2599     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2600 
2601     //-----------------------------------------------------------------------
2602     // Assembler stub will be used for this call to arraycopy
2603     // if the following conditions are met:
2604     //
2605     // (1) src and dst must not be null.
2606     // (2) src_pos must not be negative.
2607     // (3) dst_pos must not be negative.
2608     // (4) length  must not be negative.
2609     // (5) src klass and dst klass should be the same and not NULL.
2610     // (6) src and dst should be arrays.
2611     // (7) src_pos + length must not exceed length of src.
2612     // (8) dst_pos + length must not exceed length of dst.
2613     //
2614 
2615     //  if (src == NULL) return -1;
2616     __ testptr(src, src);         // src oop
2617     size_t j1off = __ offset();
2618     __ jccb(Assembler::zero, L_failed_0);
2619 
2620     //  if (src_pos < 0) return -1;
2621     __ testl(src_pos, src_pos); // src_pos (32-bits)
2622     __ jccb(Assembler::negative, L_failed_0);
2623 
2624     //  if (dst == NULL) return -1;
2625     __ testptr(dst, dst);         // dst oop
2626     __ jccb(Assembler::zero, L_failed_0);
2627 
2628     //  if (dst_pos < 0) return -1;
2629     __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2630     size_t j4off = __ offset();
2631     __ jccb(Assembler::negative, L_failed_0);
2632 
2633     // The first four tests are very dense code,
2634     // but not quite dense enough to put four
2635     // jumps in a 16-byte instruction fetch buffer.
2636     // That's good, because some branch predicters
2637     // do not like jumps so close together.
2638     // Make sure of this.
2639     guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2640 
2641     // registers used as temp
2642     const Register r11_length    = r11; // elements count to copy
2643     const Register r10_src_klass = r10; // array klass
2644 
2645     //  if (length < 0) return -1;
2646     __ movl(r11_length, length);        // length (elements count, 32-bits value)
2647     __ testl(r11_length, r11_length);
2648     __ jccb(Assembler::negative, L_failed_0);
2649 
2650     __ load_klass(r10_src_klass, src);
2651 #ifdef ASSERT
2652     //  assert(src->klass() != NULL);
2653     {
2654       BLOCK_COMMENT("assert klasses not null {");
2655       Label L1, L2;
2656       __ testptr(r10_src_klass, r10_src_klass);
2657       __ jcc(Assembler::notZero, L2);   // it is broken if klass is NULL
2658       __ bind(L1);
2659       __ stop("broken null klass");
2660       __ bind(L2);
2661       __ load_klass(rax, dst);
2662       __ cmpq(rax, 0);
2663       __ jcc(Assembler::equal, L1);     // this would be broken also
2664       BLOCK_COMMENT("} assert klasses not null done");
2665     }
2666 #endif
2667 
2668     // Load layout helper (32-bits)
2669     //
2670     //  |array_tag|     | header_size | element_type |     |log2_element_size|
2671     // 32        30    24            16              8     2                 0
2672     //
2673     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2674     //
2675 
2676     const int lh_offset = in_bytes(Klass::layout_helper_offset());
2677 
2678     // Handle objArrays completely differently...
2679     const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2680     __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2681     __ jcc(Assembler::equal, L_objArray);
2682 
2683     //  if (src->klass() != dst->klass()) return -1;
2684     __ load_klass(rax, dst);
2685     __ cmpq(r10_src_klass, rax);
2686     __ jcc(Assembler::notEqual, L_failed);
2687 
2688     const Register rax_lh = rax;  // layout helper
2689     __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2690 
2691     //  if (!src->is_Array()) return -1;
2692     __ cmpl(rax_lh, Klass::_lh_neutral_value);
2693     __ jcc(Assembler::greaterEqual, L_failed);
2694 
2695     // At this point, it is known to be a typeArray (array_tag 0x3).
2696 #ifdef ASSERT
2697     {
2698       BLOCK_COMMENT("assert primitive array {");
2699       Label L;
2700       __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2701       __ jcc(Assembler::greaterEqual, L);
2702       __ stop("must be a primitive array");
2703       __ bind(L);
2704       BLOCK_COMMENT("} assert primitive array done");
2705     }
2706 #endif
2707 
2708     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2709                            r10, L_failed);
2710 
2711     // TypeArrayKlass
2712     //
2713     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2714     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2715     //
2716 
2717     const Register r10_offset = r10;    // array offset
2718     const Register rax_elsize = rax_lh; // element size
2719 
2720     __ movl(r10_offset, rax_lh);
2721     __ shrl(r10_offset, Klass::_lh_header_size_shift);
2722     __ andptr(r10_offset, Klass::_lh_header_size_mask);   // array_offset
2723     __ addptr(src, r10_offset);           // src array offset
2724     __ addptr(dst, r10_offset);           // dst array offset
2725     BLOCK_COMMENT("choose copy loop based on element size");
2726     __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2727 
2728     // next registers should be set before the jump to corresponding stub
2729     const Register from     = c_rarg0;  // source array address
2730     const Register to       = c_rarg1;  // destination array address
2731     const Register count    = c_rarg2;  // elements count
2732 
2733     // 'from', 'to', 'count' registers should be set in such order
2734     // since they are the same as 'src', 'src_pos', 'dst'.
2735 
2736   __ BIND(L_copy_bytes);
2737     __ cmpl(rax_elsize, 0);
2738     __ jccb(Assembler::notEqual, L_copy_shorts);
2739     __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2740     __ lea(to,   Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2741     __ movl2ptr(count, r11_length); // length
2742     __ jump(RuntimeAddress(byte_copy_entry));
2743 
2744   __ BIND(L_copy_shorts);
2745     __ cmpl(rax_elsize, LogBytesPerShort);
2746     __ jccb(Assembler::notEqual, L_copy_ints);
2747     __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2748     __ lea(to,   Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2749     __ movl2ptr(count, r11_length); // length
2750     __ jump(RuntimeAddress(short_copy_entry));
2751 
2752   __ BIND(L_copy_ints);
2753     __ cmpl(rax_elsize, LogBytesPerInt);
2754     __ jccb(Assembler::notEqual, L_copy_longs);
2755     __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2756     __ lea(to,   Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2757     __ movl2ptr(count, r11_length); // length
2758     __ jump(RuntimeAddress(int_copy_entry));
2759 
2760   __ BIND(L_copy_longs);
2761 #ifdef ASSERT
2762     {
2763       BLOCK_COMMENT("assert long copy {");
2764       Label L;
2765       __ cmpl(rax_elsize, LogBytesPerLong);
2766       __ jcc(Assembler::equal, L);
2767       __ stop("must be long copy, but elsize is wrong");
2768       __ bind(L);
2769       BLOCK_COMMENT("} assert long copy done");
2770     }
2771 #endif
2772     __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2773     __ lea(to,   Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2774     __ movl2ptr(count, r11_length); // length
2775     __ jump(RuntimeAddress(long_copy_entry));
2776 
2777     // ObjArrayKlass
2778   __ BIND(L_objArray);
2779     // live at this point:  r10_src_klass, r11_length, src[_pos], dst[_pos]
2780 
2781     Label L_plain_copy, L_checkcast_copy;
2782     //  test array classes for subtyping
2783     __ load_klass(rax, dst);
2784     __ cmpq(r10_src_klass, rax); // usual case is exact equality
2785     __ jcc(Assembler::notEqual, L_checkcast_copy);
2786 
2787     // Identically typed arrays can be copied without element-wise checks.
2788     arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2789                            r10, L_failed);
2790 
2791     __ lea(from, Address(src, src_pos, TIMES_OOP,
2792                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2793     __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2794                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2795     __ movl2ptr(count, r11_length); // length
2796   __ BIND(L_plain_copy);
2797     __ jump(RuntimeAddress(oop_copy_entry));
2798 
2799   __ BIND(L_checkcast_copy);
2800     // live at this point:  r10_src_klass, r11_length, rax (dst_klass)
2801     {
2802       // Before looking at dst.length, make sure dst is also an objArray.
2803       __ cmpl(Address(rax, lh_offset), objArray_lh);
2804       __ jcc(Assembler::notEqual, L_failed);
2805 
2806       // It is safe to examine both src.length and dst.length.
2807       arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2808                              rax, L_failed);
2809 
2810       const Register r11_dst_klass = r11;
2811       __ load_klass(r11_dst_klass, dst); // reload
2812 
2813       // Marshal the base address arguments now, freeing registers.
2814       __ lea(from, Address(src, src_pos, TIMES_OOP,
2815                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2816       __ lea(to,   Address(dst, dst_pos, TIMES_OOP,
2817                    arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2818       __ movl(count, length);           // length (reloaded)
2819       Register sco_temp = c_rarg3;      // this register is free now
2820       assert_different_registers(from, to, count, sco_temp,
2821                                  r11_dst_klass, r10_src_klass);
2822       assert_clean_int(count, sco_temp);
2823 
2824       // Generate the type check.
2825       const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2826       __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2827       assert_clean_int(sco_temp, rax);
2828       generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2829 
2830       // Fetch destination element klass from the ObjArrayKlass header.
2831       int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2832       __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2833       __ movl(  sco_temp,      Address(r11_dst_klass, sco_offset));
2834       assert_clean_int(sco_temp, rax);
2835 
2836       // the checkcast_copy loop needs two extra arguments:
2837       assert(c_rarg3 == sco_temp, "#3 already in place");
2838       // Set up arguments for checkcast_copy_entry.
2839       setup_arg_regs(4);
2840       __ movptr(r8, r11_dst_klass);  // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2841       __ jump(RuntimeAddress(checkcast_copy_entry));
2842     }
2843 
2844   __ BIND(L_failed);
2845     __ xorptr(rax, rax);
2846     __ notptr(rax); // return -1
2847     __ leave();   // required for proper stackwalking of RuntimeStub frame
2848     __ ret(0);
2849 
2850     return start;
2851   }
2852 
2853   void generate_arraycopy_stubs() {
2854     address entry;
2855     address entry_jbyte_arraycopy;
2856     address entry_jshort_arraycopy;
2857     address entry_jint_arraycopy;
2858     address entry_oop_arraycopy;
2859     address entry_jlong_arraycopy;
2860     address entry_checkcast_arraycopy;
2861 
2862     StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, &entry,
2863                                                                            "jbyte_disjoint_arraycopy");
2864     StubRoutines::_jbyte_arraycopy           = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2865                                                                            "jbyte_arraycopy");
2866 
2867     StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2868                                                                             "jshort_disjoint_arraycopy");
2869     StubRoutines::_jshort_arraycopy          = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2870                                                                             "jshort_arraycopy");
2871 
2872     StubRoutines::_jint_disjoint_arraycopy   = generate_disjoint_int_oop_copy(false, false, &entry,
2873                                                                               "jint_disjoint_arraycopy");
2874     StubRoutines::_jint_arraycopy            = generate_conjoint_int_oop_copy(false, false, entry,
2875                                                                               &entry_jint_arraycopy, "jint_arraycopy");
2876 
2877     StubRoutines::_jlong_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, false, &entry,
2878                                                                                "jlong_disjoint_arraycopy");
2879     StubRoutines::_jlong_arraycopy           = generate_conjoint_long_oop_copy(false, false, entry,
2880                                                                                &entry_jlong_arraycopy, "jlong_arraycopy");
2881 
2882 
2883     if (UseCompressedOops) {
2884       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_int_oop_copy(false, true, &entry,
2885                                                                               "oop_disjoint_arraycopy");
2886       StubRoutines::_oop_arraycopy           = generate_conjoint_int_oop_copy(false, true, entry,
2887                                                                               &entry_oop_arraycopy, "oop_arraycopy");
2888       StubRoutines::_oop_disjoint_arraycopy_uninit  = generate_disjoint_int_oop_copy(false, true, &entry,
2889                                                                                      "oop_disjoint_arraycopy_uninit",
2890                                                                                      /*dest_uninitialized*/true);
2891       StubRoutines::_oop_arraycopy_uninit           = generate_conjoint_int_oop_copy(false, true, entry,
2892                                                                                      NULL, "oop_arraycopy_uninit",
2893                                                                                      /*dest_uninitialized*/true);
2894     } else {
2895       StubRoutines::_oop_disjoint_arraycopy  = generate_disjoint_long_oop_copy(false, true, &entry,
2896                                                                                "oop_disjoint_arraycopy");
2897       StubRoutines::_oop_arraycopy           = generate_conjoint_long_oop_copy(false, true, entry,
2898                                                                                &entry_oop_arraycopy, "oop_arraycopy");
2899       StubRoutines::_oop_disjoint_arraycopy_uninit  = generate_disjoint_long_oop_copy(false, true, &entry,
2900                                                                                       "oop_disjoint_arraycopy_uninit",
2901                                                                                       /*dest_uninitialized*/true);
2902       StubRoutines::_oop_arraycopy_uninit           = generate_conjoint_long_oop_copy(false, true, entry,
2903                                                                                       NULL, "oop_arraycopy_uninit",
2904                                                                                       /*dest_uninitialized*/true);
2905     }
2906 
2907     StubRoutines::_checkcast_arraycopy        = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2908     StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2909                                                                         /*dest_uninitialized*/true);
2910 
2911     StubRoutines::_unsafe_arraycopy    = generate_unsafe_copy("unsafe_arraycopy",
2912                                                               entry_jbyte_arraycopy,
2913                                                               entry_jshort_arraycopy,
2914                                                               entry_jint_arraycopy,
2915                                                               entry_jlong_arraycopy);
2916     StubRoutines::_generic_arraycopy   = generate_generic_copy("generic_arraycopy",
2917                                                                entry_jbyte_arraycopy,
2918                                                                entry_jshort_arraycopy,
2919                                                                entry_jint_arraycopy,
2920                                                                entry_oop_arraycopy,
2921                                                                entry_jlong_arraycopy,
2922                                                                entry_checkcast_arraycopy);
2923 
2924     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2925     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2926     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2927     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2928     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2929     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2930 
2931     // We don't generate specialized code for HeapWord-aligned source
2932     // arrays, so just use the code we've already generated
2933     StubRoutines::_arrayof_jbyte_disjoint_arraycopy  = StubRoutines::_jbyte_disjoint_arraycopy;
2934     StubRoutines::_arrayof_jbyte_arraycopy           = StubRoutines::_jbyte_arraycopy;
2935 
2936     StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2937     StubRoutines::_arrayof_jshort_arraycopy          = StubRoutines::_jshort_arraycopy;
2938 
2939     StubRoutines::_arrayof_jint_disjoint_arraycopy   = StubRoutines::_jint_disjoint_arraycopy;
2940     StubRoutines::_arrayof_jint_arraycopy            = StubRoutines::_jint_arraycopy;
2941 
2942     StubRoutines::_arrayof_jlong_disjoint_arraycopy  = StubRoutines::_jlong_disjoint_arraycopy;
2943     StubRoutines::_arrayof_jlong_arraycopy           = StubRoutines::_jlong_arraycopy;
2944 
2945     StubRoutines::_arrayof_oop_disjoint_arraycopy    = StubRoutines::_oop_disjoint_arraycopy;
2946     StubRoutines::_arrayof_oop_arraycopy             = StubRoutines::_oop_arraycopy;
2947 
2948     StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit    = StubRoutines::_oop_disjoint_arraycopy_uninit;
2949     StubRoutines::_arrayof_oop_arraycopy_uninit             = StubRoutines::_oop_arraycopy_uninit;
2950   }
2951 
2952   void generate_math_stubs() {
2953     {
2954       StubCodeMark mark(this, "StubRoutines", "log");
2955       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2956 
2957       __ subq(rsp, 8);
2958       __ movdbl(Address(rsp, 0), xmm0);
2959       __ fld_d(Address(rsp, 0));
2960       __ flog();
2961       __ fstp_d(Address(rsp, 0));
2962       __ movdbl(xmm0, Address(rsp, 0));
2963       __ addq(rsp, 8);
2964       __ ret(0);
2965     }
2966     {
2967       StubCodeMark mark(this, "StubRoutines", "log10");
2968       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2969 
2970       __ subq(rsp, 8);
2971       __ movdbl(Address(rsp, 0), xmm0);
2972       __ fld_d(Address(rsp, 0));
2973       __ flog10();
2974       __ fstp_d(Address(rsp, 0));
2975       __ movdbl(xmm0, Address(rsp, 0));
2976       __ addq(rsp, 8);
2977       __ ret(0);
2978     }
2979     {
2980       StubCodeMark mark(this, "StubRoutines", "sin");
2981       StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2982 
2983       __ subq(rsp, 8);
2984       __ movdbl(Address(rsp, 0), xmm0);
2985       __ fld_d(Address(rsp, 0));
2986       __ trigfunc('s');
2987       __ fstp_d(Address(rsp, 0));
2988       __ movdbl(xmm0, Address(rsp, 0));
2989       __ addq(rsp, 8);
2990       __ ret(0);
2991     }
2992     {
2993       StubCodeMark mark(this, "StubRoutines", "cos");
2994       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2995 
2996       __ subq(rsp, 8);
2997       __ movdbl(Address(rsp, 0), xmm0);
2998       __ fld_d(Address(rsp, 0));
2999       __ trigfunc('c');
3000       __ fstp_d(Address(rsp, 0));
3001       __ movdbl(xmm0, Address(rsp, 0));
3002       __ addq(rsp, 8);
3003       __ ret(0);
3004     }
3005     {
3006       StubCodeMark mark(this, "StubRoutines", "tan");
3007       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
3008 
3009       __ subq(rsp, 8);
3010       __ movdbl(Address(rsp, 0), xmm0);
3011       __ fld_d(Address(rsp, 0));
3012       __ trigfunc('t');
3013       __ fstp_d(Address(rsp, 0));
3014       __ movdbl(xmm0, Address(rsp, 0));
3015       __ addq(rsp, 8);
3016       __ ret(0);
3017     }
3018     {
3019       StubCodeMark mark(this, "StubRoutines", "exp");
3020       StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
3021 
3022       __ subq(rsp, 8);
3023       __ movdbl(Address(rsp, 0), xmm0);
3024       __ fld_d(Address(rsp, 0));
3025       __ exp_with_fallback(0);
3026       __ fstp_d(Address(rsp, 0));
3027       __ movdbl(xmm0, Address(rsp, 0));
3028       __ addq(rsp, 8);
3029       __ ret(0);
3030     }
3031     {
3032       StubCodeMark mark(this, "StubRoutines", "pow");
3033       StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
3034 
3035       __ subq(rsp, 8);
3036       __ movdbl(Address(rsp, 0), xmm1);
3037       __ fld_d(Address(rsp, 0));
3038       __ movdbl(Address(rsp, 0), xmm0);
3039       __ fld_d(Address(rsp, 0));
3040       __ pow_with_fallback(0);
3041       __ fstp_d(Address(rsp, 0));
3042       __ movdbl(xmm0, Address(rsp, 0));
3043       __ addq(rsp, 8);
3044       __ ret(0);
3045     }
3046   }
3047 
3048   // AES intrinsic stubs
3049   enum {AESBlockSize = 16};
3050 
3051   address generate_key_shuffle_mask() {
3052     __ align(16);
3053     StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3054     address start = __ pc();
3055     __ emit_data64( 0x0405060700010203, relocInfo::none );
3056     __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3057     return start;
3058   }
3059 
3060   // Utility routine for loading a 128-bit key word in little endian format
3061   // can optionally specify that the shuffle mask is already in an xmmregister
3062   void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3063     __ movdqu(xmmdst, Address(key, offset));
3064     if (xmm_shuf_mask != NULL) {
3065       __ pshufb(xmmdst, xmm_shuf_mask);
3066     } else {
3067       __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3068     }
3069   }
3070 
3071   // Arguments:
3072   //
3073   // Inputs:
3074   //   c_rarg0   - source byte array address
3075   //   c_rarg1   - destination byte array address
3076   //   c_rarg2   - K (key) in little endian int array
3077   //
3078   address generate_aescrypt_encryptBlock() {
3079     assert(UseAES, "need AES instructions and misaligned SSE support");
3080     __ align(CodeEntryAlignment);
3081     StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3082     Label L_doLast;
3083     address start = __ pc();
3084 
3085     const Register from        = c_rarg0;  // source array address
3086     const Register to          = c_rarg1;  // destination array address
3087     const Register key         = c_rarg2;  // key array address
3088     const Register keylen      = rax;
3089 
3090     const XMMRegister xmm_result = xmm0;
3091     const XMMRegister xmm_key_shuf_mask = xmm1;
3092     // On win64 xmm6-xmm15 must be preserved so don't use them.
3093     const XMMRegister xmm_temp1  = xmm2;
3094     const XMMRegister xmm_temp2  = xmm3;
3095     const XMMRegister xmm_temp3  = xmm4;
3096     const XMMRegister xmm_temp4  = xmm5;
3097 
3098     __ enter(); // required for proper stackwalking of RuntimeStub frame
3099 
3100     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3101     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3102 
3103     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3104     __ movdqu(xmm_result, Address(from, 0));  // get 16 bytes of input
3105 
3106     // For encryption, the java expanded key ordering is just what we need
3107     // we don't know if the key is aligned, hence not using load-execute form
3108 
3109     load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3110     __ pxor(xmm_result, xmm_temp1);
3111 
3112     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3113     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3114     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3115     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3116 
3117     __ aesenc(xmm_result, xmm_temp1);
3118     __ aesenc(xmm_result, xmm_temp2);
3119     __ aesenc(xmm_result, xmm_temp3);
3120     __ aesenc(xmm_result, xmm_temp4);
3121 
3122     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3123     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3124     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3125     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3126 
3127     __ aesenc(xmm_result, xmm_temp1);
3128     __ aesenc(xmm_result, xmm_temp2);
3129     __ aesenc(xmm_result, xmm_temp3);
3130     __ aesenc(xmm_result, xmm_temp4);
3131 
3132     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3133     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3134 
3135     __ cmpl(keylen, 44);
3136     __ jccb(Assembler::equal, L_doLast);
3137 
3138     __ aesenc(xmm_result, xmm_temp1);
3139     __ aesenc(xmm_result, xmm_temp2);
3140 
3141     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3142     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3143 
3144     __ cmpl(keylen, 52);
3145     __ jccb(Assembler::equal, L_doLast);
3146 
3147     __ aesenc(xmm_result, xmm_temp1);
3148     __ aesenc(xmm_result, xmm_temp2);
3149 
3150     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3151     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3152 
3153     __ BIND(L_doLast);
3154     __ aesenc(xmm_result, xmm_temp1);
3155     __ aesenclast(xmm_result, xmm_temp2);
3156     __ movdqu(Address(to, 0), xmm_result);        // store the result
3157     __ xorptr(rax, rax); // return 0
3158     __ leave(); // required for proper stackwalking of RuntimeStub frame
3159     __ ret(0);
3160 
3161     return start;
3162   }
3163 
3164 
3165   // Arguments:
3166   //
3167   // Inputs:
3168   //   c_rarg0   - source byte array address
3169   //   c_rarg1   - destination byte array address
3170   //   c_rarg2   - K (key) in little endian int array
3171   //
3172   address generate_aescrypt_decryptBlock() {
3173     assert(UseAES, "need AES instructions and misaligned SSE support");
3174     __ align(CodeEntryAlignment);
3175     StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3176     Label L_doLast;
3177     address start = __ pc();
3178 
3179     const Register from        = c_rarg0;  // source array address
3180     const Register to          = c_rarg1;  // destination array address
3181     const Register key         = c_rarg2;  // key array address
3182     const Register keylen      = rax;
3183 
3184     const XMMRegister xmm_result = xmm0;
3185     const XMMRegister xmm_key_shuf_mask = xmm1;
3186     // On win64 xmm6-xmm15 must be preserved so don't use them.
3187     const XMMRegister xmm_temp1  = xmm2;
3188     const XMMRegister xmm_temp2  = xmm3;
3189     const XMMRegister xmm_temp3  = xmm4;
3190     const XMMRegister xmm_temp4  = xmm5;
3191 
3192     __ enter(); // required for proper stackwalking of RuntimeStub frame
3193 
3194     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3195     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3196 
3197     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3198     __ movdqu(xmm_result, Address(from, 0));
3199 
3200     // for decryption java expanded key ordering is rotated one position from what we want
3201     // so we start from 0x10 here and hit 0x00 last
3202     // we don't know if the key is aligned, hence not using load-execute form
3203     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3204     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3205     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3206     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3207 
3208     __ pxor  (xmm_result, xmm_temp1);
3209     __ aesdec(xmm_result, xmm_temp2);
3210     __ aesdec(xmm_result, xmm_temp3);
3211     __ aesdec(xmm_result, xmm_temp4);
3212 
3213     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3214     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3215     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3216     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3217 
3218     __ aesdec(xmm_result, xmm_temp1);
3219     __ aesdec(xmm_result, xmm_temp2);
3220     __ aesdec(xmm_result, xmm_temp3);
3221     __ aesdec(xmm_result, xmm_temp4);
3222 
3223     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3224     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3225     load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3226 
3227     __ cmpl(keylen, 44);
3228     __ jccb(Assembler::equal, L_doLast);
3229 
3230     __ aesdec(xmm_result, xmm_temp1);
3231     __ aesdec(xmm_result, xmm_temp2);
3232 
3233     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3234     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3235 
3236     __ cmpl(keylen, 52);
3237     __ jccb(Assembler::equal, L_doLast);
3238 
3239     __ aesdec(xmm_result, xmm_temp1);
3240     __ aesdec(xmm_result, xmm_temp2);
3241 
3242     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3243     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3244 
3245     __ BIND(L_doLast);
3246     __ aesdec(xmm_result, xmm_temp1);
3247     __ aesdec(xmm_result, xmm_temp2);
3248 
3249     // for decryption the aesdeclast operation is always on key+0x00
3250     __ aesdeclast(xmm_result, xmm_temp3);
3251     __ movdqu(Address(to, 0), xmm_result);  // store the result
3252     __ xorptr(rax, rax); // return 0
3253     __ leave(); // required for proper stackwalking of RuntimeStub frame
3254     __ ret(0);
3255 
3256     return start;
3257   }
3258 
3259 
3260   // Arguments:
3261   //
3262   // Inputs:
3263   //   c_rarg0   - source byte array address
3264   //   c_rarg1   - destination byte array address
3265   //   c_rarg2   - K (key) in little endian int array
3266   //   c_rarg3   - r vector byte array address
3267   //   c_rarg4   - input length
3268   //
3269   // Output:
3270   //   rax       - input length
3271   //
3272   address generate_cipherBlockChaining_encryptAESCrypt() {
3273     assert(UseAES, "need AES instructions and misaligned SSE support");
3274     __ align(CodeEntryAlignment);
3275     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3276     address start = __ pc();
3277 
3278     Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3279     const Register from        = c_rarg0;  // source array address
3280     const Register to          = c_rarg1;  // destination array address
3281     const Register key         = c_rarg2;  // key array address
3282     const Register rvec        = c_rarg3;  // r byte array initialized from initvector array address
3283                                            // and left with the results of the last encryption block
3284 #ifndef _WIN64
3285     const Register len_reg     = c_rarg4;  // src len (must be multiple of blocksize 16)
3286 #else
3287     const Address  len_mem(rbp, 6 * wordSize);  // length is on stack on Win64
3288     const Register len_reg     = r10;      // pick the first volatile windows register
3289 #endif
3290     const Register pos         = rax;
3291 
3292     // xmm register assignments for the loops below
3293     const XMMRegister xmm_result = xmm0;
3294     const XMMRegister xmm_temp   = xmm1;
3295     // keys 0-10 preloaded into xmm2-xmm12
3296     const int XMM_REG_NUM_KEY_FIRST = 2;
3297     const int XMM_REG_NUM_KEY_LAST  = 15;
3298     const XMMRegister xmm_key0   = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3299     const XMMRegister xmm_key10  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3300     const XMMRegister xmm_key11  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3301     const XMMRegister xmm_key12  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3302     const XMMRegister xmm_key13  = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3303 
3304     __ enter(); // required for proper stackwalking of RuntimeStub frame
3305 
3306 #ifdef _WIN64
3307     // on win64, fill len_reg from stack position
3308     __ movl(len_reg, len_mem);
3309     // save the xmm registers which must be preserved 6-15
3310     __ subptr(rsp, -rsp_after_call_off * wordSize);
3311     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3312       __ movdqu(xmm_save(i), as_XMMRegister(i));
3313     }
3314 #else
3315     __ push(len_reg); // Save
3316 #endif
3317 
3318     const XMMRegister xmm_key_shuf_mask = xmm_temp;  // used temporarily to swap key bytes up front
3319     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3320     // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3321     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3322       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3323       offset += 0x10;
3324     }
3325     __ movdqu(xmm_result, Address(rvec, 0x00));   // initialize xmm_result with r vec
3326 
3327     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3328     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3329     __ cmpl(rax, 44);
3330     __ jcc(Assembler::notEqual, L_key_192_256);
3331 
3332     // 128 bit code follows here
3333     __ movptr(pos, 0);
3334     __ align(OptoLoopAlignment);
3335 
3336     __ BIND(L_loopTop_128);
3337     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3338     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3339     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3340     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3341       __ aesenc(xmm_result, as_XMMRegister(rnum));
3342     }
3343     __ aesenclast(xmm_result, xmm_key10);
3344     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3345     // no need to store r to memory until we exit
3346     __ addptr(pos, AESBlockSize);
3347     __ subptr(len_reg, AESBlockSize);
3348     __ jcc(Assembler::notEqual, L_loopTop_128);
3349 
3350     __ BIND(L_exit);
3351     __ movdqu(Address(rvec, 0), xmm_result);     // final value of r stored in rvec of CipherBlockChaining object
3352 
3353 #ifdef _WIN64
3354     // restore xmm regs belonging to calling function
3355     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3356       __ movdqu(as_XMMRegister(i), xmm_save(i));
3357     }
3358     __ movl(rax, len_mem);
3359 #else
3360     __ pop(rax); // return length
3361 #endif
3362     __ leave(); // required for proper stackwalking of RuntimeStub frame
3363     __ ret(0);
3364 
3365     __ BIND(L_key_192_256);
3366     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3367     load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3368     load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3369     __ cmpl(rax, 52);
3370     __ jcc(Assembler::notEqual, L_key_256);
3371 
3372     // 192-bit code follows here (could be changed to use more xmm registers)
3373     __ movptr(pos, 0);
3374     __ align(OptoLoopAlignment);
3375 
3376     __ BIND(L_loopTop_192);
3377     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3378     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3379     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3380     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3381       __ aesenc(xmm_result, as_XMMRegister(rnum));
3382     }
3383     __ aesenclast(xmm_result, xmm_key12);
3384     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3385     // no need to store r to memory until we exit
3386     __ addptr(pos, AESBlockSize);
3387     __ subptr(len_reg, AESBlockSize);
3388     __ jcc(Assembler::notEqual, L_loopTop_192);
3389     __ jmp(L_exit);
3390 
3391     __ BIND(L_key_256);
3392     // 256-bit code follows here (could be changed to use more xmm registers)
3393     load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3394     __ movptr(pos, 0);
3395     __ align(OptoLoopAlignment);
3396 
3397     __ BIND(L_loopTop_256);
3398     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
3399     __ pxor  (xmm_result, xmm_temp);               // xor with the current r vector
3400     __ pxor  (xmm_result, xmm_key0);               // do the aes rounds
3401     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3402       __ aesenc(xmm_result, as_XMMRegister(rnum));
3403     }
3404     load_key(xmm_temp, key, 0xe0);
3405     __ aesenclast(xmm_result, xmm_temp);
3406     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3407     // no need to store r to memory until we exit
3408     __ addptr(pos, AESBlockSize);
3409     __ subptr(len_reg, AESBlockSize);
3410     __ jcc(Assembler::notEqual, L_loopTop_256);
3411     __ jmp(L_exit);
3412 
3413     return start;
3414   }
3415 
3416   // Safefetch stubs.
3417   void generate_safefetch(const char* name, int size, address* entry,
3418                           address* fault_pc, address* continuation_pc) {
3419     // safefetch signatures:
3420     //   int      SafeFetch32(int*      adr, int      errValue);
3421     //   intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3422     //
3423     // arguments:
3424     //   c_rarg0 = adr
3425     //   c_rarg1 = errValue
3426     //
3427     // result:
3428     //   PPC_RET  = *adr or errValue
3429 
3430     StubCodeMark mark(this, "StubRoutines", name);
3431 
3432     // Entry point, pc or function descriptor.
3433     *entry = __ pc();
3434 
3435     // Load *adr into c_rarg1, may fault.
3436     *fault_pc = __ pc();
3437     switch (size) {
3438       case 4:
3439         // int32_t
3440         __ movl(c_rarg1, Address(c_rarg0, 0));
3441         break;
3442       case 8:
3443         // int64_t
3444         __ movq(c_rarg1, Address(c_rarg0, 0));
3445         break;
3446       default:
3447         ShouldNotReachHere();
3448     }
3449 
3450     // return errValue or *adr
3451     *continuation_pc = __ pc();
3452     __ movq(rax, c_rarg1);
3453     __ ret(0);
3454   }
3455 
3456   // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3457   // to hide instruction latency
3458   //
3459   // Arguments:
3460   //
3461   // Inputs:
3462   //   c_rarg0   - source byte array address
3463   //   c_rarg1   - destination byte array address
3464   //   c_rarg2   - K (key) in little endian int array
3465   //   c_rarg3   - r vector byte array address
3466   //   c_rarg4   - input length
3467   //
3468   // Output:
3469   //   rax       - input length
3470   //
3471 
3472   address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3473     assert(UseAES, "need AES instructions and misaligned SSE support");
3474     __ align(CodeEntryAlignment);
3475     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3476     address start = __ pc();
3477 
3478     Label L_exit, L_key_192_256, L_key_256;
3479     Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128;
3480     Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
3481     const Register from        = c_rarg0;  // source array address
3482     const Register to          = c_rarg1;  // destination array address
3483     const Register key         = c_rarg2;  // key array address
3484     const Register rvec        = c_rarg3;  // r byte array initialized from initvector array address
3485                                            // and left with the results of the last encryption block
3486 #ifndef _WIN64
3487     const Register len_reg     = c_rarg4;  // src len (must be multiple of blocksize 16)
3488 #else
3489     const Address  len_mem(rbp, 6 * wordSize);  // length is on stack on Win64
3490     const Register len_reg     = r10;      // pick the first volatile windows register
3491 #endif
3492     const Register pos         = rax;
3493 
3494     // keys 0-10 preloaded into xmm2-xmm12
3495     const int XMM_REG_NUM_KEY_FIRST = 5;
3496     const int XMM_REG_NUM_KEY_LAST  = 15;
3497     const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3498     const XMMRegister xmm_key_last  = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3499 
3500     __ enter(); // required for proper stackwalking of RuntimeStub frame
3501 
3502 #ifdef _WIN64
3503     // on win64, fill len_reg from stack position
3504     __ movl(len_reg, len_mem);
3505     // save the xmm registers which must be preserved 6-15
3506     __ subptr(rsp, -rsp_after_call_off * wordSize);
3507     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3508       __ movdqu(xmm_save(i), as_XMMRegister(i));
3509     }
3510 #else
3511     __ push(len_reg); // Save
3512 #endif
3513 
3514     // the java expanded key ordering is rotated one position from what we want
3515     // so we start from 0x10 here and hit 0x00 last
3516     const XMMRegister xmm_key_shuf_mask = xmm1;  // used temporarily to swap key bytes up front
3517     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3518     // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3519     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3520       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3521       offset += 0x10;
3522     }
3523     load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3524 
3525     const XMMRegister xmm_prev_block_cipher = xmm1;  // holds cipher of previous block
3526 
3527     // registers holding the four results in the parallelized loop
3528     const XMMRegister xmm_result0 = xmm0;
3529     const XMMRegister xmm_result1 = xmm2;
3530     const XMMRegister xmm_result2 = xmm3;
3531     const XMMRegister xmm_result3 = xmm4;
3532 
3533     __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00));   // initialize with initial rvec
3534 
3535     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3536     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3537     __ cmpl(rax, 44);
3538     __ jcc(Assembler::notEqual, L_key_192_256);
3539 
3540 
3541     // 128-bit code follows here, parallelized
3542     __ movptr(pos, 0);
3543     __ align(OptoLoopAlignment);
3544     __ BIND(L_multiBlock_loopTop_128);
3545     __ cmpptr(len_reg, 4*AESBlockSize);           // see if at least 4 blocks left
3546     __ jcc(Assembler::less, L_singleBlock_loopTop_128);
3547 
3548     __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize));   // get next 4 blocks into xmmresult registers
3549     __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize));
3550     __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize));
3551     __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize));
3552 
3553 #define DoFour(opc, src_reg)                    \
3554     __ opc(xmm_result0, src_reg);               \
3555     __ opc(xmm_result1, src_reg);               \
3556     __ opc(xmm_result2, src_reg);               \
3557     __ opc(xmm_result3, src_reg);
3558 
3559     DoFour(pxor, xmm_key_first);
3560     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3561       DoFour(aesdec, as_XMMRegister(rnum));
3562     }
3563     DoFour(aesdeclast, xmm_key_last);
3564     // for each result, xor with the r vector of previous cipher block
3565     __ pxor(xmm_result0, xmm_prev_block_cipher);
3566     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize));
3567     __ pxor(xmm_result1, xmm_prev_block_cipher);
3568     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize));
3569     __ pxor(xmm_result2, xmm_prev_block_cipher);
3570     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize));
3571     __ pxor(xmm_result3, xmm_prev_block_cipher);
3572     __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize));   // this will carry over to next set of blocks
3573 
3574     __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0);     // store 4 results into the next 64 bytes of output
3575     __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1);
3576     __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2);
3577     __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3);
3578 
3579     __ addptr(pos, 4*AESBlockSize);
3580     __ subptr(len_reg, 4*AESBlockSize);
3581     __ jmp(L_multiBlock_loopTop_128);
3582 
3583     // registers used in the non-parallelized loops
3584     // xmm register assignments for the loops below
3585     const XMMRegister xmm_result = xmm0;
3586     const XMMRegister xmm_prev_block_cipher_save = xmm2;
3587     const XMMRegister xmm_key11 = xmm3;
3588     const XMMRegister xmm_key12 = xmm4;
3589     const XMMRegister xmm_temp  = xmm4;
3590 
3591     __ align(OptoLoopAlignment);
3592     __ BIND(L_singleBlock_loopTop_128);
3593     __ cmpptr(len_reg, 0);           // any blocks left??
3594     __ jcc(Assembler::equal, L_exit);
3595     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
3596     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3597     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3598     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3599       __ aesdec(xmm_result, as_XMMRegister(rnum));
3600     }
3601     __ aesdeclast(xmm_result, xmm_key_last);
3602     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3603     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
3604     // no need to store r to memory until we exit
3605     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);              // set up next r vector with cipher input from this block
3606 
3607     __ addptr(pos, AESBlockSize);
3608     __ subptr(len_reg, AESBlockSize);
3609     __ jmp(L_singleBlock_loopTop_128);
3610 
3611 
3612     __ BIND(L_exit);
3613     __ movdqu(Address(rvec, 0), xmm_prev_block_cipher);     // final value of r stored in rvec of CipherBlockChaining object
3614 #ifdef _WIN64
3615     // restore regs belonging to calling function
3616     for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3617       __ movdqu(as_XMMRegister(i), xmm_save(i));
3618     }
3619     __ movl(rax, len_mem);
3620 #else
3621     __ pop(rax); // return length
3622 #endif
3623     __ leave(); // required for proper stackwalking of RuntimeStub frame
3624     __ ret(0);
3625 
3626 
3627     __ BIND(L_key_192_256);
3628     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3629     load_key(xmm_key11, key, 0xb0);
3630     __ cmpl(rax, 52);
3631     __ jcc(Assembler::notEqual, L_key_256);
3632 
3633     // 192-bit code follows here (could be optimized to use parallelism)
3634     load_key(xmm_key12, key, 0xc0);     // 192-bit key goes up to c0
3635     __ movptr(pos, 0);
3636     __ align(OptoLoopAlignment);
3637 
3638     __ BIND(L_singleBlock_loopTop_192);
3639     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
3640     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3641     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3642     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3643       __ aesdec(xmm_result, as_XMMRegister(rnum));
3644     }
3645     __ aesdec(xmm_result, xmm_key11);
3646     __ aesdec(xmm_result, xmm_key12);
3647     __ aesdeclast(xmm_result, xmm_key_last);                    // xmm15 always came from key+0
3648     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3649     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);  // store into the next 16 bytes of output
3650     // no need to store r to memory until we exit
3651     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);  // set up next r vector with cipher input from this block
3652     __ addptr(pos, AESBlockSize);
3653     __ subptr(len_reg, AESBlockSize);
3654     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
3655     __ jmp(L_exit);
3656 
3657     __ BIND(L_key_256);
3658     // 256-bit code follows here (could be optimized to use parallelism)
3659     __ movptr(pos, 0);
3660     __ align(OptoLoopAlignment);
3661 
3662     __ BIND(L_singleBlock_loopTop_256);
3663     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3664     __ movdqa(xmm_prev_block_cipher_save, xmm_result);              // save for next r vector
3665     __ pxor  (xmm_result, xmm_key_first);               // do the aes dec rounds
3666     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3667       __ aesdec(xmm_result, as_XMMRegister(rnum));
3668     }
3669     __ aesdec(xmm_result, xmm_key11);
3670     load_key(xmm_temp, key, 0xc0);
3671     __ aesdec(xmm_result, xmm_temp);
3672     load_key(xmm_temp, key, 0xd0);
3673     __ aesdec(xmm_result, xmm_temp);
3674     load_key(xmm_temp, key, 0xe0);     // 256-bit key goes up to e0
3675     __ aesdec(xmm_result, xmm_temp);
3676     __ aesdeclast(xmm_result, xmm_key_last);          // xmm15 came from key+0
3677     __ pxor  (xmm_result, xmm_prev_block_cipher);               // xor with the current r vector
3678     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);  // store into the next 16 bytes of output
3679     // no need to store r to memory until we exit
3680     __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save);  // set up next r vector with cipher input from this block
3681     __ addptr(pos, AESBlockSize);
3682     __ subptr(len_reg, AESBlockSize);
3683     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
3684     __ jmp(L_exit);
3685 
3686     return start;
3687   }
3688 
3689 
3690   // byte swap x86 long
3691   address generate_ghash_long_swap_mask() {
3692     __ align(CodeEntryAlignment);
3693     StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3694     address start = __ pc();
3695     __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
3696     __ emit_data64(0x0706050403020100, relocInfo::none );
3697   return start;
3698   }
3699 
3700   // byte swap x86 byte array
3701   address generate_ghash_byte_swap_mask() {
3702     __ align(CodeEntryAlignment);
3703     StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3704     address start = __ pc();
3705     __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
3706     __ emit_data64(0x0001020304050607, relocInfo::none );
3707   return start;
3708   }
3709 
3710   /* Single and multi-block ghash operations */
3711   address generate_ghash_processBlocks() {
3712     __ align(CodeEntryAlignment);
3713     Label L_ghash_loop, L_exit;
3714     StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3715     address start = __ pc();
3716 
3717     const Register state        = c_rarg0;
3718     const Register subkeyH      = c_rarg1;
3719     const Register data         = c_rarg2;
3720     const Register blocks       = c_rarg3;
3721 
3722 #ifdef _WIN64
3723     const int XMM_REG_LAST  = 10;
3724 #endif
3725 
3726     const XMMRegister xmm_temp0 = xmm0;
3727     const XMMRegister xmm_temp1 = xmm1;
3728     const XMMRegister xmm_temp2 = xmm2;
3729     const XMMRegister xmm_temp3 = xmm3;
3730     const XMMRegister xmm_temp4 = xmm4;
3731     const XMMRegister xmm_temp5 = xmm5;
3732     const XMMRegister xmm_temp6 = xmm6;
3733     const XMMRegister xmm_temp7 = xmm7;
3734     const XMMRegister xmm_temp8 = xmm8;
3735     const XMMRegister xmm_temp9 = xmm9;
3736     const XMMRegister xmm_temp10 = xmm10;
3737 
3738     __ enter();
3739 
3740 #ifdef _WIN64
3741     // save the xmm registers which must be preserved 6-10
3742     __ subptr(rsp, -rsp_after_call_off * wordSize);
3743     for (int i = 6; i <= XMM_REG_LAST; i++) {
3744       __ movdqu(xmm_save(i), as_XMMRegister(i));
3745     }
3746 #endif
3747 
3748     __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3749 
3750     __ movdqu(xmm_temp0, Address(state, 0));
3751     __ pshufb(xmm_temp0, xmm_temp10);
3752 
3753 
3754     __ BIND(L_ghash_loop);
3755     __ movdqu(xmm_temp2, Address(data, 0));
3756     __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3757 
3758     __ movdqu(xmm_temp1, Address(subkeyH, 0));
3759     __ pshufb(xmm_temp1, xmm_temp10);
3760 
3761     __ pxor(xmm_temp0, xmm_temp2);
3762 
3763     //
3764     // Multiply with the hash key
3765     //
3766     __ movdqu(xmm_temp3, xmm_temp0);
3767     __ pclmulqdq(xmm_temp3, xmm_temp1, 0);      // xmm3 holds a0*b0
3768     __ movdqu(xmm_temp4, xmm_temp0);
3769     __ pclmulqdq(xmm_temp4, xmm_temp1, 16);     // xmm4 holds a0*b1
3770 
3771     __ movdqu(xmm_temp5, xmm_temp0);
3772     __ pclmulqdq(xmm_temp5, xmm_temp1, 1);      // xmm5 holds a1*b0
3773     __ movdqu(xmm_temp6, xmm_temp0);
3774     __ pclmulqdq(xmm_temp6, xmm_temp1, 17);     // xmm6 holds a1*b1
3775 
3776     __ pxor(xmm_temp4, xmm_temp5);      // xmm4 holds a0*b1 + a1*b0
3777 
3778     __ movdqu(xmm_temp5, xmm_temp4);    // move the contents of xmm4 to xmm5
3779     __ psrldq(xmm_temp4, 8);    // shift by xmm4 64 bits to the right
3780     __ pslldq(xmm_temp5, 8);    // shift by xmm5 64 bits to the left
3781     __ pxor(xmm_temp3, xmm_temp5);
3782     __ pxor(xmm_temp6, xmm_temp4);      // Register pair <xmm6:xmm3> holds the result
3783                                         // of the carry-less multiplication of
3784                                         // xmm0 by xmm1.
3785 
3786     // We shift the result of the multiplication by one bit position
3787     // to the left to cope for the fact that the bits are reversed.
3788     __ movdqu(xmm_temp7, xmm_temp3);
3789     __ movdqu(xmm_temp8, xmm_temp6);
3790     __ pslld(xmm_temp3, 1);
3791     __ pslld(xmm_temp6, 1);
3792     __ psrld(xmm_temp7, 31);
3793     __ psrld(xmm_temp8, 31);
3794     __ movdqu(xmm_temp9, xmm_temp7);
3795     __ pslldq(xmm_temp8, 4);
3796     __ pslldq(xmm_temp7, 4);
3797     __ psrldq(xmm_temp9, 12);
3798     __ por(xmm_temp3, xmm_temp7);
3799     __ por(xmm_temp6, xmm_temp8);
3800     __ por(xmm_temp6, xmm_temp9);
3801 
3802     //
3803     // First phase of the reduction
3804     //
3805     // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
3806     // independently.
3807     __ movdqu(xmm_temp7, xmm_temp3);
3808     __ movdqu(xmm_temp8, xmm_temp3);
3809     __ movdqu(xmm_temp9, xmm_temp3);
3810     __ pslld(xmm_temp7, 31);    // packed right shift shifting << 31
3811     __ pslld(xmm_temp8, 30);    // packed right shift shifting << 30
3812     __ pslld(xmm_temp9, 25);    // packed right shift shifting << 25
3813     __ pxor(xmm_temp7, xmm_temp8);      // xor the shifted versions
3814     __ pxor(xmm_temp7, xmm_temp9);
3815     __ movdqu(xmm_temp8, xmm_temp7);
3816     __ pslldq(xmm_temp7, 12);
3817     __ psrldq(xmm_temp8, 4);
3818     __ pxor(xmm_temp3, xmm_temp7);      // first phase of the reduction complete
3819 
3820     //
3821     // Second phase of the reduction
3822     //
3823     // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
3824     // shift operations.
3825     __ movdqu(xmm_temp2, xmm_temp3);
3826     __ movdqu(xmm_temp4, xmm_temp3);
3827     __ movdqu(xmm_temp5, xmm_temp3);
3828     __ psrld(xmm_temp2, 1);     // packed left shifting >> 1
3829     __ psrld(xmm_temp4, 2);     // packed left shifting >> 2
3830     __ psrld(xmm_temp5, 7);     // packed left shifting >> 7
3831     __ pxor(xmm_temp2, xmm_temp4);      // xor the shifted versions
3832     __ pxor(xmm_temp2, xmm_temp5);
3833     __ pxor(xmm_temp2, xmm_temp8);
3834     __ pxor(xmm_temp3, xmm_temp2);
3835     __ pxor(xmm_temp6, xmm_temp3);      // the result is in xmm6
3836 
3837     __ decrement(blocks);
3838     __ jcc(Assembler::zero, L_exit);
3839     __ movdqu(xmm_temp0, xmm_temp6);
3840     __ addptr(data, 16);
3841     __ jmp(L_ghash_loop);
3842 
3843     __ BIND(L_exit);
3844     __ pshufb(xmm_temp6, xmm_temp10);          // Byte swap 16-byte result
3845     __ movdqu(Address(state, 0), xmm_temp6);   // store the result
3846 
3847 #ifdef _WIN64
3848     // restore xmm regs belonging to calling function
3849     for (int i = 6; i <= XMM_REG_LAST; i++) {
3850       __ movdqu(as_XMMRegister(i), xmm_save(i));
3851     }
3852 #endif
3853     __ leave();
3854     __ ret(0);
3855     return start;
3856   }
3857 
3858   /**
3859    *  Arguments:
3860    *
3861    * Inputs:
3862    *   c_rarg0   - int crc
3863    *   c_rarg1   - byte* buf
3864    *   c_rarg2   - int length
3865    *
3866    * Ouput:
3867    *       rax   - int crc result
3868    */
3869   address generate_updateBytesCRC32() {
3870     assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3871 
3872     __ align(CodeEntryAlignment);
3873     StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3874 
3875     address start = __ pc();
3876     // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3877     // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3878     // rscratch1: r10
3879     const Register crc   = c_rarg0;  // crc
3880     const Register buf   = c_rarg1;  // source java byte array address
3881     const Register len   = c_rarg2;  // length
3882     const Register table = c_rarg3;  // crc_table address (reuse register)
3883     const Register tmp   = r11;
3884     assert_different_registers(crc, buf, len, table, tmp, rax);
3885 
3886     BLOCK_COMMENT("Entry:");
3887     __ enter(); // required for proper stackwalking of RuntimeStub frame
3888 
3889     __ kernel_crc32(crc, buf, len, table, tmp);
3890 
3891     __ movl(rax, crc);
3892     __ leave(); // required for proper stackwalking of RuntimeStub frame
3893     __ ret(0);
3894 
3895     return start;
3896   }
3897 
3898   /**
3899   *  Arguments:
3900   *
3901   * Inputs:
3902   *   c_rarg0   - int crc
3903   *   c_rarg1   - byte* buf
3904   *   c_rarg2   - long length
3905   *   c_rarg3   - table_start - optional (present only when doing a library_calll, 
3906   *              not used by x86 algorithm)
3907   *
3908   * Ouput:
3909   *       rax   - int crc result
3910   */
3911   address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3912       assert(UseCRC32CIntrinsics, "need SSE4_2");
3913       __ align(CodeEntryAlignment);
3914       StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3915       address start = __ pc();
3916       //reg.arg        int#0        int#1        int#2        int#3        int#4        int#5        float regs
3917       //Windows        RCX          RDX          R8           R9           none         none         XMM0..XMM3
3918       //Lin / Sol      RDI          RSI          RDX          RCX          R8           R9           XMM0..XMM7
3919       const Register crc = c_rarg0;  // crc
3920       const Register buf = c_rarg1;  // source java byte array address
3921       const Register len = c_rarg2;  // length
3922       const Register a = rax;
3923       const Register j = r9;
3924       const Register k = r10;
3925       const Register l = r11;
3926 #ifdef _WIN64
3927       const Register y = rdi;
3928       const Register z = rsi;
3929 #else
3930       const Register y = rcx;
3931       const Register z = r8;
3932 #endif
3933       assert_different_registers(crc, buf, len, a, j, k, l, y, z);
3934 
3935       BLOCK_COMMENT("Entry:");
3936       __ enter(); // required for proper stackwalking of RuntimeStub frame
3937 #ifdef _WIN64
3938       __ push(y);
3939       __ push(z);
3940 #endif
3941       __ crc32c_ipl_alg2_alt2(crc, buf, len,
3942                               a, j, k,
3943                               l, y, z,
3944                               c_farg0, c_farg1, c_farg2,
3945                               is_pclmulqdq_supported);
3946       __ movl(rax, crc);
3947 #ifdef _WIN64
3948       __ pop(z);
3949       __ pop(y);
3950 #endif
3951       __ leave(); // required for proper stackwalking of RuntimeStub frame
3952       __ ret(0);
3953 
3954       return start;
3955   }
3956 
3957   /**
3958    *  Arguments:
3959    *
3960    *  Input:
3961    *    c_rarg0   - x address
3962    *    c_rarg1   - x length
3963    *    c_rarg2   - y address
3964    *    c_rarg3   - y lenth
3965    * not Win64
3966    *    c_rarg4   - z address
3967    *    c_rarg5   - z length
3968    * Win64
3969    *    rsp+40    - z address
3970    *    rsp+48    - z length
3971    */
3972   address generate_multiplyToLen() {
3973     __ align(CodeEntryAlignment);
3974     StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3975 
3976     address start = __ pc();
3977     // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3978     // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3979     const Register x     = rdi;
3980     const Register xlen  = rax;
3981     const Register y     = rsi;
3982     const Register ylen  = rcx;
3983     const Register z     = r8;
3984     const Register zlen  = r11;
3985 
3986     // Next registers will be saved on stack in multiply_to_len().
3987     const Register tmp1  = r12;
3988     const Register tmp2  = r13;
3989     const Register tmp3  = r14;
3990     const Register tmp4  = r15;
3991     const Register tmp5  = rbx;
3992 
3993     BLOCK_COMMENT("Entry:");
3994     __ enter(); // required for proper stackwalking of RuntimeStub frame
3995 
3996 #ifndef _WIN64
3997     __ movptr(zlen, r9); // Save r9 in r11 - zlen
3998 #endif
3999     setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
4000                        // ylen => rcx, z => r8, zlen => r11
4001                        // r9 and r10 may be used to save non-volatile registers
4002 #ifdef _WIN64
4003     // last 2 arguments (#4, #5) are on stack on Win64
4004     __ movptr(z, Address(rsp, 6 * wordSize));
4005     __ movptr(zlen, Address(rsp, 7 * wordSize));
4006 #endif
4007 
4008     __ movptr(xlen, rsi);
4009     __ movptr(y,    rdx);
4010     __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
4011 
4012     restore_arg_regs();
4013 
4014     __ leave(); // required for proper stackwalking of RuntimeStub frame
4015     __ ret(0);
4016 
4017     return start;
4018   }
4019 
4020 /**
4021    *  Arguments:
4022    *
4023   //  Input:
4024   //    c_rarg0   - x address
4025   //    c_rarg1   - x length
4026   //    c_rarg2   - z address
4027   //    c_rarg3   - z lenth
4028    *
4029    */
4030   address generate_squareToLen() {
4031 
4032     __ align(CodeEntryAlignment);
4033     StubCodeMark mark(this, "StubRoutines", "squareToLen");
4034 
4035     address start = __ pc();
4036     // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
4037     // Unix:  rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
4038     const Register x      = rdi;
4039     const Register len    = rsi;
4040     const Register z      = r8;
4041     const Register zlen   = rcx;
4042 
4043    const Register tmp1      = r12;
4044    const Register tmp2      = r13;
4045    const Register tmp3      = r14;
4046    const Register tmp4      = r15;
4047    const Register tmp5      = rbx;
4048 
4049     BLOCK_COMMENT("Entry:");
4050     __ enter(); // required for proper stackwalking of RuntimeStub frame
4051 
4052        setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
4053                           // zlen => rcx
4054                           // r9 and r10 may be used to save non-volatile registers
4055     __ movptr(r8, rdx);
4056     __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
4057 
4058     restore_arg_regs();
4059 
4060     __ leave(); // required for proper stackwalking of RuntimeStub frame
4061     __ ret(0);
4062 
4063     return start;
4064   }
4065 
4066    /**
4067    *  Arguments:
4068    *
4069    *  Input:
4070    *    c_rarg0   - out address
4071    *    c_rarg1   - in address
4072    *    c_rarg2   - offset
4073    *    c_rarg3   - len
4074    * not Win64
4075    *    c_rarg4   - k
4076    * Win64
4077    *    rsp+40    - k
4078    */
4079   address generate_mulAdd() {
4080     __ align(CodeEntryAlignment);
4081     StubCodeMark mark(this, "StubRoutines", "mulAdd");
4082 
4083     address start = __ pc();
4084     // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
4085     // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
4086     const Register out     = rdi;
4087     const Register in      = rsi;
4088     const Register offset  = r11;
4089     const Register len     = rcx;
4090     const Register k       = r8;
4091 
4092     // Next registers will be saved on stack in mul_add().
4093     const Register tmp1  = r12;
4094     const Register tmp2  = r13;
4095     const Register tmp3  = r14;
4096     const Register tmp4  = r15;
4097     const Register tmp5  = rbx;
4098 
4099     BLOCK_COMMENT("Entry:");
4100     __ enter(); // required for proper stackwalking of RuntimeStub frame
4101 
4102     setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
4103                        // len => rcx, k => r8
4104                        // r9 and r10 may be used to save non-volatile registers
4105 #ifdef _WIN64
4106     // last argument is on stack on Win64
4107     __ movl(k, Address(rsp, 6 * wordSize));
4108 #endif
4109     __ movptr(r11, rdx);  // move offset in rdx to offset(r11)
4110     __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
4111 
4112     restore_arg_regs();
4113 
4114     __ leave(); // required for proper stackwalking of RuntimeStub frame
4115     __ ret(0);
4116 
4117     return start;
4118   }
4119 
4120 
4121 #undef __
4122 #define __ masm->
4123 
4124   // Continuation point for throwing of implicit exceptions that are
4125   // not handled in the current activation. Fabricates an exception
4126   // oop and initiates normal exception dispatching in this
4127   // frame. Since we need to preserve callee-saved values (currently
4128   // only for C2, but done for C1 as well) we need a callee-saved oop
4129   // map and therefore have to make these stubs into RuntimeStubs
4130   // rather than BufferBlobs.  If the compiler needs all registers to
4131   // be preserved between the fault point and the exception handler
4132   // then it must assume responsibility for that in
4133   // AbstractCompiler::continuation_for_implicit_null_exception or
4134   // continuation_for_implicit_division_by_zero_exception. All other
4135   // implicit exceptions (e.g., NullPointerException or
4136   // AbstractMethodError on entry) are either at call sites or
4137   // otherwise assume that stack unwinding will be initiated, so
4138   // caller saved registers were assumed volatile in the compiler.
4139   address generate_throw_exception(const char* name,
4140                                    address runtime_entry,
4141                                    Register arg1 = noreg,
4142                                    Register arg2 = noreg) {
4143     // Information about frame layout at time of blocking runtime call.
4144     // Note that we only have to preserve callee-saved registers since
4145     // the compilers are responsible for supplying a continuation point
4146     // if they expect all registers to be preserved.
4147     enum layout {
4148       rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
4149       rbp_off2,
4150       return_off,
4151       return_off2,
4152       framesize // inclusive of return address
4153     };
4154 
4155     int insts_size = 512;
4156     int locs_size  = 64;
4157 
4158     CodeBuffer code(name, insts_size, locs_size);
4159     OopMapSet* oop_maps  = new OopMapSet();
4160     MacroAssembler* masm = new MacroAssembler(&code);
4161 
4162     address start = __ pc();
4163 
4164     // This is an inlined and slightly modified version of call_VM
4165     // which has the ability to fetch the return PC out of
4166     // thread-local storage and also sets up last_Java_sp slightly
4167     // differently than the real call_VM
4168 
4169     __ enter(); // required for proper stackwalking of RuntimeStub frame
4170 
4171     assert(is_even(framesize/2), "sp not 16-byte aligned");
4172 
4173     // return address and rbp are already in place
4174     __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
4175 
4176     int frame_complete = __ pc() - start;
4177 
4178     // Set up last_Java_sp and last_Java_fp
4179     address the_pc = __ pc();
4180     __ set_last_Java_frame(rsp, rbp, the_pc);
4181     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
4182 
4183     // Call runtime
4184     if (arg1 != noreg) {
4185       assert(arg2 != c_rarg1, "clobbered");
4186       __ movptr(c_rarg1, arg1);
4187     }
4188     if (arg2 != noreg) {
4189       __ movptr(c_rarg2, arg2);
4190     }
4191     __ movptr(c_rarg0, r15_thread);
4192     BLOCK_COMMENT("call runtime_entry");
4193     __ call(RuntimeAddress(runtime_entry));
4194 
4195     // Generate oop map
4196     OopMap* map = new OopMap(framesize, 0);
4197 
4198     oop_maps->add_gc_map(the_pc - start, map);
4199 
4200     __ reset_last_Java_frame(true, true);
4201 
4202     __ leave(); // required for proper stackwalking of RuntimeStub frame
4203 
4204     // check for pending exceptions
4205 #ifdef ASSERT
4206     Label L;
4207     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
4208             (int32_t) NULL_WORD);
4209     __ jcc(Assembler::notEqual, L);
4210     __ should_not_reach_here();
4211     __ bind(L);
4212 #endif // ASSERT
4213     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
4214 
4215 
4216     // codeBlob framesize is in words (not VMRegImpl::slot_size)
4217     RuntimeStub* stub =
4218       RuntimeStub::new_runtime_stub(name,
4219                                     &code,
4220                                     frame_complete,
4221                                     (framesize >> (LogBytesPerWord - LogBytesPerInt)),
4222                                     oop_maps, false);
4223     return stub->entry_point();
4224   }
4225 
4226   void create_control_words() {
4227     // Round to nearest, 53-bit mode, exceptions masked
4228     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;
4229     // Round to zero, 53-bit mode, exception mased
4230     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
4231     // Round to nearest, 24-bit mode, exceptions masked
4232     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;
4233     // Round to nearest, 64-bit mode, exceptions masked
4234     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;
4235     // Round to nearest, 64-bit mode, exceptions masked
4236     StubRoutines::_mxcsr_std           = 0x1F80;
4237     // Note: the following two constants are 80-bit values
4238     //       layout is critical for correct loading by FPU.
4239     // Bias for strict fp multiply/divide
4240     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
4241     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
4242     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
4243     // Un-Bias for strict fp multiply/divide
4244     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
4245     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
4246     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
4247   }
4248 
4249   // Initialization
4250   void generate_initial() {
4251     // Generates all stubs and initializes the entry points
4252 
4253     // This platform-specific settings are needed by generate_call_stub()
4254     create_control_words();
4255 
4256     // entry points that exist in all platforms Note: This is code
4257     // that could be shared among different platforms - however the
4258     // benefit seems to be smaller than the disadvantage of having a
4259     // much more complicated generator structure. See also comment in
4260     // stubRoutines.hpp.
4261 
4262     StubRoutines::_forward_exception_entry = generate_forward_exception();
4263 
4264     StubRoutines::_call_stub_entry =
4265       generate_call_stub(StubRoutines::_call_stub_return_address);
4266 
4267     // is referenced by megamorphic call
4268     StubRoutines::_catch_exception_entry = generate_catch_exception();
4269 
4270     // atomic calls
4271     StubRoutines::_atomic_xchg_entry         = generate_atomic_xchg();
4272     StubRoutines::_atomic_xchg_ptr_entry     = generate_atomic_xchg_ptr();
4273     StubRoutines::_atomic_cmpxchg_entry      = generate_atomic_cmpxchg();
4274     StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
4275     StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
4276     StubRoutines::_atomic_add_entry          = generate_atomic_add();
4277     StubRoutines::_atomic_add_ptr_entry      = generate_atomic_add_ptr();
4278     StubRoutines::_fence_entry               = generate_orderaccess_fence();
4279 
4280     StubRoutines::_handler_for_unsafe_access_entry =
4281       generate_handler_for_unsafe_access();
4282 
4283     // platform dependent
4284     StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
4285     StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4286 
4287     StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
4288 
4289     // Build this early so it's available for the interpreter.
4290     StubRoutines::_throw_StackOverflowError_entry =
4291       generate_throw_exception("StackOverflowError throw_exception",
4292                                CAST_FROM_FN_PTR(address,
4293                                                 SharedRuntime::
4294                                                 throw_StackOverflowError));
4295     if (UseCRC32Intrinsics) {
4296       // set table address before stub generation which use it
4297       StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4298       StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4299     }
4300 
4301     if (UseCRC32CIntrinsics) {
4302       bool supports_clmul;
4303       StubRoutines::x86::generate_CRC32C_table(supports_clmul = VM_Version::supports_clmul());
4304       StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
4305       StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
4306     }
4307   }
4308 
4309   void generate_all() {
4310     // Generates all stubs and initializes the entry points
4311 
4312     // These entry points require SharedInfo::stack0 to be set up in
4313     // non-core builds and need to be relocatable, so they each
4314     // fabricate a RuntimeStub internally.
4315     StubRoutines::_throw_AbstractMethodError_entry =
4316       generate_throw_exception("AbstractMethodError throw_exception",
4317                                CAST_FROM_FN_PTR(address,
4318                                                 SharedRuntime::
4319                                                 throw_AbstractMethodError));
4320 
4321     StubRoutines::_throw_IncompatibleClassChangeError_entry =
4322       generate_throw_exception("IncompatibleClassChangeError throw_exception",
4323                                CAST_FROM_FN_PTR(address,
4324                                                 SharedRuntime::
4325                                                 throw_IncompatibleClassChangeError));
4326 
4327     StubRoutines::_throw_NullPointerException_at_call_entry =
4328       generate_throw_exception("NullPointerException at call throw_exception",
4329                                CAST_FROM_FN_PTR(address,
4330                                                 SharedRuntime::
4331                                                 throw_NullPointerException_at_call));
4332 
4333     // entry points that are platform specific
4334     StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
4335     StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
4336     StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
4337     StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
4338 
4339     StubRoutines::x86::_float_sign_mask  = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
4340     StubRoutines::x86::_float_sign_flip  = generate_fp_mask("float_sign_flip",  0x8000000080000000);
4341     StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4342     StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
4343 
4344     // support for verify_oop (must happen after universe_init)
4345     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4346 
4347     // arraycopy stubs used by compilers
4348     generate_arraycopy_stubs();
4349 
4350     generate_math_stubs();
4351 
4352     // don't bother generating these AES intrinsic stubs unless global flag is set
4353     if (UseAESIntrinsics) {
4354       StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask();  // needed by the others
4355 
4356       StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
4357       StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
4358       StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
4359       StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
4360     }
4361 
4362     // Generate GHASH intrinsics code
4363     if (UseGHASHIntrinsics) {
4364       StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
4365       StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
4366       StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
4367     }
4368 
4369     // Safefetch stubs.
4370     generate_safefetch("SafeFetch32", sizeof(int),     &StubRoutines::_safefetch32_entry,
4371                                                        &StubRoutines::_safefetch32_fault_pc,
4372                                                        &StubRoutines::_safefetch32_continuation_pc);
4373     generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
4374                                                        &StubRoutines::_safefetchN_fault_pc,
4375                                                        &StubRoutines::_safefetchN_continuation_pc);
4376 #ifdef COMPILER2
4377     if (UseMultiplyToLenIntrinsic) {
4378       StubRoutines::_multiplyToLen = generate_multiplyToLen();
4379     }
4380     if (UseSquareToLenIntrinsic) {
4381       StubRoutines::_squareToLen = generate_squareToLen();
4382     }
4383     if (UseMulAddIntrinsic) {
4384       StubRoutines::_mulAdd = generate_mulAdd();
4385     }
4386 
4387 #ifndef _WINDOWS
4388     if (UseMontgomeryMultiplyIntrinsic) {
4389       StubRoutines::_montgomeryMultiply
4390         = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4391     }
4392     if (UseMontgomerySquareIntrinsic) {
4393       StubRoutines::_montgomerySquare
4394         = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4395     }
4396 #endif // WINDOWS
4397 #endif // COMPILER2
4398   }
4399 
4400  public:
4401   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
4402     if (all) {
4403       generate_all();
4404     } else {
4405       generate_initial();
4406     }
4407   }
4408 }; // end class declaration
4409 
4410 void StubGenerator_generate(CodeBuffer* code, bool all) {
4411   StubGenerator g(code, all);
4412 }