1 /*
   2  * Copyright (c) 1999, 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 a__ ((Assembler*)_masm)->
  52 
  53 #ifdef PRODUCT
  54 #define BLOCK_COMMENT(str) /* nothing */
  55 #else
  56 #define BLOCK_COMMENT(str) __ block_comment(str)
  57 #endif
  58 
  59 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  60 
  61 const int MXCSR_MASK  = 0xFFC0;  // Mask out any pending exceptions
  62 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
  63 
  64 // -------------------------------------------------------------------------------------------------------------------------
  65 // Stub Code definitions
  66 
  67 static address handle_unsafe_access() {
  68   JavaThread* thread = JavaThread::current();
  69   address pc  = thread->saved_exception_pc();
  70   // pc is the instruction which we must emulate
  71   // doing a no-op is fine:  return garbage from the load
  72   // therefore, compute npc
  73   address npc = Assembler::locate_next_instruction(pc);
  74 
  75   // request an async exception
  76   thread->set_pending_unsafe_access_error();
  77 
  78   // return address of next instruction to execute
  79   return npc;
  80 }
  81 
  82 class StubGenerator: public StubCodeGenerator {
  83  private:
  84 
  85 #ifdef PRODUCT
  86 #define inc_counter_np(counter) ((void)0)
  87 #else
  88   void inc_counter_np_(int& counter) {
  89     __ incrementl(ExternalAddress((address)&counter));
  90   }
  91 #define inc_counter_np(counter) \
  92   BLOCK_COMMENT("inc_counter " #counter); \
  93   inc_counter_np_(counter);
  94 #endif //PRODUCT
  95 
  96   void inc_copy_counter_np(BasicType t) {
  97 #ifndef PRODUCT
  98     switch (t) {
  99     case T_BYTE:    inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
 100     case T_SHORT:   inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
 101     case T_INT:     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
 102     case T_LONG:    inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
 103     case T_OBJECT:  inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
 104     }
 105     ShouldNotReachHere();
 106 #endif //PRODUCT
 107   }
 108 
 109   //------------------------------------------------------------------------------------------------------------------------
 110   // Call stubs are used to call Java from C
 111   //
 112   //    [ return_from_Java     ] <--- rsp
 113   //    [ argument word n      ]
 114   //      ...
 115   // -N [ argument word 1      ]
 116   // -7 [ Possible padding for stack alignment ]
 117   // -6 [ Possible padding for stack alignment ]
 118   // -5 [ Possible padding for stack alignment ]
 119   // -4 [ mxcsr save           ] <--- rsp_after_call
 120   // -3 [ saved rbx,            ]
 121   // -2 [ saved rsi            ]
 122   // -1 [ saved rdi            ]
 123   //  0 [ saved rbp,            ] <--- rbp,
 124   //  1 [ return address       ]
 125   //  2 [ ptr. to call wrapper ]
 126   //  3 [ result               ]
 127   //  4 [ result_type          ]
 128   //  5 [ method               ]
 129   //  6 [ entry_point          ]
 130   //  7 [ parameters           ]
 131   //  8 [ parameter_size       ]
 132   //  9 [ thread               ]
 133 
 134 
 135   address generate_call_stub(address& return_address) {
 136     StubCodeMark mark(this, "StubRoutines", "call_stub");
 137     address start = __ pc();
 138 
 139     // stub code parameters / addresses
 140     assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
 141     bool  sse_save = false;
 142     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
 143     const int     locals_count_in_bytes  (4*wordSize);
 144     const Address mxcsr_save    (rbp, -4 * wordSize);
 145     const Address saved_rbx     (rbp, -3 * wordSize);
 146     const Address saved_rsi     (rbp, -2 * wordSize);
 147     const Address saved_rdi     (rbp, -1 * wordSize);
 148     const Address result        (rbp,  3 * wordSize);
 149     const Address result_type   (rbp,  4 * wordSize);
 150     const Address method        (rbp,  5 * wordSize);
 151     const Address entry_point   (rbp,  6 * wordSize);
 152     const Address parameters    (rbp,  7 * wordSize);
 153     const Address parameter_size(rbp,  8 * wordSize);
 154     const Address thread        (rbp,  9 * wordSize); // same as in generate_catch_exception()!
 155     sse_save =  UseSSE > 0;
 156 
 157     // stub code
 158     __ enter();
 159     __ movptr(rcx, parameter_size);              // parameter counter
 160     __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
 161     __ addptr(rcx, locals_count_in_bytes);       // reserve space for register saves
 162     __ subptr(rsp, rcx);
 163     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
 164 
 165     // save rdi, rsi, & rbx, according to C calling conventions
 166     __ movptr(saved_rdi, rdi);
 167     __ movptr(saved_rsi, rsi);
 168     __ movptr(saved_rbx, rbx);
 169 
 170     // provide initial value for required masks
 171     if (UseAVX > 2) {
 172       __ movl(rbx, 0xffff);
 173       __ kmovdl(k1, rbx);
 174     }
 175 
 176     // save and initialize %mxcsr
 177     if (sse_save) {
 178       Label skip_ldmx;
 179       __ stmxcsr(mxcsr_save);
 180       __ movl(rax, mxcsr_save);
 181       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 182       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 183       __ cmp32(rax, mxcsr_std);
 184       __ jcc(Assembler::equal, skip_ldmx);
 185       __ ldmxcsr(mxcsr_std);
 186       __ bind(skip_ldmx);
 187     }
 188 
 189     // make sure the control word is correct.
 190     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 191 
 192 #ifdef ASSERT
 193     // make sure we have no pending exceptions
 194     { Label L;
 195       __ movptr(rcx, thread);
 196       __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 197       __ jcc(Assembler::equal, L);
 198       __ stop("StubRoutines::call_stub: entered with pending exception");
 199       __ bind(L);
 200     }
 201 #endif
 202 
 203     // pass parameters if any
 204     BLOCK_COMMENT("pass parameters if any");
 205     Label parameters_done;
 206     __ movl(rcx, parameter_size);  // parameter counter
 207     __ testl(rcx, rcx);
 208     __ jcc(Assembler::zero, parameters_done);
 209 
 210     // parameter passing loop
 211 
 212     Label loop;
 213     // Copy Java parameters in reverse order (receiver last)
 214     // Note that the argument order is inverted in the process
 215     // source is rdx[rcx: N-1..0]
 216     // dest   is rsp[rbx: 0..N-1]
 217 
 218     __ movptr(rdx, parameters);          // parameter pointer
 219     __ xorptr(rbx, rbx);
 220 
 221     __ BIND(loop);
 222 
 223     // get parameter
 224     __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
 225     __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
 226                     Interpreter::expr_offset_in_bytes(0)), rax);          // store parameter
 227     __ increment(rbx);
 228     __ decrement(rcx);
 229     __ jcc(Assembler::notZero, loop);
 230 
 231     // call Java function
 232     __ BIND(parameters_done);
 233     __ movptr(rbx, method);           // get Method*
 234     __ movptr(rax, entry_point);      // get entry_point
 235     __ mov(rsi, rsp);                 // set sender sp
 236     BLOCK_COMMENT("call Java function");
 237     __ call(rax);
 238 
 239     BLOCK_COMMENT("call_stub_return_address:");
 240     return_address = __ pc();
 241 
 242 #ifdef COMPILER2
 243     {
 244       Label L_skip;
 245       if (UseSSE >= 2) {
 246         __ verify_FPU(0, "call_stub_return");
 247       } else {
 248         for (int i = 1; i < 8; i++) {
 249           __ ffree(i);
 250         }
 251 
 252         // UseSSE <= 1 so double result should be left on TOS
 253         __ movl(rsi, result_type);
 254         __ cmpl(rsi, T_DOUBLE);
 255         __ jcc(Assembler::equal, L_skip);
 256         if (UseSSE == 0) {
 257           // UseSSE == 0 so float result should be left on TOS
 258           __ cmpl(rsi, T_FLOAT);
 259           __ jcc(Assembler::equal, L_skip);
 260         }
 261         __ ffree(0);
 262       }
 263       __ BIND(L_skip);
 264     }
 265 #endif // COMPILER2
 266 
 267     // store result depending on type
 268     // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 269     __ movptr(rdi, result);
 270     Label is_long, is_float, is_double, exit;
 271     __ movl(rsi, result_type);
 272     __ cmpl(rsi, T_LONG);
 273     __ jcc(Assembler::equal, is_long);
 274     __ cmpl(rsi, T_FLOAT);
 275     __ jcc(Assembler::equal, is_float);
 276     __ cmpl(rsi, T_DOUBLE);
 277     __ jcc(Assembler::equal, is_double);
 278 
 279     // handle T_INT case
 280     __ movl(Address(rdi, 0), rax);
 281     __ BIND(exit);
 282 
 283     // check that FPU stack is empty
 284     __ verify_FPU(0, "generate_call_stub");
 285 
 286     // pop parameters
 287     __ lea(rsp, rsp_after_call);
 288 
 289     // restore %mxcsr
 290     if (sse_save) {
 291       __ ldmxcsr(mxcsr_save);
 292     }
 293 
 294     // restore rdi, rsi and rbx,
 295     __ movptr(rbx, saved_rbx);
 296     __ movptr(rsi, saved_rsi);
 297     __ movptr(rdi, saved_rdi);
 298     __ addptr(rsp, 4*wordSize);
 299 
 300     // return
 301     __ pop(rbp);
 302     __ ret(0);
 303 
 304     // handle return types different from T_INT
 305     __ BIND(is_long);
 306     __ movl(Address(rdi, 0 * wordSize), rax);
 307     __ movl(Address(rdi, 1 * wordSize), rdx);
 308     __ jmp(exit);
 309 
 310     __ BIND(is_float);
 311     // interpreter uses xmm0 for return values
 312     if (UseSSE >= 1) {
 313       __ movflt(Address(rdi, 0), xmm0);
 314     } else {
 315       __ fstp_s(Address(rdi, 0));
 316     }
 317     __ jmp(exit);
 318 
 319     __ BIND(is_double);
 320     // interpreter uses xmm0 for return values
 321     if (UseSSE >= 2) {
 322       __ movdbl(Address(rdi, 0), xmm0);
 323     } else {
 324       __ fstp_d(Address(rdi, 0));
 325     }
 326     __ jmp(exit);
 327 
 328     return start;
 329   }
 330 
 331 
 332   //------------------------------------------------------------------------------------------------------------------------
 333   // Return point for a Java call if there's an exception thrown in Java code.
 334   // The exception is caught and transformed into a pending exception stored in
 335   // JavaThread that can be tested from within the VM.
 336   //
 337   // Note: Usually the parameters are removed by the callee. In case of an exception
 338   //       crossing an activation frame boundary, that is not the case if the callee
 339   //       is compiled code => need to setup the rsp.
 340   //
 341   // rax,: exception oop
 342 
 343   address generate_catch_exception() {
 344     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 345     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
 346     const Address thread        (rbp,  9 * wordSize); // same as in generate_call_stub()!
 347     address start = __ pc();
 348 
 349     // get thread directly
 350     __ movptr(rcx, thread);
 351 #ifdef ASSERT
 352     // verify that threads correspond
 353     { Label L;
 354       __ get_thread(rbx);
 355       __ cmpptr(rbx, rcx);
 356       __ jcc(Assembler::equal, L);
 357       __ stop("StubRoutines::catch_exception: threads must correspond");
 358       __ bind(L);
 359     }
 360 #endif
 361     // set pending exception
 362     __ verify_oop(rax);
 363     __ movptr(Address(rcx, Thread::pending_exception_offset()), rax          );
 364     __ lea(Address(rcx, Thread::exception_file_offset   ()),
 365            ExternalAddress((address)__FILE__));
 366     __ movl(Address(rcx, Thread::exception_line_offset   ()), __LINE__ );
 367     // complete return to VM
 368     assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
 369     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 370 
 371     return start;
 372   }
 373 
 374 
 375   //------------------------------------------------------------------------------------------------------------------------
 376   // Continuation point for runtime calls returning with a pending exception.
 377   // The pending exception check happened in the runtime or native call stub.
 378   // The pending exception in Thread is converted into a Java-level exception.
 379   //
 380   // Contract with Java-level exception handlers:
 381   // rax: exception
 382   // rdx: throwing pc
 383   //
 384   // NOTE: At entry of this stub, exception-pc must be on stack !!
 385 
 386   address generate_forward_exception() {
 387     StubCodeMark mark(this, "StubRoutines", "forward exception");
 388     address start = __ pc();
 389     const Register thread = rcx;
 390 
 391     // other registers used in this stub
 392     const Register exception_oop = rax;
 393     const Register handler_addr  = rbx;
 394     const Register exception_pc  = rdx;
 395 
 396     // Upon entry, the sp points to the return address returning into Java
 397     // (interpreted or compiled) code; i.e., the return address becomes the
 398     // throwing pc.
 399     //
 400     // Arguments pushed before the runtime call are still on the stack but
 401     // the exception handler will reset the stack pointer -> ignore them.
 402     // A potential result in registers can be ignored as well.
 403 
 404 #ifdef ASSERT
 405     // make sure this code is only executed if there is a pending exception
 406     { Label L;
 407       __ get_thread(thread);
 408       __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 409       __ jcc(Assembler::notEqual, L);
 410       __ stop("StubRoutines::forward exception: no pending exception (1)");
 411       __ bind(L);
 412     }
 413 #endif
 414 
 415     // compute exception handler into rbx,
 416     __ get_thread(thread);
 417     __ movptr(exception_pc, Address(rsp, 0));
 418     BLOCK_COMMENT("call exception_handler_for_return_address");
 419     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
 420     __ mov(handler_addr, rax);
 421 
 422     // setup rax & rdx, remove return address & clear pending exception
 423     __ get_thread(thread);
 424     __ pop(exception_pc);
 425     __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
 426     __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 427 
 428 #ifdef ASSERT
 429     // make sure exception is set
 430     { Label L;
 431       __ testptr(exception_oop, exception_oop);
 432       __ jcc(Assembler::notEqual, L);
 433       __ stop("StubRoutines::forward exception: no pending exception (2)");
 434       __ bind(L);
 435     }
 436 #endif
 437 
 438     // Verify that there is really a valid exception in RAX.
 439     __ verify_oop(exception_oop);
 440 
 441     // continue at exception handler (return address removed)
 442     // rax: exception
 443     // rbx: exception handler
 444     // rdx: throwing pc
 445     __ jmp(handler_addr);
 446 
 447     return start;
 448   }
 449 
 450 
 451   //----------------------------------------------------------------------------------------------------
 452   // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
 453   //
 454   // xchg exists as far back as 8086, lock needed for MP only
 455   // Stack layout immediately after call:
 456   //
 457   // 0 [ret addr ] <--- rsp
 458   // 1 [  ex     ]
 459   // 2 [  dest   ]
 460   //
 461   // Result:   *dest <- ex, return (old *dest)
 462   //
 463   // Note: win32 does not currently use this code
 464 
 465   address generate_atomic_xchg() {
 466     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 467     address start = __ pc();
 468 
 469     __ push(rdx);
 470     Address exchange(rsp, 2 * wordSize);
 471     Address dest_addr(rsp, 3 * wordSize);
 472     __ movl(rax, exchange);
 473     __ movptr(rdx, dest_addr);
 474     __ xchgl(rax, Address(rdx, 0));
 475     __ pop(rdx);
 476     __ ret(0);
 477 
 478     return start;
 479   }
 480 
 481   //----------------------------------------------------------------------------------------------------
 482   // Support for void verify_mxcsr()
 483   //
 484   // This routine is used with -Xcheck:jni to verify that native
 485   // JNI code does not return to Java code without restoring the
 486   // MXCSR register to our expected state.
 487 
 488 
 489   address generate_verify_mxcsr() {
 490     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 491     address start = __ pc();
 492 
 493     const Address mxcsr_save(rsp, 0);
 494 
 495     if (CheckJNICalls && UseSSE > 0 ) {
 496       Label ok_ret;
 497       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 498       __ push(rax);
 499       __ subptr(rsp, wordSize);      // allocate a temp location
 500       __ stmxcsr(mxcsr_save);
 501       __ movl(rax, mxcsr_save);
 502       __ andl(rax, MXCSR_MASK);
 503       __ cmp32(rax, mxcsr_std);
 504       __ jcc(Assembler::equal, ok_ret);
 505 
 506       __ warn("MXCSR changed by native JNI code.");
 507 
 508       __ ldmxcsr(mxcsr_std);
 509 
 510       __ bind(ok_ret);
 511       __ addptr(rsp, wordSize);
 512       __ pop(rax);
 513     }
 514 
 515     __ ret(0);
 516 
 517     return start;
 518   }
 519 
 520 
 521   //---------------------------------------------------------------------------
 522   // Support for void verify_fpu_cntrl_wrd()
 523   //
 524   // This routine is used with -Xcheck:jni to verify that native
 525   // JNI code does not return to Java code without restoring the
 526   // FP control word to our expected state.
 527 
 528   address generate_verify_fpu_cntrl_wrd() {
 529     StubCodeMark mark(this, "StubRoutines", "verify_spcw");
 530     address start = __ pc();
 531 
 532     const Address fpu_cntrl_wrd_save(rsp, 0);
 533 
 534     if (CheckJNICalls) {
 535       Label ok_ret;
 536       __ push(rax);
 537       __ subptr(rsp, wordSize);      // allocate a temp location
 538       __ fnstcw(fpu_cntrl_wrd_save);
 539       __ movl(rax, fpu_cntrl_wrd_save);
 540       __ andl(rax, FPU_CNTRL_WRD_MASK);
 541       ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
 542       __ cmp32(rax, fpu_std);
 543       __ jcc(Assembler::equal, ok_ret);
 544 
 545       __ warn("Floating point control word changed by native JNI code.");
 546 
 547       __ fldcw(fpu_std);
 548 
 549       __ bind(ok_ret);
 550       __ addptr(rsp, wordSize);
 551       __ pop(rax);
 552     }
 553 
 554     __ ret(0);
 555 
 556     return start;
 557   }
 558 
 559   //---------------------------------------------------------------------------
 560   // Wrapper for slow-case handling of double-to-integer conversion
 561   // d2i or f2i fast case failed either because it is nan or because
 562   // of under/overflow.
 563   // Input:  FPU TOS: float value
 564   // Output: rax, (rdx): integer (long) result
 565 
 566   address generate_d2i_wrapper(BasicType t, address fcn) {
 567     StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
 568     address start = __ pc();
 569 
 570   // Capture info about frame layout
 571   enum layout { FPUState_off         = 0,
 572                 rbp_off              = FPUStateSizeInWords,
 573                 rdi_off,
 574                 rsi_off,
 575                 rcx_off,
 576                 rbx_off,
 577                 saved_argument_off,
 578                 saved_argument_off2, // 2nd half of double
 579                 framesize
 580   };
 581 
 582   assert(FPUStateSizeInWords == 27, "update stack layout");
 583 
 584     // Save outgoing argument to stack across push_FPU_state()
 585     __ subptr(rsp, wordSize * 2);
 586     __ fstp_d(Address(rsp, 0));
 587 
 588     // Save CPU & FPU state
 589     __ push(rbx);
 590     __ push(rcx);
 591     __ push(rsi);
 592     __ push(rdi);
 593     __ push(rbp);
 594     __ push_FPU_state();
 595 
 596     // push_FPU_state() resets the FP top of stack
 597     // Load original double into FP top of stack
 598     __ fld_d(Address(rsp, saved_argument_off * wordSize));
 599     // Store double into stack as outgoing argument
 600     __ subptr(rsp, wordSize*2);
 601     __ fst_d(Address(rsp, 0));
 602 
 603     // Prepare FPU for doing math in C-land
 604     __ empty_FPU_stack();
 605     // Call the C code to massage the double.  Result in EAX
 606     if (t == T_INT)
 607       { BLOCK_COMMENT("SharedRuntime::d2i"); }
 608     else if (t == T_LONG)
 609       { BLOCK_COMMENT("SharedRuntime::d2l"); }
 610     __ call_VM_leaf( fcn, 2 );
 611 
 612     // Restore CPU & FPU state
 613     __ pop_FPU_state();
 614     __ pop(rbp);
 615     __ pop(rdi);
 616     __ pop(rsi);
 617     __ pop(rcx);
 618     __ pop(rbx);
 619     __ addptr(rsp, wordSize * 2);
 620 
 621     __ ret(0);
 622 
 623     return start;
 624   }
 625 
 626 
 627   //---------------------------------------------------------------------------
 628   // The following routine generates a subroutine to throw an asynchronous
 629   // UnknownError when an unsafe access gets a fault that could not be
 630   // reasonably prevented by the programmer.  (Example: SIGBUS/OBJERR.)
 631   address generate_handler_for_unsafe_access() {
 632     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 633     address start = __ pc();
 634 
 635     __ push(0);                       // hole for return address-to-be
 636     __ pusha();                       // push registers
 637     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 638     BLOCK_COMMENT("call handle_unsafe_access");
 639     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 640     __ movptr(next_pc, rax);          // stuff next address
 641     __ popa();
 642     __ ret(0);                        // jump to next address
 643 
 644     return start;
 645   }
 646 
 647 
 648   //----------------------------------------------------------------------------------------------------
 649   // Non-destructive plausibility checks for oops
 650 
 651   address generate_verify_oop() {
 652     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 653     address start = __ pc();
 654 
 655     // Incoming arguments on stack after saving rax,:
 656     //
 657     // [tos    ]: saved rdx
 658     // [tos + 1]: saved EFLAGS
 659     // [tos + 2]: return address
 660     // [tos + 3]: char* error message
 661     // [tos + 4]: oop   object to verify
 662     // [tos + 5]: saved rax, - saved by caller and bashed
 663 
 664     Label exit, error;
 665     __ pushf();
 666     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 667     __ push(rdx);                                // save rdx
 668     // make sure object is 'reasonable'
 669     __ movptr(rax, Address(rsp, 4 * wordSize));    // get object
 670     __ testptr(rax, rax);
 671     __ jcc(Assembler::zero, exit);               // if obj is NULL it is ok
 672 
 673     // Check if the oop is in the right area of memory
 674     const int oop_mask = Universe::verify_oop_mask();
 675     const int oop_bits = Universe::verify_oop_bits();
 676     __ mov(rdx, rax);
 677     __ andptr(rdx, oop_mask);
 678     __ cmpptr(rdx, oop_bits);
 679     __ jcc(Assembler::notZero, error);
 680 
 681     // make sure klass is 'reasonable', which is not zero.
 682     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
 683     __ testptr(rax, rax);
 684     __ jcc(Assembler::zero, error);              // if klass is NULL it is broken
 685 
 686     // return if everything seems ok
 687     __ bind(exit);
 688     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 689     __ pop(rdx);                                 // restore rdx
 690     __ popf();                                   // restore EFLAGS
 691     __ ret(3 * wordSize);                        // pop arguments
 692 
 693     // handle errors
 694     __ bind(error);
 695     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 696     __ pop(rdx);                                 // get saved rdx back
 697     __ popf();                                   // get saved EFLAGS off stack -- will be ignored
 698     __ pusha();                                  // push registers (eip = return address & msg are already pushed)
 699     BLOCK_COMMENT("call MacroAssembler::debug");
 700     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
 701     __ popa();
 702     __ ret(3 * wordSize);                        // pop arguments
 703     return start;
 704   }
 705 
 706   //
 707   //  Generate pre-barrier for array stores
 708   //
 709   //  Input:
 710   //     start   -  starting address
 711   //     count   -  element count
 712   void  gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
 713     assert_different_registers(start, count);
 714     BarrierSet* bs = Universe::heap()->barrier_set();
 715     switch (bs->kind()) {
 716       case BarrierSet::G1SATBCTLogging:
 717         // With G1, don't generate the call if we statically know that the target in uninitialized
 718         if (!uninitialized_target) {
 719            __ pusha();                      // push registers
 720            __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
 721                            start, count);
 722            __ popa();
 723          }
 724         break;
 725       case BarrierSet::CardTableModRef:
 726       case BarrierSet::CardTableExtension:
 727       case BarrierSet::ModRef:
 728         break;
 729       default      :
 730         ShouldNotReachHere();
 731 
 732     }
 733   }
 734 
 735 
 736   //
 737   // Generate a post-barrier for an array store
 738   //
 739   //     start    -  starting address
 740   //     count    -  element count
 741   //
 742   //  The two input registers are overwritten.
 743   //
 744   void  gen_write_ref_array_post_barrier(Register start, Register count) {
 745     BarrierSet* bs = Universe::heap()->barrier_set();
 746     assert_different_registers(start, count);
 747     switch (bs->kind()) {
 748       case BarrierSet::G1SATBCTLogging:
 749         {
 750           __ pusha();                      // push registers
 751           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
 752                           start, count);
 753           __ popa();
 754         }
 755         break;
 756 
 757       case BarrierSet::CardTableModRef:
 758       case BarrierSet::CardTableExtension:
 759         {
 760           CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
 761           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
 762 
 763           Label L_loop;
 764           const Register end = count;  // elements count; end == start+count-1
 765           assert_different_registers(start, end);
 766 
 767           __ lea(end,  Address(start, count, Address::times_ptr, -wordSize));
 768           __ shrptr(start, CardTableModRefBS::card_shift);
 769           __ shrptr(end,   CardTableModRefBS::card_shift);
 770           __ subptr(end, start); // end --> count
 771         __ BIND(L_loop);
 772           intptr_t disp = (intptr_t) ct->byte_map_base;
 773           Address cardtable(start, count, Address::times_1, disp);
 774           __ movb(cardtable, 0);
 775           __ decrement(count);
 776           __ jcc(Assembler::greaterEqual, L_loop);
 777         }
 778         break;
 779       case BarrierSet::ModRef:
 780         break;
 781       default      :
 782         ShouldNotReachHere();
 783 
 784     }
 785   }
 786 
 787 
 788   // Copy 64 bytes chunks
 789   //
 790   // Inputs:
 791   //   from        - source array address
 792   //   to_from     - destination array address - from
 793   //   qword_count - 8-bytes element count, negative
 794   //
 795   void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
 796     assert( UseSSE >= 2, "supported cpu only" );
 797     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 798     // Copy 64-byte chunks
 799     __ jmpb(L_copy_64_bytes);
 800     __ align(OptoLoopAlignment);
 801   __ BIND(L_copy_64_bytes_loop);
 802 
 803     if (UseUnalignedLoadStores) {
 804       if (UseAVX > 2) {
 805         __ evmovdqu(xmm0, Address(from, 0), Assembler::AVX_512bit);
 806         __ evmovdqu(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
 807       } else if (UseAVX == 2) {
 808         __ vmovdqu(xmm0, Address(from,  0));
 809         __ vmovdqu(Address(from, to_from, Address::times_1,  0), xmm0);
 810         __ vmovdqu(xmm1, Address(from, 32));
 811         __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
 812       } else {
 813         __ movdqu(xmm0, Address(from, 0));
 814         __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
 815         __ movdqu(xmm1, Address(from, 16));
 816         __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
 817         __ movdqu(xmm2, Address(from, 32));
 818         __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
 819         __ movdqu(xmm3, Address(from, 48));
 820         __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
 821       }
 822     } else {
 823       __ movq(xmm0, Address(from, 0));
 824       __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
 825       __ movq(xmm1, Address(from, 8));
 826       __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
 827       __ movq(xmm2, Address(from, 16));
 828       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 829       __ movq(xmm3, Address(from, 24));
 830       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 831       __ movq(xmm4, Address(from, 32));
 832       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 833       __ movq(xmm5, Address(from, 40));
 834       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 835       __ movq(xmm6, Address(from, 48));
 836       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 837       __ movq(xmm7, Address(from, 56));
 838       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 839     }
 840 
 841     __ addl(from, 64);
 842   __ BIND(L_copy_64_bytes);
 843     __ subl(qword_count, 8);
 844     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 845 
 846     if (UseUnalignedLoadStores && (UseAVX == 2)) {
 847       // clean upper bits of YMM registers
 848       __ vpxor(xmm0, xmm0);
 849       __ vpxor(xmm1, xmm1);
 850     }
 851     __ addl(qword_count, 8);
 852     __ jccb(Assembler::zero, L_exit);
 853     //
 854     // length is too short, just copy qwords
 855     //
 856   __ BIND(L_copy_8_bytes);
 857     __ movq(xmm0, Address(from, 0));
 858     __ movq(Address(from, to_from, Address::times_1), xmm0);
 859     __ addl(from, 8);
 860     __ decrement(qword_count);
 861     __ jcc(Assembler::greater, L_copy_8_bytes);
 862   __ BIND(L_exit);
 863   }
 864 
 865   // Copy 64 bytes chunks
 866   //
 867   // Inputs:
 868   //   from        - source array address
 869   //   to_from     - destination array address - from
 870   //   qword_count - 8-bytes element count, negative
 871   //
 872   void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
 873     assert( VM_Version::supports_mmx(), "supported cpu only" );
 874     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 875     // Copy 64-byte chunks
 876     __ jmpb(L_copy_64_bytes);
 877     __ align(OptoLoopAlignment);
 878   __ BIND(L_copy_64_bytes_loop);
 879     __ movq(mmx0, Address(from, 0));
 880     __ movq(mmx1, Address(from, 8));
 881     __ movq(mmx2, Address(from, 16));
 882     __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
 883     __ movq(mmx3, Address(from, 24));
 884     __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
 885     __ movq(mmx4, Address(from, 32));
 886     __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
 887     __ movq(mmx5, Address(from, 40));
 888     __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
 889     __ movq(mmx6, Address(from, 48));
 890     __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
 891     __ movq(mmx7, Address(from, 56));
 892     __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
 893     __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
 894     __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
 895     __ addptr(from, 64);
 896   __ BIND(L_copy_64_bytes);
 897     __ subl(qword_count, 8);
 898     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 899     __ addl(qword_count, 8);
 900     __ jccb(Assembler::zero, L_exit);
 901     //
 902     // length is too short, just copy qwords
 903     //
 904   __ BIND(L_copy_8_bytes);
 905     __ movq(mmx0, Address(from, 0));
 906     __ movq(Address(from, to_from, Address::times_1), mmx0);
 907     __ addptr(from, 8);
 908     __ decrement(qword_count);
 909     __ jcc(Assembler::greater, L_copy_8_bytes);
 910   __ BIND(L_exit);
 911     __ emms();
 912   }
 913 
 914   address generate_disjoint_copy(BasicType t, bool aligned,
 915                                  Address::ScaleFactor sf,
 916                                  address* entry, const char *name,
 917                                  bool dest_uninitialized = false) {
 918     __ align(CodeEntryAlignment);
 919     StubCodeMark mark(this, "StubRoutines", name);
 920     address start = __ pc();
 921 
 922     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
 923     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
 924 
 925     int shift = Address::times_ptr - sf;
 926 
 927     const Register from     = rsi;  // source array address
 928     const Register to       = rdi;  // destination array address
 929     const Register count    = rcx;  // elements count
 930     const Register to_from  = to;   // (to - from)
 931     const Register saved_to = rdx;  // saved destination array address
 932 
 933     __ enter(); // required for proper stackwalking of RuntimeStub frame
 934     __ push(rsi);
 935     __ push(rdi);
 936     __ movptr(from , Address(rsp, 12+ 4));
 937     __ movptr(to   , Address(rsp, 12+ 8));
 938     __ movl(count, Address(rsp, 12+ 12));
 939 
 940     if (entry != NULL) {
 941       *entry = __ pc(); // Entry point from conjoint arraycopy stub.
 942       BLOCK_COMMENT("Entry:");
 943     }
 944 
 945     if (t == T_OBJECT) {
 946       __ testl(count, count);
 947       __ jcc(Assembler::zero, L_0_count);
 948       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
 949       __ mov(saved_to, to);          // save 'to'
 950     }
 951 
 952     __ subptr(to, from); // to --> to_from
 953     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
 954     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
 955     if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 956       // align source address at 4 bytes address boundary
 957       if (t == T_BYTE) {
 958         // One byte misalignment happens only for byte arrays
 959         __ testl(from, 1);
 960         __ jccb(Assembler::zero, L_skip_align1);
 961         __ movb(rax, Address(from, 0));
 962         __ movb(Address(from, to_from, Address::times_1, 0), rax);
 963         __ increment(from);
 964         __ decrement(count);
 965       __ BIND(L_skip_align1);
 966       }
 967       // Two bytes misalignment happens only for byte and short (char) arrays
 968       __ testl(from, 2);
 969       __ jccb(Assembler::zero, L_skip_align2);
 970       __ movw(rax, Address(from, 0));
 971       __ movw(Address(from, to_from, Address::times_1, 0), rax);
 972       __ addptr(from, 2);
 973       __ subl(count, 1<<(shift-1));
 974     __ BIND(L_skip_align2);
 975     }
 976     if (!VM_Version::supports_mmx()) {
 977       __ mov(rax, count);      // save 'count'
 978       __ shrl(count, shift); // bytes count
 979       __ addptr(to_from, from);// restore 'to'
 980       __ rep_mov();
 981       __ subptr(to_from, from);// restore 'to_from'
 982       __ mov(count, rax);      // restore 'count'
 983       __ jmpb(L_copy_2_bytes); // all dwords were copied
 984     } else {
 985       if (!UseUnalignedLoadStores) {
 986         // align to 8 bytes, we know we are 4 byte aligned to start
 987         __ testptr(from, 4);
 988         __ jccb(Assembler::zero, L_copy_64_bytes);
 989         __ movl(rax, Address(from, 0));
 990         __ movl(Address(from, to_from, Address::times_1, 0), rax);
 991         __ addptr(from, 4);
 992         __ subl(count, 1<<shift);
 993       }
 994     __ BIND(L_copy_64_bytes);
 995       __ mov(rax, count);
 996       __ shrl(rax, shift+1);  // 8 bytes chunk count
 997       //
 998       // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
 999       //
1000       if (UseXMMForArrayCopy) {
1001         xmm_copy_forward(from, to_from, rax);
1002       } else {
1003         mmx_copy_forward(from, to_from, rax);
1004       }
1005     }
1006     // copy tailing dword
1007   __ BIND(L_copy_4_bytes);
1008     __ testl(count, 1<<shift);
1009     __ jccb(Assembler::zero, L_copy_2_bytes);
1010     __ movl(rax, Address(from, 0));
1011     __ movl(Address(from, to_from, Address::times_1, 0), rax);
1012     if (t == T_BYTE || t == T_SHORT) {
1013       __ addptr(from, 4);
1014     __ BIND(L_copy_2_bytes);
1015       // copy tailing word
1016       __ testl(count, 1<<(shift-1));
1017       __ jccb(Assembler::zero, L_copy_byte);
1018       __ movw(rax, Address(from, 0));
1019       __ movw(Address(from, to_from, Address::times_1, 0), rax);
1020       if (t == T_BYTE) {
1021         __ addptr(from, 2);
1022       __ BIND(L_copy_byte);
1023         // copy tailing byte
1024         __ testl(count, 1);
1025         __ jccb(Assembler::zero, L_exit);
1026         __ movb(rax, Address(from, 0));
1027         __ movb(Address(from, to_from, Address::times_1, 0), rax);
1028       __ BIND(L_exit);
1029       } else {
1030       __ BIND(L_copy_byte);
1031       }
1032     } else {
1033     __ BIND(L_copy_2_bytes);
1034     }
1035 
1036     if (t == T_OBJECT) {
1037       __ movl(count, Address(rsp, 12+12)); // reread 'count'
1038       __ mov(to, saved_to); // restore 'to'
1039       gen_write_ref_array_post_barrier(to, count);
1040     __ BIND(L_0_count);
1041     }
1042     inc_copy_counter_np(t);
1043     __ pop(rdi);
1044     __ pop(rsi);
1045     __ leave(); // required for proper stackwalking of RuntimeStub frame
1046     __ xorptr(rax, rax); // return 0
1047     __ ret(0);
1048     return start;
1049   }
1050 
1051 
1052   address generate_fill(BasicType t, bool aligned, const char *name) {
1053     __ align(CodeEntryAlignment);
1054     StubCodeMark mark(this, "StubRoutines", name);
1055     address start = __ pc();
1056 
1057     BLOCK_COMMENT("Entry:");
1058 
1059     const Register to       = rdi;  // source array address
1060     const Register value    = rdx;  // value
1061     const Register count    = rsi;  // elements count
1062 
1063     __ enter(); // required for proper stackwalking of RuntimeStub frame
1064     __ push(rsi);
1065     __ push(rdi);
1066     __ movptr(to   , Address(rsp, 12+ 4));
1067     __ movl(value, Address(rsp, 12+ 8));
1068     __ movl(count, Address(rsp, 12+ 12));
1069 
1070     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1071 
1072     __ pop(rdi);
1073     __ pop(rsi);
1074     __ leave(); // required for proper stackwalking of RuntimeStub frame
1075     __ ret(0);
1076     return start;
1077   }
1078 
1079   address generate_conjoint_copy(BasicType t, bool aligned,
1080                                  Address::ScaleFactor sf,
1081                                  address nooverlap_target,
1082                                  address* entry, const char *name,
1083                                  bool dest_uninitialized = false) {
1084     __ align(CodeEntryAlignment);
1085     StubCodeMark mark(this, "StubRoutines", name);
1086     address start = __ pc();
1087 
1088     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1089     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1090 
1091     int shift = Address::times_ptr - sf;
1092 
1093     const Register src   = rax;  // source array address
1094     const Register dst   = rdx;  // destination array address
1095     const Register from  = rsi;  // source array address
1096     const Register to    = rdi;  // destination array address
1097     const Register count = rcx;  // elements count
1098     const Register end   = rax;  // array end address
1099 
1100     __ enter(); // required for proper stackwalking of RuntimeStub frame
1101     __ push(rsi);
1102     __ push(rdi);
1103     __ movptr(src  , Address(rsp, 12+ 4));   // from
1104     __ movptr(dst  , Address(rsp, 12+ 8));   // to
1105     __ movl2ptr(count, Address(rsp, 12+12)); // count
1106 
1107     if (entry != NULL) {
1108       *entry = __ pc(); // Entry point from generic arraycopy stub.
1109       BLOCK_COMMENT("Entry:");
1110     }
1111 
1112     // nooverlap_target expects arguments in rsi and rdi.
1113     __ mov(from, src);
1114     __ mov(to  , dst);
1115 
1116     // arrays overlap test: dispatch to disjoint stub if necessary.
1117     RuntimeAddress nooverlap(nooverlap_target);
1118     __ cmpptr(dst, src);
1119     __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1120     __ jump_cc(Assembler::belowEqual, nooverlap);
1121     __ cmpptr(dst, end);
1122     __ jump_cc(Assembler::aboveEqual, nooverlap);
1123 
1124     if (t == T_OBJECT) {
1125       __ testl(count, count);
1126       __ jcc(Assembler::zero, L_0_count);
1127       gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1128     }
1129 
1130     // copy from high to low
1131     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1132     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1133     if (t == T_BYTE || t == T_SHORT) {
1134       // Align the end of destination array at 4 bytes address boundary
1135       __ lea(end, Address(dst, count, sf, 0));
1136       if (t == T_BYTE) {
1137         // One byte misalignment happens only for byte arrays
1138         __ testl(end, 1);
1139         __ jccb(Assembler::zero, L_skip_align1);
1140         __ decrement(count);
1141         __ movb(rdx, Address(from, count, sf, 0));
1142         __ movb(Address(to, count, sf, 0), rdx);
1143       __ BIND(L_skip_align1);
1144       }
1145       // Two bytes misalignment happens only for byte and short (char) arrays
1146       __ testl(end, 2);
1147       __ jccb(Assembler::zero, L_skip_align2);
1148       __ subptr(count, 1<<(shift-1));
1149       __ movw(rdx, Address(from, count, sf, 0));
1150       __ movw(Address(to, count, sf, 0), rdx);
1151     __ BIND(L_skip_align2);
1152       __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1153       __ jcc(Assembler::below, L_copy_4_bytes);
1154     }
1155 
1156     if (!VM_Version::supports_mmx()) {
1157       __ std();
1158       __ mov(rax, count); // Save 'count'
1159       __ mov(rdx, to);    // Save 'to'
1160       __ lea(rsi, Address(from, count, sf, -4));
1161       __ lea(rdi, Address(to  , count, sf, -4));
1162       __ shrptr(count, shift); // bytes count
1163       __ rep_mov();
1164       __ cld();
1165       __ mov(count, rax); // restore 'count'
1166       __ andl(count, (1<<shift)-1);      // mask the number of rest elements
1167       __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1168       __ mov(to, rdx);   // restore 'to'
1169       __ jmpb(L_copy_2_bytes); // all dword were copied
1170    } else {
1171       // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1172       __ testptr(end, 4);
1173       __ jccb(Assembler::zero, L_copy_8_bytes);
1174       __ subl(count, 1<<shift);
1175       __ movl(rdx, Address(from, count, sf, 0));
1176       __ movl(Address(to, count, sf, 0), rdx);
1177       __ jmpb(L_copy_8_bytes);
1178 
1179       __ align(OptoLoopAlignment);
1180       // Move 8 bytes
1181     __ BIND(L_copy_8_bytes_loop);
1182       if (UseXMMForArrayCopy) {
1183         __ movq(xmm0, Address(from, count, sf, 0));
1184         __ movq(Address(to, count, sf, 0), xmm0);
1185       } else {
1186         __ movq(mmx0, Address(from, count, sf, 0));
1187         __ movq(Address(to, count, sf, 0), mmx0);
1188       }
1189     __ BIND(L_copy_8_bytes);
1190       __ subl(count, 2<<shift);
1191       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1192       __ addl(count, 2<<shift);
1193       if (!UseXMMForArrayCopy) {
1194         __ emms();
1195       }
1196     }
1197   __ BIND(L_copy_4_bytes);
1198     // copy prefix qword
1199     __ testl(count, 1<<shift);
1200     __ jccb(Assembler::zero, L_copy_2_bytes);
1201     __ movl(rdx, Address(from, count, sf, -4));
1202     __ movl(Address(to, count, sf, -4), rdx);
1203 
1204     if (t == T_BYTE || t == T_SHORT) {
1205         __ subl(count, (1<<shift));
1206       __ BIND(L_copy_2_bytes);
1207         // copy prefix dword
1208         __ testl(count, 1<<(shift-1));
1209         __ jccb(Assembler::zero, L_copy_byte);
1210         __ movw(rdx, Address(from, count, sf, -2));
1211         __ movw(Address(to, count, sf, -2), rdx);
1212         if (t == T_BYTE) {
1213           __ subl(count, 1<<(shift-1));
1214         __ BIND(L_copy_byte);
1215           // copy prefix byte
1216           __ testl(count, 1);
1217           __ jccb(Assembler::zero, L_exit);
1218           __ movb(rdx, Address(from, 0));
1219           __ movb(Address(to, 0), rdx);
1220         __ BIND(L_exit);
1221         } else {
1222         __ BIND(L_copy_byte);
1223         }
1224     } else {
1225     __ BIND(L_copy_2_bytes);
1226     }
1227     if (t == T_OBJECT) {
1228       __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1229       gen_write_ref_array_post_barrier(to, count);
1230     __ BIND(L_0_count);
1231     }
1232     inc_copy_counter_np(t);
1233     __ pop(rdi);
1234     __ pop(rsi);
1235     __ leave(); // required for proper stackwalking of RuntimeStub frame
1236     __ xorptr(rax, rax); // return 0
1237     __ ret(0);
1238     return start;
1239   }
1240 
1241 
1242   address generate_disjoint_long_copy(address* entry, const char *name) {
1243     __ align(CodeEntryAlignment);
1244     StubCodeMark mark(this, "StubRoutines", name);
1245     address start = __ pc();
1246 
1247     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1248     const Register from       = rax;  // source array address
1249     const Register to         = rdx;  // destination array address
1250     const Register count      = rcx;  // elements count
1251     const Register to_from    = rdx;  // (to - from)
1252 
1253     __ enter(); // required for proper stackwalking of RuntimeStub frame
1254     __ movptr(from , Address(rsp, 8+0));       // from
1255     __ movptr(to   , Address(rsp, 8+4));       // to
1256     __ movl2ptr(count, Address(rsp, 8+8));     // count
1257 
1258     *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1259     BLOCK_COMMENT("Entry:");
1260 
1261     __ subptr(to, from); // to --> to_from
1262     if (VM_Version::supports_mmx()) {
1263       if (UseXMMForArrayCopy) {
1264         xmm_copy_forward(from, to_from, count);
1265       } else {
1266         mmx_copy_forward(from, to_from, count);
1267       }
1268     } else {
1269       __ jmpb(L_copy_8_bytes);
1270       __ align(OptoLoopAlignment);
1271     __ BIND(L_copy_8_bytes_loop);
1272       __ fild_d(Address(from, 0));
1273       __ fistp_d(Address(from, to_from, Address::times_1));
1274       __ addptr(from, 8);
1275     __ BIND(L_copy_8_bytes);
1276       __ decrement(count);
1277       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1278     }
1279     inc_copy_counter_np(T_LONG);
1280     __ leave(); // required for proper stackwalking of RuntimeStub frame
1281     __ xorptr(rax, rax); // return 0
1282     __ ret(0);
1283     return start;
1284   }
1285 
1286   address generate_conjoint_long_copy(address nooverlap_target,
1287                                       address* entry, const char *name) {
1288     __ align(CodeEntryAlignment);
1289     StubCodeMark mark(this, "StubRoutines", name);
1290     address start = __ pc();
1291 
1292     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1293     const Register from       = rax;  // source array address
1294     const Register to         = rdx;  // destination array address
1295     const Register count      = rcx;  // elements count
1296     const Register end_from   = rax;  // source array end address
1297 
1298     __ enter(); // required for proper stackwalking of RuntimeStub frame
1299     __ movptr(from , Address(rsp, 8+0));       // from
1300     __ movptr(to   , Address(rsp, 8+4));       // to
1301     __ movl2ptr(count, Address(rsp, 8+8));     // count
1302 
1303     *entry = __ pc(); // Entry point from generic arraycopy stub.
1304     BLOCK_COMMENT("Entry:");
1305 
1306     // arrays overlap test
1307     __ cmpptr(to, from);
1308     RuntimeAddress nooverlap(nooverlap_target);
1309     __ jump_cc(Assembler::belowEqual, nooverlap);
1310     __ lea(end_from, Address(from, count, Address::times_8, 0));
1311     __ cmpptr(to, end_from);
1312     __ movptr(from, Address(rsp, 8));  // from
1313     __ jump_cc(Assembler::aboveEqual, nooverlap);
1314 
1315     __ jmpb(L_copy_8_bytes);
1316 
1317     __ align(OptoLoopAlignment);
1318   __ BIND(L_copy_8_bytes_loop);
1319     if (VM_Version::supports_mmx()) {
1320       if (UseXMMForArrayCopy) {
1321         __ movq(xmm0, Address(from, count, Address::times_8));
1322         __ movq(Address(to, count, Address::times_8), xmm0);
1323       } else {
1324         __ movq(mmx0, Address(from, count, Address::times_8));
1325         __ movq(Address(to, count, Address::times_8), mmx0);
1326       }
1327     } else {
1328       __ fild_d(Address(from, count, Address::times_8));
1329       __ fistp_d(Address(to, count, Address::times_8));
1330     }
1331   __ BIND(L_copy_8_bytes);
1332     __ decrement(count);
1333     __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1334 
1335     if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1336       __ emms();
1337     }
1338     inc_copy_counter_np(T_LONG);
1339     __ leave(); // required for proper stackwalking of RuntimeStub frame
1340     __ xorptr(rax, rax); // return 0
1341     __ ret(0);
1342     return start;
1343   }
1344 
1345 
1346   // Helper for generating a dynamic type check.
1347   // The sub_klass must be one of {rbx, rdx, rsi}.
1348   // The temp is killed.
1349   void generate_type_check(Register sub_klass,
1350                            Address& super_check_offset_addr,
1351                            Address& super_klass_addr,
1352                            Register temp,
1353                            Label* L_success, Label* L_failure) {
1354     BLOCK_COMMENT("type_check:");
1355 
1356     Label L_fallthrough;
1357 #define LOCAL_JCC(assembler_con, label_ptr)                             \
1358     if (label_ptr != NULL)  __ jcc(assembler_con, *(label_ptr));        \
1359     else                    __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1360 
1361     // The following is a strange variation of the fast path which requires
1362     // one less register, because needed values are on the argument stack.
1363     // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1364     //                                  L_success, L_failure, NULL);
1365     assert_different_registers(sub_klass, temp);
1366 
1367     int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1368 
1369     // if the pointers are equal, we are done (e.g., String[] elements)
1370     __ cmpptr(sub_klass, super_klass_addr);
1371     LOCAL_JCC(Assembler::equal, L_success);
1372 
1373     // check the supertype display:
1374     __ movl2ptr(temp, super_check_offset_addr);
1375     Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1376     __ movptr(temp, super_check_addr); // load displayed supertype
1377     __ cmpptr(temp, super_klass_addr); // test the super type
1378     LOCAL_JCC(Assembler::equal, L_success);
1379 
1380     // if it was a primary super, we can just fail immediately
1381     __ cmpl(super_check_offset_addr, sc_offset);
1382     LOCAL_JCC(Assembler::notEqual, L_failure);
1383 
1384     // The repne_scan instruction uses fixed registers, which will get spilled.
1385     // We happen to know this works best when super_klass is in rax.
1386     Register super_klass = temp;
1387     __ movptr(super_klass, super_klass_addr);
1388     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1389                                      L_success, L_failure);
1390 
1391     __ bind(L_fallthrough);
1392 
1393     if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1394     if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1395 
1396 #undef LOCAL_JCC
1397   }
1398 
1399   //
1400   //  Generate checkcasting array copy stub
1401   //
1402   //  Input:
1403   //    4(rsp)   - source array address
1404   //    8(rsp)   - destination array address
1405   //   12(rsp)   - element count, can be zero
1406   //   16(rsp)   - size_t ckoff (super_check_offset)
1407   //   20(rsp)   - oop ckval (super_klass)
1408   //
1409   //  Output:
1410   //    rax, ==  0  -  success
1411   //    rax, == -1^K - failure, where K is partial transfer count
1412   //
1413   address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1414     __ align(CodeEntryAlignment);
1415     StubCodeMark mark(this, "StubRoutines", name);
1416     address start = __ pc();
1417 
1418     Label L_load_element, L_store_element, L_do_card_marks, L_done;
1419 
1420     // register use:
1421     //  rax, rdx, rcx -- loop control (end_from, end_to, count)
1422     //  rdi, rsi      -- element access (oop, klass)
1423     //  rbx,           -- temp
1424     const Register from       = rax;    // source array address
1425     const Register to         = rdx;    // destination array address
1426     const Register length     = rcx;    // elements count
1427     const Register elem       = rdi;    // each oop copied
1428     const Register elem_klass = rsi;    // each elem._klass (sub_klass)
1429     const Register temp       = rbx;    // lone remaining temp
1430 
1431     __ enter(); // required for proper stackwalking of RuntimeStub frame
1432 
1433     __ push(rsi);
1434     __ push(rdi);
1435     __ push(rbx);
1436 
1437     Address   from_arg(rsp, 16+ 4);     // from
1438     Address     to_arg(rsp, 16+ 8);     // to
1439     Address length_arg(rsp, 16+12);     // elements count
1440     Address  ckoff_arg(rsp, 16+16);     // super_check_offset
1441     Address  ckval_arg(rsp, 16+20);     // super_klass
1442 
1443     // Load up:
1444     __ movptr(from,     from_arg);
1445     __ movptr(to,         to_arg);
1446     __ movl2ptr(length, length_arg);
1447 
1448     if (entry != NULL) {
1449       *entry = __ pc(); // Entry point from generic arraycopy stub.
1450       BLOCK_COMMENT("Entry:");
1451     }
1452 
1453     //---------------------------------------------------------------
1454     // Assembler stub will be used for this call to arraycopy
1455     // if the two arrays are subtypes of Object[] but the
1456     // destination array type is not equal to or a supertype
1457     // of the source type.  Each element must be separately
1458     // checked.
1459 
1460     // Loop-invariant addresses.  They are exclusive end pointers.
1461     Address end_from_addr(from, length, Address::times_ptr, 0);
1462     Address   end_to_addr(to,   length, Address::times_ptr, 0);
1463 
1464     Register end_from = from;           // re-use
1465     Register end_to   = to;             // re-use
1466     Register count    = length;         // re-use
1467 
1468     // Loop-variant addresses.  They assume post-incremented count < 0.
1469     Address from_element_addr(end_from, count, Address::times_ptr, 0);
1470     Address   to_element_addr(end_to,   count, Address::times_ptr, 0);
1471     Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1472 
1473     // Copy from low to high addresses, indexed from the end of each array.
1474     gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1475     __ lea(end_from, end_from_addr);
1476     __ lea(end_to,   end_to_addr);
1477     assert(length == count, "");        // else fix next line:
1478     __ negptr(count);                   // negate and test the length
1479     __ jccb(Assembler::notZero, L_load_element);
1480 
1481     // Empty array:  Nothing to do.
1482     __ xorptr(rax, rax);                  // return 0 on (trivial) success
1483     __ jmp(L_done);
1484 
1485     // ======== begin loop ========
1486     // (Loop is rotated; its entry is L_load_element.)
1487     // Loop control:
1488     //   for (count = -count; count != 0; count++)
1489     // Base pointers src, dst are biased by 8*count,to last element.
1490     __ align(OptoLoopAlignment);
1491 
1492     __ BIND(L_store_element);
1493     __ movptr(to_element_addr, elem);     // store the oop
1494     __ increment(count);                // increment the count toward zero
1495     __ jccb(Assembler::zero, L_do_card_marks);
1496 
1497     // ======== loop entry is here ========
1498     __ BIND(L_load_element);
1499     __ movptr(elem, from_element_addr);   // load the oop
1500     __ testptr(elem, elem);
1501     __ jccb(Assembler::zero, L_store_element);
1502 
1503     // (Could do a trick here:  Remember last successful non-null
1504     // element stored and make a quick oop equality check on it.)
1505 
1506     __ movptr(elem_klass, elem_klass_addr); // query the object klass
1507     generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1508                         &L_store_element, NULL);
1509     // (On fall-through, we have failed the element type check.)
1510     // ======== end loop ========
1511 
1512     // It was a real error; we must depend on the caller to finish the job.
1513     // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1514     // Emit GC store barriers for the oops we have copied (length_arg + count),
1515     // and report their number to the caller.
1516     assert_different_registers(to, count, rax);
1517     Label L_post_barrier;
1518     __ addl(count, length_arg);         // transfers = (length - remaining)
1519     __ movl2ptr(rax, count);            // save the value
1520     __ notptr(rax);                     // report (-1^K) to caller (does not affect flags)
1521     __ jccb(Assembler::notZero, L_post_barrier);
1522     __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1523 
1524     // Come here on success only.
1525     __ BIND(L_do_card_marks);
1526     __ xorptr(rax, rax);                // return 0 on success
1527     __ movl2ptr(count, length_arg);
1528 
1529     __ BIND(L_post_barrier);
1530     __ movptr(to, to_arg);              // reload
1531     gen_write_ref_array_post_barrier(to, count);
1532 
1533     // Common exit point (success or failure).
1534     __ BIND(L_done);
1535     __ pop(rbx);
1536     __ pop(rdi);
1537     __ pop(rsi);
1538     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1539     __ leave(); // required for proper stackwalking of RuntimeStub frame
1540     __ ret(0);
1541 
1542     return start;
1543   }
1544 
1545   //
1546   //  Generate 'unsafe' array copy stub
1547   //  Though just as safe as the other stubs, it takes an unscaled
1548   //  size_t argument instead of an element count.
1549   //
1550   //  Input:
1551   //    4(rsp)   - source array address
1552   //    8(rsp)   - destination array address
1553   //   12(rsp)   - byte count, can be zero
1554   //
1555   //  Output:
1556   //    rax, ==  0  -  success
1557   //    rax, == -1  -  need to call System.arraycopy
1558   //
1559   // Examines the alignment of the operands and dispatches
1560   // to a long, int, short, or byte copy loop.
1561   //
1562   address generate_unsafe_copy(const char *name,
1563                                address byte_copy_entry,
1564                                address short_copy_entry,
1565                                address int_copy_entry,
1566                                address long_copy_entry) {
1567 
1568     Label L_long_aligned, L_int_aligned, L_short_aligned;
1569 
1570     __ align(CodeEntryAlignment);
1571     StubCodeMark mark(this, "StubRoutines", name);
1572     address start = __ pc();
1573 
1574     const Register from       = rax;  // source array address
1575     const Register to         = rdx;  // destination array address
1576     const Register count      = rcx;  // elements count
1577 
1578     __ enter(); // required for proper stackwalking of RuntimeStub frame
1579     __ push(rsi);
1580     __ push(rdi);
1581     Address  from_arg(rsp, 12+ 4);      // from
1582     Address    to_arg(rsp, 12+ 8);      // to
1583     Address count_arg(rsp, 12+12);      // byte count
1584 
1585     // Load up:
1586     __ movptr(from ,  from_arg);
1587     __ movptr(to   ,    to_arg);
1588     __ movl2ptr(count, count_arg);
1589 
1590     // bump this on entry, not on exit:
1591     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1592 
1593     const Register bits = rsi;
1594     __ mov(bits, from);
1595     __ orptr(bits, to);
1596     __ orptr(bits, count);
1597 
1598     __ testl(bits, BytesPerLong-1);
1599     __ jccb(Assembler::zero, L_long_aligned);
1600 
1601     __ testl(bits, BytesPerInt-1);
1602     __ jccb(Assembler::zero, L_int_aligned);
1603 
1604     __ testl(bits, BytesPerShort-1);
1605     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1606 
1607     __ BIND(L_short_aligned);
1608     __ shrptr(count, LogBytesPerShort); // size => short_count
1609     __ movl(count_arg, count);          // update 'count'
1610     __ jump(RuntimeAddress(short_copy_entry));
1611 
1612     __ BIND(L_int_aligned);
1613     __ shrptr(count, LogBytesPerInt); // size => int_count
1614     __ movl(count_arg, count);          // update 'count'
1615     __ jump(RuntimeAddress(int_copy_entry));
1616 
1617     __ BIND(L_long_aligned);
1618     __ shrptr(count, LogBytesPerLong); // size => qword_count
1619     __ movl(count_arg, count);          // update 'count'
1620     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1621     __ pop(rsi);
1622     __ jump(RuntimeAddress(long_copy_entry));
1623 
1624     return start;
1625   }
1626 
1627 
1628   // Perform range checks on the proposed arraycopy.
1629   // Smashes src_pos and dst_pos.  (Uses them up for temps.)
1630   void arraycopy_range_checks(Register src,
1631                               Register src_pos,
1632                               Register dst,
1633                               Register dst_pos,
1634                               Address& length,
1635                               Label& L_failed) {
1636     BLOCK_COMMENT("arraycopy_range_checks:");
1637     const Register src_end = src_pos;   // source array end position
1638     const Register dst_end = dst_pos;   // destination array end position
1639     __ addl(src_end, length); // src_pos + length
1640     __ addl(dst_end, length); // dst_pos + length
1641 
1642     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;
1643     __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1644     __ jcc(Assembler::above, L_failed);
1645 
1646     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1647     __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1648     __ jcc(Assembler::above, L_failed);
1649 
1650     BLOCK_COMMENT("arraycopy_range_checks done");
1651   }
1652 
1653 
1654   //
1655   //  Generate generic array copy stubs
1656   //
1657   //  Input:
1658   //     4(rsp)    -  src oop
1659   //     8(rsp)    -  src_pos
1660   //    12(rsp)    -  dst oop
1661   //    16(rsp)    -  dst_pos
1662   //    20(rsp)    -  element count
1663   //
1664   //  Output:
1665   //    rax, ==  0  -  success
1666   //    rax, == -1^K - failure, where K is partial transfer count
1667   //
1668   address generate_generic_copy(const char *name,
1669                                 address entry_jbyte_arraycopy,
1670                                 address entry_jshort_arraycopy,
1671                                 address entry_jint_arraycopy,
1672                                 address entry_oop_arraycopy,
1673                                 address entry_jlong_arraycopy,
1674                                 address entry_checkcast_arraycopy) {
1675     Label L_failed, L_failed_0, L_objArray;
1676 
1677     { int modulus = CodeEntryAlignment;
1678       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
1679       int advance = target - (__ offset() % modulus);
1680       if (advance < 0)  advance += modulus;
1681       if (advance > 0)  __ nop(advance);
1682     }
1683     StubCodeMark mark(this, "StubRoutines", name);
1684 
1685     // Short-hop target to L_failed.  Makes for denser prologue code.
1686     __ BIND(L_failed_0);
1687     __ jmp(L_failed);
1688     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1689 
1690     __ align(CodeEntryAlignment);
1691     address start = __ pc();
1692 
1693     __ enter(); // required for proper stackwalking of RuntimeStub frame
1694     __ push(rsi);
1695     __ push(rdi);
1696 
1697     // bump this on entry, not on exit:
1698     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1699 
1700     // Input values
1701     Address SRC     (rsp, 12+ 4);
1702     Address SRC_POS (rsp, 12+ 8);
1703     Address DST     (rsp, 12+12);
1704     Address DST_POS (rsp, 12+16);
1705     Address LENGTH  (rsp, 12+20);
1706 
1707     //-----------------------------------------------------------------------
1708     // Assembler stub will be used for this call to arraycopy
1709     // if the following conditions are met:
1710     //
1711     // (1) src and dst must not be null.
1712     // (2) src_pos must not be negative.
1713     // (3) dst_pos must not be negative.
1714     // (4) length  must not be negative.
1715     // (5) src klass and dst klass should be the same and not NULL.
1716     // (6) src and dst should be arrays.
1717     // (7) src_pos + length must not exceed length of src.
1718     // (8) dst_pos + length must not exceed length of dst.
1719     //
1720 
1721     const Register src     = rax;       // source array oop
1722     const Register src_pos = rsi;
1723     const Register dst     = rdx;       // destination array oop
1724     const Register dst_pos = rdi;
1725     const Register length  = rcx;       // transfer count
1726 
1727     //  if (src == NULL) return -1;
1728     __ movptr(src, SRC);      // src oop
1729     __ testptr(src, src);
1730     __ jccb(Assembler::zero, L_failed_0);
1731 
1732     //  if (src_pos < 0) return -1;
1733     __ movl2ptr(src_pos, SRC_POS);  // src_pos
1734     __ testl(src_pos, src_pos);
1735     __ jccb(Assembler::negative, L_failed_0);
1736 
1737     //  if (dst == NULL) return -1;
1738     __ movptr(dst, DST);      // dst oop
1739     __ testptr(dst, dst);
1740     __ jccb(Assembler::zero, L_failed_0);
1741 
1742     //  if (dst_pos < 0) return -1;
1743     __ movl2ptr(dst_pos, DST_POS);  // dst_pos
1744     __ testl(dst_pos, dst_pos);
1745     __ jccb(Assembler::negative, L_failed_0);
1746 
1747     //  if (length < 0) return -1;
1748     __ movl2ptr(length, LENGTH);   // length
1749     __ testl(length, length);
1750     __ jccb(Assembler::negative, L_failed_0);
1751 
1752     //  if (src->klass() == NULL) return -1;
1753     Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1754     Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1755     const Register rcx_src_klass = rcx;    // array klass
1756     __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1757 
1758 #ifdef ASSERT
1759     //  assert(src->klass() != NULL);
1760     BLOCK_COMMENT("assert klasses not null");
1761     { Label L1, L2;
1762       __ testptr(rcx_src_klass, rcx_src_klass);
1763       __ jccb(Assembler::notZero, L2);   // it is broken if klass is NULL
1764       __ bind(L1);
1765       __ stop("broken null klass");
1766       __ bind(L2);
1767       __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1768       __ jccb(Assembler::equal, L1);      // this would be broken also
1769       BLOCK_COMMENT("assert done");
1770     }
1771 #endif //ASSERT
1772 
1773     // Load layout helper (32-bits)
1774     //
1775     //  |array_tag|     | header_size | element_type |     |log2_element_size|
1776     // 32        30    24            16              8     2                 0
1777     //
1778     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1779     //
1780 
1781     int lh_offset = in_bytes(Klass::layout_helper_offset());
1782     Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1783 
1784     // Handle objArrays completely differently...
1785     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1786     __ cmpl(src_klass_lh_addr, objArray_lh);
1787     __ jcc(Assembler::equal, L_objArray);
1788 
1789     //  if (src->klass() != dst->klass()) return -1;
1790     __ cmpptr(rcx_src_klass, dst_klass_addr);
1791     __ jccb(Assembler::notEqual, L_failed_0);
1792 
1793     const Register rcx_lh = rcx;  // layout helper
1794     assert(rcx_lh == rcx_src_klass, "known alias");
1795     __ movl(rcx_lh, src_klass_lh_addr);
1796 
1797     //  if (!src->is_Array()) return -1;
1798     __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1799     __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1800 
1801     // At this point, it is known to be a typeArray (array_tag 0x3).
1802 #ifdef ASSERT
1803     { Label L;
1804       __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1805       __ jcc(Assembler::greaterEqual, L); // signed cmp
1806       __ stop("must be a primitive array");
1807       __ bind(L);
1808     }
1809 #endif
1810 
1811     assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1812     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1813 
1814     // TypeArrayKlass
1815     //
1816     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1817     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1818     //
1819     const Register rsi_offset = rsi; // array offset
1820     const Register src_array  = src; // src array offset
1821     const Register dst_array  = dst; // dst array offset
1822     const Register rdi_elsize = rdi; // log2 element size
1823 
1824     __ mov(rsi_offset, rcx_lh);
1825     __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1826     __ andptr(rsi_offset, Klass::_lh_header_size_mask);   // array_offset
1827     __ addptr(src_array, rsi_offset);  // src array offset
1828     __ addptr(dst_array, rsi_offset);  // dst array offset
1829     __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1830 
1831     // next registers should be set before the jump to corresponding stub
1832     const Register from       = src; // source array address
1833     const Register to         = dst; // destination array address
1834     const Register count      = rcx; // elements count
1835     // some of them should be duplicated on stack
1836 #define FROM   Address(rsp, 12+ 4)
1837 #define TO     Address(rsp, 12+ 8)   // Not used now
1838 #define COUNT  Address(rsp, 12+12)   // Only for oop arraycopy
1839 
1840     BLOCK_COMMENT("scale indexes to element size");
1841     __ movl2ptr(rsi, SRC_POS);  // src_pos
1842     __ shlptr(rsi);             // src_pos << rcx (log2 elsize)
1843     assert(src_array == from, "");
1844     __ addptr(from, rsi);       // from = src_array + SRC_POS << log2 elsize
1845     __ movl2ptr(rdi, DST_POS);  // dst_pos
1846     __ shlptr(rdi);             // dst_pos << rcx (log2 elsize)
1847     assert(dst_array == to, "");
1848     __ addptr(to,  rdi);        // to   = dst_array + DST_POS << log2 elsize
1849     __ movptr(FROM, from);      // src_addr
1850     __ mov(rdi_elsize, rcx_lh); // log2 elsize
1851     __ movl2ptr(count, LENGTH); // elements count
1852 
1853     BLOCK_COMMENT("choose copy loop based on element size");
1854     __ cmpl(rdi_elsize, 0);
1855 
1856     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1857     __ cmpl(rdi_elsize, LogBytesPerShort);
1858     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1859     __ cmpl(rdi_elsize, LogBytesPerInt);
1860     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1861 #ifdef ASSERT
1862     __ cmpl(rdi_elsize, LogBytesPerLong);
1863     __ jccb(Assembler::notEqual, L_failed);
1864 #endif
1865     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1866     __ pop(rsi);
1867     __ jump(RuntimeAddress(entry_jlong_arraycopy));
1868 
1869   __ BIND(L_failed);
1870     __ xorptr(rax, rax);
1871     __ notptr(rax); // return -1
1872     __ pop(rdi);
1873     __ pop(rsi);
1874     __ leave(); // required for proper stackwalking of RuntimeStub frame
1875     __ ret(0);
1876 
1877     // ObjArrayKlass
1878   __ BIND(L_objArray);
1879     // live at this point:  rcx_src_klass, src[_pos], dst[_pos]
1880 
1881     Label L_plain_copy, L_checkcast_copy;
1882     //  test array classes for subtyping
1883     __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1884     __ jccb(Assembler::notEqual, L_checkcast_copy);
1885 
1886     // Identically typed arrays can be copied without element-wise checks.
1887     assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1888     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1889 
1890   __ BIND(L_plain_copy);
1891     __ movl2ptr(count, LENGTH); // elements count
1892     __ movl2ptr(src_pos, SRC_POS);  // reload src_pos
1893     __ lea(from, Address(src, src_pos, Address::times_ptr,
1894                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1895     __ movl2ptr(dst_pos, DST_POS);  // reload dst_pos
1896     __ lea(to,   Address(dst, dst_pos, Address::times_ptr,
1897                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1898     __ movptr(FROM,  from);   // src_addr
1899     __ movptr(TO,    to);     // dst_addr
1900     __ movl(COUNT, count);  // count
1901     __ jump(RuntimeAddress(entry_oop_arraycopy));
1902 
1903   __ BIND(L_checkcast_copy);
1904     // live at this point:  rcx_src_klass, dst[_pos], src[_pos]
1905     {
1906       // Handy offsets:
1907       int  ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1908       int sco_offset = in_bytes(Klass::super_check_offset_offset());
1909 
1910       Register rsi_dst_klass = rsi;
1911       Register rdi_temp      = rdi;
1912       assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1913       assert(rdi_temp      == dst_pos, "expected alias w/ dst_pos");
1914       Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1915 
1916       // Before looking at dst.length, make sure dst is also an objArray.
1917       __ movptr(rsi_dst_klass, dst_klass_addr);
1918       __ cmpl(dst_klass_lh_addr, objArray_lh);
1919       __ jccb(Assembler::notEqual, L_failed);
1920 
1921       // It is safe to examine both src.length and dst.length.
1922       __ movl2ptr(src_pos, SRC_POS);        // reload rsi
1923       arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1924       // (Now src_pos and dst_pos are killed, but not src and dst.)
1925 
1926       // We'll need this temp (don't forget to pop it after the type check).
1927       __ push(rbx);
1928       Register rbx_src_klass = rbx;
1929 
1930       __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1931       __ movptr(rsi_dst_klass, dst_klass_addr);
1932       Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1933       Label L_fail_array_check;
1934       generate_type_check(rbx_src_klass,
1935                           super_check_offset_addr, dst_klass_addr,
1936                           rdi_temp, NULL, &L_fail_array_check);
1937       // (On fall-through, we have passed the array type check.)
1938       __ pop(rbx);
1939       __ jmp(L_plain_copy);
1940 
1941       __ BIND(L_fail_array_check);
1942       // Reshuffle arguments so we can call checkcast_arraycopy:
1943 
1944       // match initial saves for checkcast_arraycopy
1945       // push(rsi);    // already done; see above
1946       // push(rdi);    // already done; see above
1947       // push(rbx);    // already done; see above
1948 
1949       // Marshal outgoing arguments now, freeing registers.
1950       Address   from_arg(rsp, 16+ 4);   // from
1951       Address     to_arg(rsp, 16+ 8);   // to
1952       Address length_arg(rsp, 16+12);   // elements count
1953       Address  ckoff_arg(rsp, 16+16);   // super_check_offset
1954       Address  ckval_arg(rsp, 16+20);   // super_klass
1955 
1956       Address SRC_POS_arg(rsp, 16+ 8);
1957       Address DST_POS_arg(rsp, 16+16);
1958       Address  LENGTH_arg(rsp, 16+20);
1959       // push rbx, changed the incoming offsets (why not just use rbp,??)
1960       // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1961 
1962       __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1963       __ movl2ptr(length, LENGTH_arg);    // reload elements count
1964       __ movl2ptr(src_pos, SRC_POS_arg);  // reload src_pos
1965       __ movl2ptr(dst_pos, DST_POS_arg);  // reload dst_pos
1966 
1967       __ movptr(ckval_arg, rbx);          // destination element type
1968       __ movl(rbx, Address(rbx, sco_offset));
1969       __ movl(ckoff_arg, rbx);          // corresponding class check offset
1970 
1971       __ movl(length_arg, length);      // outgoing length argument
1972 
1973       __ lea(from, Address(src, src_pos, Address::times_ptr,
1974                             arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1975       __ movptr(from_arg, from);
1976 
1977       __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1978                           arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1979       __ movptr(to_arg, to);
1980       __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1981     }
1982 
1983     return start;
1984   }
1985 
1986   void generate_arraycopy_stubs() {
1987     address entry;
1988     address entry_jbyte_arraycopy;
1989     address entry_jshort_arraycopy;
1990     address entry_jint_arraycopy;
1991     address entry_oop_arraycopy;
1992     address entry_jlong_arraycopy;
1993     address entry_checkcast_arraycopy;
1994 
1995     StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1996         generate_disjoint_copy(T_BYTE,  true, Address::times_1, &entry,
1997                                "arrayof_jbyte_disjoint_arraycopy");
1998     StubRoutines::_arrayof_jbyte_arraycopy =
1999         generate_conjoint_copy(T_BYTE,  true, Address::times_1,  entry,
2000                                NULL, "arrayof_jbyte_arraycopy");
2001     StubRoutines::_jbyte_disjoint_arraycopy =
2002         generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2003                                "jbyte_disjoint_arraycopy");
2004     StubRoutines::_jbyte_arraycopy =
2005         generate_conjoint_copy(T_BYTE, false, Address::times_1,  entry,
2006                                &entry_jbyte_arraycopy, "jbyte_arraycopy");
2007 
2008     StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2009         generate_disjoint_copy(T_SHORT,  true, Address::times_2, &entry,
2010                                "arrayof_jshort_disjoint_arraycopy");
2011     StubRoutines::_arrayof_jshort_arraycopy =
2012         generate_conjoint_copy(T_SHORT,  true, Address::times_2,  entry,
2013                                NULL, "arrayof_jshort_arraycopy");
2014     StubRoutines::_jshort_disjoint_arraycopy =
2015         generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2016                                "jshort_disjoint_arraycopy");
2017     StubRoutines::_jshort_arraycopy =
2018         generate_conjoint_copy(T_SHORT, false, Address::times_2,  entry,
2019                                &entry_jshort_arraycopy, "jshort_arraycopy");
2020 
2021     // Next arrays are always aligned on 4 bytes at least.
2022     StubRoutines::_jint_disjoint_arraycopy =
2023         generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2024                                "jint_disjoint_arraycopy");
2025     StubRoutines::_jint_arraycopy =
2026         generate_conjoint_copy(T_INT, true, Address::times_4,  entry,
2027                                &entry_jint_arraycopy, "jint_arraycopy");
2028 
2029     StubRoutines::_oop_disjoint_arraycopy =
2030         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2031                                "oop_disjoint_arraycopy");
2032     StubRoutines::_oop_arraycopy =
2033         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2034                                &entry_oop_arraycopy, "oop_arraycopy");
2035 
2036     StubRoutines::_oop_disjoint_arraycopy_uninit =
2037         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2038                                "oop_disjoint_arraycopy_uninit",
2039                                /*dest_uninitialized*/true);
2040     StubRoutines::_oop_arraycopy_uninit =
2041         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2042                                NULL, "oop_arraycopy_uninit",
2043                                /*dest_uninitialized*/true);
2044 
2045     StubRoutines::_jlong_disjoint_arraycopy =
2046         generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2047     StubRoutines::_jlong_arraycopy =
2048         generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2049                                     "jlong_arraycopy");
2050 
2051     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2052     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2053     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2054     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2055     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2056     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2057 
2058     StubRoutines::_arrayof_jint_disjoint_arraycopy       = StubRoutines::_jint_disjoint_arraycopy;
2059     StubRoutines::_arrayof_oop_disjoint_arraycopy        = StubRoutines::_oop_disjoint_arraycopy;
2060     StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2061     StubRoutines::_arrayof_jlong_disjoint_arraycopy      = StubRoutines::_jlong_disjoint_arraycopy;
2062 
2063     StubRoutines::_arrayof_jint_arraycopy       = StubRoutines::_jint_arraycopy;
2064     StubRoutines::_arrayof_oop_arraycopy        = StubRoutines::_oop_arraycopy;
2065     StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2066     StubRoutines::_arrayof_jlong_arraycopy      = StubRoutines::_jlong_arraycopy;
2067 
2068     StubRoutines::_checkcast_arraycopy =
2069         generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2070     StubRoutines::_checkcast_arraycopy_uninit =
2071         generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2072 
2073     StubRoutines::_unsafe_arraycopy =
2074         generate_unsafe_copy("unsafe_arraycopy",
2075                                entry_jbyte_arraycopy,
2076                                entry_jshort_arraycopy,
2077                                entry_jint_arraycopy,
2078                                entry_jlong_arraycopy);
2079 
2080     StubRoutines::_generic_arraycopy =
2081         generate_generic_copy("generic_arraycopy",
2082                                entry_jbyte_arraycopy,
2083                                entry_jshort_arraycopy,
2084                                entry_jint_arraycopy,
2085                                entry_oop_arraycopy,
2086                                entry_jlong_arraycopy,
2087                                entry_checkcast_arraycopy);
2088   }
2089 
2090   void generate_math_stubs() {
2091     {
2092       StubCodeMark mark(this, "StubRoutines", "log");
2093       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2094 
2095       __ fld_d(Address(rsp, 4));
2096       __ flog();
2097       __ ret(0);
2098     }
2099     {
2100       StubCodeMark mark(this, "StubRoutines", "log10");
2101       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2102 
2103       __ fld_d(Address(rsp, 4));
2104       __ flog10();
2105       __ ret(0);
2106     }
2107     {
2108       StubCodeMark mark(this, "StubRoutines", "sin");
2109       StubRoutines::_intrinsic_sin = (double (*)(double))  __ pc();
2110 
2111       __ fld_d(Address(rsp, 4));
2112       __ trigfunc('s');
2113       __ ret(0);
2114     }
2115     {
2116       StubCodeMark mark(this, "StubRoutines", "cos");
2117       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2118 
2119       __ fld_d(Address(rsp, 4));
2120       __ trigfunc('c');
2121       __ ret(0);
2122     }
2123     {
2124       StubCodeMark mark(this, "StubRoutines", "tan");
2125       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2126 
2127       __ fld_d(Address(rsp, 4));
2128       __ trigfunc('t');
2129       __ ret(0);
2130     }
2131     {
2132       StubCodeMark mark(this, "StubRoutines", "exp");
2133       StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2134 
2135       __ fld_d(Address(rsp, 4));
2136       __ exp_with_fallback(0);
2137       __ ret(0);
2138     }
2139     {
2140       StubCodeMark mark(this, "StubRoutines", "pow");
2141       StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2142 
2143       __ fld_d(Address(rsp, 12));
2144       __ fld_d(Address(rsp, 4));
2145       __ pow_with_fallback(0);
2146       __ ret(0);
2147     }
2148   }
2149 
2150   // AES intrinsic stubs
2151   enum {AESBlockSize = 16};
2152 
2153   address generate_key_shuffle_mask() {
2154     __ align(16);
2155     StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2156     address start = __ pc();
2157     __ emit_data(0x00010203, relocInfo::none, 0 );
2158     __ emit_data(0x04050607, relocInfo::none, 0 );
2159     __ emit_data(0x08090a0b, relocInfo::none, 0 );
2160     __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2161     return start;
2162   }
2163 
2164   // Utility routine for loading a 128-bit key word in little endian format
2165   // can optionally specify that the shuffle mask is already in an xmmregister
2166   void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2167     __ movdqu(xmmdst, Address(key, offset));
2168     if (xmm_shuf_mask != NULL) {
2169       __ pshufb(xmmdst, xmm_shuf_mask);
2170     } else {
2171       __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2172     }
2173   }
2174 
2175   // aesenc using specified key+offset
2176   // can optionally specify that the shuffle mask is already in an xmmregister
2177   void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2178     load_key(xmmtmp, key, offset, xmm_shuf_mask);
2179     __ aesenc(xmmdst, xmmtmp);
2180   }
2181 
2182   // aesdec using specified key+offset
2183   // can optionally specify that the shuffle mask is already in an xmmregister
2184   void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2185     load_key(xmmtmp, key, offset, xmm_shuf_mask);
2186     __ aesdec(xmmdst, xmmtmp);
2187   }
2188 
2189 
2190   // Arguments:
2191   //
2192   // Inputs:
2193   //   c_rarg0   - source byte array address
2194   //   c_rarg1   - destination byte array address
2195   //   c_rarg2   - K (key) in little endian int array
2196   //
2197   address generate_aescrypt_encryptBlock() {
2198     assert(UseAES, "need AES instructions and misaligned SSE support");
2199     __ align(CodeEntryAlignment);
2200     StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2201     Label L_doLast;
2202     address start = __ pc();
2203 
2204     const Register from        = rdx;      // source array address
2205     const Register to          = rdx;      // destination array address
2206     const Register key         = rcx;      // key array address
2207     const Register keylen      = rax;
2208     const Address  from_param(rbp, 8+0);
2209     const Address  to_param  (rbp, 8+4);
2210     const Address  key_param (rbp, 8+8);
2211 
2212     const XMMRegister xmm_result = xmm0;
2213     const XMMRegister xmm_key_shuf_mask = xmm1;
2214     const XMMRegister xmm_temp1  = xmm2;
2215     const XMMRegister xmm_temp2  = xmm3;
2216     const XMMRegister xmm_temp3  = xmm4;
2217     const XMMRegister xmm_temp4  = xmm5;
2218 
2219     __ enter();   // required for proper stackwalking of RuntimeStub frame
2220     __ movptr(from, from_param);
2221     __ movptr(key, key_param);
2222 
2223     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2224     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2225 
2226     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2227     __ movdqu(xmm_result, Address(from, 0));  // get 16 bytes of input
2228     __ movptr(to, to_param);
2229 
2230     // For encryption, the java expanded key ordering is just what we need
2231 
2232     load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2233     __ pxor(xmm_result, xmm_temp1);
2234 
2235     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2236     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2237     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2238     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2239 
2240     __ aesenc(xmm_result, xmm_temp1);
2241     __ aesenc(xmm_result, xmm_temp2);
2242     __ aesenc(xmm_result, xmm_temp3);
2243     __ aesenc(xmm_result, xmm_temp4);
2244 
2245     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2246     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2247     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2248     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2249 
2250     __ aesenc(xmm_result, xmm_temp1);
2251     __ aesenc(xmm_result, xmm_temp2);
2252     __ aesenc(xmm_result, xmm_temp3);
2253     __ aesenc(xmm_result, xmm_temp4);
2254 
2255     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2256     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2257 
2258     __ cmpl(keylen, 44);
2259     __ jccb(Assembler::equal, L_doLast);
2260 
2261     __ aesenc(xmm_result, xmm_temp1);
2262     __ aesenc(xmm_result, xmm_temp2);
2263 
2264     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2265     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2266 
2267     __ cmpl(keylen, 52);
2268     __ jccb(Assembler::equal, L_doLast);
2269 
2270     __ aesenc(xmm_result, xmm_temp1);
2271     __ aesenc(xmm_result, xmm_temp2);
2272 
2273     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2274     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2275 
2276     __ BIND(L_doLast);
2277     __ aesenc(xmm_result, xmm_temp1);
2278     __ aesenclast(xmm_result, xmm_temp2);
2279     __ movdqu(Address(to, 0), xmm_result);        // store the result
2280     __ xorptr(rax, rax); // return 0
2281     __ leave(); // required for proper stackwalking of RuntimeStub frame
2282     __ ret(0);
2283 
2284     return start;
2285   }
2286 
2287 
2288   // Arguments:
2289   //
2290   // Inputs:
2291   //   c_rarg0   - source byte array address
2292   //   c_rarg1   - destination byte array address
2293   //   c_rarg2   - K (key) in little endian int array
2294   //
2295   address generate_aescrypt_decryptBlock() {
2296     assert(UseAES, "need AES instructions and misaligned SSE support");
2297     __ align(CodeEntryAlignment);
2298     StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2299     Label L_doLast;
2300     address start = __ pc();
2301 
2302     const Register from        = rdx;      // source array address
2303     const Register to          = rdx;      // destination array address
2304     const Register key         = rcx;      // key array address
2305     const Register keylen      = rax;
2306     const Address  from_param(rbp, 8+0);
2307     const Address  to_param  (rbp, 8+4);
2308     const Address  key_param (rbp, 8+8);
2309 
2310     const XMMRegister xmm_result = xmm0;
2311     const XMMRegister xmm_key_shuf_mask = xmm1;
2312     const XMMRegister xmm_temp1  = xmm2;
2313     const XMMRegister xmm_temp2  = xmm3;
2314     const XMMRegister xmm_temp3  = xmm4;
2315     const XMMRegister xmm_temp4  = xmm5;
2316 
2317     __ enter(); // required for proper stackwalking of RuntimeStub frame
2318     __ movptr(from, from_param);
2319     __ movptr(key, key_param);
2320 
2321     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2322     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2323 
2324     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2325     __ movdqu(xmm_result, Address(from, 0));
2326     __ movptr(to, to_param);
2327 
2328     // for decryption java expanded key ordering is rotated one position from what we want
2329     // so we start from 0x10 here and hit 0x00 last
2330     // we don't know if the key is aligned, hence not using load-execute form
2331     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2332     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2333     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2334     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2335 
2336     __ pxor  (xmm_result, xmm_temp1);
2337     __ aesdec(xmm_result, xmm_temp2);
2338     __ aesdec(xmm_result, xmm_temp3);
2339     __ aesdec(xmm_result, xmm_temp4);
2340 
2341     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2342     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2343     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2344     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2345 
2346     __ aesdec(xmm_result, xmm_temp1);
2347     __ aesdec(xmm_result, xmm_temp2);
2348     __ aesdec(xmm_result, xmm_temp3);
2349     __ aesdec(xmm_result, xmm_temp4);
2350 
2351     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2352     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2353     load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2354 
2355     __ cmpl(keylen, 44);
2356     __ jccb(Assembler::equal, L_doLast);
2357 
2358     __ aesdec(xmm_result, xmm_temp1);
2359     __ aesdec(xmm_result, xmm_temp2);
2360 
2361     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2362     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2363 
2364     __ cmpl(keylen, 52);
2365     __ jccb(Assembler::equal, L_doLast);
2366 
2367     __ aesdec(xmm_result, xmm_temp1);
2368     __ aesdec(xmm_result, xmm_temp2);
2369 
2370     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2371     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2372 
2373     __ BIND(L_doLast);
2374     __ aesdec(xmm_result, xmm_temp1);
2375     __ aesdec(xmm_result, xmm_temp2);
2376 
2377     // for decryption the aesdeclast operation is always on key+0x00
2378     __ aesdeclast(xmm_result, xmm_temp3);
2379     __ movdqu(Address(to, 0), xmm_result);  // store the result
2380     __ xorptr(rax, rax); // return 0
2381     __ leave(); // required for proper stackwalking of RuntimeStub frame
2382     __ ret(0);
2383 
2384     return start;
2385   }
2386 
2387   void handleSOERegisters(bool saving) {
2388     const int saveFrameSizeInBytes = 4 * wordSize;
2389     const Address saved_rbx     (rbp, -3 * wordSize);
2390     const Address saved_rsi     (rbp, -2 * wordSize);
2391     const Address saved_rdi     (rbp, -1 * wordSize);
2392 
2393     if (saving) {
2394       __ subptr(rsp, saveFrameSizeInBytes);
2395       __ movptr(saved_rsi, rsi);
2396       __ movptr(saved_rdi, rdi);
2397       __ movptr(saved_rbx, rbx);
2398     } else {
2399       // restoring
2400       __ movptr(rsi, saved_rsi);
2401       __ movptr(rdi, saved_rdi);
2402       __ movptr(rbx, saved_rbx);
2403     }
2404   }
2405 
2406   // Arguments:
2407   //
2408   // Inputs:
2409   //   c_rarg0   - source byte array address
2410   //   c_rarg1   - destination byte array address
2411   //   c_rarg2   - K (key) in little endian int array
2412   //   c_rarg3   - r vector byte array address
2413   //   c_rarg4   - input length
2414   //
2415   // Output:
2416   //   rax       - input length
2417   //
2418   address generate_cipherBlockChaining_encryptAESCrypt() {
2419     assert(UseAES, "need AES instructions and misaligned SSE support");
2420     __ align(CodeEntryAlignment);
2421     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2422     address start = __ pc();
2423 
2424     Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2425     const Register from        = rsi;      // source array address
2426     const Register to          = rdx;      // destination array address
2427     const Register key         = rcx;      // key array address
2428     const Register rvec        = rdi;      // r byte array initialized from initvector array address
2429                                            // and left with the results of the last encryption block
2430     const Register len_reg     = rbx;      // src len (must be multiple of blocksize 16)
2431     const Register pos         = rax;
2432 
2433     // xmm register assignments for the loops below
2434     const XMMRegister xmm_result = xmm0;
2435     const XMMRegister xmm_temp   = xmm1;
2436     // first 6 keys preloaded into xmm2-xmm7
2437     const int XMM_REG_NUM_KEY_FIRST = 2;
2438     const int XMM_REG_NUM_KEY_LAST  = 7;
2439     const XMMRegister xmm_key0   = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2440 
2441     __ enter(); // required for proper stackwalking of RuntimeStub frame
2442     handleSOERegisters(true /*saving*/);
2443 
2444     // load registers from incoming parameters
2445     const Address  from_param(rbp, 8+0);
2446     const Address  to_param  (rbp, 8+4);
2447     const Address  key_param (rbp, 8+8);
2448     const Address  rvec_param (rbp, 8+12);
2449     const Address  len_param  (rbp, 8+16);
2450     __ movptr(from , from_param);
2451     __ movptr(to   , to_param);
2452     __ movptr(key  , key_param);
2453     __ movptr(rvec , rvec_param);
2454     __ movptr(len_reg , len_param);
2455 
2456     const XMMRegister xmm_key_shuf_mask = xmm_temp;  // used temporarily to swap key bytes up front
2457     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2458     // load up xmm regs 2 thru 7 with keys 0-5
2459     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2460       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2461       offset += 0x10;
2462     }
2463 
2464     __ movdqu(xmm_result, Address(rvec, 0x00));   // initialize xmm_result with r vec
2465 
2466     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2467     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2468     __ cmpl(rax, 44);
2469     __ jcc(Assembler::notEqual, L_key_192_256);
2470 
2471     // 128 bit code follows here
2472     __ movl(pos, 0);
2473     __ align(OptoLoopAlignment);
2474     __ BIND(L_loopTop_128);
2475     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2476     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2477 
2478     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2479     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2480       __ aesenc(xmm_result, as_XMMRegister(rnum));
2481     }
2482     for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2483       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2484     }
2485     load_key(xmm_temp, key, 0xa0);
2486     __ aesenclast(xmm_result, xmm_temp);
2487 
2488     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
2489     // no need to store r to memory until we exit
2490     __ addptr(pos, AESBlockSize);
2491     __ subptr(len_reg, AESBlockSize);
2492     __ jcc(Assembler::notEqual, L_loopTop_128);
2493 
2494     __ BIND(L_exit);
2495     __ movdqu(Address(rvec, 0), xmm_result);     // final value of r stored in rvec of CipherBlockChaining object
2496 
2497     handleSOERegisters(false /*restoring*/);
2498     __ movptr(rax, len_param); // return length
2499     __ leave();                                  // required for proper stackwalking of RuntimeStub frame
2500     __ ret(0);
2501 
2502     __ BIND(L_key_192_256);
2503     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2504     __ cmpl(rax, 52);
2505     __ jcc(Assembler::notEqual, L_key_256);
2506 
2507     // 192-bit code follows here (could be changed to use more xmm registers)
2508     __ movl(pos, 0);
2509     __ align(OptoLoopAlignment);
2510     __ BIND(L_loopTop_192);
2511     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2512     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2513 
2514     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2515     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2516       __ aesenc(xmm_result, as_XMMRegister(rnum));
2517     }
2518     for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2519       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2520     }
2521     load_key(xmm_temp, key, 0xc0);
2522     __ aesenclast(xmm_result, xmm_temp);
2523 
2524     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);   // store into the next 16 bytes of output
2525     // no need to store r to memory until we exit
2526     __ addptr(pos, AESBlockSize);
2527     __ subptr(len_reg, AESBlockSize);
2528     __ jcc(Assembler::notEqual, L_loopTop_192);
2529     __ jmp(L_exit);
2530 
2531     __ BIND(L_key_256);
2532     // 256-bit code follows here (could be changed to use more xmm registers)
2533     __ movl(pos, 0);
2534     __ align(OptoLoopAlignment);
2535     __ BIND(L_loopTop_256);
2536     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2537     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2538 
2539     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2540     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2541       __ aesenc(xmm_result, as_XMMRegister(rnum));
2542     }
2543     for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2544       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2545     }
2546     load_key(xmm_temp, key, 0xe0);
2547     __ aesenclast(xmm_result, xmm_temp);
2548 
2549     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);   // store into the next 16 bytes of output
2550     // no need to store r to memory until we exit
2551     __ addptr(pos, AESBlockSize);
2552     __ subptr(len_reg, AESBlockSize);
2553     __ jcc(Assembler::notEqual, L_loopTop_256);
2554     __ jmp(L_exit);
2555 
2556     return start;
2557   }
2558 
2559 
2560   // CBC AES Decryption.
2561   // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2562   //
2563   // Arguments:
2564   //
2565   // Inputs:
2566   //   c_rarg0   - source byte array address
2567   //   c_rarg1   - destination byte array address
2568   //   c_rarg2   - K (key) in little endian int array
2569   //   c_rarg3   - r vector byte array address
2570   //   c_rarg4   - input length
2571   //
2572   // Output:
2573   //   rax       - input length
2574   //
2575 
2576   address generate_cipherBlockChaining_decryptAESCrypt() {
2577     assert(UseAES, "need AES instructions and misaligned SSE support");
2578     __ align(CodeEntryAlignment);
2579     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2580     address start = __ pc();
2581 
2582     Label L_exit, L_key_192_256, L_key_256;
2583     Label L_singleBlock_loopTop_128;
2584     Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2585     const Register from        = rsi;      // source array address
2586     const Register to          = rdx;      // destination array address
2587     const Register key         = rcx;      // key array address
2588     const Register rvec        = rdi;      // r byte array initialized from initvector array address
2589                                            // and left with the results of the last encryption block
2590     const Register len_reg     = rbx;      // src len (must be multiple of blocksize 16)
2591     const Register pos         = rax;
2592 
2593     // xmm register assignments for the loops below
2594     const XMMRegister xmm_result = xmm0;
2595     const XMMRegister xmm_temp   = xmm1;
2596     // first 6 keys preloaded into xmm2-xmm7
2597     const int XMM_REG_NUM_KEY_FIRST = 2;
2598     const int XMM_REG_NUM_KEY_LAST  = 7;
2599     const int FIRST_NON_REG_KEY_offset = 0x70;
2600     const XMMRegister xmm_key_first   = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2601 
2602     __ enter(); // required for proper stackwalking of RuntimeStub frame
2603     handleSOERegisters(true /*saving*/);
2604 
2605     // load registers from incoming parameters
2606     const Address  from_param(rbp, 8+0);
2607     const Address  to_param  (rbp, 8+4);
2608     const Address  key_param (rbp, 8+8);
2609     const Address  rvec_param (rbp, 8+12);
2610     const Address  len_param  (rbp, 8+16);
2611     __ movptr(from , from_param);
2612     __ movptr(to   , to_param);
2613     __ movptr(key  , key_param);
2614     __ movptr(rvec , rvec_param);
2615     __ movptr(len_reg , len_param);
2616 
2617     // the java expanded key ordering is rotated one position from what we want
2618     // so we start from 0x10 here and hit 0x00 last
2619     const XMMRegister xmm_key_shuf_mask = xmm1;  // used temporarily to swap key bytes up front
2620     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2621     // load up xmm regs 2 thru 6 with first 5 keys
2622     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2623       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2624       offset += 0x10;
2625     }
2626 
2627     // inside here, use the rvec register to point to previous block cipher
2628     // with which we xor at the end of each newly decrypted block
2629     const Register  prev_block_cipher_ptr = rvec;
2630 
2631     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2632     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2633     __ cmpl(rax, 44);
2634     __ jcc(Assembler::notEqual, L_key_192_256);
2635 
2636 
2637     // 128-bit code follows here, parallelized
2638     __ movl(pos, 0);
2639     __ align(OptoLoopAlignment);
2640     __ BIND(L_singleBlock_loopTop_128);
2641     __ cmpptr(len_reg, 0);           // any blocks left??
2642     __ jcc(Assembler::equal, L_exit);
2643     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
2644     __ pxor  (xmm_result, xmm_key_first);                             // do the aes dec rounds
2645     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2646       __ aesdec(xmm_result, as_XMMRegister(rnum));
2647     }
2648     for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) {   // 128-bit runs up to key offset a0
2649       aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2650     }
2651     load_key(xmm_temp, key, 0x00);                                     // final key is stored in java expanded array at offset 0
2652     __ aesdeclast(xmm_result, xmm_temp);
2653     __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2654     __ pxor  (xmm_result, xmm_temp);                                  // xor with the current r vector
2655     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
2656     // no need to store r to memory until we exit
2657     __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0));     // set up new ptr
2658     __ addptr(pos, AESBlockSize);
2659     __ subptr(len_reg, AESBlockSize);
2660     __ jmp(L_singleBlock_loopTop_128);
2661 
2662 
2663     __ BIND(L_exit);
2664     __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2665     __ movptr(rvec , rvec_param);                                     // restore this since used in loop
2666     __ movdqu(Address(rvec, 0), xmm_temp);                            // final value of r stored in rvec of CipherBlockChaining object
2667     handleSOERegisters(false /*restoring*/);
2668     __ movptr(rax, len_param); // return length
2669     __ leave();                                                       // required for proper stackwalking of RuntimeStub frame
2670     __ ret(0);
2671 
2672 
2673     __ BIND(L_key_192_256);
2674     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2675     __ cmpl(rax, 52);
2676     __ jcc(Assembler::notEqual, L_key_256);
2677 
2678     // 192-bit code follows here (could be optimized to use parallelism)
2679     __ movl(pos, 0);
2680     __ align(OptoLoopAlignment);
2681     __ BIND(L_singleBlock_loopTop_192);
2682     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
2683     __ pxor  (xmm_result, xmm_key_first);                             // do the aes dec rounds
2684     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2685       __ aesdec(xmm_result, as_XMMRegister(rnum));
2686     }
2687     for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) {   // 192-bit runs up to key offset c0
2688       aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2689     }
2690     load_key(xmm_temp, key, 0x00);                                     // final key is stored in java expanded array at offset 0
2691     __ aesdeclast(xmm_result, xmm_temp);
2692     __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2693     __ pxor  (xmm_result, xmm_temp);                                  // xor with the current r vector
2694     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
2695     // no need to store r to memory until we exit
2696     __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0));     // set up new ptr
2697     __ addptr(pos, AESBlockSize);
2698     __ subptr(len_reg, AESBlockSize);
2699     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2700     __ jmp(L_exit);
2701 
2702     __ BIND(L_key_256);
2703     // 256-bit code follows here (could be optimized to use parallelism)
2704     __ movl(pos, 0);
2705     __ align(OptoLoopAlignment);
2706     __ BIND(L_singleBlock_loopTop_256);
2707     __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of cipher input
2708     __ pxor  (xmm_result, xmm_key_first);                             // do the aes dec rounds
2709     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2710       __ aesdec(xmm_result, as_XMMRegister(rnum));
2711     }
2712     for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) {   // 256-bit runs up to key offset e0
2713       aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2714     }
2715     load_key(xmm_temp, key, 0x00);                                     // final key is stored in java expanded array at offset 0
2716     __ aesdeclast(xmm_result, xmm_temp);
2717     __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2718     __ pxor  (xmm_result, xmm_temp);                                  // xor with the current r vector
2719     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
2720     // no need to store r to memory until we exit
2721     __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0));     // set up new ptr
2722     __ addptr(pos, AESBlockSize);
2723     __ subptr(len_reg, AESBlockSize);
2724     __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2725     __ jmp(L_exit);
2726 
2727     return start;
2728   }
2729 
2730   // byte swap x86 long
2731   address generate_ghash_long_swap_mask() {
2732     __ align(CodeEntryAlignment);
2733     StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
2734     address start = __ pc();
2735     __ emit_data(0x0b0a0908, relocInfo::none, 0);
2736     __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
2737     __ emit_data(0x03020100, relocInfo::none, 0);
2738     __ emit_data(0x07060504, relocInfo::none, 0);
2739 
2740   return start;
2741   }
2742 
2743   // byte swap x86 byte array
2744   address generate_ghash_byte_swap_mask() {
2745     __ align(CodeEntryAlignment);
2746     StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
2747     address start = __ pc();
2748     __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2749     __ emit_data(0x08090a0b, relocInfo::none, 0);
2750     __ emit_data(0x04050607, relocInfo::none, 0);
2751     __ emit_data(0x00010203, relocInfo::none, 0);
2752   return start;
2753   }
2754 
2755   /* Single and multi-block ghash operations */
2756   address generate_ghash_processBlocks() {
2757     assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
2758     __ align(CodeEntryAlignment);
2759     Label L_ghash_loop, L_exit;
2760     StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
2761     address start = __ pc();
2762 
2763     const Register state        = rdi;
2764     const Register subkeyH      = rsi;
2765     const Register data         = rdx;
2766     const Register blocks       = rcx;
2767 
2768     const Address  state_param(rbp, 8+0);
2769     const Address  subkeyH_param(rbp, 8+4);
2770     const Address  data_param(rbp, 8+8);
2771     const Address  blocks_param(rbp, 8+12);
2772 
2773     const XMMRegister xmm_temp0 = xmm0;
2774     const XMMRegister xmm_temp1 = xmm1;
2775     const XMMRegister xmm_temp2 = xmm2;
2776     const XMMRegister xmm_temp3 = xmm3;
2777     const XMMRegister xmm_temp4 = xmm4;
2778     const XMMRegister xmm_temp5 = xmm5;
2779     const XMMRegister xmm_temp6 = xmm6;
2780     const XMMRegister xmm_temp7 = xmm7;
2781 
2782     __ enter();
2783     handleSOERegisters(true);  // Save registers
2784 
2785     __ movptr(state, state_param);
2786     __ movptr(subkeyH, subkeyH_param);
2787     __ movptr(data, data_param);
2788     __ movptr(blocks, blocks_param);
2789 
2790     __ movdqu(xmm_temp0, Address(state, 0));
2791     __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
2792 
2793     __ movdqu(xmm_temp1, Address(subkeyH, 0));
2794     __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
2795 
2796     __ BIND(L_ghash_loop);
2797     __ movdqu(xmm_temp2, Address(data, 0));
2798     __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
2799 
2800     __ pxor(xmm_temp0, xmm_temp2);
2801 
2802     //
2803     // Multiply with the hash key
2804     //
2805     __ movdqu(xmm_temp3, xmm_temp0);
2806     __ pclmulqdq(xmm_temp3, xmm_temp1, 0);      // xmm3 holds a0*b0
2807     __ movdqu(xmm_temp4, xmm_temp0);
2808     __ pclmulqdq(xmm_temp4, xmm_temp1, 16);     // xmm4 holds a0*b1
2809 
2810     __ movdqu(xmm_temp5, xmm_temp0);
2811     __ pclmulqdq(xmm_temp5, xmm_temp1, 1);      // xmm5 holds a1*b0
2812     __ movdqu(xmm_temp6, xmm_temp0);
2813     __ pclmulqdq(xmm_temp6, xmm_temp1, 17);     // xmm6 holds a1*b1
2814 
2815     __ pxor(xmm_temp4, xmm_temp5);      // xmm4 holds a0*b1 + a1*b0
2816 
2817     __ movdqu(xmm_temp5, xmm_temp4);    // move the contents of xmm4 to xmm5
2818     __ psrldq(xmm_temp4, 8);    // shift by xmm4 64 bits to the right
2819     __ pslldq(xmm_temp5, 8);    // shift by xmm5 64 bits to the left
2820     __ pxor(xmm_temp3, xmm_temp5);
2821     __ pxor(xmm_temp6, xmm_temp4);      // Register pair <xmm6:xmm3> holds the result
2822                                         // of the carry-less multiplication of
2823                                         // xmm0 by xmm1.
2824 
2825     // We shift the result of the multiplication by one bit position
2826     // to the left to cope for the fact that the bits are reversed.
2827     __ movdqu(xmm_temp7, xmm_temp3);
2828     __ movdqu(xmm_temp4, xmm_temp6);
2829     __ pslld (xmm_temp3, 1);
2830     __ pslld(xmm_temp6, 1);
2831     __ psrld(xmm_temp7, 31);
2832     __ psrld(xmm_temp4, 31);
2833     __ movdqu(xmm_temp5, xmm_temp7);
2834     __ pslldq(xmm_temp4, 4);
2835     __ pslldq(xmm_temp7, 4);
2836     __ psrldq(xmm_temp5, 12);
2837     __ por(xmm_temp3, xmm_temp7);
2838     __ por(xmm_temp6, xmm_temp4);
2839     __ por(xmm_temp6, xmm_temp5);
2840 
2841     //
2842     // First phase of the reduction
2843     //
2844     // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
2845     // independently.
2846     __ movdqu(xmm_temp7, xmm_temp3);
2847     __ movdqu(xmm_temp4, xmm_temp3);
2848     __ movdqu(xmm_temp5, xmm_temp3);
2849     __ pslld(xmm_temp7, 31);    // packed right shift shifting << 31
2850     __ pslld(xmm_temp4, 30);    // packed right shift shifting << 30
2851     __ pslld(xmm_temp5, 25);    // packed right shift shifting << 25
2852     __ pxor(xmm_temp7, xmm_temp4);      // xor the shifted versions
2853     __ pxor(xmm_temp7, xmm_temp5);
2854     __ movdqu(xmm_temp4, xmm_temp7);
2855     __ pslldq(xmm_temp7, 12);
2856     __ psrldq(xmm_temp4, 4);
2857     __ pxor(xmm_temp3, xmm_temp7);      // first phase of the reduction complete
2858 
2859     //
2860     // Second phase of the reduction
2861     //
2862     // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
2863     // shift operations.
2864     __ movdqu(xmm_temp2, xmm_temp3);
2865     __ movdqu(xmm_temp7, xmm_temp3);
2866     __ movdqu(xmm_temp5, xmm_temp3);
2867     __ psrld(xmm_temp2, 1);     // packed left shifting >> 1
2868     __ psrld(xmm_temp7, 2);     // packed left shifting >> 2
2869     __ psrld(xmm_temp5, 7);     // packed left shifting >> 7
2870     __ pxor(xmm_temp2, xmm_temp7);      // xor the shifted versions
2871     __ pxor(xmm_temp2, xmm_temp5);
2872     __ pxor(xmm_temp2, xmm_temp4);
2873     __ pxor(xmm_temp3, xmm_temp2);
2874     __ pxor(xmm_temp6, xmm_temp3);      // the result is in xmm6
2875 
2876     __ decrement(blocks);
2877     __ jcc(Assembler::zero, L_exit);
2878     __ movdqu(xmm_temp0, xmm_temp6);
2879     __ addptr(data, 16);
2880     __ jmp(L_ghash_loop);
2881 
2882     __ BIND(L_exit);
2883        // Byte swap 16-byte result
2884     __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
2885     __ movdqu(Address(state, 0), xmm_temp6);   // store the result
2886 
2887     handleSOERegisters(false);  // restore registers
2888     __ leave();
2889     __ ret(0);
2890     return start;
2891   }
2892 
2893   /**
2894    *  Arguments:
2895    *
2896    * Inputs:
2897    *   rsp(4)   - int crc
2898    *   rsp(8)   - byte* buf
2899    *   rsp(12)  - int length
2900    *
2901    * Ouput:
2902    *       rax   - int crc result
2903    */
2904   address generate_updateBytesCRC32() {
2905     assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2906 
2907     __ align(CodeEntryAlignment);
2908     StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2909 
2910     address start = __ pc();
2911 
2912     const Register crc   = rdx;  // crc
2913     const Register buf   = rsi;  // source java byte array address
2914     const Register len   = rcx;  // length
2915     const Register table = rdi;  // crc_table address (reuse register)
2916     const Register tmp   = rbx;
2917     assert_different_registers(crc, buf, len, table, tmp, rax);
2918 
2919     BLOCK_COMMENT("Entry:");
2920     __ enter(); // required for proper stackwalking of RuntimeStub frame
2921     __ push(rsi);
2922     __ push(rdi);
2923     __ push(rbx);
2924 
2925     Address crc_arg(rbp, 8 + 0);
2926     Address buf_arg(rbp, 8 + 4);
2927     Address len_arg(rbp, 8 + 8);
2928 
2929     // Load up:
2930     __ movl(crc,   crc_arg);
2931     __ movptr(buf, buf_arg);
2932     __ movl(len,   len_arg);
2933 
2934     __ kernel_crc32(crc, buf, len, table, tmp);
2935 
2936     __ movl(rax, crc);
2937     __ pop(rbx);
2938     __ pop(rdi);
2939     __ pop(rsi);
2940     __ leave(); // required for proper stackwalking of RuntimeStub frame
2941     __ ret(0);
2942 
2943     return start;
2944   }
2945 
2946   // Safefetch stubs.
2947   void generate_safefetch(const char* name, int size, address* entry,
2948                           address* fault_pc, address* continuation_pc) {
2949     // safefetch signatures:
2950     //   int      SafeFetch32(int*      adr, int      errValue);
2951     //   intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2952 
2953     StubCodeMark mark(this, "StubRoutines", name);
2954 
2955     // Entry point, pc or function descriptor.
2956     *entry = __ pc();
2957 
2958     __ movl(rax, Address(rsp, 0x8));
2959     __ movl(rcx, Address(rsp, 0x4));
2960     // Load *adr into eax, may fault.
2961     *fault_pc = __ pc();
2962     switch (size) {
2963       case 4:
2964         // int32_t
2965         __ movl(rax, Address(rcx, 0));
2966         break;
2967       case 8:
2968         // int64_t
2969         Unimplemented();
2970         break;
2971       default:
2972         ShouldNotReachHere();
2973     }
2974 
2975     // Return errValue or *adr.
2976     *continuation_pc = __ pc();
2977     __ ret(0);
2978   }
2979 
2980  public:
2981   // Information about frame layout at time of blocking runtime call.
2982   // Note that we only have to preserve callee-saved registers since
2983   // the compilers are responsible for supplying a continuation point
2984   // if they expect all registers to be preserved.
2985   enum layout {
2986     thread_off,    // last_java_sp
2987     arg1_off,
2988     arg2_off,
2989     rbp_off,       // callee saved register
2990     ret_pc,
2991     framesize
2992   };
2993 
2994  private:
2995 
2996 #undef  __
2997 #define __ masm->
2998 
2999   //------------------------------------------------------------------------------------------------------------------------
3000   // Continuation point for throwing of implicit exceptions that are not handled in
3001   // the current activation. Fabricates an exception oop and initiates normal
3002   // exception dispatching in this frame.
3003   //
3004   // Previously the compiler (c2) allowed for callee save registers on Java calls.
3005   // This is no longer true after adapter frames were removed but could possibly
3006   // be brought back in the future if the interpreter code was reworked and it
3007   // was deemed worthwhile. The comment below was left to describe what must
3008   // happen here if callee saves were resurrected. As it stands now this stub
3009   // could actually be a vanilla BufferBlob and have now oopMap at all.
3010   // Since it doesn't make much difference we've chosen to leave it the
3011   // way it was in the callee save days and keep the comment.
3012 
3013   // If we need to preserve callee-saved values we need a callee-saved oop map and
3014   // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3015   // If the compiler needs all registers to be preserved between the fault
3016   // point and the exception handler then it must assume responsibility for that in
3017   // AbstractCompiler::continuation_for_implicit_null_exception or
3018   // continuation_for_implicit_division_by_zero_exception. All other implicit
3019   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3020   // either at call sites or otherwise assume that stack unwinding will be initiated,
3021   // so caller saved registers were assumed volatile in the compiler.
3022   address generate_throw_exception(const char* name, address runtime_entry,
3023                                    Register arg1 = noreg, Register arg2 = noreg) {
3024 
3025     int insts_size = 256;
3026     int locs_size  = 32;
3027 
3028     CodeBuffer code(name, insts_size, locs_size);
3029     OopMapSet* oop_maps  = new OopMapSet();
3030     MacroAssembler* masm = new MacroAssembler(&code);
3031 
3032     address start = __ pc();
3033 
3034     // This is an inlined and slightly modified version of call_VM
3035     // which has the ability to fetch the return PC out of
3036     // thread-local storage and also sets up last_Java_sp slightly
3037     // differently than the real call_VM
3038     Register java_thread = rbx;
3039     __ get_thread(java_thread);
3040 
3041     __ enter(); // required for proper stackwalking of RuntimeStub frame
3042 
3043     // pc and rbp, already pushed
3044     __ subptr(rsp, (framesize-2) * wordSize); // prolog
3045 
3046     // Frame is now completed as far as size and linkage.
3047 
3048     int frame_complete = __ pc() - start;
3049 
3050     // push java thread (becomes first argument of C function)
3051     __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3052     if (arg1 != noreg) {
3053       __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3054     }
3055     if (arg2 != noreg) {
3056       assert(arg1 != noreg, "missing reg arg");
3057       __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3058     }
3059 
3060     // Set up last_Java_sp and last_Java_fp
3061     __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3062 
3063     // Call runtime
3064     BLOCK_COMMENT("call runtime_entry");
3065     __ call(RuntimeAddress(runtime_entry));
3066     // Generate oop map
3067     OopMap* map =  new OopMap(framesize, 0);
3068     oop_maps->add_gc_map(__ pc() - start, map);
3069 
3070     // restore the thread (cannot use the pushed argument since arguments
3071     // may be overwritten by C code generated by an optimizing compiler);
3072     // however can use the register value directly if it is callee saved.
3073     __ get_thread(java_thread);
3074 
3075     __ reset_last_Java_frame(java_thread, true, false);
3076 
3077     __ leave(); // required for proper stackwalking of RuntimeStub frame
3078 
3079     // check for pending exceptions
3080 #ifdef ASSERT
3081     Label L;
3082     __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3083     __ jcc(Assembler::notEqual, L);
3084     __ should_not_reach_here();
3085     __ bind(L);
3086 #endif /* ASSERT */
3087     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3088 
3089 
3090     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3091     return stub->entry_point();
3092   }
3093 
3094 
3095   void create_control_words() {
3096     // Round to nearest, 53-bit mode, exceptions masked
3097     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;
3098     // Round to zero, 53-bit mode, exception mased
3099     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3100     // Round to nearest, 24-bit mode, exceptions masked
3101     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;
3102     // Round to nearest, 64-bit mode, exceptions masked
3103     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;
3104     // Round to nearest, 64-bit mode, exceptions masked
3105     StubRoutines::_mxcsr_std           = 0x1F80;
3106     // Note: the following two constants are 80-bit values
3107     //       layout is critical for correct loading by FPU.
3108     // Bias for strict fp multiply/divide
3109     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3110     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3111     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3112     // Un-Bias for strict fp multiply/divide
3113     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3114     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3115     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3116   }
3117 
3118   //---------------------------------------------------------------------------
3119   // Initialization
3120 
3121   void generate_initial() {
3122     // Generates all stubs and initializes the entry points
3123 
3124     //------------------------------------------------------------------------------------------------------------------------
3125     // entry points that exist in all platforms
3126     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3127     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3128     StubRoutines::_forward_exception_entry      = generate_forward_exception();
3129 
3130     StubRoutines::_call_stub_entry              =
3131       generate_call_stub(StubRoutines::_call_stub_return_address);
3132     // is referenced by megamorphic call
3133     StubRoutines::_catch_exception_entry        = generate_catch_exception();
3134 
3135     // These are currently used by Solaris/Intel
3136     StubRoutines::_atomic_xchg_entry            = generate_atomic_xchg();
3137 
3138     StubRoutines::_handler_for_unsafe_access_entry =
3139       generate_handler_for_unsafe_access();
3140 
3141     // platform dependent
3142     create_control_words();
3143 
3144     StubRoutines::x86::_verify_mxcsr_entry                 = generate_verify_mxcsr();
3145     StubRoutines::x86::_verify_fpu_cntrl_wrd_entry         = generate_verify_fpu_cntrl_wrd();
3146     StubRoutines::_d2i_wrapper                              = generate_d2i_wrapper(T_INT,
3147                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3148     StubRoutines::_d2l_wrapper                              = generate_d2i_wrapper(T_LONG,
3149                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3150 
3151     // Build this early so it's available for the interpreter
3152     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",           CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3153 
3154     if (UseCRC32Intrinsics) {
3155       // set table address before stub generation which use it
3156       StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3157       StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3158     }
3159   }
3160 
3161 
3162   void generate_all() {
3163     // Generates all stubs and initializes the entry points
3164 
3165     // These entry points require SharedInfo::stack0 to be set up in non-core builds
3166     // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3167     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3168     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3169     StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3170 
3171     //------------------------------------------------------------------------------------------------------------------------
3172     // entry points that are platform specific
3173 
3174     // support for verify_oop (must happen after universe_init)
3175     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop();
3176 
3177     // arraycopy stubs used by compilers
3178     generate_arraycopy_stubs();
3179 
3180     generate_math_stubs();
3181 
3182     // don't bother generating these AES intrinsic stubs unless global flag is set
3183     if (UseAESIntrinsics) {
3184       StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask();  // might be needed by the others
3185 
3186       StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3187       StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3188       StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3189       StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
3190     }
3191 
3192     // Generate GHASH intrinsics code
3193     if (UseGHASHIntrinsics) {
3194       StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
3195       StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
3196       StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3197     }
3198 
3199     // Safefetch stubs.
3200     generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3201                                                    &StubRoutines::_safefetch32_fault_pc,
3202                                                    &StubRoutines::_safefetch32_continuation_pc);
3203     StubRoutines::_safefetchN_entry           = StubRoutines::_safefetch32_entry;
3204     StubRoutines::_safefetchN_fault_pc        = StubRoutines::_safefetch32_fault_pc;
3205     StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3206   }
3207 
3208 
3209  public:
3210   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3211     if (all) {
3212       generate_all();
3213     } else {
3214       generate_initial();
3215     }
3216   }
3217 }; // end class declaration
3218 
3219 
3220 void StubGenerator_generate(CodeBuffer* code, bool all) {
3221   StubGenerator g(code, all);
3222 }