1 /*
   2  * Copyright (c) 1999, 2011, 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/assembler.hpp"
  27 #include "assembler_x86.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "nativeInst_x86.hpp"
  30 #include "oops/instanceOop.hpp"
  31 #include "oops/methodOop.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 "utilities/top.hpp"
  41 #ifdef TARGET_OS_FAMILY_linux
  42 # include "thread_linux.inline.hpp"
  43 #endif
  44 #ifdef TARGET_OS_FAMILY_solaris
  45 # include "thread_solaris.inline.hpp"
  46 #endif
  47 #ifdef TARGET_OS_FAMILY_windows
  48 # include "thread_windows.inline.hpp"
  49 #endif
  50 #ifdef COMPILER2
  51 #include "opto/runtime.hpp"
  52 #endif
  53 
  54 // Declaration and definition of StubGenerator (no .hpp file).
  55 // For a more detailed description of the stub routine structure
  56 // see the comment in stubRoutines.hpp
  57 
  58 #define __ _masm->
  59 #define a__ ((Assembler*)_masm)->
  60 
  61 #ifdef PRODUCT
  62 #define BLOCK_COMMENT(str) /* nothing */
  63 #else
  64 #define BLOCK_COMMENT(str) __ block_comment(str)
  65 #endif
  66 
  67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  68 
  69 const int MXCSR_MASK  = 0xFFC0;  // Mask out any pending exceptions
  70 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
  71 
  72 // -------------------------------------------------------------------------------------------------------------------------
  73 // Stub Code definitions
  74 
  75 static address handle_unsafe_access() {
  76   JavaThread* thread = JavaThread::current();
  77   address pc  = thread->saved_exception_pc();
  78   // pc is the instruction which we must emulate
  79   // doing a no-op is fine:  return garbage from the load
  80   // therefore, compute npc
  81   address npc = Assembler::locate_next_instruction(pc);
  82 
  83   // request an async exception
  84   thread->set_pending_unsafe_access_error();
  85 
  86   // return address of next instruction to execute
  87   return npc;
  88 }
  89 
  90 class StubGenerator: public StubCodeGenerator {
  91  private:
  92 
  93 #ifdef PRODUCT
  94 #define inc_counter_np(counter) (0)
  95 #else
  96   void inc_counter_np_(int& counter) {
  97     __ incrementl(ExternalAddress((address)&counter));
  98   }
  99 #define inc_counter_np(counter) \
 100   BLOCK_COMMENT("inc_counter " #counter); \
 101   inc_counter_np_(counter);
 102 #endif //PRODUCT
 103 
 104   void inc_copy_counter_np(BasicType t) {
 105 #ifndef PRODUCT
 106     switch (t) {
 107     case T_BYTE:    inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
 108     case T_SHORT:   inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
 109     case T_INT:     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
 110     case T_LONG:    inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
 111     case T_OBJECT:  inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
 112     }
 113     ShouldNotReachHere();
 114 #endif //PRODUCT
 115   }
 116 
 117   //------------------------------------------------------------------------------------------------------------------------
 118   // Call stubs are used to call Java from C
 119   //
 120   //    [ return_from_Java     ] <--- rsp
 121   //    [ argument word n      ]
 122   //      ...
 123   // -N [ argument word 1      ]
 124   // -7 [ Possible padding for stack alignment ]
 125   // -6 [ Possible padding for stack alignment ]
 126   // -5 [ Possible padding for stack alignment ]
 127   // -4 [ mxcsr save           ] <--- rsp_after_call
 128   // -3 [ saved rbx,            ]
 129   // -2 [ saved rsi            ]
 130   // -1 [ saved rdi            ]
 131   //  0 [ saved rbp,            ] <--- rbp,
 132   //  1 [ return address       ]
 133   //  2 [ ptr. to call wrapper ]
 134   //  3 [ result               ]
 135   //  4 [ result_type          ]
 136   //  5 [ method               ]
 137   //  6 [ entry_point          ]
 138   //  7 [ parameters           ]
 139   //  8 [ parameter_size       ]
 140   //  9 [ thread               ]
 141 
 142 
 143   address generate_call_stub(address& return_address) {
 144     StubCodeMark mark(this, "StubRoutines", "call_stub");
 145     address start = __ pc();
 146 
 147     // stub code parameters / addresses
 148     assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
 149     bool  sse_save = false;
 150     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
 151     const int     locals_count_in_bytes  (4*wordSize);
 152     const Address mxcsr_save    (rbp, -4 * wordSize);
 153     const Address saved_rbx     (rbp, -3 * wordSize);
 154     const Address saved_rsi     (rbp, -2 * wordSize);
 155     const Address saved_rdi     (rbp, -1 * wordSize);
 156     const Address result        (rbp,  3 * wordSize);
 157     const Address result_type   (rbp,  4 * wordSize);
 158     const Address method        (rbp,  5 * wordSize);
 159     const Address entry_point   (rbp,  6 * wordSize);
 160     const Address parameters    (rbp,  7 * wordSize);
 161     const Address parameter_size(rbp,  8 * wordSize);
 162     const Address thread        (rbp,  9 * wordSize); // same as in generate_catch_exception()!
 163     sse_save =  UseSSE > 0;
 164 
 165     // stub code
 166     __ enter();
 167     __ movptr(rcx, parameter_size);              // parameter counter
 168     __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
 169     __ addptr(rcx, locals_count_in_bytes);       // reserve space for register saves
 170     __ subptr(rsp, rcx);
 171     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
 172 
 173     // save rdi, rsi, & rbx, according to C calling conventions
 174     __ movptr(saved_rdi, rdi);
 175     __ movptr(saved_rsi, rsi);
 176     __ movptr(saved_rbx, rbx);
 177     // save and initialize %mxcsr
 178     if (sse_save) {
 179       Label skip_ldmx;
 180       __ stmxcsr(mxcsr_save);
 181       __ movl(rax, mxcsr_save);
 182       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 183       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 184       __ cmp32(rax, mxcsr_std);
 185       __ jcc(Assembler::equal, skip_ldmx);
 186       __ ldmxcsr(mxcsr_std);
 187       __ bind(skip_ldmx);
 188     }
 189 
 190     // make sure the control word is correct.
 191     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 192 
 193 #ifdef ASSERT
 194     // make sure we have no pending exceptions
 195     { Label L;
 196       __ movptr(rcx, thread);
 197       __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 198       __ jcc(Assembler::equal, L);
 199       __ stop("StubRoutines::call_stub: entered with pending exception");
 200       __ bind(L);
 201     }
 202 #endif
 203 
 204     // pass parameters if any
 205     BLOCK_COMMENT("pass parameters if any");
 206     Label parameters_done;
 207     __ movl(rcx, parameter_size);  // parameter counter
 208     __ testl(rcx, rcx);
 209     __ jcc(Assembler::zero, parameters_done);
 210 
 211     // parameter passing loop
 212 
 213     Label loop;
 214     // Copy Java parameters in reverse order (receiver last)
 215     // Note that the argument order is inverted in the process
 216     // source is rdx[rcx: N-1..0]
 217     // dest   is rsp[rbx: 0..N-1]
 218 
 219     __ movptr(rdx, parameters);          // parameter pointer
 220     __ xorptr(rbx, rbx);
 221 
 222     __ BIND(loop);
 223 
 224     // get parameter
 225     __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
 226     __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
 227                     Interpreter::expr_offset_in_bytes(0)), rax);          // store parameter
 228     __ increment(rbx);
 229     __ decrement(rcx);
 230     __ jcc(Assembler::notZero, loop);
 231 
 232     // call Java function
 233     __ BIND(parameters_done);
 234     __ movptr(rbx, method);           // get methodOop
 235     __ movptr(rax, entry_point);      // get entry_point
 236     __ mov(rsi, rsp);                 // set sender sp
 237     BLOCK_COMMENT("call Java function");
 238     __ call(rax);
 239 
 240     BLOCK_COMMENT("call_stub_return_address:");
 241     return_address = __ pc();
 242 
 243 #ifdef COMPILER2
 244     {
 245       Label L_skip;
 246       if (UseSSE >= 2) {
 247         __ verify_FPU(0, "call_stub_return");
 248       } else {
 249         for (int i = 1; i < 8; i++) {
 250           __ ffree(i);
 251         }
 252 
 253         // UseSSE <= 1 so double result should be left on TOS
 254         __ movl(rsi, result_type);
 255         __ cmpl(rsi, T_DOUBLE);
 256         __ jcc(Assembler::equal, L_skip);
 257         if (UseSSE == 0) {
 258           // UseSSE == 0 so float result should be left on TOS
 259           __ cmpl(rsi, T_FLOAT);
 260           __ jcc(Assembler::equal, L_skip);
 261         }
 262         __ ffree(0);
 263       }
 264       __ BIND(L_skip);
 265     }
 266 #endif // COMPILER2
 267 
 268     // store result depending on type
 269     // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 270     __ movptr(rdi, result);
 271     Label is_long, is_float, is_double, exit;
 272     __ movl(rsi, result_type);
 273     __ cmpl(rsi, T_LONG);
 274     __ jcc(Assembler::equal, is_long);
 275     __ cmpl(rsi, T_FLOAT);
 276     __ jcc(Assembler::equal, is_float);
 277     __ cmpl(rsi, T_DOUBLE);
 278     __ jcc(Assembler::equal, is_double);
 279 
 280     // handle T_INT case
 281     __ movl(Address(rdi, 0), rax);
 282     __ BIND(exit);
 283 
 284     // check that FPU stack is empty
 285     __ verify_FPU(0, "generate_call_stub");
 286 
 287     // pop parameters
 288     __ lea(rsp, rsp_after_call);
 289 
 290     // restore %mxcsr
 291     if (sse_save) {
 292       __ ldmxcsr(mxcsr_save);
 293     }
 294 
 295     // restore rdi, rsi and rbx,
 296     __ movptr(rbx, saved_rbx);
 297     __ movptr(rsi, saved_rsi);
 298     __ movptr(rdi, saved_rdi);
 299     __ addptr(rsp, 4*wordSize);
 300 
 301     // return
 302     __ pop(rbp);
 303     __ ret(0);
 304 
 305     // handle return types different from T_INT
 306     __ BIND(is_long);
 307     __ movl(Address(rdi, 0 * wordSize), rax);
 308     __ movl(Address(rdi, 1 * wordSize), rdx);
 309     __ jmp(exit);
 310 
 311     __ BIND(is_float);
 312     // interpreter uses xmm0 for return values
 313     if (UseSSE >= 1) {
 314       __ movflt(Address(rdi, 0), xmm0);
 315     } else {
 316       __ fstp_s(Address(rdi, 0));
 317     }
 318     __ jmp(exit);
 319 
 320     __ BIND(is_double);
 321     // interpreter uses xmm0 for return values
 322     if (UseSSE >= 2) {
 323       __ movdbl(Address(rdi, 0), xmm0);
 324     } else {
 325       __ fstp_d(Address(rdi, 0));
 326     }
 327     __ jmp(exit);
 328 
 329     return start;
 330   }
 331 
 332 
 333   //------------------------------------------------------------------------------------------------------------------------
 334   // Return point for a Java call if there's an exception thrown in Java code.
 335   // The exception is caught and transformed into a pending exception stored in
 336   // JavaThread that can be tested from within the VM.
 337   //
 338   // Note: Usually the parameters are removed by the callee. In case of an exception
 339   //       crossing an activation frame boundary, that is not the case if the callee
 340   //       is compiled code => need to setup the rsp.
 341   //
 342   // rax,: exception oop
 343 
 344   address generate_catch_exception() {
 345     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 346     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
 347     const Address thread        (rbp,  9 * wordSize); // same as in generate_call_stub()!
 348     address start = __ pc();
 349 
 350     // get thread directly
 351     __ movptr(rcx, thread);
 352 #ifdef ASSERT
 353     // verify that threads correspond
 354     { Label L;
 355       __ get_thread(rbx);
 356       __ cmpptr(rbx, rcx);
 357       __ jcc(Assembler::equal, L);
 358       __ stop("StubRoutines::catch_exception: threads must correspond");
 359       __ bind(L);
 360     }
 361 #endif
 362     // set pending exception
 363     __ verify_oop(rax);
 364     __ movptr(Address(rcx, Thread::pending_exception_offset()), rax          );
 365     __ lea(Address(rcx, Thread::exception_file_offset   ()),
 366            ExternalAddress((address)__FILE__));
 367     __ movl(Address(rcx, Thread::exception_line_offset   ()), __LINE__ );
 368     // complete return to VM
 369     assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
 370     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 371 
 372     return start;
 373   }
 374 
 375 
 376   //------------------------------------------------------------------------------------------------------------------------
 377   // Continuation point for runtime calls returning with a pending exception.
 378   // The pending exception check happened in the runtime or native call stub.
 379   // The pending exception in Thread is converted into a Java-level exception.
 380   //
 381   // Contract with Java-level exception handlers:
 382   // rax: exception
 383   // rdx: throwing pc
 384   //
 385   // NOTE: At entry of this stub, exception-pc must be on stack !!
 386 
 387   address generate_forward_exception() {
 388     StubCodeMark mark(this, "StubRoutines", "forward exception");
 389     address start = __ pc();
 390     const Register thread = rcx;
 391 
 392     // other registers used in this stub
 393     const Register exception_oop = rax;
 394     const Register handler_addr  = rbx;
 395     const Register exception_pc  = rdx;
 396 
 397     // Upon entry, the sp points to the return address returning into Java
 398     // (interpreted or compiled) code; i.e., the return address becomes the
 399     // throwing pc.
 400     //
 401     // Arguments pushed before the runtime call are still on the stack but
 402     // the exception handler will reset the stack pointer -> ignore them.
 403     // A potential result in registers can be ignored as well.
 404 
 405 #ifdef ASSERT
 406     // make sure this code is only executed if there is a pending exception
 407     { Label L;
 408       __ get_thread(thread);
 409       __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 410       __ jcc(Assembler::notEqual, L);
 411       __ stop("StubRoutines::forward exception: no pending exception (1)");
 412       __ bind(L);
 413     }
 414 #endif
 415 
 416     // compute exception handler into rbx,
 417     __ get_thread(thread);
 418     __ movptr(exception_pc, Address(rsp, 0));
 419     BLOCK_COMMENT("call exception_handler_for_return_address");
 420     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
 421     __ mov(handler_addr, rax);
 422 
 423     // setup rax & rdx, remove return address & clear pending exception
 424     __ get_thread(thread);
 425     __ pop(exception_pc);
 426     __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
 427     __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 428 
 429 #ifdef ASSERT
 430     // make sure exception is set
 431     { Label L;
 432       __ testptr(exception_oop, exception_oop);
 433       __ jcc(Assembler::notEqual, L);
 434       __ stop("StubRoutines::forward exception: no pending exception (2)");
 435       __ bind(L);
 436     }
 437 #endif
 438 
 439     // Verify that there is really a valid exception in RAX.
 440     __ verify_oop(exception_oop);
 441 
 442     // Restore SP from BP if the exception PC is a MethodHandle call site.
 443     __ cmpl(Address(thread, JavaThread::is_method_handle_return_offset()), 0);
 444     __ cmovptr(Assembler::notEqual, rsp, rbp);
 445 
 446     // continue at exception handler (return address removed)
 447     // rax: exception
 448     // rbx: exception handler
 449     // rdx: throwing pc
 450     __ jmp(handler_addr);
 451 
 452     return start;
 453   }
 454 
 455 
 456   //----------------------------------------------------------------------------------------------------
 457   // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
 458   //
 459   // xchg exists as far back as 8086, lock needed for MP only
 460   // Stack layout immediately after call:
 461   //
 462   // 0 [ret addr ] <--- rsp
 463   // 1 [  ex     ]
 464   // 2 [  dest   ]
 465   //
 466   // Result:   *dest <- ex, return (old *dest)
 467   //
 468   // Note: win32 does not currently use this code
 469 
 470   address generate_atomic_xchg() {
 471     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 472     address start = __ pc();
 473 
 474     __ push(rdx);
 475     Address exchange(rsp, 2 * wordSize);
 476     Address dest_addr(rsp, 3 * wordSize);
 477     __ movl(rax, exchange);
 478     __ movptr(rdx, dest_addr);
 479     __ xchgl(rax, Address(rdx, 0));
 480     __ pop(rdx);
 481     __ ret(0);
 482 
 483     return start;
 484   }
 485 
 486   //----------------------------------------------------------------------------------------------------
 487   // Support for void verify_mxcsr()
 488   //
 489   // This routine is used with -Xcheck:jni to verify that native
 490   // JNI code does not return to Java code without restoring the
 491   // MXCSR register to our expected state.
 492 
 493 
 494   address generate_verify_mxcsr() {
 495     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 496     address start = __ pc();
 497 
 498     const Address mxcsr_save(rsp, 0);
 499 
 500     if (CheckJNICalls && UseSSE > 0 ) {
 501       Label ok_ret;
 502       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 503       __ push(rax);
 504       __ subptr(rsp, wordSize);      // allocate a temp location
 505       __ stmxcsr(mxcsr_save);
 506       __ movl(rax, mxcsr_save);
 507       __ andl(rax, MXCSR_MASK);
 508       __ cmp32(rax, mxcsr_std);
 509       __ jcc(Assembler::equal, ok_ret);
 510 
 511       __ warn("MXCSR changed by native JNI code.");
 512 
 513       __ ldmxcsr(mxcsr_std);
 514 
 515       __ bind(ok_ret);
 516       __ addptr(rsp, wordSize);
 517       __ pop(rax);
 518     }
 519 
 520     __ ret(0);
 521 
 522     return start;
 523   }
 524 
 525 
 526   //---------------------------------------------------------------------------
 527   // Support for void verify_fpu_cntrl_wrd()
 528   //
 529   // This routine is used with -Xcheck:jni to verify that native
 530   // JNI code does not return to Java code without restoring the
 531   // FP control word to our expected state.
 532 
 533   address generate_verify_fpu_cntrl_wrd() {
 534     StubCodeMark mark(this, "StubRoutines", "verify_spcw");
 535     address start = __ pc();
 536 
 537     const Address fpu_cntrl_wrd_save(rsp, 0);
 538 
 539     if (CheckJNICalls) {
 540       Label ok_ret;
 541       __ push(rax);
 542       __ subptr(rsp, wordSize);      // allocate a temp location
 543       __ fnstcw(fpu_cntrl_wrd_save);
 544       __ movl(rax, fpu_cntrl_wrd_save);
 545       __ andl(rax, FPU_CNTRL_WRD_MASK);
 546       ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
 547       __ cmp32(rax, fpu_std);
 548       __ jcc(Assembler::equal, ok_ret);
 549 
 550       __ warn("Floating point control word changed by native JNI code.");
 551 
 552       __ fldcw(fpu_std);
 553 
 554       __ bind(ok_ret);
 555       __ addptr(rsp, wordSize);
 556       __ pop(rax);
 557     }
 558 
 559     __ ret(0);
 560 
 561     return start;
 562   }
 563 
 564   //---------------------------------------------------------------------------
 565   // Wrapper for slow-case handling of double-to-integer conversion
 566   // d2i or f2i fast case failed either because it is nan or because
 567   // of under/overflow.
 568   // Input:  FPU TOS: float value
 569   // Output: rax, (rdx): integer (long) result
 570 
 571   address generate_d2i_wrapper(BasicType t, address fcn) {
 572     StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
 573     address start = __ pc();
 574 
 575   // Capture info about frame layout
 576   enum layout { FPUState_off         = 0,
 577                 rbp_off              = FPUStateSizeInWords,
 578                 rdi_off,
 579                 rsi_off,
 580                 rcx_off,
 581                 rbx_off,
 582                 saved_argument_off,
 583                 saved_argument_off2, // 2nd half of double
 584                 framesize
 585   };
 586 
 587   assert(FPUStateSizeInWords == 27, "update stack layout");
 588 
 589     // Save outgoing argument to stack across push_FPU_state()
 590     __ subptr(rsp, wordSize * 2);
 591     __ fstp_d(Address(rsp, 0));
 592 
 593     // Save CPU & FPU state
 594     __ push(rbx);
 595     __ push(rcx);
 596     __ push(rsi);
 597     __ push(rdi);
 598     __ push(rbp);
 599     __ push_FPU_state();
 600 
 601     // push_FPU_state() resets the FP top of stack
 602     // Load original double into FP top of stack
 603     __ fld_d(Address(rsp, saved_argument_off * wordSize));
 604     // Store double into stack as outgoing argument
 605     __ subptr(rsp, wordSize*2);
 606     __ fst_d(Address(rsp, 0));
 607 
 608     // Prepare FPU for doing math in C-land
 609     __ empty_FPU_stack();
 610     // Call the C code to massage the double.  Result in EAX
 611     if (t == T_INT)
 612       { BLOCK_COMMENT("SharedRuntime::d2i"); }
 613     else if (t == T_LONG)
 614       { BLOCK_COMMENT("SharedRuntime::d2l"); }
 615     __ call_VM_leaf( fcn, 2 );
 616 
 617     // Restore CPU & FPU state
 618     __ pop_FPU_state();
 619     __ pop(rbp);
 620     __ pop(rdi);
 621     __ pop(rsi);
 622     __ pop(rcx);
 623     __ pop(rbx);
 624     __ addptr(rsp, wordSize * 2);
 625 
 626     __ ret(0);
 627 
 628     return start;
 629   }
 630 
 631 
 632   //---------------------------------------------------------------------------
 633   // The following routine generates a subroutine to throw an asynchronous
 634   // UnknownError when an unsafe access gets a fault that could not be
 635   // reasonably prevented by the programmer.  (Example: SIGBUS/OBJERR.)
 636   address generate_handler_for_unsafe_access() {
 637     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 638     address start = __ pc();
 639 
 640     __ push(0);                       // hole for return address-to-be
 641     __ pusha();                       // push registers
 642     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 643     BLOCK_COMMENT("call handle_unsafe_access");
 644     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 645     __ movptr(next_pc, rax);          // stuff next address
 646     __ popa();
 647     __ ret(0);                        // jump to next address
 648 
 649     return start;
 650   }
 651 
 652 
 653   //----------------------------------------------------------------------------------------------------
 654   // Non-destructive plausibility checks for oops
 655 
 656   address generate_verify_oop() {
 657     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 658     address start = __ pc();
 659 
 660     // Incoming arguments on stack after saving rax,:
 661     //
 662     // [tos    ]: saved rdx
 663     // [tos + 1]: saved EFLAGS
 664     // [tos + 2]: return address
 665     // [tos + 3]: char* error message
 666     // [tos + 4]: oop   object to verify
 667     // [tos + 5]: saved rax, - saved by caller and bashed
 668 
 669     Label exit, error;
 670     __ pushf();
 671     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 672     __ push(rdx);                                // save rdx
 673     // make sure object is 'reasonable'
 674     __ movptr(rax, Address(rsp, 4 * wordSize));    // get object
 675     __ testptr(rax, rax);
 676     __ jcc(Assembler::zero, exit);               // if obj is NULL it is ok
 677 
 678     // Check if the oop is in the right area of memory
 679     const int oop_mask = Universe::verify_oop_mask();
 680     const int oop_bits = Universe::verify_oop_bits();
 681     __ mov(rdx, rax);
 682     __ andptr(rdx, oop_mask);
 683     __ cmpptr(rdx, oop_bits);
 684     __ jcc(Assembler::notZero, error);
 685 
 686     // make sure klass is 'reasonable'
 687     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
 688     __ testptr(rax, rax);
 689     __ jcc(Assembler::zero, error);              // if klass is NULL it is broken
 690 
 691     // Check if the klass is in the right area of memory
 692     const int klass_mask = Universe::verify_klass_mask();
 693     const int klass_bits = Universe::verify_klass_bits();
 694     __ mov(rdx, rax);
 695     __ andptr(rdx, klass_mask);
 696     __ cmpptr(rdx, klass_bits);
 697     __ jcc(Assembler::notZero, error);
 698 
 699     // make sure klass' klass is 'reasonable'
 700     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
 701     __ testptr(rax, rax);
 702     __ jcc(Assembler::zero, error);              // if klass' klass is NULL it is broken
 703 
 704     __ mov(rdx, rax);
 705     __ andptr(rdx, klass_mask);
 706     __ cmpptr(rdx, klass_bits);
 707     __ jcc(Assembler::notZero, error);           // if klass not in right area
 708                                                  // of memory it is broken too.
 709 
 710     // return if everything seems ok
 711     __ bind(exit);
 712     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 713     __ pop(rdx);                                 // restore rdx
 714     __ popf();                                   // restore EFLAGS
 715     __ ret(3 * wordSize);                        // pop arguments
 716 
 717     // handle errors
 718     __ bind(error);
 719     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 720     __ pop(rdx);                                 // get saved rdx back
 721     __ popf();                                   // get saved EFLAGS off stack -- will be ignored
 722     __ pusha();                                  // push registers (eip = return address & msg are already pushed)
 723     BLOCK_COMMENT("call MacroAssembler::debug");
 724     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
 725     __ popa();
 726     __ ret(3 * wordSize);                        // pop arguments
 727     return start;
 728   }
 729 
 730   //
 731   //  Generate pre-barrier for array stores
 732   //
 733   //  Input:
 734   //     start   -  starting address
 735   //     count   -  element count
 736   void  gen_write_ref_array_pre_barrier(Register start, Register count) {
 737     assert_different_registers(start, count);
 738     BarrierSet* bs = Universe::heap()->barrier_set();
 739     switch (bs->kind()) {
 740       case BarrierSet::G1SATBCT:
 741       case BarrierSet::G1SATBCTLogging:
 742         {
 743           __ pusha();                      // push registers
 744           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
 745                           start, count);
 746           __ popa();
 747         }
 748         break;
 749       case BarrierSet::CardTableModRef:
 750       case BarrierSet::CardTableExtension:
 751       case BarrierSet::ModRef:
 752         break;
 753       default      :
 754         ShouldNotReachHere();
 755 
 756     }
 757   }
 758 
 759 
 760   //
 761   // Generate a post-barrier for an array store
 762   //
 763   //     start    -  starting address
 764   //     count    -  element count
 765   //
 766   //  The two input registers are overwritten.
 767   //
 768   void  gen_write_ref_array_post_barrier(Register start, Register count) {
 769     BarrierSet* bs = Universe::heap()->barrier_set();
 770     assert_different_registers(start, count);
 771     switch (bs->kind()) {
 772       case BarrierSet::G1SATBCT:
 773       case BarrierSet::G1SATBCTLogging:
 774         {
 775           __ pusha();                      // push registers
 776           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
 777                           start, count);
 778           __ popa();
 779         }
 780         break;
 781 
 782       case BarrierSet::CardTableModRef:
 783       case BarrierSet::CardTableExtension:
 784         {
 785           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
 786           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
 787 
 788           Label L_loop;
 789           const Register end = count;  // elements count; end == start+count-1
 790           assert_different_registers(start, end);
 791 
 792           __ lea(end,  Address(start, count, Address::times_ptr, -wordSize));
 793           __ shrptr(start, CardTableModRefBS::card_shift);
 794           __ shrptr(end,   CardTableModRefBS::card_shift);
 795           __ subptr(end, start); // end --> count
 796         __ BIND(L_loop);
 797           intptr_t disp = (intptr_t) ct->byte_map_base;
 798           Address cardtable(start, count, Address::times_1, disp);
 799           __ movb(cardtable, 0);
 800           __ decrement(count);
 801           __ jcc(Assembler::greaterEqual, L_loop);
 802         }
 803         break;
 804       case BarrierSet::ModRef:
 805         break;
 806       default      :
 807         ShouldNotReachHere();
 808 
 809     }
 810   }
 811 
 812 
 813   // Copy 64 bytes chunks
 814   //
 815   // Inputs:
 816   //   from        - source array address
 817   //   to_from     - destination array address - from
 818   //   qword_count - 8-bytes element count, negative
 819   //
 820   void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
 821     assert( UseSSE >= 2, "supported cpu only" );
 822     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 823     // Copy 64-byte chunks
 824     __ jmpb(L_copy_64_bytes);
 825     __ align(OptoLoopAlignment);
 826   __ BIND(L_copy_64_bytes_loop);
 827 
 828     if(UseUnalignedLoadStores) {
 829       __ movdqu(xmm0, Address(from, 0));
 830       __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
 831       __ movdqu(xmm1, Address(from, 16));
 832       __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
 833       __ movdqu(xmm2, Address(from, 32));
 834       __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
 835       __ movdqu(xmm3, Address(from, 48));
 836       __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
 837 
 838     } else {
 839       __ movq(xmm0, Address(from, 0));
 840       __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
 841       __ movq(xmm1, Address(from, 8));
 842       __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
 843       __ movq(xmm2, Address(from, 16));
 844       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 845       __ movq(xmm3, Address(from, 24));
 846       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 847       __ movq(xmm4, Address(from, 32));
 848       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 849       __ movq(xmm5, Address(from, 40));
 850       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 851       __ movq(xmm6, Address(from, 48));
 852       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 853       __ movq(xmm7, Address(from, 56));
 854       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 855     }
 856 
 857     __ addl(from, 64);
 858   __ BIND(L_copy_64_bytes);
 859     __ subl(qword_count, 8);
 860     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 861     __ addl(qword_count, 8);
 862     __ jccb(Assembler::zero, L_exit);
 863     //
 864     // length is too short, just copy qwords
 865     //
 866   __ BIND(L_copy_8_bytes);
 867     __ movq(xmm0, Address(from, 0));
 868     __ movq(Address(from, to_from, Address::times_1), xmm0);
 869     __ addl(from, 8);
 870     __ decrement(qword_count);
 871     __ jcc(Assembler::greater, L_copy_8_bytes);
 872   __ BIND(L_exit);
 873   }
 874 
 875   // Copy 64 bytes chunks
 876   //
 877   // Inputs:
 878   //   from        - source array address
 879   //   to_from     - destination array address - from
 880   //   qword_count - 8-bytes element count, negative
 881   //
 882   void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
 883     assert( VM_Version::supports_mmx(), "supported cpu only" );
 884     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 885     // Copy 64-byte chunks
 886     __ jmpb(L_copy_64_bytes);
 887     __ align(OptoLoopAlignment);
 888   __ BIND(L_copy_64_bytes_loop);
 889     __ movq(mmx0, Address(from, 0));
 890     __ movq(mmx1, Address(from, 8));
 891     __ movq(mmx2, Address(from, 16));
 892     __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
 893     __ movq(mmx3, Address(from, 24));
 894     __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
 895     __ movq(mmx4, Address(from, 32));
 896     __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
 897     __ movq(mmx5, Address(from, 40));
 898     __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
 899     __ movq(mmx6, Address(from, 48));
 900     __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
 901     __ movq(mmx7, Address(from, 56));
 902     __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
 903     __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
 904     __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
 905     __ addptr(from, 64);
 906   __ BIND(L_copy_64_bytes);
 907     __ subl(qword_count, 8);
 908     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 909     __ addl(qword_count, 8);
 910     __ jccb(Assembler::zero, L_exit);
 911     //
 912     // length is too short, just copy qwords
 913     //
 914   __ BIND(L_copy_8_bytes);
 915     __ movq(mmx0, Address(from, 0));
 916     __ movq(Address(from, to_from, Address::times_1), mmx0);
 917     __ addptr(from, 8);
 918     __ decrement(qword_count);
 919     __ jcc(Assembler::greater, L_copy_8_bytes);
 920   __ BIND(L_exit);
 921     __ emms();
 922   }
 923 
 924   address generate_disjoint_copy(BasicType t, bool aligned,
 925                                  Address::ScaleFactor sf,
 926                                  address* entry, const char *name,
 927                                  bool need_pre_barrier = true) {
 928     __ align(CodeEntryAlignment);
 929     StubCodeMark mark(this, "StubRoutines", name);
 930     address start = __ pc();
 931 
 932     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
 933     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
 934 
 935     int shift = Address::times_ptr - sf;
 936 
 937     const Register from     = rsi;  // source array address
 938     const Register to       = rdi;  // destination array address
 939     const Register count    = rcx;  // elements count
 940     const Register to_from  = to;   // (to - from)
 941     const Register saved_to = rdx;  // saved destination array address
 942 
 943     __ enter(); // required for proper stackwalking of RuntimeStub frame
 944     __ push(rsi);
 945     __ push(rdi);
 946     __ movptr(from , Address(rsp, 12+ 4));
 947     __ movptr(to   , Address(rsp, 12+ 8));
 948     __ movl(count, Address(rsp, 12+ 12));
 949 
 950     if (entry != NULL) {
 951       *entry = __ pc(); // Entry point from conjoint arraycopy stub.
 952       BLOCK_COMMENT("Entry:");
 953     }
 954 
 955     if (t == T_OBJECT) {
 956       __ testl(count, count);
 957       __ jcc(Assembler::zero, L_0_count);
 958       if (need_pre_barrier) {
 959         gen_write_ref_array_pre_barrier(to, count);
 960       }
 961       __ mov(saved_to, to);          // save 'to'
 962     }
 963 
 964     __ subptr(to, from); // to --> to_from
 965     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
 966     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
 967     if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 968       // align source address at 4 bytes address boundary
 969       if (t == T_BYTE) {
 970         // One byte misalignment happens only for byte arrays
 971         __ testl(from, 1);
 972         __ jccb(Assembler::zero, L_skip_align1);
 973         __ movb(rax, Address(from, 0));
 974         __ movb(Address(from, to_from, Address::times_1, 0), rax);
 975         __ increment(from);
 976         __ decrement(count);
 977       __ BIND(L_skip_align1);
 978       }
 979       // Two bytes misalignment happens only for byte and short (char) arrays
 980       __ testl(from, 2);
 981       __ jccb(Assembler::zero, L_skip_align2);
 982       __ movw(rax, Address(from, 0));
 983       __ movw(Address(from, to_from, Address::times_1, 0), rax);
 984       __ addptr(from, 2);
 985       __ subl(count, 1<<(shift-1));
 986     __ BIND(L_skip_align2);
 987     }
 988     if (!VM_Version::supports_mmx()) {
 989       __ mov(rax, count);      // save 'count'
 990       __ shrl(count, shift); // bytes count
 991       __ addptr(to_from, from);// restore 'to'
 992       __ rep_mov();
 993       __ subptr(to_from, from);// restore 'to_from'
 994       __ mov(count, rax);      // restore 'count'
 995       __ jmpb(L_copy_2_bytes); // all dwords were copied
 996     } else {
 997       if (!UseUnalignedLoadStores) {
 998         // align to 8 bytes, we know we are 4 byte aligned to start
 999         __ testptr(from, 4);
1000         __ jccb(Assembler::zero, L_copy_64_bytes);
1001         __ movl(rax, Address(from, 0));
1002         __ movl(Address(from, to_from, Address::times_1, 0), rax);
1003         __ addptr(from, 4);
1004         __ subl(count, 1<<shift);
1005       }
1006     __ BIND(L_copy_64_bytes);
1007       __ mov(rax, count);
1008       __ shrl(rax, shift+1);  // 8 bytes chunk count
1009       //
1010       // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
1011       //
1012       if (UseXMMForArrayCopy) {
1013         xmm_copy_forward(from, to_from, rax);
1014       } else {
1015         mmx_copy_forward(from, to_from, rax);
1016       }
1017     }
1018     // copy tailing dword
1019   __ BIND(L_copy_4_bytes);
1020     __ testl(count, 1<<shift);
1021     __ jccb(Assembler::zero, L_copy_2_bytes);
1022     __ movl(rax, Address(from, 0));
1023     __ movl(Address(from, to_from, Address::times_1, 0), rax);
1024     if (t == T_BYTE || t == T_SHORT) {
1025       __ addptr(from, 4);
1026     __ BIND(L_copy_2_bytes);
1027       // copy tailing word
1028       __ testl(count, 1<<(shift-1));
1029       __ jccb(Assembler::zero, L_copy_byte);
1030       __ movw(rax, Address(from, 0));
1031       __ movw(Address(from, to_from, Address::times_1, 0), rax);
1032       if (t == T_BYTE) {
1033         __ addptr(from, 2);
1034       __ BIND(L_copy_byte);
1035         // copy tailing byte
1036         __ testl(count, 1);
1037         __ jccb(Assembler::zero, L_exit);
1038         __ movb(rax, Address(from, 0));
1039         __ movb(Address(from, to_from, Address::times_1, 0), rax);
1040       __ BIND(L_exit);
1041       } else {
1042       __ BIND(L_copy_byte);
1043       }
1044     } else {
1045     __ BIND(L_copy_2_bytes);
1046     }
1047 
1048     if (t == T_OBJECT) {
1049       __ movl(count, Address(rsp, 12+12)); // reread 'count'
1050       __ mov(to, saved_to); // restore 'to'
1051       gen_write_ref_array_post_barrier(to, count);
1052     __ BIND(L_0_count);
1053     }
1054     inc_copy_counter_np(t);
1055     __ pop(rdi);
1056     __ pop(rsi);
1057     __ leave(); // required for proper stackwalking of RuntimeStub frame
1058     __ xorptr(rax, rax); // return 0
1059     __ ret(0);
1060     return start;
1061   }
1062 
1063 
1064   address generate_fill(BasicType t, bool aligned, const char *name) {
1065     __ align(CodeEntryAlignment);
1066     StubCodeMark mark(this, "StubRoutines", name);
1067     address start = __ pc();
1068 
1069     BLOCK_COMMENT("Entry:");
1070 
1071     const Register to       = rdi;  // source array address
1072     const Register value    = rdx;  // value
1073     const Register count    = rsi;  // elements count
1074 
1075     __ enter(); // required for proper stackwalking of RuntimeStub frame
1076     __ push(rsi);
1077     __ push(rdi);
1078     __ movptr(to   , Address(rsp, 12+ 4));
1079     __ movl(value, Address(rsp, 12+ 8));
1080     __ movl(count, Address(rsp, 12+ 12));
1081 
1082     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1083 
1084     __ pop(rdi);
1085     __ pop(rsi);
1086     __ leave(); // required for proper stackwalking of RuntimeStub frame
1087     __ ret(0);
1088     return start;
1089   }
1090 
1091   address generate_conjoint_copy(BasicType t, bool aligned,
1092                                  Address::ScaleFactor sf,
1093                                  address nooverlap_target,
1094                                  address* entry, const char *name,
1095                                  bool need_pre_barrier = true) {
1096     __ align(CodeEntryAlignment);
1097     StubCodeMark mark(this, "StubRoutines", name);
1098     address start = __ pc();
1099 
1100     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1101     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1102 
1103     int shift = Address::times_ptr - sf;
1104 
1105     const Register src   = rax;  // source array address
1106     const Register dst   = rdx;  // destination array address
1107     const Register from  = rsi;  // source array address
1108     const Register to    = rdi;  // destination array address
1109     const Register count = rcx;  // elements count
1110     const Register end   = rax;  // array end address
1111 
1112     __ enter(); // required for proper stackwalking of RuntimeStub frame
1113     __ push(rsi);
1114     __ push(rdi);
1115     __ movptr(src  , Address(rsp, 12+ 4));   // from
1116     __ movptr(dst  , Address(rsp, 12+ 8));   // to
1117     __ movl2ptr(count, Address(rsp, 12+12)); // count
1118 
1119     if (entry != NULL) {
1120       *entry = __ pc(); // Entry point from generic arraycopy stub.
1121       BLOCK_COMMENT("Entry:");
1122     }
1123 
1124     // nooverlap_target expects arguments in rsi and rdi.
1125     __ mov(from, src);
1126     __ mov(to  , dst);
1127 
1128     // arrays overlap test: dispatch to disjoint stub if necessary.
1129     RuntimeAddress nooverlap(nooverlap_target);
1130     __ cmpptr(dst, src);
1131     __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1132     __ jump_cc(Assembler::belowEqual, nooverlap);
1133     __ cmpptr(dst, end);
1134     __ jump_cc(Assembler::aboveEqual, nooverlap);
1135 
1136     if (t == T_OBJECT) {
1137       __ testl(count, count);
1138       __ jcc(Assembler::zero, L_0_count);
1139       if (need_pre_barrier) {
1140         gen_write_ref_array_pre_barrier(dst, count);
1141       }
1142     }
1143 
1144     // copy from high to low
1145     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1146     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1147     if (t == T_BYTE || t == T_SHORT) {
1148       // Align the end of destination array at 4 bytes address boundary
1149       __ lea(end, Address(dst, count, sf, 0));
1150       if (t == T_BYTE) {
1151         // One byte misalignment happens only for byte arrays
1152         __ testl(end, 1);
1153         __ jccb(Assembler::zero, L_skip_align1);
1154         __ decrement(count);
1155         __ movb(rdx, Address(from, count, sf, 0));
1156         __ movb(Address(to, count, sf, 0), rdx);
1157       __ BIND(L_skip_align1);
1158       }
1159       // Two bytes misalignment happens only for byte and short (char) arrays
1160       __ testl(end, 2);
1161       __ jccb(Assembler::zero, L_skip_align2);
1162       __ subptr(count, 1<<(shift-1));
1163       __ movw(rdx, Address(from, count, sf, 0));
1164       __ movw(Address(to, count, sf, 0), rdx);
1165     __ BIND(L_skip_align2);
1166       __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1167       __ jcc(Assembler::below, L_copy_4_bytes);
1168     }
1169 
1170     if (!VM_Version::supports_mmx()) {
1171       __ std();
1172       __ mov(rax, count); // Save 'count'
1173       __ mov(rdx, to);    // Save 'to'
1174       __ lea(rsi, Address(from, count, sf, -4));
1175       __ lea(rdi, Address(to  , count, sf, -4));
1176       __ shrptr(count, shift); // bytes count
1177       __ rep_mov();
1178       __ cld();
1179       __ mov(count, rax); // restore 'count'
1180       __ andl(count, (1<<shift)-1);      // mask the number of rest elements
1181       __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1182       __ mov(to, rdx);   // restore 'to'
1183       __ jmpb(L_copy_2_bytes); // all dword were copied
1184    } else {
1185       // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1186       __ testptr(end, 4);
1187       __ jccb(Assembler::zero, L_copy_8_bytes);
1188       __ subl(count, 1<<shift);
1189       __ movl(rdx, Address(from, count, sf, 0));
1190       __ movl(Address(to, count, sf, 0), rdx);
1191       __ jmpb(L_copy_8_bytes);
1192 
1193       __ align(OptoLoopAlignment);
1194       // Move 8 bytes
1195     __ BIND(L_copy_8_bytes_loop);
1196       if (UseXMMForArrayCopy) {
1197         __ movq(xmm0, Address(from, count, sf, 0));
1198         __ movq(Address(to, count, sf, 0), xmm0);
1199       } else {
1200         __ movq(mmx0, Address(from, count, sf, 0));
1201         __ movq(Address(to, count, sf, 0), mmx0);
1202       }
1203     __ BIND(L_copy_8_bytes);
1204       __ subl(count, 2<<shift);
1205       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1206       __ addl(count, 2<<shift);
1207       if (!UseXMMForArrayCopy) {
1208         __ emms();
1209       }
1210     }
1211   __ BIND(L_copy_4_bytes);
1212     // copy prefix qword
1213     __ testl(count, 1<<shift);
1214     __ jccb(Assembler::zero, L_copy_2_bytes);
1215     __ movl(rdx, Address(from, count, sf, -4));
1216     __ movl(Address(to, count, sf, -4), rdx);
1217 
1218     if (t == T_BYTE || t == T_SHORT) {
1219         __ subl(count, (1<<shift));
1220       __ BIND(L_copy_2_bytes);
1221         // copy prefix dword
1222         __ testl(count, 1<<(shift-1));
1223         __ jccb(Assembler::zero, L_copy_byte);
1224         __ movw(rdx, Address(from, count, sf, -2));
1225         __ movw(Address(to, count, sf, -2), rdx);
1226         if (t == T_BYTE) {
1227           __ subl(count, 1<<(shift-1));
1228         __ BIND(L_copy_byte);
1229           // copy prefix byte
1230           __ testl(count, 1);
1231           __ jccb(Assembler::zero, L_exit);
1232           __ movb(rdx, Address(from, 0));
1233           __ movb(Address(to, 0), rdx);
1234         __ BIND(L_exit);
1235         } else {
1236         __ BIND(L_copy_byte);
1237         }
1238     } else {
1239     __ BIND(L_copy_2_bytes);
1240     }
1241     if (t == T_OBJECT) {
1242       __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1243       gen_write_ref_array_post_barrier(to, count);
1244     __ BIND(L_0_count);
1245     }
1246     inc_copy_counter_np(t);
1247     __ pop(rdi);
1248     __ pop(rsi);
1249     __ leave(); // required for proper stackwalking of RuntimeStub frame
1250     __ xorptr(rax, rax); // return 0
1251     __ ret(0);
1252     return start;
1253   }
1254 
1255 
1256   address generate_disjoint_long_copy(address* entry, const char *name) {
1257     __ align(CodeEntryAlignment);
1258     StubCodeMark mark(this, "StubRoutines", name);
1259     address start = __ pc();
1260 
1261     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1262     const Register from       = rax;  // source array address
1263     const Register to         = rdx;  // destination array address
1264     const Register count      = rcx;  // elements count
1265     const Register to_from    = rdx;  // (to - from)
1266 
1267     __ enter(); // required for proper stackwalking of RuntimeStub frame
1268     __ movptr(from , Address(rsp, 8+0));       // from
1269     __ movptr(to   , Address(rsp, 8+4));       // to
1270     __ movl2ptr(count, Address(rsp, 8+8));     // count
1271 
1272     *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1273     BLOCK_COMMENT("Entry:");
1274 
1275     __ subptr(to, from); // to --> to_from
1276     if (VM_Version::supports_mmx()) {
1277       if (UseXMMForArrayCopy) {
1278         xmm_copy_forward(from, to_from, count);
1279       } else {
1280         mmx_copy_forward(from, to_from, count);
1281       }
1282     } else {
1283       __ jmpb(L_copy_8_bytes);
1284       __ align(OptoLoopAlignment);
1285     __ BIND(L_copy_8_bytes_loop);
1286       __ fild_d(Address(from, 0));
1287       __ fistp_d(Address(from, to_from, Address::times_1));
1288       __ addptr(from, 8);
1289     __ BIND(L_copy_8_bytes);
1290       __ decrement(count);
1291       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1292     }
1293     inc_copy_counter_np(T_LONG);
1294     __ leave(); // required for proper stackwalking of RuntimeStub frame
1295     __ xorptr(rax, rax); // return 0
1296     __ ret(0);
1297     return start;
1298   }
1299 
1300   address generate_conjoint_long_copy(address nooverlap_target,
1301                                       address* entry, const char *name) {
1302     __ align(CodeEntryAlignment);
1303     StubCodeMark mark(this, "StubRoutines", name);
1304     address start = __ pc();
1305 
1306     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1307     const Register from       = rax;  // source array address
1308     const Register to         = rdx;  // destination array address
1309     const Register count      = rcx;  // elements count
1310     const Register end_from   = rax;  // source array end address
1311 
1312     __ enter(); // required for proper stackwalking of RuntimeStub frame
1313     __ movptr(from , Address(rsp, 8+0));       // from
1314     __ movptr(to   , Address(rsp, 8+4));       // to
1315     __ movl2ptr(count, Address(rsp, 8+8));     // count
1316 
1317     *entry = __ pc(); // Entry point from generic arraycopy stub.
1318     BLOCK_COMMENT("Entry:");
1319 
1320     // arrays overlap test
1321     __ cmpptr(to, from);
1322     RuntimeAddress nooverlap(nooverlap_target);
1323     __ jump_cc(Assembler::belowEqual, nooverlap);
1324     __ lea(end_from, Address(from, count, Address::times_8, 0));
1325     __ cmpptr(to, end_from);
1326     __ movptr(from, Address(rsp, 8));  // from
1327     __ jump_cc(Assembler::aboveEqual, nooverlap);
1328 
1329     __ jmpb(L_copy_8_bytes);
1330 
1331     __ align(OptoLoopAlignment);
1332   __ BIND(L_copy_8_bytes_loop);
1333     if (VM_Version::supports_mmx()) {
1334       if (UseXMMForArrayCopy) {
1335         __ movq(xmm0, Address(from, count, Address::times_8));
1336         __ movq(Address(to, count, Address::times_8), xmm0);
1337       } else {
1338         __ movq(mmx0, Address(from, count, Address::times_8));
1339         __ movq(Address(to, count, Address::times_8), mmx0);
1340       }
1341     } else {
1342       __ fild_d(Address(from, count, Address::times_8));
1343       __ fistp_d(Address(to, count, Address::times_8));
1344     }
1345   __ BIND(L_copy_8_bytes);
1346     __ decrement(count);
1347     __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1348 
1349     if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1350       __ emms();
1351     }
1352     inc_copy_counter_np(T_LONG);
1353     __ leave(); // required for proper stackwalking of RuntimeStub frame
1354     __ xorptr(rax, rax); // return 0
1355     __ ret(0);
1356     return start;
1357   }
1358 
1359 
1360   // Helper for generating a dynamic type check.
1361   // The sub_klass must be one of {rbx, rdx, rsi}.
1362   // The temp is killed.
1363   void generate_type_check(Register sub_klass,
1364                            Address& super_check_offset_addr,
1365                            Address& super_klass_addr,
1366                            Register temp,
1367                            Label* L_success, Label* L_failure) {
1368     BLOCK_COMMENT("type_check:");
1369 
1370     Label L_fallthrough;
1371 #define LOCAL_JCC(assembler_con, label_ptr)                             \
1372     if (label_ptr != NULL)  __ jcc(assembler_con, *(label_ptr));        \
1373     else                    __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1374 
1375     // The following is a strange variation of the fast path which requires
1376     // one less register, because needed values are on the argument stack.
1377     // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1378     //                                  L_success, L_failure, NULL);
1379     assert_different_registers(sub_klass, temp);
1380 
1381     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1382                      Klass::secondary_super_cache_offset_in_bytes());
1383 
1384     // if the pointers are equal, we are done (e.g., String[] elements)
1385     __ cmpptr(sub_klass, super_klass_addr);
1386     LOCAL_JCC(Assembler::equal, L_success);
1387 
1388     // check the supertype display:
1389     __ movl2ptr(temp, super_check_offset_addr);
1390     Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1391     __ movptr(temp, super_check_addr); // load displayed supertype
1392     __ cmpptr(temp, super_klass_addr); // test the super type
1393     LOCAL_JCC(Assembler::equal, L_success);
1394 
1395     // if it was a primary super, we can just fail immediately
1396     __ cmpl(super_check_offset_addr, sc_offset);
1397     LOCAL_JCC(Assembler::notEqual, L_failure);
1398 
1399     // The repne_scan instruction uses fixed registers, which will get spilled.
1400     // We happen to know this works best when super_klass is in rax.
1401     Register super_klass = temp;
1402     __ movptr(super_klass, super_klass_addr);
1403     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1404                                      L_success, L_failure);
1405 
1406     __ bind(L_fallthrough);
1407 
1408     if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1409     if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1410 
1411 #undef LOCAL_JCC
1412   }
1413 
1414   //
1415   //  Generate checkcasting array copy stub
1416   //
1417   //  Input:
1418   //    4(rsp)   - source array address
1419   //    8(rsp)   - destination array address
1420   //   12(rsp)   - element count, can be zero
1421   //   16(rsp)   - size_t ckoff (super_check_offset)
1422   //   20(rsp)   - oop ckval (super_klass)
1423   //
1424   //  Output:
1425   //    rax, ==  0  -  success
1426   //    rax, == -1^K - failure, where K is partial transfer count
1427   //
1428   address generate_checkcast_copy(const char *name, address* entry, bool need_pre_barrier = true) {
1429     __ align(CodeEntryAlignment);
1430     StubCodeMark mark(this, "StubRoutines", name);
1431     address start = __ pc();
1432 
1433     Label L_load_element, L_store_element, L_do_card_marks, L_done;
1434 
1435     // register use:
1436     //  rax, rdx, rcx -- loop control (end_from, end_to, count)
1437     //  rdi, rsi      -- element access (oop, klass)
1438     //  rbx,           -- temp
1439     const Register from       = rax;    // source array address
1440     const Register to         = rdx;    // destination array address
1441     const Register length     = rcx;    // elements count
1442     const Register elem       = rdi;    // each oop copied
1443     const Register elem_klass = rsi;    // each elem._klass (sub_klass)
1444     const Register temp       = rbx;    // lone remaining temp
1445 
1446     __ enter(); // required for proper stackwalking of RuntimeStub frame
1447 
1448     __ push(rsi);
1449     __ push(rdi);
1450     __ push(rbx);
1451 
1452     Address   from_arg(rsp, 16+ 4);     // from
1453     Address     to_arg(rsp, 16+ 8);     // to
1454     Address length_arg(rsp, 16+12);     // elements count
1455     Address  ckoff_arg(rsp, 16+16);     // super_check_offset
1456     Address  ckval_arg(rsp, 16+20);     // super_klass
1457 
1458     // Load up:
1459     __ movptr(from,     from_arg);
1460     __ movptr(to,         to_arg);
1461     __ movl2ptr(length, length_arg);
1462 
1463     if (entry != NULL) {
1464       *entry = __ pc(); // Entry point from generic arraycopy stub.
1465       BLOCK_COMMENT("Entry:");
1466     }
1467 
1468     //---------------------------------------------------------------
1469     // Assembler stub will be used for this call to arraycopy
1470     // if the two arrays are subtypes of Object[] but the
1471     // destination array type is not equal to or a supertype
1472     // of the source type.  Each element must be separately
1473     // checked.
1474 
1475     // Loop-invariant addresses.  They are exclusive end pointers.
1476     Address end_from_addr(from, length, Address::times_ptr, 0);
1477     Address   end_to_addr(to,   length, Address::times_ptr, 0);
1478 
1479     Register end_from = from;           // re-use
1480     Register end_to   = to;             // re-use
1481     Register count    = length;         // re-use
1482 
1483     // Loop-variant addresses.  They assume post-incremented count < 0.
1484     Address from_element_addr(end_from, count, Address::times_ptr, 0);
1485     Address   to_element_addr(end_to,   count, Address::times_ptr, 0);
1486     Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1487 
1488     // Copy from low to high addresses, indexed from the end of each array.
1489     if (need_pre_barrier) {
1490       gen_write_ref_array_pre_barrier(to, count);
1491     }
1492     __ lea(end_from, end_from_addr);
1493     __ lea(end_to,   end_to_addr);
1494     assert(length == count, "");        // else fix next line:
1495     __ negptr(count);                   // negate and test the length
1496     __ jccb(Assembler::notZero, L_load_element);
1497 
1498     // Empty array:  Nothing to do.
1499     __ xorptr(rax, rax);                  // return 0 on (trivial) success
1500     __ jmp(L_done);
1501 
1502     // ======== begin loop ========
1503     // (Loop is rotated; its entry is L_load_element.)
1504     // Loop control:
1505     //   for (count = -count; count != 0; count++)
1506     // Base pointers src, dst are biased by 8*count,to last element.
1507     __ align(OptoLoopAlignment);
1508 
1509     __ BIND(L_store_element);
1510     __ movptr(to_element_addr, elem);     // store the oop
1511     __ increment(count);                // increment the count toward zero
1512     __ jccb(Assembler::zero, L_do_card_marks);
1513 
1514     // ======== loop entry is here ========
1515     __ BIND(L_load_element);
1516     __ movptr(elem, from_element_addr);   // load the oop
1517     __ testptr(elem, elem);
1518     __ jccb(Assembler::zero, L_store_element);
1519 
1520     // (Could do a trick here:  Remember last successful non-null
1521     // element stored and make a quick oop equality check on it.)
1522 
1523     __ movptr(elem_klass, elem_klass_addr); // query the object klass
1524     generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1525                         &L_store_element, NULL);
1526       // (On fall-through, we have failed the element type check.)
1527     // ======== end loop ========
1528 
1529     // It was a real error; we must depend on the caller to finish the job.
1530     // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1531     // Emit GC store barriers for the oops we have copied (length_arg + count),
1532     // and report their number to the caller.
1533     __ addl(count, length_arg);         // transfers = (length - remaining)
1534     __ movl2ptr(rax, count);            // save the value
1535     __ notptr(rax);                     // report (-1^K) to caller
1536     __ movptr(to, to_arg);              // reload
1537     assert_different_registers(to, count, rax);
1538     gen_write_ref_array_post_barrier(to, count);
1539     __ jmpb(L_done);
1540 
1541     // Come here on success only.
1542     __ BIND(L_do_card_marks);
1543     __ movl2ptr(count, length_arg);
1544     __ movptr(to, to_arg);                // reload
1545     gen_write_ref_array_post_barrier(to, count);
1546     __ xorptr(rax, rax);                  // return 0 on success
1547 
1548     // Common exit point (success or failure).
1549     __ BIND(L_done);
1550     __ pop(rbx);
1551     __ pop(rdi);
1552     __ pop(rsi);
1553     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1554     __ leave(); // required for proper stackwalking of RuntimeStub frame
1555     __ ret(0);
1556 
1557     return start;
1558   }
1559 
1560   //
1561   //  Generate 'unsafe' array copy stub
1562   //  Though just as safe as the other stubs, it takes an unscaled
1563   //  size_t argument instead of an element count.
1564   //
1565   //  Input:
1566   //    4(rsp)   - source array address
1567   //    8(rsp)   - destination array address
1568   //   12(rsp)   - byte count, can be zero
1569   //
1570   //  Output:
1571   //    rax, ==  0  -  success
1572   //    rax, == -1  -  need to call System.arraycopy
1573   //
1574   // Examines the alignment of the operands and dispatches
1575   // to a long, int, short, or byte copy loop.
1576   //
1577   address generate_unsafe_copy(const char *name,
1578                                address byte_copy_entry,
1579                                address short_copy_entry,
1580                                address int_copy_entry,
1581                                address long_copy_entry) {
1582 
1583     Label L_long_aligned, L_int_aligned, L_short_aligned;
1584 
1585     __ align(CodeEntryAlignment);
1586     StubCodeMark mark(this, "StubRoutines", name);
1587     address start = __ pc();
1588 
1589     const Register from       = rax;  // source array address
1590     const Register to         = rdx;  // destination array address
1591     const Register count      = rcx;  // elements count
1592 
1593     __ enter(); // required for proper stackwalking of RuntimeStub frame
1594     __ push(rsi);
1595     __ push(rdi);
1596     Address  from_arg(rsp, 12+ 4);      // from
1597     Address    to_arg(rsp, 12+ 8);      // to
1598     Address count_arg(rsp, 12+12);      // byte count
1599 
1600     // Load up:
1601     __ movptr(from ,  from_arg);
1602     __ movptr(to   ,    to_arg);
1603     __ movl2ptr(count, count_arg);
1604 
1605     // bump this on entry, not on exit:
1606     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1607 
1608     const Register bits = rsi;
1609     __ mov(bits, from);
1610     __ orptr(bits, to);
1611     __ orptr(bits, count);
1612 
1613     __ testl(bits, BytesPerLong-1);
1614     __ jccb(Assembler::zero, L_long_aligned);
1615 
1616     __ testl(bits, BytesPerInt-1);
1617     __ jccb(Assembler::zero, L_int_aligned);
1618 
1619     __ testl(bits, BytesPerShort-1);
1620     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1621 
1622     __ BIND(L_short_aligned);
1623     __ shrptr(count, LogBytesPerShort); // size => short_count
1624     __ movl(count_arg, count);          // update 'count'
1625     __ jump(RuntimeAddress(short_copy_entry));
1626 
1627     __ BIND(L_int_aligned);
1628     __ shrptr(count, LogBytesPerInt); // size => int_count
1629     __ movl(count_arg, count);          // update 'count'
1630     __ jump(RuntimeAddress(int_copy_entry));
1631 
1632     __ BIND(L_long_aligned);
1633     __ shrptr(count, LogBytesPerLong); // size => qword_count
1634     __ movl(count_arg, count);          // update 'count'
1635     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1636     __ pop(rsi);
1637     __ jump(RuntimeAddress(long_copy_entry));
1638 
1639     return start;
1640   }
1641 
1642 
1643   // Perform range checks on the proposed arraycopy.
1644   // Smashes src_pos and dst_pos.  (Uses them up for temps.)
1645   void arraycopy_range_checks(Register src,
1646                               Register src_pos,
1647                               Register dst,
1648                               Register dst_pos,
1649                               Address& length,
1650                               Label& L_failed) {
1651     BLOCK_COMMENT("arraycopy_range_checks:");
1652     const Register src_end = src_pos;   // source array end position
1653     const Register dst_end = dst_pos;   // destination array end position
1654     __ addl(src_end, length); // src_pos + length
1655     __ addl(dst_end, length); // dst_pos + length
1656 
1657     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;
1658     __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1659     __ jcc(Assembler::above, L_failed);
1660 
1661     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1662     __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1663     __ jcc(Assembler::above, L_failed);
1664 
1665     BLOCK_COMMENT("arraycopy_range_checks done");
1666   }
1667 
1668 
1669   //
1670   //  Generate generic array copy stubs
1671   //
1672   //  Input:
1673   //     4(rsp)    -  src oop
1674   //     8(rsp)    -  src_pos
1675   //    12(rsp)    -  dst oop
1676   //    16(rsp)    -  dst_pos
1677   //    20(rsp)    -  element count
1678   //
1679   //  Output:
1680   //    rax, ==  0  -  success
1681   //    rax, == -1^K - failure, where K is partial transfer count
1682   //
1683   address generate_generic_copy(const char *name,
1684                                 address entry_jbyte_arraycopy,
1685                                 address entry_jshort_arraycopy,
1686                                 address entry_jint_arraycopy,
1687                                 address entry_oop_arraycopy,
1688                                 address entry_jlong_arraycopy,
1689                                 address entry_checkcast_arraycopy) {
1690     Label L_failed, L_failed_0, L_objArray;
1691 
1692     { int modulus = CodeEntryAlignment;
1693       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
1694       int advance = target - (__ offset() % modulus);
1695       if (advance < 0)  advance += modulus;
1696       if (advance > 0)  __ nop(advance);
1697     }
1698     StubCodeMark mark(this, "StubRoutines", name);
1699 
1700     // Short-hop target to L_failed.  Makes for denser prologue code.
1701     __ BIND(L_failed_0);
1702     __ jmp(L_failed);
1703     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1704 
1705     __ align(CodeEntryAlignment);
1706     address start = __ pc();
1707 
1708     __ enter(); // required for proper stackwalking of RuntimeStub frame
1709     __ push(rsi);
1710     __ push(rdi);
1711 
1712     // bump this on entry, not on exit:
1713     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1714 
1715     // Input values
1716     Address SRC     (rsp, 12+ 4);
1717     Address SRC_POS (rsp, 12+ 8);
1718     Address DST     (rsp, 12+12);
1719     Address DST_POS (rsp, 12+16);
1720     Address LENGTH  (rsp, 12+20);
1721 
1722     //-----------------------------------------------------------------------
1723     // Assembler stub will be used for this call to arraycopy
1724     // if the following conditions are met:
1725     //
1726     // (1) src and dst must not be null.
1727     // (2) src_pos must not be negative.
1728     // (3) dst_pos must not be negative.
1729     // (4) length  must not be negative.
1730     // (5) src klass and dst klass should be the same and not NULL.
1731     // (6) src and dst should be arrays.
1732     // (7) src_pos + length must not exceed length of src.
1733     // (8) dst_pos + length must not exceed length of dst.
1734     //
1735 
1736     const Register src     = rax;       // source array oop
1737     const Register src_pos = rsi;
1738     const Register dst     = rdx;       // destination array oop
1739     const Register dst_pos = rdi;
1740     const Register length  = rcx;       // transfer count
1741 
1742     //  if (src == NULL) return -1;
1743     __ movptr(src, SRC);      // src oop
1744     __ testptr(src, src);
1745     __ jccb(Assembler::zero, L_failed_0);
1746 
1747     //  if (src_pos < 0) return -1;
1748     __ movl2ptr(src_pos, SRC_POS);  // src_pos
1749     __ testl(src_pos, src_pos);
1750     __ jccb(Assembler::negative, L_failed_0);
1751 
1752     //  if (dst == NULL) return -1;
1753     __ movptr(dst, DST);      // dst oop
1754     __ testptr(dst, dst);
1755     __ jccb(Assembler::zero, L_failed_0);
1756 
1757     //  if (dst_pos < 0) return -1;
1758     __ movl2ptr(dst_pos, DST_POS);  // dst_pos
1759     __ testl(dst_pos, dst_pos);
1760     __ jccb(Assembler::negative, L_failed_0);
1761 
1762     //  if (length < 0) return -1;
1763     __ movl2ptr(length, LENGTH);   // length
1764     __ testl(length, length);
1765     __ jccb(Assembler::negative, L_failed_0);
1766 
1767     //  if (src->klass() == NULL) return -1;
1768     Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1769     Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1770     const Register rcx_src_klass = rcx;    // array klass
1771     __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1772 
1773 #ifdef ASSERT
1774     //  assert(src->klass() != NULL);
1775     BLOCK_COMMENT("assert klasses not null");
1776     { Label L1, L2;
1777       __ testptr(rcx_src_klass, rcx_src_klass);
1778       __ jccb(Assembler::notZero, L2);   // it is broken if klass is NULL
1779       __ bind(L1);
1780       __ stop("broken null klass");
1781       __ bind(L2);
1782       __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1783       __ jccb(Assembler::equal, L1);      // this would be broken also
1784       BLOCK_COMMENT("assert done");
1785     }
1786 #endif //ASSERT
1787 
1788     // Load layout helper (32-bits)
1789     //
1790     //  |array_tag|     | header_size | element_type |     |log2_element_size|
1791     // 32        30    24            16              8     2                 0
1792     //
1793     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1794     //
1795 
1796     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1797                     Klass::layout_helper_offset_in_bytes();
1798     Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1799 
1800     // Handle objArrays completely differently...
1801     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1802     __ cmpl(src_klass_lh_addr, objArray_lh);
1803     __ jcc(Assembler::equal, L_objArray);
1804 
1805     //  if (src->klass() != dst->klass()) return -1;
1806     __ cmpptr(rcx_src_klass, dst_klass_addr);
1807     __ jccb(Assembler::notEqual, L_failed_0);
1808 
1809     const Register rcx_lh = rcx;  // layout helper
1810     assert(rcx_lh == rcx_src_klass, "known alias");
1811     __ movl(rcx_lh, src_klass_lh_addr);
1812 
1813     //  if (!src->is_Array()) return -1;
1814     __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1815     __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1816 
1817     // At this point, it is known to be a typeArray (array_tag 0x3).
1818 #ifdef ASSERT
1819     { Label L;
1820       __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1821       __ jcc(Assembler::greaterEqual, L); // signed cmp
1822       __ stop("must be a primitive array");
1823       __ bind(L);
1824     }
1825 #endif
1826 
1827     assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1828     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1829 
1830     // typeArrayKlass
1831     //
1832     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1833     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1834     //
1835     const Register rsi_offset = rsi; // array offset
1836     const Register src_array  = src; // src array offset
1837     const Register dst_array  = dst; // dst array offset
1838     const Register rdi_elsize = rdi; // log2 element size
1839 
1840     __ mov(rsi_offset, rcx_lh);
1841     __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1842     __ andptr(rsi_offset, Klass::_lh_header_size_mask);   // array_offset
1843     __ addptr(src_array, rsi_offset);  // src array offset
1844     __ addptr(dst_array, rsi_offset);  // dst array offset
1845     __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1846 
1847     // next registers should be set before the jump to corresponding stub
1848     const Register from       = src; // source array address
1849     const Register to         = dst; // destination array address
1850     const Register count      = rcx; // elements count
1851     // some of them should be duplicated on stack
1852 #define FROM   Address(rsp, 12+ 4)
1853 #define TO     Address(rsp, 12+ 8)   // Not used now
1854 #define COUNT  Address(rsp, 12+12)   // Only for oop arraycopy
1855 
1856     BLOCK_COMMENT("scale indexes to element size");
1857     __ movl2ptr(rsi, SRC_POS);  // src_pos
1858     __ shlptr(rsi);             // src_pos << rcx (log2 elsize)
1859     assert(src_array == from, "");
1860     __ addptr(from, rsi);       // from = src_array + SRC_POS << log2 elsize
1861     __ movl2ptr(rdi, DST_POS);  // dst_pos
1862     __ shlptr(rdi);             // dst_pos << rcx (log2 elsize)
1863     assert(dst_array == to, "");
1864     __ addptr(to,  rdi);        // to   = dst_array + DST_POS << log2 elsize
1865     __ movptr(FROM, from);      // src_addr
1866     __ mov(rdi_elsize, rcx_lh); // log2 elsize
1867     __ movl2ptr(count, LENGTH); // elements count
1868 
1869     BLOCK_COMMENT("choose copy loop based on element size");
1870     __ cmpl(rdi_elsize, 0);
1871 
1872     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1873     __ cmpl(rdi_elsize, LogBytesPerShort);
1874     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1875     __ cmpl(rdi_elsize, LogBytesPerInt);
1876     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1877 #ifdef ASSERT
1878     __ cmpl(rdi_elsize, LogBytesPerLong);
1879     __ jccb(Assembler::notEqual, L_failed);
1880 #endif
1881     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1882     __ pop(rsi);
1883     __ jump(RuntimeAddress(entry_jlong_arraycopy));
1884 
1885   __ BIND(L_failed);
1886     __ xorptr(rax, rax);
1887     __ notptr(rax); // return -1
1888     __ pop(rdi);
1889     __ pop(rsi);
1890     __ leave(); // required for proper stackwalking of RuntimeStub frame
1891     __ ret(0);
1892 
1893     // objArrayKlass
1894   __ BIND(L_objArray);
1895     // live at this point:  rcx_src_klass, src[_pos], dst[_pos]
1896 
1897     Label L_plain_copy, L_checkcast_copy;
1898     //  test array classes for subtyping
1899     __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1900     __ jccb(Assembler::notEqual, L_checkcast_copy);
1901 
1902     // Identically typed arrays can be copied without element-wise checks.
1903     assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1904     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1905 
1906   __ BIND(L_plain_copy);
1907     __ movl2ptr(count, LENGTH); // elements count
1908     __ movl2ptr(src_pos, SRC_POS);  // reload src_pos
1909     __ lea(from, Address(src, src_pos, Address::times_ptr,
1910                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1911     __ movl2ptr(dst_pos, DST_POS);  // reload dst_pos
1912     __ lea(to,   Address(dst, dst_pos, Address::times_ptr,
1913                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1914     __ movptr(FROM,  from);   // src_addr
1915     __ movptr(TO,    to);     // dst_addr
1916     __ movl(COUNT, count);  // count
1917     __ jump(RuntimeAddress(entry_oop_arraycopy));
1918 
1919   __ BIND(L_checkcast_copy);
1920     // live at this point:  rcx_src_klass, dst[_pos], src[_pos]
1921     {
1922       // Handy offsets:
1923       int  ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1924                         objArrayKlass::element_klass_offset_in_bytes());
1925       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1926                         Klass::super_check_offset_offset_in_bytes());
1927 
1928       Register rsi_dst_klass = rsi;
1929       Register rdi_temp      = rdi;
1930       assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1931       assert(rdi_temp      == dst_pos, "expected alias w/ dst_pos");
1932       Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1933 
1934       // Before looking at dst.length, make sure dst is also an objArray.
1935       __ movptr(rsi_dst_klass, dst_klass_addr);
1936       __ cmpl(dst_klass_lh_addr, objArray_lh);
1937       __ jccb(Assembler::notEqual, L_failed);
1938 
1939       // It is safe to examine both src.length and dst.length.
1940       __ movl2ptr(src_pos, SRC_POS);        // reload rsi
1941       arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1942       // (Now src_pos and dst_pos are killed, but not src and dst.)
1943 
1944       // We'll need this temp (don't forget to pop it after the type check).
1945       __ push(rbx);
1946       Register rbx_src_klass = rbx;
1947 
1948       __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1949       __ movptr(rsi_dst_klass, dst_klass_addr);
1950       Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1951       Label L_fail_array_check;
1952       generate_type_check(rbx_src_klass,
1953                           super_check_offset_addr, dst_klass_addr,
1954                           rdi_temp, NULL, &L_fail_array_check);
1955       // (On fall-through, we have passed the array type check.)
1956       __ pop(rbx);
1957       __ jmp(L_plain_copy);
1958 
1959       __ BIND(L_fail_array_check);
1960       // Reshuffle arguments so we can call checkcast_arraycopy:
1961 
1962       // match initial saves for checkcast_arraycopy
1963       // push(rsi);    // already done; see above
1964       // push(rdi);    // already done; see above
1965       // push(rbx);    // already done; see above
1966 
1967       // Marshal outgoing arguments now, freeing registers.
1968       Address   from_arg(rsp, 16+ 4);   // from
1969       Address     to_arg(rsp, 16+ 8);   // to
1970       Address length_arg(rsp, 16+12);   // elements count
1971       Address  ckoff_arg(rsp, 16+16);   // super_check_offset
1972       Address  ckval_arg(rsp, 16+20);   // super_klass
1973 
1974       Address SRC_POS_arg(rsp, 16+ 8);
1975       Address DST_POS_arg(rsp, 16+16);
1976       Address  LENGTH_arg(rsp, 16+20);
1977       // push rbx, changed the incoming offsets (why not just use rbp,??)
1978       // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1979 
1980       __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1981       __ movl2ptr(length, LENGTH_arg);    // reload elements count
1982       __ movl2ptr(src_pos, SRC_POS_arg);  // reload src_pos
1983       __ movl2ptr(dst_pos, DST_POS_arg);  // reload dst_pos
1984 
1985       __ movptr(ckval_arg, rbx);          // destination element type
1986       __ movl(rbx, Address(rbx, sco_offset));
1987       __ movl(ckoff_arg, rbx);          // corresponding class check offset
1988 
1989       __ movl(length_arg, length);      // outgoing length argument
1990 
1991       __ lea(from, Address(src, src_pos, Address::times_ptr,
1992                             arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1993       __ movptr(from_arg, from);
1994 
1995       __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1996                           arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1997       __ movptr(to_arg, to);
1998       __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1999     }
2000 
2001     return start;
2002   }
2003 
2004   void generate_arraycopy_stubs() {
2005     address entry;
2006     address entry_jbyte_arraycopy;
2007     address entry_jshort_arraycopy;
2008     address entry_jint_arraycopy;
2009     address entry_oop_arraycopy;
2010     address entry_jlong_arraycopy;
2011     address entry_checkcast_arraycopy;
2012 
2013     StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
2014         generate_disjoint_copy(T_BYTE,  true, Address::times_1, &entry,
2015                                "arrayof_jbyte_disjoint_arraycopy");
2016     StubRoutines::_arrayof_jbyte_arraycopy =
2017         generate_conjoint_copy(T_BYTE,  true, Address::times_1,  entry,
2018                                NULL, "arrayof_jbyte_arraycopy");
2019     StubRoutines::_jbyte_disjoint_arraycopy =
2020         generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2021                                "jbyte_disjoint_arraycopy");
2022     StubRoutines::_jbyte_arraycopy =
2023         generate_conjoint_copy(T_BYTE, false, Address::times_1,  entry,
2024                                &entry_jbyte_arraycopy, "jbyte_arraycopy");
2025 
2026     StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2027         generate_disjoint_copy(T_SHORT,  true, Address::times_2, &entry,
2028                                "arrayof_jshort_disjoint_arraycopy");
2029     StubRoutines::_arrayof_jshort_arraycopy =
2030         generate_conjoint_copy(T_SHORT,  true, Address::times_2,  entry,
2031                                NULL, "arrayof_jshort_arraycopy");
2032     StubRoutines::_jshort_disjoint_arraycopy =
2033         generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2034                                "jshort_disjoint_arraycopy");
2035     StubRoutines::_jshort_arraycopy =
2036         generate_conjoint_copy(T_SHORT, false, Address::times_2,  entry,
2037                                &entry_jshort_arraycopy, "jshort_arraycopy");
2038 
2039     // Next arrays are always aligned on 4 bytes at least.
2040     StubRoutines::_jint_disjoint_arraycopy =
2041         generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2042                                "jint_disjoint_arraycopy");
2043     StubRoutines::_jint_arraycopy =
2044         generate_conjoint_copy(T_INT, true, Address::times_4,  entry,
2045                                &entry_jint_arraycopy, "jint_arraycopy");
2046 
2047     StubRoutines::_oop_disjoint_arraycopy =
2048         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2049                                "oop_disjoint_arraycopy");
2050     StubRoutines::_oop_arraycopy =
2051         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2052                                &entry_oop_arraycopy, "oop_arraycopy");
2053 
2054     StubRoutines::_oop_disjoint_arraycopy_no_pre =
2055         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2056                                "oop_disjoint_arraycopy_no_pre", false);
2057     StubRoutines::_oop_arraycopy_no_pre =
2058         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2059                                NULL, "oop_arraycopy_no_pre", false);
2060 
2061     StubRoutines::_jlong_disjoint_arraycopy =
2062         generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2063     StubRoutines::_jlong_arraycopy =
2064         generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2065                                     "jlong_arraycopy");
2066 
2067     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2068     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2069     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2070     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2071     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2072     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2073 
2074     StubRoutines::_arrayof_jint_disjoint_arraycopy       = StubRoutines::_jint_disjoint_arraycopy;
2075     StubRoutines::_arrayof_oop_disjoint_arraycopy        = StubRoutines::_oop_disjoint_arraycopy;
2076     StubRoutines::_arrayof_oop_disjoint_arraycopy_no_pre = StubRoutines::_oop_disjoint_arraycopy_no_pre;
2077     StubRoutines::_arrayof_jlong_disjoint_arraycopy      = StubRoutines::_jlong_disjoint_arraycopy;
2078 
2079     StubRoutines::_arrayof_jint_arraycopy       = StubRoutines::_jint_arraycopy;
2080     StubRoutines::_arrayof_oop_arraycopy        = StubRoutines::_oop_arraycopy;
2081     StubRoutines::_arrayof_oop_arraycopy_no_pre = StubRoutines::_oop_arraycopy_no_pre;
2082     StubRoutines::_arrayof_jlong_arraycopy      = StubRoutines::_jlong_arraycopy;
2083 
2084     StubRoutines::_checkcast_arraycopy =
2085         generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2086     StubRoutines::_checkcast_arraycopy_no_pre =
2087         generate_checkcast_copy("checkcast_arraycopy_no_pre", NULL, false);
2088 
2089     StubRoutines::_unsafe_arraycopy =
2090         generate_unsafe_copy("unsafe_arraycopy",
2091                                entry_jbyte_arraycopy,
2092                                entry_jshort_arraycopy,
2093                                entry_jint_arraycopy,
2094                                entry_jlong_arraycopy);
2095 
2096     StubRoutines::_generic_arraycopy =
2097         generate_generic_copy("generic_arraycopy",
2098                                entry_jbyte_arraycopy,
2099                                entry_jshort_arraycopy,
2100                                entry_jint_arraycopy,
2101                                entry_oop_arraycopy,
2102                                entry_jlong_arraycopy,
2103                                entry_checkcast_arraycopy);
2104   }
2105 
2106   void generate_math_stubs() {
2107     {
2108       StubCodeMark mark(this, "StubRoutines", "log");
2109       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2110 
2111       __ fld_d(Address(rsp, 4));
2112       __ flog();
2113       __ ret(0);
2114     }
2115     {
2116       StubCodeMark mark(this, "StubRoutines", "log10");
2117       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2118 
2119       __ fld_d(Address(rsp, 4));
2120       __ flog10();
2121       __ ret(0);
2122     }
2123     {
2124       StubCodeMark mark(this, "StubRoutines", "sin");
2125       StubRoutines::_intrinsic_sin = (double (*)(double))  __ pc();
2126 
2127       __ fld_d(Address(rsp, 4));
2128       __ trigfunc('s');
2129       __ ret(0);
2130     }
2131     {
2132       StubCodeMark mark(this, "StubRoutines", "cos");
2133       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2134 
2135       __ fld_d(Address(rsp, 4));
2136       __ trigfunc('c');
2137       __ ret(0);
2138     }
2139     {
2140       StubCodeMark mark(this, "StubRoutines", "tan");
2141       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2142 
2143       __ fld_d(Address(rsp, 4));
2144       __ trigfunc('t');
2145       __ ret(0);
2146     }
2147 
2148     // The intrinsic version of these seem to return the same value as
2149     // the strict version.
2150     StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2151     StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2152   }
2153 
2154  public:
2155   // Information about frame layout at time of blocking runtime call.
2156   // Note that we only have to preserve callee-saved registers since
2157   // the compilers are responsible for supplying a continuation point
2158   // if they expect all registers to be preserved.
2159   enum layout {
2160     thread_off,    // last_java_sp
2161     rbp_off,       // callee saved register
2162     ret_pc,
2163     framesize
2164   };
2165 
2166  private:
2167 
2168 #undef  __
2169 #define __ masm->
2170 
2171   //------------------------------------------------------------------------------------------------------------------------
2172   // Continuation point for throwing of implicit exceptions that are not handled in
2173   // the current activation. Fabricates an exception oop and initiates normal
2174   // exception dispatching in this frame.
2175   //
2176   // Previously the compiler (c2) allowed for callee save registers on Java calls.
2177   // This is no longer true after adapter frames were removed but could possibly
2178   // be brought back in the future if the interpreter code was reworked and it
2179   // was deemed worthwhile. The comment below was left to describe what must
2180   // happen here if callee saves were resurrected. As it stands now this stub
2181   // could actually be a vanilla BufferBlob and have now oopMap at all.
2182   // Since it doesn't make much difference we've chosen to leave it the
2183   // way it was in the callee save days and keep the comment.
2184 
2185   // If we need to preserve callee-saved values we need a callee-saved oop map and
2186   // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2187   // If the compiler needs all registers to be preserved between the fault
2188   // point and the exception handler then it must assume responsibility for that in
2189   // AbstractCompiler::continuation_for_implicit_null_exception or
2190   // continuation_for_implicit_division_by_zero_exception. All other implicit
2191   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2192   // either at call sites or otherwise assume that stack unwinding will be initiated,
2193   // so caller saved registers were assumed volatile in the compiler.
2194   address generate_throw_exception(const char* name, address runtime_entry,
2195                                    bool restore_saved_exception_pc) {
2196 
2197     int insts_size = 256;
2198     int locs_size  = 32;
2199 
2200     CodeBuffer code(name, insts_size, locs_size);
2201     OopMapSet* oop_maps  = new OopMapSet();
2202     MacroAssembler* masm = new MacroAssembler(&code);
2203 
2204     address start = __ pc();
2205 
2206     // This is an inlined and slightly modified version of call_VM
2207     // which has the ability to fetch the return PC out of
2208     // thread-local storage and also sets up last_Java_sp slightly
2209     // differently than the real call_VM
2210     Register java_thread = rbx;
2211     __ get_thread(java_thread);
2212     if (restore_saved_exception_pc) {
2213       __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2214       __ push(rax);
2215     }
2216 
2217     __ enter(); // required for proper stackwalking of RuntimeStub frame
2218 
2219     // pc and rbp, already pushed
2220     __ subptr(rsp, (framesize-2) * wordSize); // prolog
2221 
2222     // Frame is now completed as far as size and linkage.
2223 
2224     int frame_complete = __ pc() - start;
2225 
2226     // push java thread (becomes first argument of C function)
2227     __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2228 
2229     // Set up last_Java_sp and last_Java_fp
2230     __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2231 
2232     // Call runtime
2233     BLOCK_COMMENT("call runtime_entry");
2234     __ call(RuntimeAddress(runtime_entry));
2235     // Generate oop map
2236     OopMap* map =  new OopMap(framesize, 0);
2237     oop_maps->add_gc_map(__ pc() - start, map);
2238 
2239     // restore the thread (cannot use the pushed argument since arguments
2240     // may be overwritten by C code generated by an optimizing compiler);
2241     // however can use the register value directly if it is callee saved.
2242     __ get_thread(java_thread);
2243 
2244     __ reset_last_Java_frame(java_thread, true, false);
2245 
2246     __ leave(); // required for proper stackwalking of RuntimeStub frame
2247 
2248     // check for pending exceptions
2249 #ifdef ASSERT
2250     Label L;
2251     __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2252     __ jcc(Assembler::notEqual, L);
2253     __ should_not_reach_here();
2254     __ bind(L);
2255 #endif /* ASSERT */
2256     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2257 
2258 
2259     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2260     return stub->entry_point();
2261   }
2262 
2263 
2264   void create_control_words() {
2265     // Round to nearest, 53-bit mode, exceptions masked
2266     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;
2267     // Round to zero, 53-bit mode, exception mased
2268     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2269     // Round to nearest, 24-bit mode, exceptions masked
2270     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;
2271     // Round to nearest, 64-bit mode, exceptions masked
2272     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;
2273     // Round to nearest, 64-bit mode, exceptions masked
2274     StubRoutines::_mxcsr_std           = 0x1F80;
2275     // Note: the following two constants are 80-bit values
2276     //       layout is critical for correct loading by FPU.
2277     // Bias for strict fp multiply/divide
2278     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2279     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2280     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2281     // Un-Bias for strict fp multiply/divide
2282     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2283     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2284     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2285   }
2286 
2287   //---------------------------------------------------------------------------
2288   // Initialization
2289 
2290   void generate_initial() {
2291     // Generates all stubs and initializes the entry points
2292 
2293     //------------------------------------------------------------------------------------------------------------------------
2294     // entry points that exist in all platforms
2295     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2296     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2297     StubRoutines::_forward_exception_entry      = generate_forward_exception();
2298 
2299     StubRoutines::_call_stub_entry              =
2300       generate_call_stub(StubRoutines::_call_stub_return_address);
2301     // is referenced by megamorphic call
2302     StubRoutines::_catch_exception_entry        = generate_catch_exception();
2303 
2304     // These are currently used by Solaris/Intel
2305     StubRoutines::_atomic_xchg_entry            = generate_atomic_xchg();
2306 
2307     StubRoutines::_handler_for_unsafe_access_entry =
2308       generate_handler_for_unsafe_access();
2309 
2310     // platform dependent
2311     create_control_words();
2312 
2313     StubRoutines::x86::_verify_mxcsr_entry                 = generate_verify_mxcsr();
2314     StubRoutines::x86::_verify_fpu_cntrl_wrd_entry         = generate_verify_fpu_cntrl_wrd();
2315     StubRoutines::_d2i_wrapper                              = generate_d2i_wrapper(T_INT,
2316                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2317     StubRoutines::_d2l_wrapper                              = generate_d2i_wrapper(T_LONG,
2318                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2319   }
2320 
2321 
2322   void generate_all() {
2323     // Generates all stubs and initializes the entry points
2324 
2325     // These entry points require SharedInfo::stack0 to be set up in non-core builds
2326     // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2327     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError),  false);
2328     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError),  false);
2329     StubRoutines::_throw_ArithmeticException_entry         = generate_throw_exception("ArithmeticException throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException),  true);
2330     StubRoutines::_throw_NullPointerException_entry        = generate_throw_exception("NullPointerException throw_exception",         CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2331     StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
2332     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",           CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError),   false);
2333 
2334     //------------------------------------------------------------------------------------------------------------------------
2335     // entry points that are platform specific
2336 
2337     // support for verify_oop (must happen after universe_init)
2338     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop();
2339 
2340     // arraycopy stubs used by compilers
2341     generate_arraycopy_stubs();
2342 
2343     generate_math_stubs();
2344   }
2345 
2346 
2347  public:
2348   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2349     if (all) {
2350       generate_all();
2351     } else {
2352       generate_initial();
2353     }
2354   }
2355 }; // end class declaration
2356 
2357 
2358 void StubGenerator_generate(CodeBuffer* code, bool all) {
2359   StubGenerator g(code, all);
2360 }