jdk8-mips64-public/hotspot: src/cpu/x86/vm/stubGenerator_x86

     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  */

    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

    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

    58 #define __ _masm->

    59 #define a__ ((Assembler*)_masm)->

    61 #ifdef PRODUCT

    62 #define BLOCK_COMMENT(str) /* nothing */

    63 #else

    64 #define BLOCK_COMMENT(str) __ block_comment(str)

    65 #endif

    67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")

    69 const int MXCSR_MASK  = 0xFFC0;  // Mask out any pending exceptions

    70 const int FPU_CNTRL_WRD_MASK = 0xFFFF;

    72 // -------------------------------------------------------------------------------------------------------------------------

    73 // Stub Code definitions

    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);

    83   // request an async exception

    84   thread->set_pending_unsafe_access_error();

    86   // return address of next instruction to execute

    87   return npc;

    88 }

    90 class StubGenerator: public StubCodeGenerator {

    91  private:

    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

   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   }

   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               ]

   143   address generate_call_stub(address& return_address) {

   144     StubCodeMark mark(this, "StubRoutines", "call_stub");

   145     address start = __ pc();

   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;

   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

   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     }

   190     // make sure the control word is correct.

   191     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));

   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

   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);

   211     // parameter passing loop

   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]

   219     __ movptr(rdx, parameters);          // parameter pointer

   220     __ xorptr(rbx, rbx);

   222     __ BIND(loop);

   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);

   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);

   240     BLOCK_COMMENT("call_stub_return_address:");

   241     return_address = __ pc();

   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         }

   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

   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);

   280     // handle T_INT case

   281     __ movl(Address(rdi, 0), rax);

   282     __ BIND(exit);

   284     // check that FPU stack is empty

   285     __ verify_FPU(0, "generate_call_stub");

   287     // pop parameters

   288     __ lea(rsp, rsp_after_call);

   290     // restore %mxcsr

   291     if (sse_save) {

   292       __ ldmxcsr(mxcsr_save);

   293     }

   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);

   301     // return

   302     __ pop(rbp);

   303     __ ret(0);

   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);

   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);

   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);

   329     return start;

   330   }

   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

   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();

   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));

   372     return start;

   373   }

   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 !!

   387   address generate_forward_exception() {

   388     StubCodeMark mark(this, "StubRoutines", "forward exception");

   389     address start = __ pc();

   390     const Register thread = rcx;

   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;

   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.

   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

   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);

   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);

   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

   439     // Verify that there is really a valid exception in RAX.

   440     __ verify_oop(exception_oop);

   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);

   446     // continue at exception handler (return address removed)

   447     // rax: exception

   448     // rbx: exception handler

   449     // rdx: throwing pc

   450     __ jmp(handler_addr);

   452     return start;

   453   }

   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

   470   address generate_atomic_xchg() {

   471     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");

   472     address start = __ pc();

   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);

   483     return start;

   484   }

   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.

   494   address generate_verify_mxcsr() {

   495     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");

   496     address start = __ pc();

   498     const Address mxcsr_save(rsp, 0);

   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);

   511       __ warn("MXCSR changed by native JNI code.");

   513       __ ldmxcsr(mxcsr_std);

   515       __ bind(ok_ret);

   516       __ addptr(rsp, wordSize);

   517       __ pop(rax);

   518     }

   520     __ ret(0);

   522     return start;

   523   }

   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.

   533   address generate_verify_fpu_cntrl_wrd() {

   534     StubCodeMark mark(this, "StubRoutines", "verify_spcw");

   535     address start = __ pc();

   537     const Address fpu_cntrl_wrd_save(rsp, 0);

   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);

   550       __ warn("Floating point control word changed by native JNI code.");

   552       __ fldcw(fpu_std);

   554       __ bind(ok_ret);

   555       __ addptr(rsp, wordSize);

   556       __ pop(rax);

   557     }

   559     __ ret(0);

   561     return start;

   562   }

   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

   571   address generate_d2i_wrapper(BasicType t, address fcn) {

   572     StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");

   573     address start = __ pc();

   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   };

   587   assert(FPUStateSizeInWords == 27, "update stack layout");

   589     // Save outgoing argument to stack across push_FPU_state()

   590     __ subptr(rsp, wordSize * 2);

   591     __ fstp_d(Address(rsp, 0));

   593     // Save CPU & FPU state

   594     __ push(rbx);

   595     __ push(rcx);

   596     __ push(rsi);

   597     __ push(rdi);

   598     __ push(rbp);

   599     __ push_FPU_state();

   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));

   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 );

   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);

   626     __ ret(0);

   628     return start;

   629   }

   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();

   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

   649     return start;

   650   }

   653   //----------------------------------------------------------------------------------------------------

   654   // Non-destructive plausibility checks for oops

   656   address generate_verify_oop() {

   657     StubCodeMark mark(this, "StubRoutines", "verify_oop");

   658     address start = __ pc();

   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

   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

   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);

   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

   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);

   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

   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.

   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

   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   }

   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();

   756     }

   757   }

   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;

   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");

   788           Label L_loop;

   789           const Register end = count;  // elements count; end == start+count-1

   790           assert_different_registers(start, end);

   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();

   809     }

   810   }

   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);

   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);

   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     }

   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   }

   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   }

   924   address generate_disjoint_copy(BasicType t, bool aligned,

   925                                  Address::ScaleFactor sf,

   926                                  address* entry, const char *name) {

   927     __ align(CodeEntryAlignment);

   928     StubCodeMark mark(this, "StubRoutines", name);

   929     address start = __ pc();

   931     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;

   932     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;

   934     int shift = Address::times_ptr - sf;

   936     const Register from     = rsi;  // source array address

   937     const Register to       = rdi;  // destination array address

   938     const Register count    = rcx;  // elements count

   939     const Register to_from  = to;   // (to - from)

   940     const Register saved_to = rdx;  // saved destination array address

   942     __ enter(); // required for proper stackwalking of RuntimeStub frame

   943     __ push(rsi);

   944     __ push(rdi);

   945     __ movptr(from , Address(rsp, 12+ 4));

   946     __ movptr(to   , Address(rsp, 12+ 8));

   947     __ movl(count, Address(rsp, 12+ 12));

   948     if (t == T_OBJECT) {

   949       __ testl(count, count);

   950       __ jcc(Assembler::zero, L_0_count);

   951       gen_write_ref_array_pre_barrier(to, count);

   952       __ mov(saved_to, to);          // save 'to'

   953     }

   955     *entry = __ pc(); // Entry point from conjoint arraycopy stub.

   956     BLOCK_COMMENT("Entry:");

   958     __ subptr(to, from); // to --> to_from

   959     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element

   960     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp

   961     if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {

   962       // align source address at 4 bytes address boundary

   963       if (t == T_BYTE) {

   964         // One byte misalignment happens only for byte arrays

   965         __ testl(from, 1);

   966         __ jccb(Assembler::zero, L_skip_align1);

   967         __ movb(rax, Address(from, 0));

   968         __ movb(Address(from, to_from, Address::times_1, 0), rax);

   969         __ increment(from);

   970         __ decrement(count);

   971       __ BIND(L_skip_align1);

   972       }

   973       // Two bytes misalignment happens only for byte and short (char) arrays

   974       __ testl(from, 2);

   975       __ jccb(Assembler::zero, L_skip_align2);

   976       __ movw(rax, Address(from, 0));

   977       __ movw(Address(from, to_from, Address::times_1, 0), rax);

   978       __ addptr(from, 2);

   979       __ subl(count, 1<<(shift-1));

   980     __ BIND(L_skip_align2);

   981     }

   982     if (!VM_Version::supports_mmx()) {

   983       __ mov(rax, count);      // save 'count'

   984       __ shrl(count, shift); // bytes count

   985       __ addptr(to_from, from);// restore 'to'

   986       __ rep_mov();

   987       __ subptr(to_from, from);// restore 'to_from'

   988       __ mov(count, rax);      // restore 'count'

   989       __ jmpb(L_copy_2_bytes); // all dwords were copied

   990     } else {

   991       if (!UseUnalignedLoadStores) {

   992         // align to 8 bytes, we know we are 4 byte aligned to start

   993         __ testptr(from, 4);

   994         __ jccb(Assembler::zero, L_copy_64_bytes);

   995         __ movl(rax, Address(from, 0));

   996         __ movl(Address(from, to_from, Address::times_1, 0), rax);

   997         __ addptr(from, 4);

   998         __ subl(count, 1<<shift);

   999       }

  1000     __ BIND(L_copy_64_bytes);

  1001       __ mov(rax, count);

  1002       __ shrl(rax, shift+1);  // 8 bytes chunk count

  1003       //

  1004       // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop

  1005       //

  1006       if (UseXMMForArrayCopy) {

  1007         xmm_copy_forward(from, to_from, rax);

  1008       } else {

  1009         mmx_copy_forward(from, to_from, rax);

  1010       }

  1011     }

  1012     // copy tailing dword

  1013   __ BIND(L_copy_4_bytes);

  1014     __ testl(count, 1<<shift);

  1015     __ jccb(Assembler::zero, L_copy_2_bytes);

  1016     __ movl(rax, Address(from, 0));

  1017     __ movl(Address(from, to_from, Address::times_1, 0), rax);

  1018     if (t == T_BYTE || t == T_SHORT) {

  1019       __ addptr(from, 4);

  1020     __ BIND(L_copy_2_bytes);

  1021       // copy tailing word

  1022       __ testl(count, 1<<(shift-1));

  1023       __ jccb(Assembler::zero, L_copy_byte);

  1024       __ movw(rax, Address(from, 0));

  1025       __ movw(Address(from, to_from, Address::times_1, 0), rax);

  1026       if (t == T_BYTE) {

  1027         __ addptr(from, 2);

  1028       __ BIND(L_copy_byte);

  1029         // copy tailing byte

  1030         __ testl(count, 1);

  1031         __ jccb(Assembler::zero, L_exit);

  1032         __ movb(rax, Address(from, 0));

  1033         __ movb(Address(from, to_from, Address::times_1, 0), rax);

  1034       __ BIND(L_exit);

  1035       } else {

  1036       __ BIND(L_copy_byte);

  1037       }

  1038     } else {

  1039     __ BIND(L_copy_2_bytes);

  1040     }

  1042     if (t == T_OBJECT) {

  1043       __ movl(count, Address(rsp, 12+12)); // reread 'count'

  1044       __ mov(to, saved_to); // restore 'to'

  1045       gen_write_ref_array_post_barrier(to, count);

  1046     __ BIND(L_0_count);

  1047     }

  1048     inc_copy_counter_np(t);

  1049     __ pop(rdi);

  1050     __ pop(rsi);

  1051     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1052     __ xorptr(rax, rax); // return 0

  1053     __ ret(0);

  1054     return start;

  1055   }

  1058   address generate_fill(BasicType t, bool aligned, const char *name) {

  1059     __ align(CodeEntryAlignment);

  1060     StubCodeMark mark(this, "StubRoutines", name);

  1061     address start = __ pc();

  1063     BLOCK_COMMENT("Entry:");

  1065     const Register to       = rdi;  // source array address

  1066     const Register value    = rdx;  // value

  1067     const Register count    = rsi;  // elements count

  1069     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1070     __ push(rsi);

  1071     __ push(rdi);

  1072     __ movptr(to   , Address(rsp, 12+ 4));

  1073     __ movl(value, Address(rsp, 12+ 8));

  1074     __ movl(count, Address(rsp, 12+ 12));

  1076     __ generate_fill(t, aligned, to, value, count, rax, xmm0);

  1078     __ pop(rdi);

  1079     __ pop(rsi);

  1080     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1081     __ ret(0);

  1082     return start;

  1083   }

  1085   address generate_conjoint_copy(BasicType t, bool aligned,

  1086                                  Address::ScaleFactor sf,

  1087                                  address nooverlap_target,

  1088                                  address* entry, const char *name) {

  1089     __ align(CodeEntryAlignment);

  1090     StubCodeMark mark(this, "StubRoutines", name);

  1091     address start = __ pc();

  1093     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;

  1094     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;

  1096     int shift = Address::times_ptr - sf;

  1098     const Register src   = rax;  // source array address

  1099     const Register dst   = rdx;  // destination array address

  1100     const Register from  = rsi;  // source array address

  1101     const Register to    = rdi;  // destination array address

  1102     const Register count = rcx;  // elements count

  1103     const Register end   = rax;  // array end address

  1105     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1106     __ push(rsi);

  1107     __ push(rdi);

  1108     __ movptr(src  , Address(rsp, 12+ 4));   // from

  1109     __ movptr(dst  , Address(rsp, 12+ 8));   // to

  1110     __ movl2ptr(count, Address(rsp, 12+12)); // count

  1111     if (t == T_OBJECT) {

  1112        gen_write_ref_array_pre_barrier(dst, count);

  1113     }

  1115     if (entry != NULL) {

  1116       *entry = __ pc(); // Entry point from generic arraycopy stub.

  1117       BLOCK_COMMENT("Entry:");

  1118     }

  1120     if (t == T_OBJECT) {

  1121       __ testl(count, count);

  1122       __ jcc(Assembler::zero, L_0_count);

  1123     }

  1124     __ mov(from, src);

  1125     __ mov(to  , dst);

  1127     // arrays overlap test

  1128     RuntimeAddress nooverlap(nooverlap_target);

  1129     __ cmpptr(dst, src);

  1130     __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size

  1131     __ jump_cc(Assembler::belowEqual, nooverlap);

  1132     __ cmpptr(dst, end);

  1133     __ jump_cc(Assembler::aboveEqual, nooverlap);

  1135     // copy from high to low

  1136     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element

  1137     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp

  1138     if (t == T_BYTE || t == T_SHORT) {

  1139       // Align the end of destination array at 4 bytes address boundary

  1140       __ lea(end, Address(dst, count, sf, 0));

  1141       if (t == T_BYTE) {

  1142         // One byte misalignment happens only for byte arrays

  1143         __ testl(end, 1);

  1144         __ jccb(Assembler::zero, L_skip_align1);

  1145         __ decrement(count);

  1146         __ movb(rdx, Address(from, count, sf, 0));

  1147         __ movb(Address(to, count, sf, 0), rdx);

  1148       __ BIND(L_skip_align1);

  1149       }

  1150       // Two bytes misalignment happens only for byte and short (char) arrays

  1151       __ testl(end, 2);

  1152       __ jccb(Assembler::zero, L_skip_align2);

  1153       __ subptr(count, 1<<(shift-1));

  1154       __ movw(rdx, Address(from, count, sf, 0));

  1155       __ movw(Address(to, count, sf, 0), rdx);

  1156     __ BIND(L_skip_align2);

  1157       __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element

  1158       __ jcc(Assembler::below, L_copy_4_bytes);

  1159     }

  1161     if (!VM_Version::supports_mmx()) {

  1162       __ std();

  1163       __ mov(rax, count); // Save 'count'

  1164       __ mov(rdx, to);    // Save 'to'

  1165       __ lea(rsi, Address(from, count, sf, -4));

  1166       __ lea(rdi, Address(to  , count, sf, -4));

  1167       __ shrptr(count, shift); // bytes count

  1168       __ rep_mov();

  1169       __ cld();

  1170       __ mov(count, rax); // restore 'count'

  1171       __ andl(count, (1<<shift)-1);      // mask the number of rest elements

  1172       __ movptr(from, Address(rsp, 12+4)); // reread 'from'

  1173       __ mov(to, rdx);   // restore 'to'

  1174       __ jmpb(L_copy_2_bytes); // all dword were copied

  1175    } else {

  1176       // Align to 8 bytes the end of array. It is aligned to 4 bytes already.

  1177       __ testptr(end, 4);

  1178       __ jccb(Assembler::zero, L_copy_8_bytes);

  1179       __ subl(count, 1<<shift);

  1180       __ movl(rdx, Address(from, count, sf, 0));

  1181       __ movl(Address(to, count, sf, 0), rdx);

  1182       __ jmpb(L_copy_8_bytes);

  1184       __ align(OptoLoopAlignment);

  1185       // Move 8 bytes

  1186     __ BIND(L_copy_8_bytes_loop);

  1187       if (UseXMMForArrayCopy) {

  1188         __ movq(xmm0, Address(from, count, sf, 0));

  1189         __ movq(Address(to, count, sf, 0), xmm0);

  1190       } else {

  1191         __ movq(mmx0, Address(from, count, sf, 0));

  1192         __ movq(Address(to, count, sf, 0), mmx0);

  1193       }

  1194     __ BIND(L_copy_8_bytes);

  1195       __ subl(count, 2<<shift);

  1196       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);

  1197       __ addl(count, 2<<shift);

  1198       if (!UseXMMForArrayCopy) {

  1199         __ emms();

  1200       }

  1201     }

  1202   __ BIND(L_copy_4_bytes);

  1203     // copy prefix qword

  1204     __ testl(count, 1<<shift);

  1205     __ jccb(Assembler::zero, L_copy_2_bytes);

  1206     __ movl(rdx, Address(from, count, sf, -4));

  1207     __ movl(Address(to, count, sf, -4), rdx);

  1209     if (t == T_BYTE || t == T_SHORT) {

  1210         __ subl(count, (1<<shift));

  1211       __ BIND(L_copy_2_bytes);

  1212         // copy prefix dword

  1213         __ testl(count, 1<<(shift-1));

  1214         __ jccb(Assembler::zero, L_copy_byte);

  1215         __ movw(rdx, Address(from, count, sf, -2));

  1216         __ movw(Address(to, count, sf, -2), rdx);

  1217         if (t == T_BYTE) {

  1218           __ subl(count, 1<<(shift-1));

  1219         __ BIND(L_copy_byte);

  1220           // copy prefix byte

  1221           __ testl(count, 1);

  1222           __ jccb(Assembler::zero, L_exit);

  1223           __ movb(rdx, Address(from, 0));

  1224           __ movb(Address(to, 0), rdx);

  1225         __ BIND(L_exit);

  1226         } else {

  1227         __ BIND(L_copy_byte);

  1228         }

  1229     } else {

  1230     __ BIND(L_copy_2_bytes);

  1231     }

  1232     if (t == T_OBJECT) {

  1233       __ movl2ptr(count, Address(rsp, 12+12)); // reread count

  1234       gen_write_ref_array_post_barrier(to, count);

  1235     __ BIND(L_0_count);

  1236     }

  1237     inc_copy_counter_np(t);

  1238     __ pop(rdi);

  1239     __ pop(rsi);

  1240     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1241     __ xorptr(rax, rax); // return 0

  1242     __ ret(0);

  1243     return start;

  1244   }

  1247   address generate_disjoint_long_copy(address* entry, const char *name) {

  1248     __ align(CodeEntryAlignment);

  1249     StubCodeMark mark(this, "StubRoutines", name);

  1250     address start = __ pc();

  1252     Label L_copy_8_bytes, L_copy_8_bytes_loop;

  1253     const Register from       = rax;  // source array address

  1254     const Register to         = rdx;  // destination array address

  1255     const Register count      = rcx;  // elements count

  1256     const Register to_from    = rdx;  // (to - from)

  1258     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1259     __ movptr(from , Address(rsp, 8+0));       // from

  1260     __ movptr(to   , Address(rsp, 8+4));       // to

  1261     __ movl2ptr(count, Address(rsp, 8+8));     // count

  1263     *entry = __ pc(); // Entry point from conjoint arraycopy stub.

  1264     BLOCK_COMMENT("Entry:");

  1266     __ subptr(to, from); // to --> to_from

  1267     if (VM_Version::supports_mmx()) {

  1268       if (UseXMMForArrayCopy) {

  1269         xmm_copy_forward(from, to_from, count);

  1270       } else {

  1271         mmx_copy_forward(from, to_from, count);

  1272       }

  1273     } else {

  1274       __ jmpb(L_copy_8_bytes);

  1275       __ align(OptoLoopAlignment);

  1276     __ BIND(L_copy_8_bytes_loop);

  1277       __ fild_d(Address(from, 0));

  1278       __ fistp_d(Address(from, to_from, Address::times_1));

  1279       __ addptr(from, 8);

  1280     __ BIND(L_copy_8_bytes);

  1281       __ decrement(count);

  1282       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);

  1283     }

  1284     inc_copy_counter_np(T_LONG);

  1285     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1286     __ xorptr(rax, rax); // return 0

  1287     __ ret(0);

  1288     return start;

  1289   }

  1291   address generate_conjoint_long_copy(address nooverlap_target,

  1292                                       address* entry, const char *name) {

  1293     __ align(CodeEntryAlignment);

  1294     StubCodeMark mark(this, "StubRoutines", name);

  1295     address start = __ pc();

  1297     Label L_copy_8_bytes, L_copy_8_bytes_loop;

  1298     const Register from       = rax;  // source array address

  1299     const Register to         = rdx;  // destination array address

  1300     const Register count      = rcx;  // elements count

  1301     const Register end_from   = rax;  // source array end address

  1303     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1304     __ movptr(from , Address(rsp, 8+0));       // from

  1305     __ movptr(to   , Address(rsp, 8+4));       // to

  1306     __ movl2ptr(count, Address(rsp, 8+8));     // count

  1308     *entry = __ pc(); // Entry point from generic arraycopy stub.

  1309     BLOCK_COMMENT("Entry:");

  1311     // arrays overlap test

  1312     __ cmpptr(to, from);

  1313     RuntimeAddress nooverlap(nooverlap_target);

  1314     __ jump_cc(Assembler::belowEqual, nooverlap);

  1315     __ lea(end_from, Address(from, count, Address::times_8, 0));

  1316     __ cmpptr(to, end_from);

  1317     __ movptr(from, Address(rsp, 8));  // from

  1318     __ jump_cc(Assembler::aboveEqual, nooverlap);

  1320     __ jmpb(L_copy_8_bytes);

  1322     __ align(OptoLoopAlignment);

  1323   __ BIND(L_copy_8_bytes_loop);

  1324     if (VM_Version::supports_mmx()) {

  1325       if (UseXMMForArrayCopy) {

  1326         __ movq(xmm0, Address(from, count, Address::times_8));

  1327         __ movq(Address(to, count, Address::times_8), xmm0);

  1328       } else {

  1329         __ movq(mmx0, Address(from, count, Address::times_8));

  1330         __ movq(Address(to, count, Address::times_8), mmx0);

  1331       }

  1332     } else {

  1333       __ fild_d(Address(from, count, Address::times_8));

  1334       __ fistp_d(Address(to, count, Address::times_8));

  1335     }

  1336   __ BIND(L_copy_8_bytes);

  1337     __ decrement(count);

  1338     __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);

  1340     if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {

  1341       __ emms();

  1342     }

  1343     inc_copy_counter_np(T_LONG);

  1344     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1345     __ xorptr(rax, rax); // return 0

  1346     __ ret(0);

  1347     return start;

  1348   }

  1351   // Helper for generating a dynamic type check.

  1352   // The sub_klass must be one of {rbx, rdx, rsi}.

  1353   // The temp is killed.

  1354   void generate_type_check(Register sub_klass,

  1355                            Address& super_check_offset_addr,

  1356                            Address& super_klass_addr,

  1357                            Register temp,

  1358                            Label* L_success, Label* L_failure) {

  1359     BLOCK_COMMENT("type_check:");

  1361     Label L_fallthrough;

  1362 #define LOCAL_JCC(assembler_con, label_ptr)                             \

  1363     if (label_ptr != NULL)  __ jcc(assembler_con, *(label_ptr));        \

  1364     else                    __ jcc(assembler_con, L_fallthrough) /*omit semi*/

  1366     // The following is a strange variation of the fast path which requires

  1367     // one less register, because needed values are on the argument stack.

  1368     // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,

  1369     //                                  L_success, L_failure, NULL);

  1370     assert_different_registers(sub_klass, temp);

  1372     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +

  1373                      Klass::secondary_super_cache_offset_in_bytes());

  1375     // if the pointers are equal, we are done (e.g., String[] elements)

  1376     __ cmpptr(sub_klass, super_klass_addr);

  1377     LOCAL_JCC(Assembler::equal, L_success);

  1379     // check the supertype display:

  1380     __ movl2ptr(temp, super_check_offset_addr);

  1381     Address super_check_addr(sub_klass, temp, Address::times_1, 0);

  1382     __ movptr(temp, super_check_addr); // load displayed supertype

  1383     __ cmpptr(temp, super_klass_addr); // test the super type

  1384     LOCAL_JCC(Assembler::equal, L_success);

  1386     // if it was a primary super, we can just fail immediately

  1387     __ cmpl(super_check_offset_addr, sc_offset);

  1388     LOCAL_JCC(Assembler::notEqual, L_failure);

  1390     // The repne_scan instruction uses fixed registers, which will get spilled.

  1391     // We happen to know this works best when super_klass is in rax.

  1392     Register super_klass = temp;

  1393     __ movptr(super_klass, super_klass_addr);

  1394     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,

  1395                                      L_success, L_failure);

  1397     __ bind(L_fallthrough);

  1399     if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }

  1400     if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }

  1402 #undef LOCAL_JCC

  1403   }

  1405   //

  1406   //  Generate checkcasting array copy stub

  1407   //

  1408   //  Input:

  1409   //    4(rsp)   - source array address

  1410   //    8(rsp)   - destination array address

  1411   //   12(rsp)   - element count, can be zero

  1412   //   16(rsp)   - size_t ckoff (super_check_offset)

  1413   //   20(rsp)   - oop ckval (super_klass)

  1414   //

  1415   //  Output:

  1416   //    rax, ==  0  -  success

  1417   //    rax, == -1^K - failure, where K is partial transfer count

  1418   //

  1419   address generate_checkcast_copy(const char *name, address* entry) {

  1420     __ align(CodeEntryAlignment);

  1421     StubCodeMark mark(this, "StubRoutines", name);

  1422     address start = __ pc();

  1424     Label L_load_element, L_store_element, L_do_card_marks, L_done;

  1426     // register use:

  1427     //  rax, rdx, rcx -- loop control (end_from, end_to, count)

  1428     //  rdi, rsi      -- element access (oop, klass)

  1429     //  rbx,           -- temp

  1430     const Register from       = rax;    // source array address

  1431     const Register to         = rdx;    // destination array address

  1432     const Register length     = rcx;    // elements count

  1433     const Register elem       = rdi;    // each oop copied

  1434     const Register elem_klass = rsi;    // each elem._klass (sub_klass)

  1435     const Register temp       = rbx;    // lone remaining temp

  1437     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1439     __ push(rsi);

  1440     __ push(rdi);

  1441     __ push(rbx);

  1443     Address   from_arg(rsp, 16+ 4);     // from

  1444     Address     to_arg(rsp, 16+ 8);     // to

  1445     Address length_arg(rsp, 16+12);     // elements count

  1446     Address  ckoff_arg(rsp, 16+16);     // super_check_offset

  1447     Address  ckval_arg(rsp, 16+20);     // super_klass

  1449     // Load up:

  1450     __ movptr(from,     from_arg);

  1451     __ movptr(to,         to_arg);

  1452     __ movl2ptr(length, length_arg);

  1454     *entry = __ pc(); // Entry point from generic arraycopy stub.

  1455     BLOCK_COMMENT("Entry:");

  1457     //---------------------------------------------------------------

  1458     // Assembler stub will be used for this call to arraycopy

  1459     // if the two arrays are subtypes of Object[] but the

  1460     // destination array type is not equal to or a supertype

  1461     // of the source type.  Each element must be separately

  1462     // checked.

  1464     // Loop-invariant addresses.  They are exclusive end pointers.

  1465     Address end_from_addr(from, length, Address::times_ptr, 0);

  1466     Address   end_to_addr(to,   length, Address::times_ptr, 0);

  1468     Register end_from = from;           // re-use

  1469     Register end_to   = to;             // re-use

  1470     Register count    = length;         // re-use

  1472     // Loop-variant addresses.  They assume post-incremented count < 0.

  1473     Address from_element_addr(end_from, count, Address::times_ptr, 0);

  1474     Address   to_element_addr(end_to,   count, Address::times_ptr, 0);

  1475     Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());

  1477     // Copy from low to high addresses, indexed from the end of each array.

  1478     gen_write_ref_array_pre_barrier(to, count);

  1479     __ lea(end_from, end_from_addr);

  1480     __ lea(end_to,   end_to_addr);

  1481     assert(length == count, "");        // else fix next line:

  1482     __ negptr(count);                   // negate and test the length

  1483     __ jccb(Assembler::notZero, L_load_element);

  1485     // Empty array:  Nothing to do.

  1486     __ xorptr(rax, rax);                  // return 0 on (trivial) success

  1487     __ jmp(L_done);

  1489     // ======== begin loop ========

  1490     // (Loop is rotated; its entry is L_load_element.)

  1491     // Loop control:

  1492     //   for (count = -count; count != 0; count++)

  1493     // Base pointers src, dst are biased by 8*count,to last element.

  1494     __ align(OptoLoopAlignment);

  1496     __ BIND(L_store_element);

  1497     __ movptr(to_element_addr, elem);     // store the oop

  1498     __ increment(count);                // increment the count toward zero

  1499     __ jccb(Assembler::zero, L_do_card_marks);

  1501     // ======== loop entry is here ========

  1502     __ BIND(L_load_element);

  1503     __ movptr(elem, from_element_addr);   // load the oop

  1504     __ testptr(elem, elem);

  1505     __ jccb(Assembler::zero, L_store_element);

  1507     // (Could do a trick here:  Remember last successful non-null

  1508     // element stored and make a quick oop equality check on it.)

  1510     __ movptr(elem_klass, elem_klass_addr); // query the object klass

  1511     generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,

  1512                         &L_store_element, NULL);

  1513       // (On fall-through, we have failed the element type check.)

  1514     // ======== end loop ========

  1516     // It was a real error; we must depend on the caller to finish the job.

  1517     // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.

  1518     // Emit GC store barriers for the oops we have copied (length_arg + count),

  1519     // and report their number to the caller.

  1520     __ addl(count, length_arg);         // transfers = (length - remaining)

  1521     __ movl2ptr(rax, count);            // save the value

  1522     __ notptr(rax);                     // report (-1^K) to caller

  1523     __ movptr(to, to_arg);              // reload

  1524     assert_different_registers(to, count, rax);

  1525     gen_write_ref_array_post_barrier(to, count);

  1526     __ jmpb(L_done);

  1528     // Come here on success only.

  1529     __ BIND(L_do_card_marks);

  1530     __ movl2ptr(count, length_arg);

  1531     __ movptr(to, to_arg);                // reload

  1532     gen_write_ref_array_post_barrier(to, count);

  1533     __ xorptr(rax, rax);                  // return 0 on success

  1535     // Common exit point (success or failure).

  1536     __ BIND(L_done);

  1537     __ pop(rbx);

  1538     __ pop(rdi);

  1539     __ pop(rsi);

  1540     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);

  1541     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1542     __ ret(0);

  1544     return start;

  1545   }

  1547   //

  1548   //  Generate 'unsafe' array copy stub

  1549   //  Though just as safe as the other stubs, it takes an unscaled

  1550   //  size_t argument instead of an element count.

  1551   //

  1552   //  Input:

  1553   //    4(rsp)   - source array address

  1554   //    8(rsp)   - destination array address

  1555   //   12(rsp)   - byte count, can be zero

  1556   //

  1557   //  Output:

  1558   //    rax, ==  0  -  success

  1559   //    rax, == -1  -  need to call System.arraycopy

  1560   //

  1561   // Examines the alignment of the operands and dispatches

  1562   // to a long, int, short, or byte copy loop.

  1563   //

  1564   address generate_unsafe_copy(const char *name,

  1565                                address byte_copy_entry,

  1566                                address short_copy_entry,

  1567                                address int_copy_entry,

  1568                                address long_copy_entry) {

  1570     Label L_long_aligned, L_int_aligned, L_short_aligned;

  1572     __ align(CodeEntryAlignment);

  1573     StubCodeMark mark(this, "StubRoutines", name);

  1574     address start = __ pc();

  1576     const Register from       = rax;  // source array address

  1577     const Register to         = rdx;  // destination array address

  1578     const Register count      = rcx;  // elements count

  1580     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1581     __ push(rsi);

  1582     __ push(rdi);

  1583     Address  from_arg(rsp, 12+ 4);      // from

  1584     Address    to_arg(rsp, 12+ 8);      // to

  1585     Address count_arg(rsp, 12+12);      // byte count

  1587     // Load up:

  1588     __ movptr(from ,  from_arg);

  1589     __ movptr(to   ,    to_arg);

  1590     __ movl2ptr(count, count_arg);

  1592     // bump this on entry, not on exit:

  1593     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);

  1595     const Register bits = rsi;

  1596     __ mov(bits, from);

  1597     __ orptr(bits, to);

  1598     __ orptr(bits, count);

  1600     __ testl(bits, BytesPerLong-1);

  1601     __ jccb(Assembler::zero, L_long_aligned);

  1603     __ testl(bits, BytesPerInt-1);

  1604     __ jccb(Assembler::zero, L_int_aligned);

  1606     __ testl(bits, BytesPerShort-1);

  1607     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));

  1609     __ BIND(L_short_aligned);

  1610     __ shrptr(count, LogBytesPerShort); // size => short_count

  1611     __ movl(count_arg, count);          // update 'count'

  1612     __ jump(RuntimeAddress(short_copy_entry));

  1614     __ BIND(L_int_aligned);

  1615     __ shrptr(count, LogBytesPerInt); // size => int_count

  1616     __ movl(count_arg, count);          // update 'count'

  1617     __ jump(RuntimeAddress(int_copy_entry));

  1619     __ BIND(L_long_aligned);

  1620     __ shrptr(count, LogBytesPerLong); // size => qword_count

  1621     __ movl(count_arg, count);          // update 'count'

  1622     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.

  1623     __ pop(rsi);

  1624     __ jump(RuntimeAddress(long_copy_entry));

  1626     return start;

  1627   }

  1630   // Perform range checks on the proposed arraycopy.

  1631   // Smashes src_pos and dst_pos.  (Uses them up for temps.)

  1632   void arraycopy_range_checks(Register src,

  1633                               Register src_pos,

  1634                               Register dst,

  1635                               Register dst_pos,

  1636                               Address& length,

  1637                               Label& L_failed) {

  1638     BLOCK_COMMENT("arraycopy_range_checks:");

  1639     const Register src_end = src_pos;   // source array end position

  1640     const Register dst_end = dst_pos;   // destination array end position

  1641     __ addl(src_end, length); // src_pos + length

  1642     __ addl(dst_end, length); // dst_pos + length

  1644     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;

  1645     __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));

  1646     __ jcc(Assembler::above, L_failed);

  1648     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;

  1649     __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));

  1650     __ jcc(Assembler::above, L_failed);

  1652     BLOCK_COMMENT("arraycopy_range_checks done");

  1653   }

  1656   //

  1657   //  Generate generic array copy stubs

  1658   //

  1659   //  Input:

  1660   //     4(rsp)    -  src oop

  1661   //     8(rsp)    -  src_pos

  1662   //    12(rsp)    -  dst oop

  1663   //    16(rsp)    -  dst_pos

  1664   //    20(rsp)    -  element count

  1665   //

  1666   //  Output:

  1667   //    rax, ==  0  -  success

  1668   //    rax, == -1^K - failure, where K is partial transfer count

  1669   //

  1670   address generate_generic_copy(const char *name,

  1671                                 address entry_jbyte_arraycopy,

  1672                                 address entry_jshort_arraycopy,

  1673                                 address entry_jint_arraycopy,

  1674                                 address entry_oop_arraycopy,

  1675                                 address entry_jlong_arraycopy,

  1676                                 address entry_checkcast_arraycopy) {

  1677     Label L_failed, L_failed_0, L_objArray;

  1679     { int modulus = CodeEntryAlignment;

  1680       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)

  1681       int advance = target - (__ offset() % modulus);

  1682       if (advance < 0)  advance += modulus;

  1683       if (advance > 0)  __ nop(advance);

  1684     }

  1685     StubCodeMark mark(this, "StubRoutines", name);

  1687     // Short-hop target to L_failed.  Makes for denser prologue code.

  1688     __ BIND(L_failed_0);

  1689     __ jmp(L_failed);

  1690     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");

  1692     __ align(CodeEntryAlignment);

  1693     address start = __ pc();

  1695     __ enter(); // required for proper stackwalking of RuntimeStub frame

  1696     __ push(rsi);

  1697     __ push(rdi);

  1699     // bump this on entry, not on exit:

  1700     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);

  1702     // Input values

  1703     Address SRC     (rsp, 12+ 4);

  1704     Address SRC_POS (rsp, 12+ 8);

  1705     Address DST     (rsp, 12+12);

  1706     Address DST_POS (rsp, 12+16);

  1707     Address LENGTH  (rsp, 12+20);

  1709     //-----------------------------------------------------------------------

  1710     // Assembler stub will be used for this call to arraycopy

  1711     // if the following conditions are met:

  1712     //

  1713     // (1) src and dst must not be null.

  1714     // (2) src_pos must not be negative.

  1715     // (3) dst_pos must not be negative.

  1716     // (4) length  must not be negative.

  1717     // (5) src klass and dst klass should be the same and not NULL.

  1718     // (6) src and dst should be arrays.

  1719     // (7) src_pos + length must not exceed length of src.

  1720     // (8) dst_pos + length must not exceed length of dst.

  1721     //

  1723     const Register src     = rax;       // source array oop

  1724     const Register src_pos = rsi;

  1725     const Register dst     = rdx;       // destination array oop

  1726     const Register dst_pos = rdi;

  1727     const Register length  = rcx;       // transfer count

  1729     //  if (src == NULL) return -1;

  1730     __ movptr(src, SRC);      // src oop

  1731     __ testptr(src, src);

  1732     __ jccb(Assembler::zero, L_failed_0);

  1734     //  if (src_pos < 0) return -1;

  1735     __ movl2ptr(src_pos, SRC_POS);  // src_pos

  1736     __ testl(src_pos, src_pos);

  1737     __ jccb(Assembler::negative, L_failed_0);

  1739     //  if (dst == NULL) return -1;

  1740     __ movptr(dst, DST);      // dst oop

  1741     __ testptr(dst, dst);

  1742     __ jccb(Assembler::zero, L_failed_0);

  1744     //  if (dst_pos < 0) return -1;

  1745     __ movl2ptr(dst_pos, DST_POS);  // dst_pos

  1746     __ testl(dst_pos, dst_pos);

  1747     __ jccb(Assembler::negative, L_failed_0);

  1749     //  if (length < 0) return -1;

  1750     __ movl2ptr(length, LENGTH);   // length

  1751     __ testl(length, length);

  1752     __ jccb(Assembler::negative, L_failed_0);

  1754     //  if (src->klass() == NULL) return -1;

  1755     Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());

  1756     Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());

  1757     const Register rcx_src_klass = rcx;    // array klass

  1758     __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));

  1760 #ifdef ASSERT

  1761     //  assert(src->klass() != NULL);

  1762     BLOCK_COMMENT("assert klasses not null");

  1763     { Label L1, L2;

  1764       __ testptr(rcx_src_klass, rcx_src_klass);

  1765       __ jccb(Assembler::notZero, L2);   // it is broken if klass is NULL

  1766       __ bind(L1);

  1767       __ stop("broken null klass");

  1768       __ bind(L2);

  1769       __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);

  1770       __ jccb(Assembler::equal, L1);      // this would be broken also

  1771       BLOCK_COMMENT("assert done");

  1772     }

  1773 #endif //ASSERT

  1775     // Load layout helper (32-bits)

  1776     //

  1777     //  |array_tag|     | header_size | element_type |     |log2_element_size|

  1778     // 32        30    24            16              8     2                 0

  1779     //

  1780     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0

  1781     //

  1783     int lh_offset = klassOopDesc::header_size() * HeapWordSize +

  1784                     Klass::layout_helper_offset_in_bytes();

  1785     Address src_klass_lh_addr(rcx_src_klass, lh_offset);

  1787     // Handle objArrays completely differently...

  1788     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);

  1789     __ cmpl(src_klass_lh_addr, objArray_lh);

  1790     __ jcc(Assembler::equal, L_objArray);

  1792     //  if (src->klass() != dst->klass()) return -1;

  1793     __ cmpptr(rcx_src_klass, dst_klass_addr);

  1794     __ jccb(Assembler::notEqual, L_failed_0);

  1796     const Register rcx_lh = rcx;  // layout helper

  1797     assert(rcx_lh == rcx_src_klass, "known alias");

  1798     __ movl(rcx_lh, src_klass_lh_addr);

  1800     //  if (!src->is_Array()) return -1;

  1801     __ cmpl(rcx_lh, Klass::_lh_neutral_value);

  1802     __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp

  1804     // At this point, it is known to be a typeArray (array_tag 0x3).

  1805 #ifdef ASSERT

  1806     { Label L;

  1807       __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));

  1808       __ jcc(Assembler::greaterEqual, L); // signed cmp

  1809       __ stop("must be a primitive array");

  1810       __ bind(L);

  1811     }

  1812 #endif

  1814     assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);

  1815     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);

  1817     // typeArrayKlass

  1818     //

  1819     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);

  1820     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);

  1821     //

  1822     const Register rsi_offset = rsi; // array offset

  1823     const Register src_array  = src; // src array offset

  1824     const Register dst_array  = dst; // dst array offset

  1825     const Register rdi_elsize = rdi; // log2 element size

  1827     __ mov(rsi_offset, rcx_lh);

  1828     __ shrptr(rsi_offset, Klass::_lh_header_size_shift);

  1829     __ andptr(rsi_offset, Klass::_lh_header_size_mask);   // array_offset

  1830     __ addptr(src_array, rsi_offset);  // src array offset

  1831     __ addptr(dst_array, rsi_offset);  // dst array offset

  1832     __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize

  1834     // next registers should be set before the jump to corresponding stub

  1835     const Register from       = src; // source array address

  1836     const Register to         = dst; // destination array address

  1837     const Register count      = rcx; // elements count

  1838     // some of them should be duplicated on stack

  1839 #define FROM   Address(rsp, 12+ 4)

  1840 #define TO     Address(rsp, 12+ 8)   // Not used now

  1841 #define COUNT  Address(rsp, 12+12)   // Only for oop arraycopy

  1843     BLOCK_COMMENT("scale indexes to element size");

  1844     __ movl2ptr(rsi, SRC_POS);  // src_pos

  1845     __ shlptr(rsi);             // src_pos << rcx (log2 elsize)

  1846     assert(src_array == from, "");

  1847     __ addptr(from, rsi);       // from = src_array + SRC_POS << log2 elsize

  1848     __ movl2ptr(rdi, DST_POS);  // dst_pos

  1849     __ shlptr(rdi);             // dst_pos << rcx (log2 elsize)

  1850     assert(dst_array == to, "");

  1851     __ addptr(to,  rdi);        // to   = dst_array + DST_POS << log2 elsize

  1852     __ movptr(FROM, from);      // src_addr

  1853     __ mov(rdi_elsize, rcx_lh); // log2 elsize

  1854     __ movl2ptr(count, LENGTH); // elements count

  1856     BLOCK_COMMENT("choose copy loop based on element size");

  1857     __ cmpl(rdi_elsize, 0);

  1859     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));

  1860     __ cmpl(rdi_elsize, LogBytesPerShort);

  1861     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));

  1862     __ cmpl(rdi_elsize, LogBytesPerInt);

  1863     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));

  1864 #ifdef ASSERT

  1865     __ cmpl(rdi_elsize, LogBytesPerLong);

  1866     __ jccb(Assembler::notEqual, L_failed);

  1867 #endif

  1868     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.

  1869     __ pop(rsi);

  1870     __ jump(RuntimeAddress(entry_jlong_arraycopy));

  1872   __ BIND(L_failed);

  1873     __ xorptr(rax, rax);

  1874     __ notptr(rax); // return -1

  1875     __ pop(rdi);

  1876     __ pop(rsi);

  1877     __ leave(); // required for proper stackwalking of RuntimeStub frame

  1878     __ ret(0);

  1880     // objArrayKlass

  1881   __ BIND(L_objArray);

  1882     // live at this point:  rcx_src_klass, src[_pos], dst[_pos]

  1884     Label L_plain_copy, L_checkcast_copy;

  1885     //  test array classes for subtyping

  1886     __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality

  1887     __ jccb(Assembler::notEqual, L_checkcast_copy);

  1889     // Identically typed arrays can be copied without element-wise checks.

  1890     assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);

  1891     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);

  1893   __ BIND(L_plain_copy);

  1894     __ movl2ptr(count, LENGTH); // elements count

  1895     __ movl2ptr(src_pos, SRC_POS);  // reload src_pos

  1896     __ lea(from, Address(src, src_pos, Address::times_ptr,

  1897                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr

  1898     __ movl2ptr(dst_pos, DST_POS);  // reload dst_pos

  1899     __ lea(to,   Address(dst, dst_pos, Address::times_ptr,

  1900                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr

  1901     __ movptr(FROM,  from);   // src_addr

  1902     __ movptr(TO,    to);     // dst_addr

  1903     __ movl(COUNT, count);  // count

  1904     __ jump(RuntimeAddress(entry_oop_arraycopy));

  1906   __ BIND(L_checkcast_copy);

  1907     // live at this point:  rcx_src_klass, dst[_pos], src[_pos]

  1908     {

  1909       // Handy offsets:

  1910       int  ek_offset = (klassOopDesc::header_size() * HeapWordSize +

  1911                         objArrayKlass::element_klass_offset_in_bytes());

  1912       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +

  1913                         Klass::super_check_offset_offset_in_bytes());

  1915       Register rsi_dst_klass = rsi;

  1916       Register rdi_temp      = rdi;

  1917       assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");

  1918       assert(rdi_temp      == dst_pos, "expected alias w/ dst_pos");

  1919       Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);

  1921       // Before looking at dst.length, make sure dst is also an objArray.

  1922       __ movptr(rsi_dst_klass, dst_klass_addr);

  1923       __ cmpl(dst_klass_lh_addr, objArray_lh);

  1924       __ jccb(Assembler::notEqual, L_failed);

  1926       // It is safe to examine both src.length and dst.length.

  1927       __ movl2ptr(src_pos, SRC_POS);        // reload rsi

  1928       arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);

  1929       // (Now src_pos and dst_pos are killed, but not src and dst.)

  1931       // We'll need this temp (don't forget to pop it after the type check).

  1932       __ push(rbx);

  1933       Register rbx_src_klass = rbx;

  1935       __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx

  1936       __ movptr(rsi_dst_klass, dst_klass_addr);

  1937       Address super_check_offset_addr(rsi_dst_klass, sco_offset);

  1938       Label L_fail_array_check;

  1939       generate_type_check(rbx_src_klass,

  1940                           super_check_offset_addr, dst_klass_addr,

  1941                           rdi_temp, NULL, &L_fail_array_check);

  1942       // (On fall-through, we have passed the array type check.)

  1943       __ pop(rbx);

  1944       __ jmp(L_plain_copy);

  1946       __ BIND(L_fail_array_check);

  1947       // Reshuffle arguments so we can call checkcast_arraycopy:

  1949       // match initial saves for checkcast_arraycopy

  1950       // push(rsi);    // already done; see above

  1951       // push(rdi);    // already done; see above

  1952       // push(rbx);    // already done; see above

  1954       // Marshal outgoing arguments now, freeing registers.

  1955       Address   from_arg(rsp, 16+ 4);   // from

  1956       Address     to_arg(rsp, 16+ 8);   // to

  1957       Address length_arg(rsp, 16+12);   // elements count

  1958       Address  ckoff_arg(rsp, 16+16);   // super_check_offset

  1959       Address  ckval_arg(rsp, 16+20);   // super_klass

  1961       Address SRC_POS_arg(rsp, 16+ 8);

  1962       Address DST_POS_arg(rsp, 16+16);

  1963       Address  LENGTH_arg(rsp, 16+20);

  1964       // push rbx, changed the incoming offsets (why not just use rbp,??)

  1965       // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");

  1967       __ movptr(rbx, Address(rsi_dst_klass, ek_offset));

  1968       __ movl2ptr(length, LENGTH_arg);    // reload elements count

  1969       __ movl2ptr(src_pos, SRC_POS_arg);  // reload src_pos

  1970       __ movl2ptr(dst_pos, DST_POS_arg);  // reload dst_pos

  1972       __ movptr(ckval_arg, rbx);          // destination element type

  1973       __ movl(rbx, Address(rbx, sco_offset));

  1974       __ movl(ckoff_arg, rbx);          // corresponding class check offset

  1976       __ movl(length_arg, length);      // outgoing length argument

  1978       __ lea(from, Address(src, src_pos, Address::times_ptr,

  1979                             arrayOopDesc::base_offset_in_bytes(T_OBJECT)));

  1980       __ movptr(from_arg, from);

  1982       __ lea(to, Address(dst, dst_pos, Address::times_ptr,

  1983                           arrayOopDesc::base_offset_in_bytes(T_OBJECT)));

  1984       __ movptr(to_arg, to);

  1985       __ jump(RuntimeAddress(entry_checkcast_arraycopy));

  1986     }

  1988     return start;

  1989   }

  1991   void generate_arraycopy_stubs() {

  1992     address entry;

  1993     address entry_jbyte_arraycopy;

  1994     address entry_jshort_arraycopy;

  1995     address entry_jint_arraycopy;

  1996     address entry_oop_arraycopy;

  1997     address entry_jlong_arraycopy;

  1998     address entry_checkcast_arraycopy;

  2000     StubRoutines::_arrayof_jbyte_disjoint_arraycopy =

  2001         generate_disjoint_copy(T_BYTE,  true, Address::times_1, &entry,

  2002                                "arrayof_jbyte_disjoint_arraycopy");

  2003     StubRoutines::_arrayof_jbyte_arraycopy =

  2004         generate_conjoint_copy(T_BYTE,  true, Address::times_1,  entry,

  2005                                NULL, "arrayof_jbyte_arraycopy");

  2006     StubRoutines::_jbyte_disjoint_arraycopy =

  2007         generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,

  2008                                "jbyte_disjoint_arraycopy");

  2009     StubRoutines::_jbyte_arraycopy =

  2010         generate_conjoint_copy(T_BYTE, false, Address::times_1,  entry,

  2011                                &entry_jbyte_arraycopy, "jbyte_arraycopy");

  2013     StubRoutines::_arrayof_jshort_disjoint_arraycopy =

  2014         generate_disjoint_copy(T_SHORT,  true, Address::times_2, &entry,

  2015                                "arrayof_jshort_disjoint_arraycopy");

  2016     StubRoutines::_arrayof_jshort_arraycopy =

  2017         generate_conjoint_copy(T_SHORT,  true, Address::times_2,  entry,

  2018                                NULL, "arrayof_jshort_arraycopy");

  2019     StubRoutines::_jshort_disjoint_arraycopy =

  2020         generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,

  2021                                "jshort_disjoint_arraycopy");

  2022     StubRoutines::_jshort_arraycopy =

  2023         generate_conjoint_copy(T_SHORT, false, Address::times_2,  entry,

  2024                                &entry_jshort_arraycopy, "jshort_arraycopy");

  2026     // Next arrays are always aligned on 4 bytes at least.

  2027     StubRoutines::_jint_disjoint_arraycopy =

  2028         generate_disjoint_copy(T_INT, true, Address::times_4, &entry,

  2029                                "jint_disjoint_arraycopy");

  2030     StubRoutines::_jint_arraycopy =

  2031         generate_conjoint_copy(T_INT, true, Address::times_4,  entry,

  2032                                &entry_jint_arraycopy, "jint_arraycopy");

  2034     StubRoutines::_oop_disjoint_arraycopy =

  2035         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,

  2036                                "oop_disjoint_arraycopy");

  2037     StubRoutines::_oop_arraycopy =

  2038         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,

  2039                                &entry_oop_arraycopy, "oop_arraycopy");

  2041     StubRoutines::_jlong_disjoint_arraycopy =

  2042         generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");

  2043     StubRoutines::_jlong_arraycopy =

  2044         generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,

  2045                                     "jlong_arraycopy");

  2047     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");

  2048     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");

  2049     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");

  2050     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");

  2051     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");

  2052     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");

  2054     StubRoutines::_arrayof_jint_disjoint_arraycopy  =

  2055         StubRoutines::_jint_disjoint_arraycopy;

  2056     StubRoutines::_arrayof_oop_disjoint_arraycopy   =

  2057         StubRoutines::_oop_disjoint_arraycopy;

  2058     StubRoutines::_arrayof_jlong_disjoint_arraycopy =

  2059         StubRoutines::_jlong_disjoint_arraycopy;

  2061     StubRoutines::_arrayof_jint_arraycopy  = StubRoutines::_jint_arraycopy;

  2062     StubRoutines::_arrayof_oop_arraycopy   = StubRoutines::_oop_arraycopy;

  2063     StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;

  2065     StubRoutines::_checkcast_arraycopy =

  2066         generate_checkcast_copy("checkcast_arraycopy",

  2067                                   &entry_checkcast_arraycopy);

  2069     StubRoutines::_unsafe_arraycopy =

  2070         generate_unsafe_copy("unsafe_arraycopy",

  2071                                entry_jbyte_arraycopy,

  2072                                entry_jshort_arraycopy,

  2073                                entry_jint_arraycopy,

  2074                                entry_jlong_arraycopy);

  2076     StubRoutines::_generic_arraycopy =

  2077         generate_generic_copy("generic_arraycopy",

  2078                                entry_jbyte_arraycopy,

  2079                                entry_jshort_arraycopy,

  2080                                entry_jint_arraycopy,

  2081                                entry_oop_arraycopy,

  2082                                entry_jlong_arraycopy,

  2083                                entry_checkcast_arraycopy);

  2084   }

  2086   void generate_math_stubs() {

  2087     {

  2088       StubCodeMark mark(this, "StubRoutines", "log");

  2089       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();

  2091       __ fld_d(Address(rsp, 4));

  2092       __ flog();

  2093       __ ret(0);

  2094     }

  2095     {

  2096       StubCodeMark mark(this, "StubRoutines", "log10");

  2097       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();

  2099       __ fld_d(Address(rsp, 4));

  2100       __ flog10();

  2101       __ ret(0);

  2102     }

  2103     {

  2104       StubCodeMark mark(this, "StubRoutines", "sin");

  2105       StubRoutines::_intrinsic_sin = (double (*)(double))  __ pc();

  2107       __ fld_d(Address(rsp, 4));

  2108       __ trigfunc('s');

  2109       __ ret(0);

  2110     }

  2111     {

  2112       StubCodeMark mark(this, "StubRoutines", "cos");

  2113       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();

  2115       __ fld_d(Address(rsp, 4));

  2116       __ trigfunc('c');

  2117       __ ret(0);

  2118     }

  2119     {

  2120       StubCodeMark mark(this, "StubRoutines", "tan");

  2121       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();

  2123       __ fld_d(Address(rsp, 4));

  2124       __ trigfunc('t');

  2125       __ ret(0);

  2126     }

  2128     // The intrinsic version of these seem to return the same value as

  2129     // the strict version.

  2130     StubRoutines::_intrinsic_exp = SharedRuntime::dexp;

  2131     StubRoutines::_intrinsic_pow = SharedRuntime::dpow;

  2132   }

  2134  public:

  2135   // Information about frame layout at time of blocking runtime call.

  2136   // Note that we only have to preserve callee-saved registers since

  2137   // the compilers are responsible for supplying a continuation point

  2138   // if they expect all registers to be preserved.

  2139   enum layout {

  2140     thread_off,    // last_java_sp

  2141     rbp_off,       // callee saved register

  2142     ret_pc,

  2143     framesize

  2144   };

  2146  private:

  2148 #undef  __

  2149 #define __ masm->

  2151   //------------------------------------------------------------------------------------------------------------------------

  2152   // Continuation point for throwing of implicit exceptions that are not handled in

  2153   // the current activation. Fabricates an exception oop and initiates normal

  2154   // exception dispatching in this frame.

  2155   //

  2156   // Previously the compiler (c2) allowed for callee save registers on Java calls.

  2157   // This is no longer true after adapter frames were removed but could possibly

  2158   // be brought back in the future if the interpreter code was reworked and it

  2159   // was deemed worthwhile. The comment below was left to describe what must

  2160   // happen here if callee saves were resurrected. As it stands now this stub

  2161   // could actually be a vanilla BufferBlob and have now oopMap at all.

  2162   // Since it doesn't make much difference we've chosen to leave it the

  2163   // way it was in the callee save days and keep the comment.

  2165   // If we need to preserve callee-saved values we need a callee-saved oop map and

  2166   // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.

  2167   // If the compiler needs all registers to be preserved between the fault

  2168   // point and the exception handler then it must assume responsibility for that in

  2169   // AbstractCompiler::continuation_for_implicit_null_exception or

  2170   // continuation_for_implicit_division_by_zero_exception. All other implicit

  2171   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are

  2172   // either at call sites or otherwise assume that stack unwinding will be initiated,

  2173   // so caller saved registers were assumed volatile in the compiler.

  2174   address generate_throw_exception(const char* name, address runtime_entry,

  2175                                    bool restore_saved_exception_pc) {

  2177     int insts_size = 256;

  2178     int locs_size  = 32;

  2180     CodeBuffer code(name, insts_size, locs_size);

  2181     OopMapSet* oop_maps  = new OopMapSet();

  2182     MacroAssembler* masm = new MacroAssembler(&code);

  2184     address start = __ pc();

  2186     // This is an inlined and slightly modified version of call_VM

  2187     // which has the ability to fetch the return PC out of

  2188     // thread-local storage and also sets up last_Java_sp slightly

  2189     // differently than the real call_VM

  2190     Register java_thread = rbx;

  2191     __ get_thread(java_thread);

  2192     if (restore_saved_exception_pc) {

  2193       __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));

  2194       __ push(rax);

  2195     }

  2197     __ enter(); // required for proper stackwalking of RuntimeStub frame

  2199     // pc and rbp, already pushed

  2200     __ subptr(rsp, (framesize-2) * wordSize); // prolog

  2202     // Frame is now completed as far as size and linkage.

  2204     int frame_complete = __ pc() - start;

  2206     // push java thread (becomes first argument of C function)

  2207     __ movptr(Address(rsp, thread_off * wordSize), java_thread);

  2209     // Set up last_Java_sp and last_Java_fp

  2210     __ set_last_Java_frame(java_thread, rsp, rbp, NULL);

  2212     // Call runtime

  2213     BLOCK_COMMENT("call runtime_entry");

  2214     __ call(RuntimeAddress(runtime_entry));

  2215     // Generate oop map

  2216     OopMap* map =  new OopMap(framesize, 0);

  2217     oop_maps->add_gc_map(__ pc() - start, map);

  2219     // restore the thread (cannot use the pushed argument since arguments

  2220     // may be overwritten by C code generated by an optimizing compiler);

  2221     // however can use the register value directly if it is callee saved.

  2222     __ get_thread(java_thread);

  2224     __ reset_last_Java_frame(java_thread, true, false);

  2226     __ leave(); // required for proper stackwalking of RuntimeStub frame

  2228     // check for pending exceptions

  2229 #ifdef ASSERT

  2230     Label L;

  2231     __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);

  2232     __ jcc(Assembler::notEqual, L);

  2233     __ should_not_reach_here();

  2234     __ bind(L);

  2235 #endif /* ASSERT */

  2236     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));

  2239     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);

  2240     return stub->entry_point();

  2241   }

  2244   void create_control_words() {

  2245     // Round to nearest, 53-bit mode, exceptions masked

  2246     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;

  2247     // Round to zero, 53-bit mode, exception mased

  2248     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;

  2249     // Round to nearest, 24-bit mode, exceptions masked

  2250     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;

  2251     // Round to nearest, 64-bit mode, exceptions masked

  2252     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;

  2253     // Round to nearest, 64-bit mode, exceptions masked

  2254     StubRoutines::_mxcsr_std           = 0x1F80;

  2255     // Note: the following two constants are 80-bit values

  2256     //       layout is critical for correct loading by FPU.

  2257     // Bias for strict fp multiply/divide

  2258     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000

  2259     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;

  2260     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;

  2261     // Un-Bias for strict fp multiply/divide

  2262     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000

  2263     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;

  2264     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;

  2265   }

  2267   //---------------------------------------------------------------------------

  2268   // Initialization

  2270   void generate_initial() {

  2271     // Generates all stubs and initializes the entry points

  2273     //------------------------------------------------------------------------------------------------------------------------

  2274     // entry points that exist in all platforms

  2275     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than

  2276     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.

  2277     StubRoutines::_forward_exception_entry      = generate_forward_exception();

  2279     StubRoutines::_call_stub_entry              =

  2280       generate_call_stub(StubRoutines::_call_stub_return_address);

  2281     // is referenced by megamorphic call

  2282     StubRoutines::_catch_exception_entry        = generate_catch_exception();

  2284     // These are currently used by Solaris/Intel

  2285     StubRoutines::_atomic_xchg_entry            = generate_atomic_xchg();

  2287     StubRoutines::_handler_for_unsafe_access_entry =

  2288       generate_handler_for_unsafe_access();

  2290     // platform dependent

  2291     create_control_words();

  2293     StubRoutines::x86::_verify_mxcsr_entry                 = generate_verify_mxcsr();

  2294     StubRoutines::x86::_verify_fpu_cntrl_wrd_entry         = generate_verify_fpu_cntrl_wrd();

  2295     StubRoutines::_d2i_wrapper                              = generate_d2i_wrapper(T_INT,

  2296                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2i));

  2297     StubRoutines::_d2l_wrapper                              = generate_d2i_wrapper(T_LONG,

  2298                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2l));

  2299   }

  2302   void generate_all() {

  2303     // Generates all stubs and initializes the entry points

  2305     // These entry points require SharedInfo::stack0 to be set up in non-core builds

  2306     // and need to be relocatable, so they each fabricate a RuntimeStub internally.

  2307     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError),  false);

  2308     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError),  false);

  2309     StubRoutines::_throw_ArithmeticException_entry         = generate_throw_exception("ArithmeticException throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException),  true);

  2310     StubRoutines::_throw_NullPointerException_entry        = generate_throw_exception("NullPointerException throw_exception",         CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);

  2311     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);

  2312     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",           CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError),   false);

  2314     //------------------------------------------------------------------------------------------------------------------------

  2315     // entry points that are platform specific

  2317     // support for verify_oop (must happen after universe_init)

  2318     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop();

  2320     // arraycopy stubs used by compilers

  2321     generate_arraycopy_stubs();

  2323     generate_math_stubs();

  2324   }

  2327  public:

  2328   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {

  2329     if (all) {

  2330       generate_all();

  2331     } else {

  2332       generate_initial();

  2333     }

  2334   }

  2335 }; // end class declaration

  2338 void StubGenerator_generate(CodeBuffer* code, bool all) {

  2339   StubGenerator g(code, all);

  2340 }

src/cpu/x86/vm/stubGenerator_x86_32.cpp@6bbaedb03534

src/cpu/x86/vm/stubGenerator_x86_32.cpp

Mercurial > jdk8-mips64-public > hotspot / file revision

src/cpu/x86/vm/stubGenerator_x86_32.cpp@6bbaedb03534

src/cpu/x86/vm/stubGenerator_x86_32.cpp