1 /*
   2  * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/assembler.hpp"
  27 #include "assembler_x86.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "nativeInst_x86.hpp"
  30 #include "oops/instanceOop.hpp"
  31 #include "oops/methodOop.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "oops/oop.inline.hpp"
  34 #include "prims/methodHandles.hpp"
  35 #include "runtime/frame.inline.hpp"
  36 #include "runtime/handles.inline.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/stubCodeGenerator.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "utilities/top.hpp"
  41 #ifdef TARGET_OS_FAMILY_linux
  42 # include "thread_linux.inline.hpp"
  43 #endif
  44 #ifdef TARGET_OS_FAMILY_solaris
  45 # include "thread_solaris.inline.hpp"
  46 #endif
  47 #ifdef TARGET_OS_FAMILY_windows
  48 # include "thread_windows.inline.hpp"
  49 #endif
  50 #ifdef COMPILER2
  51 #include "opto/runtime.hpp"
  52 #endif
  53 
  54 // Declaration and definition of StubGenerator (no .hpp file).
  55 // For a more detailed description of the stub routine structure
  56 // see the comment in stubRoutines.hpp
  57 
  58 #define __ _masm->
  59 #define a__ ((Assembler*)_masm)->
  60 
  61 #ifdef PRODUCT
  62 #define BLOCK_COMMENT(str) /* nothing */
  63 #else
  64 #define BLOCK_COMMENT(str) __ block_comment(str)
  65 #endif
  66 
  67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  68 
  69 const int MXCSR_MASK  = 0xFFC0;  // Mask out any pending exceptions
  70 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
  71 
  72 // -------------------------------------------------------------------------------------------------------------------------
  73 // Stub Code definitions
  74 
  75 static address handle_unsafe_access() {
  76   JavaThread* thread = JavaThread::current();
  77   address pc  = thread->saved_exception_pc();
  78   // pc is the instruction which we must emulate
  79   // doing a no-op is fine:  return garbage from the load
  80   // therefore, compute npc
  81   address npc = Assembler::locate_next_instruction(pc);
  82 
  83   // request an async exception
  84   thread->set_pending_unsafe_access_error();
  85 
  86   // return address of next instruction to execute
  87   return npc;
  88 }
  89 
  90 class StubGenerator: public StubCodeGenerator {
  91  private:
  92 
  93 #ifdef PRODUCT
  94 #define inc_counter_np(counter) (0)
  95 #else
  96   void inc_counter_np_(int& counter) {
  97     __ incrementl(ExternalAddress((address)&counter));
  98   }
  99 #define inc_counter_np(counter) \
 100   BLOCK_COMMENT("inc_counter " #counter); \
 101   inc_counter_np_(counter);
 102 #endif //PRODUCT
 103 
 104   void inc_copy_counter_np(BasicType t) {
 105 #ifndef PRODUCT
 106     switch (t) {
 107     case T_BYTE:    inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
 108     case T_SHORT:   inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
 109     case T_INT:     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
 110     case T_LONG:    inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
 111     case T_OBJECT:  inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
 112     }
 113     ShouldNotReachHere();
 114 #endif //PRODUCT
 115   }
 116 
 117   //------------------------------------------------------------------------------------------------------------------------
 118   // Call stubs are used to call Java from C
 119   //
 120   //    [ return_from_Java     ] <--- rsp
 121   //    [ argument word n      ]
 122   //      ...
 123   // -N [ argument word 1      ]
 124   // -7 [ Possible padding for stack alignment ]
 125   // -6 [ Possible padding for stack alignment ]
 126   // -5 [ Possible padding for stack alignment ]
 127   // -4 [ mxcsr save           ] <--- rsp_after_call
 128   // -3 [ saved rbx,            ]
 129   // -2 [ saved rsi            ]
 130   // -1 [ saved rdi            ]
 131   //  0 [ saved rbp,            ] <--- rbp,
 132   //  1 [ return address       ]
 133   //  2 [ ptr. to call wrapper ]
 134   //  3 [ result               ]
 135   //  4 [ result_type          ]
 136   //  5 [ method               ]
 137   //  6 [ entry_point          ]
 138   //  7 [ parameters           ]
 139   //  8 [ parameter_size       ]
 140   //  9 [ thread               ]
 141 
 142 
 143   address generate_call_stub(address& return_address) {
 144     StubCodeMark mark(this, "StubRoutines", "call_stub");
 145     address start = __ pc();
 146 
 147     // stub code parameters / addresses
 148     assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
 149     bool  sse_save = false;
 150     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
 151     const int     locals_count_in_bytes  (4*wordSize);
 152     const Address mxcsr_save    (rbp, -4 * wordSize);
 153     const Address saved_rbx     (rbp, -3 * wordSize);
 154     const Address saved_rsi     (rbp, -2 * wordSize);
 155     const Address saved_rdi     (rbp, -1 * wordSize);
 156     const Address result        (rbp,  3 * wordSize);
 157     const Address result_type   (rbp,  4 * wordSize);
 158     const Address method        (rbp,  5 * wordSize);
 159     const Address entry_point   (rbp,  6 * wordSize);
 160     const Address parameters    (rbp,  7 * wordSize);
 161     const Address parameter_size(rbp,  8 * wordSize);
 162     const Address thread        (rbp,  9 * wordSize); // same as in generate_catch_exception()!
 163     sse_save =  UseSSE > 0;
 164 
 165     // stub code
 166     __ enter();
 167     __ movptr(rcx, parameter_size);              // parameter counter
 168     __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
 169     __ addptr(rcx, locals_count_in_bytes);       // reserve space for register saves
 170     __ subptr(rsp, rcx);
 171     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
 172 
 173     // save rdi, rsi, & rbx, according to C calling conventions
 174     __ movptr(saved_rdi, rdi);
 175     __ movptr(saved_rsi, rsi);
 176     __ movptr(saved_rbx, rbx);
 177     // save and initialize %mxcsr
 178     if (sse_save) {
 179       Label skip_ldmx;
 180       __ stmxcsr(mxcsr_save);
 181       __ movl(rax, mxcsr_save);
 182       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 183       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 184       __ cmp32(rax, mxcsr_std);
 185       __ jcc(Assembler::equal, skip_ldmx);
 186       __ ldmxcsr(mxcsr_std);
 187       __ bind(skip_ldmx);
 188     }
 189 
 190     // make sure the control word is correct.
 191     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 192 
 193 #ifdef ASSERT
 194     // make sure we have no pending exceptions
 195     { Label L;
 196       __ movptr(rcx, thread);
 197       __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 198       __ jcc(Assembler::equal, L);
 199       __ stop("StubRoutines::call_stub: entered with pending exception");
 200       __ bind(L);
 201     }
 202 #endif
 203 
 204     // pass parameters if any
 205     BLOCK_COMMENT("pass parameters if any");
 206     Label parameters_done;
 207     __ movl(rcx, parameter_size);  // parameter counter
 208     __ testl(rcx, rcx);
 209     __ jcc(Assembler::zero, parameters_done);
 210 
 211     // parameter passing loop
 212 
 213     Label loop;
 214     // Copy Java parameters in reverse order (receiver last)
 215     // Note that the argument order is inverted in the process
 216     // source is rdx[rcx: N-1..0]
 217     // dest   is rsp[rbx: 0..N-1]
 218 
 219     __ movptr(rdx, parameters);          // parameter pointer
 220     __ xorptr(rbx, rbx);
 221 
 222     __ BIND(loop);
 223 
 224     // get parameter
 225     __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
 226     __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
 227                     Interpreter::expr_offset_in_bytes(0)), rax);          // store parameter
 228     __ increment(rbx);
 229     __ decrement(rcx);
 230     __ jcc(Assembler::notZero, loop);
 231 
 232     // call Java function
 233     __ BIND(parameters_done);
 234     __ movptr(rbx, method);           // get methodOop
 235     __ movptr(rax, entry_point);      // get entry_point
 236     __ mov(rsi, rsp);                 // set sender sp
 237     BLOCK_COMMENT("call Java function");
 238     __ call(rax);
 239 
 240     BLOCK_COMMENT("call_stub_return_address:");
 241     return_address = __ pc();
 242 
 243 #ifdef COMPILER2
 244     {
 245       Label L_skip;
 246       if (UseSSE >= 2) {
 247         __ verify_FPU(0, "call_stub_return");
 248       } else {
 249         for (int i = 1; i < 8; i++) {
 250           __ ffree(i);
 251         }
 252 
 253         // UseSSE <= 1 so double result should be left on TOS
 254         __ movl(rsi, result_type);
 255         __ cmpl(rsi, T_DOUBLE);
 256         __ jcc(Assembler::equal, L_skip);
 257         if (UseSSE == 0) {
 258           // UseSSE == 0 so float result should be left on TOS
 259           __ cmpl(rsi, T_FLOAT);
 260           __ jcc(Assembler::equal, L_skip);
 261         }
 262         __ ffree(0);
 263       }
 264       __ BIND(L_skip);
 265     }
 266 #endif // COMPILER2
 267 
 268     // store result depending on type
 269     // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 270     __ movptr(rdi, result);
 271     Label is_long, is_float, is_double, exit;
 272     __ movl(rsi, result_type);
 273     __ cmpl(rsi, T_LONG);
 274     __ jcc(Assembler::equal, is_long);
 275     __ cmpl(rsi, T_FLOAT);
 276     __ jcc(Assembler::equal, is_float);
 277     __ cmpl(rsi, T_DOUBLE);
 278     __ jcc(Assembler::equal, is_double);
 279 
 280     // handle T_INT case
 281     __ movl(Address(rdi, 0), rax);
 282     __ BIND(exit);
 283 
 284     // check that FPU stack is empty
 285     __ verify_FPU(0, "generate_call_stub");
 286 
 287     // pop parameters
 288     __ lea(rsp, rsp_after_call);
 289 
 290     // restore %mxcsr
 291     if (sse_save) {
 292       __ ldmxcsr(mxcsr_save);
 293     }
 294 
 295     // restore rdi, rsi and rbx,
 296     __ movptr(rbx, saved_rbx);
 297     __ movptr(rsi, saved_rsi);
 298     __ movptr(rdi, saved_rdi);
 299     __ addptr(rsp, 4*wordSize);
 300 
 301     // return
 302     __ pop(rbp);
 303     __ ret(0);
 304 
 305     // handle return types different from T_INT
 306     __ BIND(is_long);
 307     __ movl(Address(rdi, 0 * wordSize), rax);
 308     __ movl(Address(rdi, 1 * wordSize), rdx);
 309     __ jmp(exit);
 310 
 311     __ BIND(is_float);
 312     // interpreter uses xmm0 for return values
 313     if (UseSSE >= 1) {
 314       __ movflt(Address(rdi, 0), xmm0);
 315     } else {
 316       __ fstp_s(Address(rdi, 0));
 317     }
 318     __ jmp(exit);
 319 
 320     __ BIND(is_double);
 321     // interpreter uses xmm0 for return values
 322     if (UseSSE >= 2) {
 323       __ movdbl(Address(rdi, 0), xmm0);
 324     } else {
 325       __ fstp_d(Address(rdi, 0));
 326     }
 327     __ jmp(exit);
 328 
 329     return start;
 330   }
 331 
 332 
 333   //------------------------------------------------------------------------------------------------------------------------
 334   // Return point for a Java call if there's an exception thrown in Java code.
 335   // The exception is caught and transformed into a pending exception stored in
 336   // JavaThread that can be tested from within the VM.
 337   //
 338   // Note: Usually the parameters are removed by the callee. In case of an exception
 339   //       crossing an activation frame boundary, that is not the case if the callee
 340   //       is compiled code => need to setup the rsp.
 341   //
 342   // rax,: exception oop
 343 
 344   address generate_catch_exception() {
 345     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 346     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
 347     const Address thread        (rbp,  9 * wordSize); // same as in generate_call_stub()!
 348     address start = __ pc();
 349 
 350     // get thread directly
 351     __ movptr(rcx, thread);
 352 #ifdef ASSERT
 353     // verify that threads correspond
 354     { Label L;
 355       __ get_thread(rbx);
 356       __ cmpptr(rbx, rcx);
 357       __ jcc(Assembler::equal, L);
 358       __ stop("StubRoutines::catch_exception: threads must correspond");
 359       __ bind(L);
 360     }
 361 #endif
 362     // set pending exception
 363     __ verify_oop(rax);
 364     __ movptr(Address(rcx, Thread::pending_exception_offset()), rax          );
 365     __ lea(Address(rcx, Thread::exception_file_offset   ()),
 366            ExternalAddress((address)__FILE__));
 367     __ movl(Address(rcx, Thread::exception_line_offset   ()), __LINE__ );
 368     // complete return to VM
 369     assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
 370     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 371 
 372     return start;
 373   }
 374 
 375 
 376   //------------------------------------------------------------------------------------------------------------------------
 377   // Continuation point for runtime calls returning with a pending exception.
 378   // The pending exception check happened in the runtime or native call stub.
 379   // The pending exception in Thread is converted into a Java-level exception.
 380   //
 381   // Contract with Java-level exception handlers:
 382   // rax: exception
 383   // rdx: throwing pc
 384   //
 385   // NOTE: At entry of this stub, exception-pc must be on stack !!
 386 
 387   address generate_forward_exception() {
 388     StubCodeMark mark(this, "StubRoutines", "forward exception");
 389     address start = __ pc();
 390     const Register thread = rcx;
 391 
 392     // other registers used in this stub
 393     const Register exception_oop = rax;
 394     const Register handler_addr  = rbx;
 395     const Register exception_pc  = rdx;
 396 
 397     // Upon entry, the sp points to the return address returning into Java
 398     // (interpreted or compiled) code; i.e., the return address becomes the
 399     // throwing pc.
 400     //
 401     // Arguments pushed before the runtime call are still on the stack but
 402     // the exception handler will reset the stack pointer -> ignore them.
 403     // A potential result in registers can be ignored as well.
 404 
 405 #ifdef ASSERT
 406     // make sure this code is only executed if there is a pending exception
 407     { Label L;
 408       __ get_thread(thread);
 409       __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 410       __ jcc(Assembler::notEqual, L);
 411       __ stop("StubRoutines::forward exception: no pending exception (1)");
 412       __ bind(L);
 413     }
 414 #endif
 415 
 416     // compute exception handler into rbx,
 417     __ get_thread(thread);
 418     __ movptr(exception_pc, Address(rsp, 0));
 419     BLOCK_COMMENT("call exception_handler_for_return_address");
 420     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
 421     __ mov(handler_addr, rax);
 422 
 423     // setup rax & rdx, remove return address & clear pending exception
 424     __ get_thread(thread);
 425     __ pop(exception_pc);
 426     __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
 427     __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 428 
 429 #ifdef ASSERT
 430     // make sure exception is set
 431     { Label L;
 432       __ testptr(exception_oop, exception_oop);
 433       __ jcc(Assembler::notEqual, L);
 434       __ stop("StubRoutines::forward exception: no pending exception (2)");
 435       __ bind(L);
 436     }
 437 #endif
 438 
 439     // Verify that there is really a valid exception in RAX.
 440     __ verify_oop(exception_oop);
 441 
 442     // Restore SP from BP if the exception PC is a MethodHandle call site.
 443     __ cmpl(Address(thread, JavaThread::is_method_handle_return_offset()), 0);
 444     __ cmovptr(Assembler::notEqual, rsp, rbp);
 445 
 446     // continue at exception handler (return address removed)
 447     // rax: exception
 448     // rbx: exception handler
 449     // rdx: throwing pc
 450     __ jmp(handler_addr);
 451 
 452     return start;
 453   }
 454 
 455 
 456   //----------------------------------------------------------------------------------------------------
 457   // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
 458   //
 459   // xchg exists as far back as 8086, lock needed for MP only
 460   // Stack layout immediately after call:
 461   //
 462   // 0 [ret addr ] <--- rsp
 463   // 1 [  ex     ]
 464   // 2 [  dest   ]
 465   //
 466   // Result:   *dest <- ex, return (old *dest)
 467   //
 468   // Note: win32 does not currently use this code
 469 
 470   address generate_atomic_xchg() {
 471     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 472     address start = __ pc();
 473 
 474     __ push(rdx);
 475     Address exchange(rsp, 2 * wordSize);
 476     Address dest_addr(rsp, 3 * wordSize);
 477     __ movl(rax, exchange);
 478     __ movptr(rdx, dest_addr);
 479     __ xchgl(rax, Address(rdx, 0));
 480     __ pop(rdx);
 481     __ ret(0);
 482 
 483     return start;
 484   }
 485 
 486   //----------------------------------------------------------------------------------------------------
 487   // Support for void verify_mxcsr()
 488   //
 489   // This routine is used with -Xcheck:jni to verify that native
 490   // JNI code does not return to Java code without restoring the
 491   // MXCSR register to our expected state.
 492 
 493 
 494   address generate_verify_mxcsr() {
 495     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 496     address start = __ pc();
 497 
 498     const Address mxcsr_save(rsp, 0);
 499 
 500     if (CheckJNICalls && UseSSE > 0 ) {
 501       Label ok_ret;
 502       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 503       __ push(rax);
 504       __ subptr(rsp, wordSize);      // allocate a temp location
 505       __ stmxcsr(mxcsr_save);
 506       __ movl(rax, mxcsr_save);
 507       __ andl(rax, MXCSR_MASK);
 508       __ cmp32(rax, mxcsr_std);
 509       __ jcc(Assembler::equal, ok_ret);
 510 
 511       __ warn("MXCSR changed by native JNI code.");
 512 
 513       __ ldmxcsr(mxcsr_std);
 514 
 515       __ bind(ok_ret);
 516       __ addptr(rsp, wordSize);
 517       __ pop(rax);
 518     }
 519 
 520     __ ret(0);
 521 
 522     return start;
 523   }
 524 
 525 
 526   //---------------------------------------------------------------------------
 527   // Support for void verify_fpu_cntrl_wrd()
 528   //
 529   // This routine is used with -Xcheck:jni to verify that native
 530   // JNI code does not return to Java code without restoring the
 531   // FP control word to our expected state.
 532 
 533   address generate_verify_fpu_cntrl_wrd() {
 534     StubCodeMark mark(this, "StubRoutines", "verify_spcw");
 535     address start = __ pc();
 536 
 537     const Address fpu_cntrl_wrd_save(rsp, 0);
 538 
 539     if (CheckJNICalls) {
 540       Label ok_ret;
 541       __ push(rax);
 542       __ subptr(rsp, wordSize);      // allocate a temp location
 543       __ fnstcw(fpu_cntrl_wrd_save);
 544       __ movl(rax, fpu_cntrl_wrd_save);
 545       __ andl(rax, FPU_CNTRL_WRD_MASK);
 546       ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
 547       __ cmp32(rax, fpu_std);
 548       __ jcc(Assembler::equal, ok_ret);
 549 
 550       __ warn("Floating point control word changed by native JNI code.");
 551 
 552       __ fldcw(fpu_std);
 553 
 554       __ bind(ok_ret);
 555       __ addptr(rsp, wordSize);
 556       __ pop(rax);
 557     }
 558 
 559     __ ret(0);
 560 
 561     return start;
 562   }
 563 
 564   //---------------------------------------------------------------------------
 565   // Wrapper for slow-case handling of double-to-integer conversion
 566   // d2i or f2i fast case failed either because it is nan or because
 567   // of under/overflow.
 568   // Input:  FPU TOS: float value
 569   // Output: rax, (rdx): integer (long) result
 570 
 571   address generate_d2i_wrapper(BasicType t, address fcn) {
 572     StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
 573     address start = __ pc();
 574 
 575   // Capture info about frame layout
 576   enum layout { FPUState_off         = 0,
 577                 rbp_off              = FPUStateSizeInWords,
 578                 rdi_off,
 579                 rsi_off,
 580                 rcx_off,
 581                 rbx_off,
 582                 saved_argument_off,
 583                 saved_argument_off2, // 2nd half of double
 584                 framesize
 585   };
 586 
 587   assert(FPUStateSizeInWords == 27, "update stack layout");
 588 
 589     // Save outgoing argument to stack across push_FPU_state()
 590     __ subptr(rsp, wordSize * 2);
 591     __ fstp_d(Address(rsp, 0));
 592 
 593     // Save CPU & FPU state
 594     __ push(rbx);
 595     __ push(rcx);
 596     __ push(rsi);
 597     __ push(rdi);
 598     __ push(rbp);
 599     __ push_FPU_state();
 600 
 601     // push_FPU_state() resets the FP top of stack
 602     // Load original double into FP top of stack
 603     __ fld_d(Address(rsp, saved_argument_off * wordSize));
 604     // Store double into stack as outgoing argument
 605     __ subptr(rsp, wordSize*2);
 606     __ fst_d(Address(rsp, 0));
 607 
 608     // Prepare FPU for doing math in C-land
 609     __ empty_FPU_stack();
 610     // Call the C code to massage the double.  Result in EAX
 611     if (t == T_INT)
 612       { BLOCK_COMMENT("SharedRuntime::d2i"); }
 613     else if (t == T_LONG)
 614       { BLOCK_COMMENT("SharedRuntime::d2l"); }
 615     __ call_VM_leaf( fcn, 2 );
 616 
 617     // Restore CPU & FPU state
 618     __ pop_FPU_state();
 619     __ pop(rbp);
 620     __ pop(rdi);
 621     __ pop(rsi);
 622     __ pop(rcx);
 623     __ pop(rbx);
 624     __ addptr(rsp, wordSize * 2);
 625 
 626     __ ret(0);
 627 
 628     return start;
 629   }
 630 
 631 
 632   //---------------------------------------------------------------------------
 633   // The following routine generates a subroutine to throw an asynchronous
 634   // UnknownError when an unsafe access gets a fault that could not be
 635   // reasonably prevented by the programmer.  (Example: SIGBUS/OBJERR.)
 636   address generate_handler_for_unsafe_access() {
 637     StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
 638     address start = __ pc();
 639 
 640     __ push(0);                       // hole for return address-to-be
 641     __ pusha();                       // push registers
 642     Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
 643     BLOCK_COMMENT("call handle_unsafe_access");
 644     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
 645     __ movptr(next_pc, rax);          // stuff next address
 646     __ popa();
 647     __ ret(0);                        // jump to next address
 648 
 649     return start;
 650   }
 651 
 652 
 653   //----------------------------------------------------------------------------------------------------
 654   // Non-destructive plausibility checks for oops
 655 
 656   address generate_verify_oop() {
 657     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 658     address start = __ pc();
 659 
 660     // Incoming arguments on stack after saving rax,:
 661     //
 662     // [tos    ]: saved rdx
 663     // [tos + 1]: saved EFLAGS
 664     // [tos + 2]: return address
 665     // [tos + 3]: char* error message
 666     // [tos + 4]: oop   object to verify
 667     // [tos + 5]: saved rax, - saved by caller and bashed
 668 
 669     Label exit, error;
 670     __ pushf();
 671     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 672     __ push(rdx);                                // save rdx
 673     // make sure object is 'reasonable'
 674     __ movptr(rax, Address(rsp, 4 * wordSize));    // get object
 675     __ testptr(rax, rax);
 676     __ jcc(Assembler::zero, exit);               // if obj is NULL it is ok
 677 
 678     // Check if the oop is in the right area of memory
 679     const int oop_mask = Universe::verify_oop_mask();
 680     const int oop_bits = Universe::verify_oop_bits();
 681     __ mov(rdx, rax);
 682     __ andptr(rdx, oop_mask);
 683     __ cmpptr(rdx, oop_bits);
 684     __ jcc(Assembler::notZero, error);
 685 
 686     // make sure klass is 'reasonable'
 687     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
 688     __ testptr(rax, rax);
 689     __ jcc(Assembler::zero, error);              // if klass is NULL it is broken
 690 
 691     // Check if the klass is in the right area of memory
 692     const int klass_mask = Universe::verify_klass_mask();
 693     const int klass_bits = Universe::verify_klass_bits();
 694     __ mov(rdx, rax);
 695     __ andptr(rdx, klass_mask);
 696     __ cmpptr(rdx, klass_bits);
 697     __ jcc(Assembler::notZero, error);
 698 
 699     // make sure klass' klass is 'reasonable'
 700     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
 701     __ testptr(rax, rax);
 702     __ jcc(Assembler::zero, error);              // if klass' klass is NULL it is broken
 703 
 704     __ mov(rdx, rax);
 705     __ andptr(rdx, klass_mask);
 706     __ cmpptr(rdx, klass_bits);
 707     __ jcc(Assembler::notZero, error);           // if klass not in right area
 708                                                  // of memory it is broken too.
 709 
 710     // return if everything seems ok
 711     __ bind(exit);
 712     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 713     __ pop(rdx);                                 // restore rdx
 714     __ popf();                                   // restore EFLAGS
 715     __ ret(3 * wordSize);                        // pop arguments
 716 
 717     // handle errors
 718     __ bind(error);
 719     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 720     __ pop(rdx);                                 // get saved rdx back
 721     __ popf();                                   // get saved EFLAGS off stack -- will be ignored
 722     __ pusha();                                  // push registers (eip = return address & msg are already pushed)
 723     BLOCK_COMMENT("call MacroAssembler::debug");
 724     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
 725     __ popa();
 726     __ ret(3 * wordSize);                        // pop arguments
 727     return start;
 728   }
 729 
 730   //
 731   //  Generate pre-barrier for array stores
 732   //
 733   //  Input:
 734   //     start   -  starting address
 735   //     count   -  element count
 736   void  gen_write_ref_array_pre_barrier(Register start, Register count) {
 737     assert_different_registers(start, count);
 738     BarrierSet* bs = Universe::heap()->barrier_set();
 739     switch (bs->kind()) {
 740       case BarrierSet::G1SATBCT:
 741       case BarrierSet::G1SATBCTLogging:
 742         {
 743           __ pusha();                      // push registers
 744           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
 745                           start, count);
 746           __ popa();
 747         }
 748         break;
 749       case BarrierSet::CardTableModRef:
 750       case BarrierSet::CardTableExtension:
 751       case BarrierSet::ModRef:
 752         break;
 753       default      :
 754         ShouldNotReachHere();
 755 
 756     }
 757   }
 758 
 759 
 760   //
 761   // Generate a post-barrier for an array store
 762   //
 763   //     start    -  starting address
 764   //     count    -  element count
 765   //
 766   //  The two input registers are overwritten.
 767   //
 768   void  gen_write_ref_array_post_barrier(Register start, Register count) {
 769     BarrierSet* bs = Universe::heap()->barrier_set();
 770     assert_different_registers(start, count);
 771     switch (bs->kind()) {
 772       case BarrierSet::G1SATBCT:
 773       case BarrierSet::G1SATBCTLogging:
 774         {
 775           __ pusha();                      // push registers
 776           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
 777                           start, count);
 778           __ popa();
 779         }
 780         break;
 781 
 782       case BarrierSet::CardTableModRef:
 783       case BarrierSet::CardTableExtension:
 784         {
 785           CardTableModRefBS* ct = (CardTableModRefBS*)bs;
 786           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
 787 
 788           Label L_loop;
 789           const Register end = count;  // elements count; end == start+count-1
 790           assert_different_registers(start, end);
 791 
 792           __ lea(end,  Address(start, count, Address::times_ptr, -wordSize));
 793           __ shrptr(start, CardTableModRefBS::card_shift);
 794           __ shrptr(end,   CardTableModRefBS::card_shift);
 795           __ subptr(end, start); // end --> count
 796         __ BIND(L_loop);
 797           intptr_t disp = (intptr_t) ct->byte_map_base;
 798           Address cardtable(start, count, Address::times_1, disp);
 799           __ movb(cardtable, 0);
 800           __ decrement(count);
 801           __ jcc(Assembler::greaterEqual, L_loop);
 802         }
 803         break;
 804       case BarrierSet::ModRef:
 805         break;
 806       default      :
 807         ShouldNotReachHere();
 808 
 809     }
 810   }
 811 
 812 
 813   // Copy 64 bytes chunks
 814   //
 815   // Inputs:
 816   //   from        - source array address
 817   //   to_from     - destination array address - from
 818   //   qword_count - 8-bytes element count, negative
 819   //
 820   void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
 821     assert( UseSSE >= 2, "supported cpu only" );
 822     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 823     // Copy 64-byte chunks
 824     __ jmpb(L_copy_64_bytes);
 825     __ align(OptoLoopAlignment);
 826   __ BIND(L_copy_64_bytes_loop);
 827 
 828     if(UseUnalignedLoadStores) {
 829       __ movdqu(xmm0, Address(from, 0));
 830       __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
 831       __ movdqu(xmm1, Address(from, 16));
 832       __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
 833       __ movdqu(xmm2, Address(from, 32));
 834       __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
 835       __ movdqu(xmm3, Address(from, 48));
 836       __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
 837 
 838     } else {
 839       __ movq(xmm0, Address(from, 0));
 840       __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
 841       __ movq(xmm1, Address(from, 8));
 842       __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
 843       __ movq(xmm2, Address(from, 16));
 844       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 845       __ movq(xmm3, Address(from, 24));
 846       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 847       __ movq(xmm4, Address(from, 32));
 848       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 849       __ movq(xmm5, Address(from, 40));
 850       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 851       __ movq(xmm6, Address(from, 48));
 852       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 853       __ movq(xmm7, Address(from, 56));
 854       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 855     }
 856 
 857     __ addl(from, 64);
 858   __ BIND(L_copy_64_bytes);
 859     __ subl(qword_count, 8);
 860     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 861     __ addl(qword_count, 8);
 862     __ jccb(Assembler::zero, L_exit);
 863     //
 864     // length is too short, just copy qwords
 865     //
 866   __ BIND(L_copy_8_bytes);
 867     __ movq(xmm0, Address(from, 0));
 868     __ movq(Address(from, to_from, Address::times_1), xmm0);
 869     __ addl(from, 8);
 870     __ decrement(qword_count);
 871     __ jcc(Assembler::greater, L_copy_8_bytes);
 872   __ BIND(L_exit);
 873   }
 874 
 875   // Copy 64 bytes chunks
 876   //
 877   // Inputs:
 878   //   from        - source array address
 879   //   to_from     - destination array address - from
 880   //   qword_count - 8-bytes element count, negative
 881   //
 882   void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
 883     assert( VM_Version::supports_mmx(), "supported cpu only" );
 884     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 885     // Copy 64-byte chunks
 886     __ jmpb(L_copy_64_bytes);
 887     __ align(OptoLoopAlignment);
 888   __ BIND(L_copy_64_bytes_loop);
 889     __ movq(mmx0, Address(from, 0));
 890     __ movq(mmx1, Address(from, 8));
 891     __ movq(mmx2, Address(from, 16));
 892     __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
 893     __ movq(mmx3, Address(from, 24));
 894     __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
 895     __ movq(mmx4, Address(from, 32));
 896     __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
 897     __ movq(mmx5, Address(from, 40));
 898     __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
 899     __ movq(mmx6, Address(from, 48));
 900     __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
 901     __ movq(mmx7, Address(from, 56));
 902     __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
 903     __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
 904     __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
 905     __ addptr(from, 64);
 906   __ BIND(L_copy_64_bytes);
 907     __ subl(qword_count, 8);
 908     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 909     __ addl(qword_count, 8);
 910     __ jccb(Assembler::zero, L_exit);
 911     //
 912     // length is too short, just copy qwords
 913     //
 914   __ BIND(L_copy_8_bytes);
 915     __ movq(mmx0, Address(from, 0));
 916     __ movq(Address(from, to_from, Address::times_1), mmx0);
 917     __ addptr(from, 8);
 918     __ decrement(qword_count);
 919     __ jcc(Assembler::greater, L_copy_8_bytes);
 920   __ BIND(L_exit);
 921     __ emms();
 922   }
 923 
 924   address generate_disjoint_copy(BasicType t, bool aligned,
 925                                  Address::ScaleFactor sf,
 926                                  address* entry, const char *name) {

 927     __ align(CodeEntryAlignment);
 928     StubCodeMark mark(this, "StubRoutines", name);
 929     address start = __ pc();
 930 
 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;
 933 
 934     int shift = Address::times_ptr - sf;
 935 
 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
 941 
 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 
 949     if (entry != NULL) {
 950       *entry = __ pc(); // Entry point from conjoint arraycopy stub.
 951       BLOCK_COMMENT("Entry:");
 952     }
 953 
 954     if (t == T_OBJECT) {
 955       __ testl(count, count);
 956       __ jcc(Assembler::zero, L_0_count);

 957       gen_write_ref_array_pre_barrier(to, count);

 958       __ mov(saved_to, to);          // save 'to'
 959     }
 960 
 961     __ subptr(to, from); // to --> to_from
 962     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
 963     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
 964     if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 965       // align source address at 4 bytes address boundary
 966       if (t == T_BYTE) {
 967         // One byte misalignment happens only for byte arrays
 968         __ testl(from, 1);
 969         __ jccb(Assembler::zero, L_skip_align1);
 970         __ movb(rax, Address(from, 0));
 971         __ movb(Address(from, to_from, Address::times_1, 0), rax);
 972         __ increment(from);
 973         __ decrement(count);
 974       __ BIND(L_skip_align1);
 975       }
 976       // Two bytes misalignment happens only for byte and short (char) arrays
 977       __ testl(from, 2);
 978       __ jccb(Assembler::zero, L_skip_align2);
 979       __ movw(rax, Address(from, 0));
 980       __ movw(Address(from, to_from, Address::times_1, 0), rax);
 981       __ addptr(from, 2);
 982       __ subl(count, 1<<(shift-1));
 983     __ BIND(L_skip_align2);
 984     }
 985     if (!VM_Version::supports_mmx()) {
 986       __ mov(rax, count);      // save 'count'
 987       __ shrl(count, shift); // bytes count
 988       __ addptr(to_from, from);// restore 'to'
 989       __ rep_mov();
 990       __ subptr(to_from, from);// restore 'to_from'
 991       __ mov(count, rax);      // restore 'count'
 992       __ jmpb(L_copy_2_bytes); // all dwords were copied
 993     } else {
 994       if (!UseUnalignedLoadStores) {
 995         // align to 8 bytes, we know we are 4 byte aligned to start
 996         __ testptr(from, 4);
 997         __ jccb(Assembler::zero, L_copy_64_bytes);
 998         __ movl(rax, Address(from, 0));
 999         __ movl(Address(from, to_from, Address::times_1, 0), rax);
1000         __ addptr(from, 4);
1001         __ subl(count, 1<<shift);
1002       }
1003     __ BIND(L_copy_64_bytes);
1004       __ mov(rax, count);
1005       __ shrl(rax, shift+1);  // 8 bytes chunk count
1006       //
1007       // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
1008       //
1009       if (UseXMMForArrayCopy) {
1010         xmm_copy_forward(from, to_from, rax);
1011       } else {
1012         mmx_copy_forward(from, to_from, rax);
1013       }
1014     }
1015     // copy tailing dword
1016   __ BIND(L_copy_4_bytes);
1017     __ testl(count, 1<<shift);
1018     __ jccb(Assembler::zero, L_copy_2_bytes);
1019     __ movl(rax, Address(from, 0));
1020     __ movl(Address(from, to_from, Address::times_1, 0), rax);
1021     if (t == T_BYTE || t == T_SHORT) {
1022       __ addptr(from, 4);
1023     __ BIND(L_copy_2_bytes);
1024       // copy tailing word
1025       __ testl(count, 1<<(shift-1));
1026       __ jccb(Assembler::zero, L_copy_byte);
1027       __ movw(rax, Address(from, 0));
1028       __ movw(Address(from, to_from, Address::times_1, 0), rax);
1029       if (t == T_BYTE) {
1030         __ addptr(from, 2);
1031       __ BIND(L_copy_byte);
1032         // copy tailing byte
1033         __ testl(count, 1);
1034         __ jccb(Assembler::zero, L_exit);
1035         __ movb(rax, Address(from, 0));
1036         __ movb(Address(from, to_from, Address::times_1, 0), rax);
1037       __ BIND(L_exit);
1038       } else {
1039       __ BIND(L_copy_byte);
1040       }
1041     } else {
1042     __ BIND(L_copy_2_bytes);
1043     }
1044 
1045     if (t == T_OBJECT) {
1046       __ movl(count, Address(rsp, 12+12)); // reread 'count'
1047       __ mov(to, saved_to); // restore 'to'
1048       gen_write_ref_array_post_barrier(to, count);
1049     __ BIND(L_0_count);
1050     }
1051     inc_copy_counter_np(t);
1052     __ pop(rdi);
1053     __ pop(rsi);
1054     __ leave(); // required for proper stackwalking of RuntimeStub frame
1055     __ xorptr(rax, rax); // return 0
1056     __ ret(0);
1057     return start;
1058   }
1059 
1060 
1061   address generate_fill(BasicType t, bool aligned, const char *name) {
1062     __ align(CodeEntryAlignment);
1063     StubCodeMark mark(this, "StubRoutines", name);
1064     address start = __ pc();
1065 
1066     BLOCK_COMMENT("Entry:");
1067 
1068     const Register to       = rdi;  // source array address
1069     const Register value    = rdx;  // value
1070     const Register count    = rsi;  // elements count
1071 
1072     __ enter(); // required for proper stackwalking of RuntimeStub frame
1073     __ push(rsi);
1074     __ push(rdi);
1075     __ movptr(to   , Address(rsp, 12+ 4));
1076     __ movl(value, Address(rsp, 12+ 8));
1077     __ movl(count, Address(rsp, 12+ 12));
1078 
1079     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1080 
1081     __ pop(rdi);
1082     __ pop(rsi);
1083     __ leave(); // required for proper stackwalking of RuntimeStub frame
1084     __ ret(0);
1085     return start;
1086   }
1087 
1088   address generate_conjoint_copy(BasicType t, bool aligned,
1089                                  Address::ScaleFactor sf,
1090                                  address nooverlap_target,
1091                                  address* entry, const char *name) {

1092     __ align(CodeEntryAlignment);
1093     StubCodeMark mark(this, "StubRoutines", name);
1094     address start = __ pc();
1095 
1096     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1097     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1098 
1099     int shift = Address::times_ptr - sf;
1100 
1101     const Register src   = rax;  // source array address
1102     const Register dst   = rdx;  // destination array address
1103     const Register from  = rsi;  // source array address
1104     const Register to    = rdi;  // destination array address
1105     const Register count = rcx;  // elements count
1106     const Register end   = rax;  // array end address
1107 
1108     __ enter(); // required for proper stackwalking of RuntimeStub frame
1109     __ push(rsi);
1110     __ push(rdi);
1111     __ movptr(src  , Address(rsp, 12+ 4));   // from
1112     __ movptr(dst  , Address(rsp, 12+ 8));   // to
1113     __ movl2ptr(count, Address(rsp, 12+12)); // count
1114 
1115     if (entry != NULL) {
1116       *entry = __ pc(); // Entry point from generic arraycopy stub.
1117       BLOCK_COMMENT("Entry:");
1118     }
1119 
1120     // nooverlap_target expects arguments in rsi and rdi.
1121     __ mov(from, src);
1122     __ mov(to  , dst);
1123 
1124     // arrays overlap test: dispatch to disjoint stub if necessary.
1125     RuntimeAddress nooverlap(nooverlap_target);
1126     __ cmpptr(dst, src);
1127     __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1128     __ jump_cc(Assembler::belowEqual, nooverlap);
1129     __ cmpptr(dst, end);
1130     __ jump_cc(Assembler::aboveEqual, nooverlap);
1131 
1132     if (t == T_OBJECT) {
1133       __ testl(count, count);
1134       __ jcc(Assembler::zero, L_0_count);

1135        gen_write_ref_array_pre_barrier(dst, count);
1136     }

1137 
1138     // copy from high to low
1139     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1140     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1141     if (t == T_BYTE || t == T_SHORT) {
1142       // Align the end of destination array at 4 bytes address boundary
1143       __ lea(end, Address(dst, count, sf, 0));
1144       if (t == T_BYTE) {
1145         // One byte misalignment happens only for byte arrays
1146         __ testl(end, 1);
1147         __ jccb(Assembler::zero, L_skip_align1);
1148         __ decrement(count);
1149         __ movb(rdx, Address(from, count, sf, 0));
1150         __ movb(Address(to, count, sf, 0), rdx);
1151       __ BIND(L_skip_align1);
1152       }
1153       // Two bytes misalignment happens only for byte and short (char) arrays
1154       __ testl(end, 2);
1155       __ jccb(Assembler::zero, L_skip_align2);
1156       __ subptr(count, 1<<(shift-1));
1157       __ movw(rdx, Address(from, count, sf, 0));
1158       __ movw(Address(to, count, sf, 0), rdx);
1159     __ BIND(L_skip_align2);
1160       __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1161       __ jcc(Assembler::below, L_copy_4_bytes);
1162     }
1163 
1164     if (!VM_Version::supports_mmx()) {
1165       __ std();
1166       __ mov(rax, count); // Save 'count'
1167       __ mov(rdx, to);    // Save 'to'
1168       __ lea(rsi, Address(from, count, sf, -4));
1169       __ lea(rdi, Address(to  , count, sf, -4));
1170       __ shrptr(count, shift); // bytes count
1171       __ rep_mov();
1172       __ cld();
1173       __ mov(count, rax); // restore 'count'
1174       __ andl(count, (1<<shift)-1);      // mask the number of rest elements
1175       __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1176       __ mov(to, rdx);   // restore 'to'
1177       __ jmpb(L_copy_2_bytes); // all dword were copied
1178    } else {
1179       // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1180       __ testptr(end, 4);
1181       __ jccb(Assembler::zero, L_copy_8_bytes);
1182       __ subl(count, 1<<shift);
1183       __ movl(rdx, Address(from, count, sf, 0));
1184       __ movl(Address(to, count, sf, 0), rdx);
1185       __ jmpb(L_copy_8_bytes);
1186 
1187       __ align(OptoLoopAlignment);
1188       // Move 8 bytes
1189     __ BIND(L_copy_8_bytes_loop);
1190       if (UseXMMForArrayCopy) {
1191         __ movq(xmm0, Address(from, count, sf, 0));
1192         __ movq(Address(to, count, sf, 0), xmm0);
1193       } else {
1194         __ movq(mmx0, Address(from, count, sf, 0));
1195         __ movq(Address(to, count, sf, 0), mmx0);
1196       }
1197     __ BIND(L_copy_8_bytes);
1198       __ subl(count, 2<<shift);
1199       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1200       __ addl(count, 2<<shift);
1201       if (!UseXMMForArrayCopy) {
1202         __ emms();
1203       }
1204     }
1205   __ BIND(L_copy_4_bytes);
1206     // copy prefix qword
1207     __ testl(count, 1<<shift);
1208     __ jccb(Assembler::zero, L_copy_2_bytes);
1209     __ movl(rdx, Address(from, count, sf, -4));
1210     __ movl(Address(to, count, sf, -4), rdx);
1211 
1212     if (t == T_BYTE || t == T_SHORT) {
1213         __ subl(count, (1<<shift));
1214       __ BIND(L_copy_2_bytes);
1215         // copy prefix dword
1216         __ testl(count, 1<<(shift-1));
1217         __ jccb(Assembler::zero, L_copy_byte);
1218         __ movw(rdx, Address(from, count, sf, -2));
1219         __ movw(Address(to, count, sf, -2), rdx);
1220         if (t == T_BYTE) {
1221           __ subl(count, 1<<(shift-1));
1222         __ BIND(L_copy_byte);
1223           // copy prefix byte
1224           __ testl(count, 1);
1225           __ jccb(Assembler::zero, L_exit);
1226           __ movb(rdx, Address(from, 0));
1227           __ movb(Address(to, 0), rdx);
1228         __ BIND(L_exit);
1229         } else {
1230         __ BIND(L_copy_byte);
1231         }
1232     } else {
1233     __ BIND(L_copy_2_bytes);
1234     }
1235     if (t == T_OBJECT) {
1236       __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1237       gen_write_ref_array_post_barrier(to, count);
1238     __ BIND(L_0_count);
1239     }
1240     inc_copy_counter_np(t);
1241     __ pop(rdi);
1242     __ pop(rsi);
1243     __ leave(); // required for proper stackwalking of RuntimeStub frame
1244     __ xorptr(rax, rax); // return 0
1245     __ ret(0);
1246     return start;
1247   }
1248 
1249 
1250   address generate_disjoint_long_copy(address* entry, const char *name) {
1251     __ align(CodeEntryAlignment);
1252     StubCodeMark mark(this, "StubRoutines", name);
1253     address start = __ pc();
1254 
1255     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1256     const Register from       = rax;  // source array address
1257     const Register to         = rdx;  // destination array address
1258     const Register count      = rcx;  // elements count
1259     const Register to_from    = rdx;  // (to - from)
1260 
1261     __ enter(); // required for proper stackwalking of RuntimeStub frame
1262     __ movptr(from , Address(rsp, 8+0));       // from
1263     __ movptr(to   , Address(rsp, 8+4));       // to
1264     __ movl2ptr(count, Address(rsp, 8+8));     // count
1265 
1266     *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1267     BLOCK_COMMENT("Entry:");
1268 
1269     __ subptr(to, from); // to --> to_from
1270     if (VM_Version::supports_mmx()) {
1271       if (UseXMMForArrayCopy) {
1272         xmm_copy_forward(from, to_from, count);
1273       } else {
1274         mmx_copy_forward(from, to_from, count);
1275       }
1276     } else {
1277       __ jmpb(L_copy_8_bytes);
1278       __ align(OptoLoopAlignment);
1279     __ BIND(L_copy_8_bytes_loop);
1280       __ fild_d(Address(from, 0));
1281       __ fistp_d(Address(from, to_from, Address::times_1));
1282       __ addptr(from, 8);
1283     __ BIND(L_copy_8_bytes);
1284       __ decrement(count);
1285       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1286     }
1287     inc_copy_counter_np(T_LONG);
1288     __ leave(); // required for proper stackwalking of RuntimeStub frame
1289     __ xorptr(rax, rax); // return 0
1290     __ ret(0);
1291     return start;
1292   }
1293 
1294   address generate_conjoint_long_copy(address nooverlap_target,
1295                                       address* entry, const char *name) {
1296     __ align(CodeEntryAlignment);
1297     StubCodeMark mark(this, "StubRoutines", name);
1298     address start = __ pc();
1299 
1300     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1301     const Register from       = rax;  // source array address
1302     const Register to         = rdx;  // destination array address
1303     const Register count      = rcx;  // elements count
1304     const Register end_from   = rax;  // source array end address
1305 
1306     __ enter(); // required for proper stackwalking of RuntimeStub frame
1307     __ movptr(from , Address(rsp, 8+0));       // from
1308     __ movptr(to   , Address(rsp, 8+4));       // to
1309     __ movl2ptr(count, Address(rsp, 8+8));     // count
1310 
1311     *entry = __ pc(); // Entry point from generic arraycopy stub.
1312     BLOCK_COMMENT("Entry:");
1313 
1314     // arrays overlap test
1315     __ cmpptr(to, from);
1316     RuntimeAddress nooverlap(nooverlap_target);
1317     __ jump_cc(Assembler::belowEqual, nooverlap);
1318     __ lea(end_from, Address(from, count, Address::times_8, 0));
1319     __ cmpptr(to, end_from);
1320     __ movptr(from, Address(rsp, 8));  // from
1321     __ jump_cc(Assembler::aboveEqual, nooverlap);
1322 
1323     __ jmpb(L_copy_8_bytes);
1324 
1325     __ align(OptoLoopAlignment);
1326   __ BIND(L_copy_8_bytes_loop);
1327     if (VM_Version::supports_mmx()) {
1328       if (UseXMMForArrayCopy) {
1329         __ movq(xmm0, Address(from, count, Address::times_8));
1330         __ movq(Address(to, count, Address::times_8), xmm0);
1331       } else {
1332         __ movq(mmx0, Address(from, count, Address::times_8));
1333         __ movq(Address(to, count, Address::times_8), mmx0);
1334       }
1335     } else {
1336       __ fild_d(Address(from, count, Address::times_8));
1337       __ fistp_d(Address(to, count, Address::times_8));
1338     }
1339   __ BIND(L_copy_8_bytes);
1340     __ decrement(count);
1341     __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1342 
1343     if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1344       __ emms();
1345     }
1346     inc_copy_counter_np(T_LONG);
1347     __ leave(); // required for proper stackwalking of RuntimeStub frame
1348     __ xorptr(rax, rax); // return 0
1349     __ ret(0);
1350     return start;
1351   }
1352 
1353 
1354   // Helper for generating a dynamic type check.
1355   // The sub_klass must be one of {rbx, rdx, rsi}.
1356   // The temp is killed.
1357   void generate_type_check(Register sub_klass,
1358                            Address& super_check_offset_addr,
1359                            Address& super_klass_addr,
1360                            Register temp,
1361                            Label* L_success, Label* L_failure) {
1362     BLOCK_COMMENT("type_check:");
1363 
1364     Label L_fallthrough;
1365 #define LOCAL_JCC(assembler_con, label_ptr)                             \
1366     if (label_ptr != NULL)  __ jcc(assembler_con, *(label_ptr));        \
1367     else                    __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1368 
1369     // The following is a strange variation of the fast path which requires
1370     // one less register, because needed values are on the argument stack.
1371     // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1372     //                                  L_success, L_failure, NULL);
1373     assert_different_registers(sub_klass, temp);
1374 
1375     int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1376                      Klass::secondary_super_cache_offset_in_bytes());
1377 
1378     // if the pointers are equal, we are done (e.g., String[] elements)
1379     __ cmpptr(sub_klass, super_klass_addr);
1380     LOCAL_JCC(Assembler::equal, L_success);
1381 
1382     // check the supertype display:
1383     __ movl2ptr(temp, super_check_offset_addr);
1384     Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1385     __ movptr(temp, super_check_addr); // load displayed supertype
1386     __ cmpptr(temp, super_klass_addr); // test the super type
1387     LOCAL_JCC(Assembler::equal, L_success);
1388 
1389     // if it was a primary super, we can just fail immediately
1390     __ cmpl(super_check_offset_addr, sc_offset);
1391     LOCAL_JCC(Assembler::notEqual, L_failure);
1392 
1393     // The repne_scan instruction uses fixed registers, which will get spilled.
1394     // We happen to know this works best when super_klass is in rax.
1395     Register super_klass = temp;
1396     __ movptr(super_klass, super_klass_addr);
1397     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1398                                      L_success, L_failure);
1399 
1400     __ bind(L_fallthrough);
1401 
1402     if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1403     if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1404 
1405 #undef LOCAL_JCC
1406   }
1407 
1408   //
1409   //  Generate checkcasting array copy stub
1410   //
1411   //  Input:
1412   //    4(rsp)   - source array address
1413   //    8(rsp)   - destination array address
1414   //   12(rsp)   - element count, can be zero
1415   //   16(rsp)   - size_t ckoff (super_check_offset)
1416   //   20(rsp)   - oop ckval (super_klass)
1417   //
1418   //  Output:
1419   //    rax, ==  0  -  success
1420   //    rax, == -1^K - failure, where K is partial transfer count
1421   //
1422   address generate_checkcast_copy(const char *name, address* entry) {
1423     __ align(CodeEntryAlignment);
1424     StubCodeMark mark(this, "StubRoutines", name);
1425     address start = __ pc();
1426 
1427     Label L_load_element, L_store_element, L_do_card_marks, L_done;
1428 
1429     // register use:
1430     //  rax, rdx, rcx -- loop control (end_from, end_to, count)
1431     //  rdi, rsi      -- element access (oop, klass)
1432     //  rbx,           -- temp
1433     const Register from       = rax;    // source array address
1434     const Register to         = rdx;    // destination array address
1435     const Register length     = rcx;    // elements count
1436     const Register elem       = rdi;    // each oop copied
1437     const Register elem_klass = rsi;    // each elem._klass (sub_klass)
1438     const Register temp       = rbx;    // lone remaining temp
1439 
1440     __ enter(); // required for proper stackwalking of RuntimeStub frame
1441 
1442     __ push(rsi);
1443     __ push(rdi);
1444     __ push(rbx);
1445 
1446     Address   from_arg(rsp, 16+ 4);     // from
1447     Address     to_arg(rsp, 16+ 8);     // to
1448     Address length_arg(rsp, 16+12);     // elements count
1449     Address  ckoff_arg(rsp, 16+16);     // super_check_offset
1450     Address  ckval_arg(rsp, 16+20);     // super_klass
1451 
1452     // Load up:
1453     __ movptr(from,     from_arg);
1454     __ movptr(to,         to_arg);
1455     __ movl2ptr(length, length_arg);
1456 
1457     if (entry != NULL) {
1458       *entry = __ pc(); // Entry point from generic arraycopy stub.
1459       BLOCK_COMMENT("Entry:");
1460     }
1461 
1462     //---------------------------------------------------------------
1463     // Assembler stub will be used for this call to arraycopy
1464     // if the two arrays are subtypes of Object[] but the
1465     // destination array type is not equal to or a supertype
1466     // of the source type.  Each element must be separately
1467     // checked.
1468 
1469     // Loop-invariant addresses.  They are exclusive end pointers.
1470     Address end_from_addr(from, length, Address::times_ptr, 0);
1471     Address   end_to_addr(to,   length, Address::times_ptr, 0);
1472 
1473     Register end_from = from;           // re-use
1474     Register end_to   = to;             // re-use
1475     Register count    = length;         // re-use
1476 
1477     // Loop-variant addresses.  They assume post-incremented count < 0.
1478     Address from_element_addr(end_from, count, Address::times_ptr, 0);
1479     Address   to_element_addr(end_to,   count, Address::times_ptr, 0);
1480     Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1481 
1482     // Copy from low to high addresses, indexed from the end of each array.

1483     gen_write_ref_array_pre_barrier(to, count);

1484     __ lea(end_from, end_from_addr);
1485     __ lea(end_to,   end_to_addr);
1486     assert(length == count, "");        // else fix next line:
1487     __ negptr(count);                   // negate and test the length
1488     __ jccb(Assembler::notZero, L_load_element);
1489 
1490     // Empty array:  Nothing to do.
1491     __ xorptr(rax, rax);                  // return 0 on (trivial) success
1492     __ jmp(L_done);
1493 
1494     // ======== begin loop ========
1495     // (Loop is rotated; its entry is L_load_element.)
1496     // Loop control:
1497     //   for (count = -count; count != 0; count++)
1498     // Base pointers src, dst are biased by 8*count,to last element.
1499     __ align(OptoLoopAlignment);
1500 
1501     __ BIND(L_store_element);
1502     __ movptr(to_element_addr, elem);     // store the oop
1503     __ increment(count);                // increment the count toward zero
1504     __ jccb(Assembler::zero, L_do_card_marks);
1505 
1506     // ======== loop entry is here ========
1507     __ BIND(L_load_element);
1508     __ movptr(elem, from_element_addr);   // load the oop
1509     __ testptr(elem, elem);
1510     __ jccb(Assembler::zero, L_store_element);
1511 
1512     // (Could do a trick here:  Remember last successful non-null
1513     // element stored and make a quick oop equality check on it.)
1514 
1515     __ movptr(elem_klass, elem_klass_addr); // query the object klass
1516     generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1517                         &L_store_element, NULL);
1518       // (On fall-through, we have failed the element type check.)
1519     // ======== end loop ========
1520 
1521     // It was a real error; we must depend on the caller to finish the job.
1522     // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1523     // Emit GC store barriers for the oops we have copied (length_arg + count),
1524     // and report their number to the caller.
1525     __ addl(count, length_arg);         // transfers = (length - remaining)
1526     __ movl2ptr(rax, count);            // save the value
1527     __ notptr(rax);                     // report (-1^K) to caller
1528     __ movptr(to, to_arg);              // reload
1529     assert_different_registers(to, count, rax);
1530     gen_write_ref_array_post_barrier(to, count);
1531     __ jmpb(L_done);
1532 
1533     // Come here on success only.
1534     __ BIND(L_do_card_marks);
1535     __ movl2ptr(count, length_arg);
1536     __ movptr(to, to_arg);                // reload
1537     gen_write_ref_array_post_barrier(to, count);
1538     __ xorptr(rax, rax);                  // return 0 on success
1539 
1540     // Common exit point (success or failure).
1541     __ BIND(L_done);
1542     __ pop(rbx);
1543     __ pop(rdi);
1544     __ pop(rsi);
1545     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1546     __ leave(); // required for proper stackwalking of RuntimeStub frame
1547     __ ret(0);
1548 
1549     return start;
1550   }
1551 
1552   //
1553   //  Generate 'unsafe' array copy stub
1554   //  Though just as safe as the other stubs, it takes an unscaled
1555   //  size_t argument instead of an element count.
1556   //
1557   //  Input:
1558   //    4(rsp)   - source array address
1559   //    8(rsp)   - destination array address
1560   //   12(rsp)   - byte count, can be zero
1561   //
1562   //  Output:
1563   //    rax, ==  0  -  success
1564   //    rax, == -1  -  need to call System.arraycopy
1565   //
1566   // Examines the alignment of the operands and dispatches
1567   // to a long, int, short, or byte copy loop.
1568   //
1569   address generate_unsafe_copy(const char *name,
1570                                address byte_copy_entry,
1571                                address short_copy_entry,
1572                                address int_copy_entry,
1573                                address long_copy_entry) {
1574 
1575     Label L_long_aligned, L_int_aligned, L_short_aligned;
1576 
1577     __ align(CodeEntryAlignment);
1578     StubCodeMark mark(this, "StubRoutines", name);
1579     address start = __ pc();
1580 
1581     const Register from       = rax;  // source array address
1582     const Register to         = rdx;  // destination array address
1583     const Register count      = rcx;  // elements count
1584 
1585     __ enter(); // required for proper stackwalking of RuntimeStub frame
1586     __ push(rsi);
1587     __ push(rdi);
1588     Address  from_arg(rsp, 12+ 4);      // from
1589     Address    to_arg(rsp, 12+ 8);      // to
1590     Address count_arg(rsp, 12+12);      // byte count
1591 
1592     // Load up:
1593     __ movptr(from ,  from_arg);
1594     __ movptr(to   ,    to_arg);
1595     __ movl2ptr(count, count_arg);
1596 
1597     // bump this on entry, not on exit:
1598     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1599 
1600     const Register bits = rsi;
1601     __ mov(bits, from);
1602     __ orptr(bits, to);
1603     __ orptr(bits, count);
1604 
1605     __ testl(bits, BytesPerLong-1);
1606     __ jccb(Assembler::zero, L_long_aligned);
1607 
1608     __ testl(bits, BytesPerInt-1);
1609     __ jccb(Assembler::zero, L_int_aligned);
1610 
1611     __ testl(bits, BytesPerShort-1);
1612     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1613 
1614     __ BIND(L_short_aligned);
1615     __ shrptr(count, LogBytesPerShort); // size => short_count
1616     __ movl(count_arg, count);          // update 'count'
1617     __ jump(RuntimeAddress(short_copy_entry));
1618 
1619     __ BIND(L_int_aligned);
1620     __ shrptr(count, LogBytesPerInt); // size => int_count
1621     __ movl(count_arg, count);          // update 'count'
1622     __ jump(RuntimeAddress(int_copy_entry));
1623 
1624     __ BIND(L_long_aligned);
1625     __ shrptr(count, LogBytesPerLong); // size => qword_count
1626     __ movl(count_arg, count);          // update 'count'
1627     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1628     __ pop(rsi);
1629     __ jump(RuntimeAddress(long_copy_entry));
1630 
1631     return start;
1632   }
1633 
1634 
1635   // Perform range checks on the proposed arraycopy.
1636   // Smashes src_pos and dst_pos.  (Uses them up for temps.)
1637   void arraycopy_range_checks(Register src,
1638                               Register src_pos,
1639                               Register dst,
1640                               Register dst_pos,
1641                               Address& length,
1642                               Label& L_failed) {
1643     BLOCK_COMMENT("arraycopy_range_checks:");
1644     const Register src_end = src_pos;   // source array end position
1645     const Register dst_end = dst_pos;   // destination array end position
1646     __ addl(src_end, length); // src_pos + length
1647     __ addl(dst_end, length); // dst_pos + length
1648 
1649     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;
1650     __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1651     __ jcc(Assembler::above, L_failed);
1652 
1653     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1654     __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1655     __ jcc(Assembler::above, L_failed);
1656 
1657     BLOCK_COMMENT("arraycopy_range_checks done");
1658   }
1659 
1660 
1661   //
1662   //  Generate generic array copy stubs
1663   //
1664   //  Input:
1665   //     4(rsp)    -  src oop
1666   //     8(rsp)    -  src_pos
1667   //    12(rsp)    -  dst oop
1668   //    16(rsp)    -  dst_pos
1669   //    20(rsp)    -  element count
1670   //
1671   //  Output:
1672   //    rax, ==  0  -  success
1673   //    rax, == -1^K - failure, where K is partial transfer count
1674   //
1675   address generate_generic_copy(const char *name,
1676                                 address entry_jbyte_arraycopy,
1677                                 address entry_jshort_arraycopy,
1678                                 address entry_jint_arraycopy,
1679                                 address entry_oop_arraycopy,
1680                                 address entry_jlong_arraycopy,
1681                                 address entry_checkcast_arraycopy) {
1682     Label L_failed, L_failed_0, L_objArray;
1683 
1684     { int modulus = CodeEntryAlignment;
1685       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
1686       int advance = target - (__ offset() % modulus);
1687       if (advance < 0)  advance += modulus;
1688       if (advance > 0)  __ nop(advance);
1689     }
1690     StubCodeMark mark(this, "StubRoutines", name);
1691 
1692     // Short-hop target to L_failed.  Makes for denser prologue code.
1693     __ BIND(L_failed_0);
1694     __ jmp(L_failed);
1695     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1696 
1697     __ align(CodeEntryAlignment);
1698     address start = __ pc();
1699 
1700     __ enter(); // required for proper stackwalking of RuntimeStub frame
1701     __ push(rsi);
1702     __ push(rdi);
1703 
1704     // bump this on entry, not on exit:
1705     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1706 
1707     // Input values
1708     Address SRC     (rsp, 12+ 4);
1709     Address SRC_POS (rsp, 12+ 8);
1710     Address DST     (rsp, 12+12);
1711     Address DST_POS (rsp, 12+16);
1712     Address LENGTH  (rsp, 12+20);
1713 
1714     //-----------------------------------------------------------------------
1715     // Assembler stub will be used for this call to arraycopy
1716     // if the following conditions are met:
1717     //
1718     // (1) src and dst must not be null.
1719     // (2) src_pos must not be negative.
1720     // (3) dst_pos must not be negative.
1721     // (4) length  must not be negative.
1722     // (5) src klass and dst klass should be the same and not NULL.
1723     // (6) src and dst should be arrays.
1724     // (7) src_pos + length must not exceed length of src.
1725     // (8) dst_pos + length must not exceed length of dst.
1726     //
1727 
1728     const Register src     = rax;       // source array oop
1729     const Register src_pos = rsi;
1730     const Register dst     = rdx;       // destination array oop
1731     const Register dst_pos = rdi;
1732     const Register length  = rcx;       // transfer count
1733 
1734     //  if (src == NULL) return -1;
1735     __ movptr(src, SRC);      // src oop
1736     __ testptr(src, src);
1737     __ jccb(Assembler::zero, L_failed_0);
1738 
1739     //  if (src_pos < 0) return -1;
1740     __ movl2ptr(src_pos, SRC_POS);  // src_pos
1741     __ testl(src_pos, src_pos);
1742     __ jccb(Assembler::negative, L_failed_0);
1743 
1744     //  if (dst == NULL) return -1;
1745     __ movptr(dst, DST);      // dst oop
1746     __ testptr(dst, dst);
1747     __ jccb(Assembler::zero, L_failed_0);
1748 
1749     //  if (dst_pos < 0) return -1;
1750     __ movl2ptr(dst_pos, DST_POS);  // dst_pos
1751     __ testl(dst_pos, dst_pos);
1752     __ jccb(Assembler::negative, L_failed_0);
1753 
1754     //  if (length < 0) return -1;
1755     __ movl2ptr(length, LENGTH);   // length
1756     __ testl(length, length);
1757     __ jccb(Assembler::negative, L_failed_0);
1758 
1759     //  if (src->klass() == NULL) return -1;
1760     Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1761     Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1762     const Register rcx_src_klass = rcx;    // array klass
1763     __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1764 
1765 #ifdef ASSERT
1766     //  assert(src->klass() != NULL);
1767     BLOCK_COMMENT("assert klasses not null");
1768     { Label L1, L2;
1769       __ testptr(rcx_src_klass, rcx_src_klass);
1770       __ jccb(Assembler::notZero, L2);   // it is broken if klass is NULL
1771       __ bind(L1);
1772       __ stop("broken null klass");
1773       __ bind(L2);
1774       __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1775       __ jccb(Assembler::equal, L1);      // this would be broken also
1776       BLOCK_COMMENT("assert done");
1777     }
1778 #endif //ASSERT
1779 
1780     // Load layout helper (32-bits)
1781     //
1782     //  |array_tag|     | header_size | element_type |     |log2_element_size|
1783     // 32        30    24            16              8     2                 0
1784     //
1785     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1786     //
1787 
1788     int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1789                     Klass::layout_helper_offset_in_bytes();
1790     Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1791 
1792     // Handle objArrays completely differently...
1793     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1794     __ cmpl(src_klass_lh_addr, objArray_lh);
1795     __ jcc(Assembler::equal, L_objArray);
1796 
1797     //  if (src->klass() != dst->klass()) return -1;
1798     __ cmpptr(rcx_src_klass, dst_klass_addr);
1799     __ jccb(Assembler::notEqual, L_failed_0);
1800 
1801     const Register rcx_lh = rcx;  // layout helper
1802     assert(rcx_lh == rcx_src_klass, "known alias");
1803     __ movl(rcx_lh, src_klass_lh_addr);
1804 
1805     //  if (!src->is_Array()) return -1;
1806     __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1807     __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1808 
1809     // At this point, it is known to be a typeArray (array_tag 0x3).
1810 #ifdef ASSERT
1811     { Label L;
1812       __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1813       __ jcc(Assembler::greaterEqual, L); // signed cmp
1814       __ stop("must be a primitive array");
1815       __ bind(L);
1816     }
1817 #endif
1818 
1819     assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1820     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1821 
1822     // typeArrayKlass
1823     //
1824     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1825     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1826     //
1827     const Register rsi_offset = rsi; // array offset
1828     const Register src_array  = src; // src array offset
1829     const Register dst_array  = dst; // dst array offset
1830     const Register rdi_elsize = rdi; // log2 element size
1831 
1832     __ mov(rsi_offset, rcx_lh);
1833     __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1834     __ andptr(rsi_offset, Klass::_lh_header_size_mask);   // array_offset
1835     __ addptr(src_array, rsi_offset);  // src array offset
1836     __ addptr(dst_array, rsi_offset);  // dst array offset
1837     __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1838 
1839     // next registers should be set before the jump to corresponding stub
1840     const Register from       = src; // source array address
1841     const Register to         = dst; // destination array address
1842     const Register count      = rcx; // elements count
1843     // some of them should be duplicated on stack
1844 #define FROM   Address(rsp, 12+ 4)
1845 #define TO     Address(rsp, 12+ 8)   // Not used now
1846 #define COUNT  Address(rsp, 12+12)   // Only for oop arraycopy
1847 
1848     BLOCK_COMMENT("scale indexes to element size");
1849     __ movl2ptr(rsi, SRC_POS);  // src_pos
1850     __ shlptr(rsi);             // src_pos << rcx (log2 elsize)
1851     assert(src_array == from, "");
1852     __ addptr(from, rsi);       // from = src_array + SRC_POS << log2 elsize
1853     __ movl2ptr(rdi, DST_POS);  // dst_pos
1854     __ shlptr(rdi);             // dst_pos << rcx (log2 elsize)
1855     assert(dst_array == to, "");
1856     __ addptr(to,  rdi);        // to   = dst_array + DST_POS << log2 elsize
1857     __ movptr(FROM, from);      // src_addr
1858     __ mov(rdi_elsize, rcx_lh); // log2 elsize
1859     __ movl2ptr(count, LENGTH); // elements count
1860 
1861     BLOCK_COMMENT("choose copy loop based on element size");
1862     __ cmpl(rdi_elsize, 0);
1863 
1864     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1865     __ cmpl(rdi_elsize, LogBytesPerShort);
1866     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1867     __ cmpl(rdi_elsize, LogBytesPerInt);
1868     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1869 #ifdef ASSERT
1870     __ cmpl(rdi_elsize, LogBytesPerLong);
1871     __ jccb(Assembler::notEqual, L_failed);
1872 #endif
1873     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1874     __ pop(rsi);
1875     __ jump(RuntimeAddress(entry_jlong_arraycopy));
1876 
1877   __ BIND(L_failed);
1878     __ xorptr(rax, rax);
1879     __ notptr(rax); // return -1
1880     __ pop(rdi);
1881     __ pop(rsi);
1882     __ leave(); // required for proper stackwalking of RuntimeStub frame
1883     __ ret(0);
1884 
1885     // objArrayKlass
1886   __ BIND(L_objArray);
1887     // live at this point:  rcx_src_klass, src[_pos], dst[_pos]
1888 
1889     Label L_plain_copy, L_checkcast_copy;
1890     //  test array classes for subtyping
1891     __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1892     __ jccb(Assembler::notEqual, L_checkcast_copy);
1893 
1894     // Identically typed arrays can be copied without element-wise checks.
1895     assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1896     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1897 
1898   __ BIND(L_plain_copy);
1899     __ movl2ptr(count, LENGTH); // elements count
1900     __ movl2ptr(src_pos, SRC_POS);  // reload src_pos
1901     __ lea(from, Address(src, src_pos, Address::times_ptr,
1902                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1903     __ movl2ptr(dst_pos, DST_POS);  // reload dst_pos
1904     __ lea(to,   Address(dst, dst_pos, Address::times_ptr,
1905                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1906     __ movptr(FROM,  from);   // src_addr
1907     __ movptr(TO,    to);     // dst_addr
1908     __ movl(COUNT, count);  // count
1909     __ jump(RuntimeAddress(entry_oop_arraycopy));
1910 
1911   __ BIND(L_checkcast_copy);
1912     // live at this point:  rcx_src_klass, dst[_pos], src[_pos]
1913     {
1914       // Handy offsets:
1915       int  ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1916                         objArrayKlass::element_klass_offset_in_bytes());
1917       int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1918                         Klass::super_check_offset_offset_in_bytes());
1919 
1920       Register rsi_dst_klass = rsi;
1921       Register rdi_temp      = rdi;
1922       assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1923       assert(rdi_temp      == dst_pos, "expected alias w/ dst_pos");
1924       Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1925 
1926       // Before looking at dst.length, make sure dst is also an objArray.
1927       __ movptr(rsi_dst_klass, dst_klass_addr);
1928       __ cmpl(dst_klass_lh_addr, objArray_lh);
1929       __ jccb(Assembler::notEqual, L_failed);
1930 
1931       // It is safe to examine both src.length and dst.length.
1932       __ movl2ptr(src_pos, SRC_POS);        // reload rsi
1933       arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1934       // (Now src_pos and dst_pos are killed, but not src and dst.)
1935 
1936       // We'll need this temp (don't forget to pop it after the type check).
1937       __ push(rbx);
1938       Register rbx_src_klass = rbx;
1939 
1940       __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1941       __ movptr(rsi_dst_klass, dst_klass_addr);
1942       Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1943       Label L_fail_array_check;
1944       generate_type_check(rbx_src_klass,
1945                           super_check_offset_addr, dst_klass_addr,
1946                           rdi_temp, NULL, &L_fail_array_check);
1947       // (On fall-through, we have passed the array type check.)
1948       __ pop(rbx);
1949       __ jmp(L_plain_copy);
1950 
1951       __ BIND(L_fail_array_check);
1952       // Reshuffle arguments so we can call checkcast_arraycopy:
1953 
1954       // match initial saves for checkcast_arraycopy
1955       // push(rsi);    // already done; see above
1956       // push(rdi);    // already done; see above
1957       // push(rbx);    // already done; see above
1958 
1959       // Marshal outgoing arguments now, freeing registers.
1960       Address   from_arg(rsp, 16+ 4);   // from
1961       Address     to_arg(rsp, 16+ 8);   // to
1962       Address length_arg(rsp, 16+12);   // elements count
1963       Address  ckoff_arg(rsp, 16+16);   // super_check_offset
1964       Address  ckval_arg(rsp, 16+20);   // super_klass
1965 
1966       Address SRC_POS_arg(rsp, 16+ 8);
1967       Address DST_POS_arg(rsp, 16+16);
1968       Address  LENGTH_arg(rsp, 16+20);
1969       // push rbx, changed the incoming offsets (why not just use rbp,??)
1970       // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1971 
1972       __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1973       __ movl2ptr(length, LENGTH_arg);    // reload elements count
1974       __ movl2ptr(src_pos, SRC_POS_arg);  // reload src_pos
1975       __ movl2ptr(dst_pos, DST_POS_arg);  // reload dst_pos
1976 
1977       __ movptr(ckval_arg, rbx);          // destination element type
1978       __ movl(rbx, Address(rbx, sco_offset));
1979       __ movl(ckoff_arg, rbx);          // corresponding class check offset
1980 
1981       __ movl(length_arg, length);      // outgoing length argument
1982 
1983       __ lea(from, Address(src, src_pos, Address::times_ptr,
1984                             arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1985       __ movptr(from_arg, from);
1986 
1987       __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1988                           arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1989       __ movptr(to_arg, to);
1990       __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1991     }
1992 
1993     return start;
1994   }
1995 
1996   void generate_arraycopy_stubs() {
1997     address entry;
1998     address entry_jbyte_arraycopy;
1999     address entry_jshort_arraycopy;
2000     address entry_jint_arraycopy;
2001     address entry_oop_arraycopy;
2002     address entry_jlong_arraycopy;
2003     address entry_checkcast_arraycopy;
2004 
2005     StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
2006         generate_disjoint_copy(T_BYTE,  true, Address::times_1, &entry,
2007                                "arrayof_jbyte_disjoint_arraycopy");
2008     StubRoutines::_arrayof_jbyte_arraycopy =
2009         generate_conjoint_copy(T_BYTE,  true, Address::times_1,  entry,
2010                                NULL, "arrayof_jbyte_arraycopy");
2011     StubRoutines::_jbyte_disjoint_arraycopy =
2012         generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2013                                "jbyte_disjoint_arraycopy");
2014     StubRoutines::_jbyte_arraycopy =
2015         generate_conjoint_copy(T_BYTE, false, Address::times_1,  entry,
2016                                &entry_jbyte_arraycopy, "jbyte_arraycopy");
2017 
2018     StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2019         generate_disjoint_copy(T_SHORT,  true, Address::times_2, &entry,
2020                                "arrayof_jshort_disjoint_arraycopy");
2021     StubRoutines::_arrayof_jshort_arraycopy =
2022         generate_conjoint_copy(T_SHORT,  true, Address::times_2,  entry,
2023                                NULL, "arrayof_jshort_arraycopy");
2024     StubRoutines::_jshort_disjoint_arraycopy =
2025         generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2026                                "jshort_disjoint_arraycopy");
2027     StubRoutines::_jshort_arraycopy =
2028         generate_conjoint_copy(T_SHORT, false, Address::times_2,  entry,
2029                                &entry_jshort_arraycopy, "jshort_arraycopy");
2030 
2031     // Next arrays are always aligned on 4 bytes at least.
2032     StubRoutines::_jint_disjoint_arraycopy =
2033         generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2034                                "jint_disjoint_arraycopy");
2035     StubRoutines::_jint_arraycopy =
2036         generate_conjoint_copy(T_INT, true, Address::times_4,  entry,
2037                                &entry_jint_arraycopy, "jint_arraycopy");
2038 
2039     StubRoutines::_oop_disjoint_arraycopy =
2040         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2041                                "oop_disjoint_arraycopy");
2042     StubRoutines::_oop_arraycopy =
2043         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2044                                &entry_oop_arraycopy, "oop_arraycopy");
2045 







2046     StubRoutines::_jlong_disjoint_arraycopy =
2047         generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2048     StubRoutines::_jlong_arraycopy =
2049         generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2050                                     "jlong_arraycopy");
2051 
2052     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2053     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2054     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2055     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2056     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2057     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2058 
2059     StubRoutines::_arrayof_jint_disjoint_arraycopy  =
2060         StubRoutines::_jint_disjoint_arraycopy;
2061     StubRoutines::_arrayof_oop_disjoint_arraycopy   =
2062         StubRoutines::_oop_disjoint_arraycopy;
2063     StubRoutines::_arrayof_jlong_disjoint_arraycopy =
2064         StubRoutines::_jlong_disjoint_arraycopy;
2065 
2066     StubRoutines::_arrayof_jint_arraycopy  = StubRoutines::_jint_arraycopy;
2067     StubRoutines::_arrayof_oop_arraycopy   = StubRoutines::_oop_arraycopy;

2068     StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2069 
2070     StubRoutines::_checkcast_arraycopy =
2071         generate_checkcast_copy("checkcast_arraycopy",
2072                                   &entry_checkcast_arraycopy);

2073 
2074     StubRoutines::_unsafe_arraycopy =
2075         generate_unsafe_copy("unsafe_arraycopy",
2076                                entry_jbyte_arraycopy,
2077                                entry_jshort_arraycopy,
2078                                entry_jint_arraycopy,
2079                                entry_jlong_arraycopy);
2080 
2081     StubRoutines::_generic_arraycopy =
2082         generate_generic_copy("generic_arraycopy",
2083                                entry_jbyte_arraycopy,
2084                                entry_jshort_arraycopy,
2085                                entry_jint_arraycopy,
2086                                entry_oop_arraycopy,
2087                                entry_jlong_arraycopy,
2088                                entry_checkcast_arraycopy);
2089   }
2090 
2091   void generate_math_stubs() {
2092     {
2093       StubCodeMark mark(this, "StubRoutines", "log");
2094       StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2095 
2096       __ fld_d(Address(rsp, 4));
2097       __ flog();
2098       __ ret(0);
2099     }
2100     {
2101       StubCodeMark mark(this, "StubRoutines", "log10");
2102       StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2103 
2104       __ fld_d(Address(rsp, 4));
2105       __ flog10();
2106       __ ret(0);
2107     }
2108     {
2109       StubCodeMark mark(this, "StubRoutines", "sin");
2110       StubRoutines::_intrinsic_sin = (double (*)(double))  __ pc();
2111 
2112       __ fld_d(Address(rsp, 4));
2113       __ trigfunc('s');
2114       __ ret(0);
2115     }
2116     {
2117       StubCodeMark mark(this, "StubRoutines", "cos");
2118       StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2119 
2120       __ fld_d(Address(rsp, 4));
2121       __ trigfunc('c');
2122       __ ret(0);
2123     }
2124     {
2125       StubCodeMark mark(this, "StubRoutines", "tan");
2126       StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2127 
2128       __ fld_d(Address(rsp, 4));
2129       __ trigfunc('t');
2130       __ ret(0);
2131     }
2132 
2133     // The intrinsic version of these seem to return the same value as
2134     // the strict version.
2135     StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2136     StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2137   }
2138 
2139  public:
2140   // Information about frame layout at time of blocking runtime call.
2141   // Note that we only have to preserve callee-saved registers since
2142   // the compilers are responsible for supplying a continuation point
2143   // if they expect all registers to be preserved.
2144   enum layout {
2145     thread_off,    // last_java_sp
2146     rbp_off,       // callee saved register
2147     ret_pc,
2148     framesize
2149   };
2150 
2151  private:
2152 
2153 #undef  __
2154 #define __ masm->
2155 
2156   //------------------------------------------------------------------------------------------------------------------------
2157   // Continuation point for throwing of implicit exceptions that are not handled in
2158   // the current activation. Fabricates an exception oop and initiates normal
2159   // exception dispatching in this frame.
2160   //
2161   // Previously the compiler (c2) allowed for callee save registers on Java calls.
2162   // This is no longer true after adapter frames were removed but could possibly
2163   // be brought back in the future if the interpreter code was reworked and it
2164   // was deemed worthwhile. The comment below was left to describe what must
2165   // happen here if callee saves were resurrected. As it stands now this stub
2166   // could actually be a vanilla BufferBlob and have now oopMap at all.
2167   // Since it doesn't make much difference we've chosen to leave it the
2168   // way it was in the callee save days and keep the comment.
2169 
2170   // If we need to preserve callee-saved values we need a callee-saved oop map and
2171   // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2172   // If the compiler needs all registers to be preserved between the fault
2173   // point and the exception handler then it must assume responsibility for that in
2174   // AbstractCompiler::continuation_for_implicit_null_exception or
2175   // continuation_for_implicit_division_by_zero_exception. All other implicit
2176   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2177   // either at call sites or otherwise assume that stack unwinding will be initiated,
2178   // so caller saved registers were assumed volatile in the compiler.
2179   address generate_throw_exception(const char* name, address runtime_entry,
2180                                    bool restore_saved_exception_pc) {
2181 
2182     int insts_size = 256;
2183     int locs_size  = 32;
2184 
2185     CodeBuffer code(name, insts_size, locs_size);
2186     OopMapSet* oop_maps  = new OopMapSet();
2187     MacroAssembler* masm = new MacroAssembler(&code);
2188 
2189     address start = __ pc();
2190 
2191     // This is an inlined and slightly modified version of call_VM
2192     // which has the ability to fetch the return PC out of
2193     // thread-local storage and also sets up last_Java_sp slightly
2194     // differently than the real call_VM
2195     Register java_thread = rbx;
2196     __ get_thread(java_thread);
2197     if (restore_saved_exception_pc) {
2198       __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2199       __ push(rax);
2200     }
2201 
2202     __ enter(); // required for proper stackwalking of RuntimeStub frame
2203 
2204     // pc and rbp, already pushed
2205     __ subptr(rsp, (framesize-2) * wordSize); // prolog
2206 
2207     // Frame is now completed as far as size and linkage.
2208 
2209     int frame_complete = __ pc() - start;
2210 
2211     // push java thread (becomes first argument of C function)
2212     __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2213 
2214     // Set up last_Java_sp and last_Java_fp
2215     __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2216 
2217     // Call runtime
2218     BLOCK_COMMENT("call runtime_entry");
2219     __ call(RuntimeAddress(runtime_entry));
2220     // Generate oop map
2221     OopMap* map =  new OopMap(framesize, 0);
2222     oop_maps->add_gc_map(__ pc() - start, map);
2223 
2224     // restore the thread (cannot use the pushed argument since arguments
2225     // may be overwritten by C code generated by an optimizing compiler);
2226     // however can use the register value directly if it is callee saved.
2227     __ get_thread(java_thread);
2228 
2229     __ reset_last_Java_frame(java_thread, true, false);
2230 
2231     __ leave(); // required for proper stackwalking of RuntimeStub frame
2232 
2233     // check for pending exceptions
2234 #ifdef ASSERT
2235     Label L;
2236     __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2237     __ jcc(Assembler::notEqual, L);
2238     __ should_not_reach_here();
2239     __ bind(L);
2240 #endif /* ASSERT */
2241     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2242 
2243 
2244     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2245     return stub->entry_point();
2246   }
2247 
2248 
2249   void create_control_words() {
2250     // Round to nearest, 53-bit mode, exceptions masked
2251     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;
2252     // Round to zero, 53-bit mode, exception mased
2253     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2254     // Round to nearest, 24-bit mode, exceptions masked
2255     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;
2256     // Round to nearest, 64-bit mode, exceptions masked
2257     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;
2258     // Round to nearest, 64-bit mode, exceptions masked
2259     StubRoutines::_mxcsr_std           = 0x1F80;
2260     // Note: the following two constants are 80-bit values
2261     //       layout is critical for correct loading by FPU.
2262     // Bias for strict fp multiply/divide
2263     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2264     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2265     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2266     // Un-Bias for strict fp multiply/divide
2267     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2268     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2269     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2270   }
2271 
2272   //---------------------------------------------------------------------------
2273   // Initialization
2274 
2275   void generate_initial() {
2276     // Generates all stubs and initializes the entry points
2277 
2278     //------------------------------------------------------------------------------------------------------------------------
2279     // entry points that exist in all platforms
2280     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2281     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2282     StubRoutines::_forward_exception_entry      = generate_forward_exception();
2283 
2284     StubRoutines::_call_stub_entry              =
2285       generate_call_stub(StubRoutines::_call_stub_return_address);
2286     // is referenced by megamorphic call
2287     StubRoutines::_catch_exception_entry        = generate_catch_exception();
2288 
2289     // These are currently used by Solaris/Intel
2290     StubRoutines::_atomic_xchg_entry            = generate_atomic_xchg();
2291 
2292     StubRoutines::_handler_for_unsafe_access_entry =
2293       generate_handler_for_unsafe_access();
2294 
2295     // platform dependent
2296     create_control_words();
2297 
2298     StubRoutines::x86::_verify_mxcsr_entry                 = generate_verify_mxcsr();
2299     StubRoutines::x86::_verify_fpu_cntrl_wrd_entry         = generate_verify_fpu_cntrl_wrd();
2300     StubRoutines::_d2i_wrapper                              = generate_d2i_wrapper(T_INT,
2301                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2302     StubRoutines::_d2l_wrapper                              = generate_d2i_wrapper(T_LONG,
2303                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2304   }
2305 
2306 
2307   void generate_all() {
2308     // Generates all stubs and initializes the entry points
2309 
2310     // These entry points require SharedInfo::stack0 to be set up in non-core builds
2311     // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2312     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError),  false);
2313     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError),  false);
2314     StubRoutines::_throw_ArithmeticException_entry         = generate_throw_exception("ArithmeticException throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException),  true);
2315     StubRoutines::_throw_NullPointerException_entry        = generate_throw_exception("NullPointerException throw_exception",         CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2316     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);
2317     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",           CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError),   false);
2318 
2319     //------------------------------------------------------------------------------------------------------------------------
2320     // entry points that are platform specific
2321 
2322     // support for verify_oop (must happen after universe_init)
2323     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop();
2324 
2325     // arraycopy stubs used by compilers
2326     generate_arraycopy_stubs();
2327 
2328     generate_math_stubs();
2329   }
2330 
2331 
2332  public:
2333   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2334     if (all) {
2335       generate_all();
2336     } else {
2337       generate_initial();
2338     }
2339   }
2340 }; // end class declaration
2341 
2342 
2343 void StubGenerator_generate(CodeBuffer* code, bool all) {
2344   StubGenerator g(code, all);
2345 }
--- EOF ---