1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "nativeInst_x86.hpp"
  30 #include "oops/instanceOop.hpp"
  31 #include "oops/method.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "oops/oop.inline.hpp"
  34 #include "prims/methodHandles.hpp"
  35 #include "runtime/frame.inline.hpp"
  36 #include "runtime/handles.inline.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/stubCodeGenerator.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "runtime/thread.inline.hpp"
  41 #ifdef COMPILER2
  42 #include "opto/runtime.hpp"
  43 #endif
  44 
  45 // Declaration and definition of StubGenerator (no .hpp file).
  46 // For a more detailed description of the stub routine structure
  47 // see the comment in stubRoutines.hpp
  48 
  49 #define __ _masm->
  50 #define a__ ((Assembler*)_masm)->
  51 
  52 #ifdef PRODUCT
  53 #define BLOCK_COMMENT(str) /* nothing */
  54 #else
  55 #define BLOCK_COMMENT(str) __ block_comment(str)
  56 #endif
  57 
  58 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  59 
  60 const int MXCSR_MASK  = 0xFFC0;  // Mask out any pending exceptions
  61 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
  62 
  63 // -------------------------------------------------------------------------------------------------------------------------
  64 // Stub Code definitions
  65 
  66 class StubGenerator: public StubCodeGenerator {
  67  private:
  68 
  69 #ifdef PRODUCT
  70 #define inc_counter_np(counter) ((void)0)
  71 #else
  72   void inc_counter_np_(int& counter) {
  73     __ incrementl(ExternalAddress((address)&counter));
  74   }
  75 #define inc_counter_np(counter) \
  76   BLOCK_COMMENT("inc_counter " #counter); \
  77   inc_counter_np_(counter);
  78 #endif //PRODUCT
  79 
  80   void inc_copy_counter_np(BasicType t) {
  81 #ifndef PRODUCT
  82     switch (t) {
  83     case T_BYTE:    inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
  84     case T_SHORT:   inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
  85     case T_INT:     inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
  86     case T_LONG:    inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
  87     case T_OBJECT:  inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
  88     }
  89     ShouldNotReachHere();
  90 #endif //PRODUCT
  91   }
  92 
  93   //------------------------------------------------------------------------------------------------------------------------
  94   // Call stubs are used to call Java from C
  95   //
  96   //    [ return_from_Java     ] <--- rsp
  97   //    [ argument word n      ]
  98   //      ...
  99   // -N [ argument word 1      ]
 100   // -7 [ Possible padding for stack alignment ]
 101   // -6 [ Possible padding for stack alignment ]
 102   // -5 [ Possible padding for stack alignment ]
 103   // -4 [ mxcsr save           ] <--- rsp_after_call
 104   // -3 [ saved rbx,            ]
 105   // -2 [ saved rsi            ]
 106   // -1 [ saved rdi            ]
 107   //  0 [ saved rbp,            ] <--- rbp,
 108   //  1 [ return address       ]
 109   //  2 [ ptr. to call wrapper ]
 110   //  3 [ result               ]
 111   //  4 [ result_type          ]
 112   //  5 [ method               ]
 113   //  6 [ entry_point          ]
 114   //  7 [ parameters           ]
 115   //  8 [ parameter_size       ]
 116   //  9 [ thread               ]
 117 
 118 
 119   address generate_call_stub(address& return_address) {
 120     StubCodeMark mark(this, "StubRoutines", "call_stub");
 121     address start = __ pc();
 122 
 123     // stub code parameters / addresses
 124     assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
 125     bool  sse_save = false;
 126     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
 127     const int     locals_count_in_bytes  (4*wordSize);
 128     const Address mxcsr_save    (rbp, -4 * wordSize);
 129     const Address saved_rbx     (rbp, -3 * wordSize);
 130     const Address saved_rsi     (rbp, -2 * wordSize);
 131     const Address saved_rdi     (rbp, -1 * wordSize);
 132     const Address result        (rbp,  3 * wordSize);
 133     const Address result_type   (rbp,  4 * wordSize);
 134     const Address method        (rbp,  5 * wordSize);
 135     const Address entry_point   (rbp,  6 * wordSize);
 136     const Address parameters    (rbp,  7 * wordSize);
 137     const Address parameter_size(rbp,  8 * wordSize);
 138     const Address thread        (rbp,  9 * wordSize); // same as in generate_catch_exception()!
 139     sse_save =  UseSSE > 0;
 140 
 141     // stub code
 142     __ enter();
 143     __ movptr(rcx, parameter_size);              // parameter counter
 144     __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
 145     __ addptr(rcx, locals_count_in_bytes);       // reserve space for register saves
 146     __ subptr(rsp, rcx);
 147     __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
 148 
 149     // save rdi, rsi, & rbx, according to C calling conventions
 150     __ movptr(saved_rdi, rdi);
 151     __ movptr(saved_rsi, rsi);
 152     __ movptr(saved_rbx, rbx);
 153 
 154     // provide initial value for required masks
 155     if (UseAVX > 2) {
 156       __ movl(rbx, 0xffff);
 157       __ kmovwl(k1, rbx);
 158     }
 159 
 160     // save and initialize %mxcsr
 161     if (sse_save) {
 162       Label skip_ldmx;
 163       __ stmxcsr(mxcsr_save);
 164       __ movl(rax, mxcsr_save);
 165       __ andl(rax, MXCSR_MASK);    // Only check control and mask bits
 166       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 167       __ cmp32(rax, mxcsr_std);
 168       __ jcc(Assembler::equal, skip_ldmx);
 169       __ ldmxcsr(mxcsr_std);
 170       __ bind(skip_ldmx);
 171     }
 172 
 173     // make sure the control word is correct.
 174     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 175 
 176 #ifdef ASSERT
 177     // make sure we have no pending exceptions
 178     { Label L;
 179       __ movptr(rcx, thread);
 180       __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 181       __ jcc(Assembler::equal, L);
 182       __ stop("StubRoutines::call_stub: entered with pending exception");
 183       __ bind(L);
 184     }
 185 #endif
 186 
 187     // pass parameters if any
 188     BLOCK_COMMENT("pass parameters if any");
 189     Label parameters_done;
 190     __ movl(rcx, parameter_size);  // parameter counter
 191     __ testl(rcx, rcx);
 192     __ jcc(Assembler::zero, parameters_done);
 193 
 194     // parameter passing loop
 195 
 196     Label loop;
 197     // Copy Java parameters in reverse order (receiver last)
 198     // Note that the argument order is inverted in the process
 199     // source is rdx[rcx: N-1..0]
 200     // dest   is rsp[rbx: 0..N-1]
 201 
 202     __ movptr(rdx, parameters);          // parameter pointer
 203     __ xorptr(rbx, rbx);
 204 
 205     __ BIND(loop);
 206 
 207     // get parameter
 208     __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
 209     __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
 210                     Interpreter::expr_offset_in_bytes(0)), rax);          // store parameter
 211     __ increment(rbx);
 212     __ decrement(rcx);
 213     __ jcc(Assembler::notZero, loop);
 214 
 215     // call Java function
 216     __ BIND(parameters_done);
 217     __ movptr(rbx, method);           // get Method*
 218     __ movptr(rax, entry_point);      // get entry_point
 219     __ mov(rsi, rsp);                 // set sender sp
 220     BLOCK_COMMENT("call Java function");
 221     __ call(rax);
 222 
 223     BLOCK_COMMENT("call_stub_return_address:");
 224     return_address = __ pc();
 225 
 226 #ifdef COMPILER2
 227     {
 228       Label L_skip;
 229       if (UseSSE >= 2) {
 230         __ verify_FPU(0, "call_stub_return");
 231       } else {
 232         for (int i = 1; i < 8; i++) {
 233           __ ffree(i);
 234         }
 235 
 236         // UseSSE <= 1 so double result should be left on TOS
 237         __ movl(rsi, result_type);
 238         __ cmpl(rsi, T_DOUBLE);
 239         __ jcc(Assembler::equal, L_skip);
 240         if (UseSSE == 0) {
 241           // UseSSE == 0 so float result should be left on TOS
 242           __ cmpl(rsi, T_FLOAT);
 243           __ jcc(Assembler::equal, L_skip);
 244         }
 245         __ ffree(0);
 246       }
 247       __ BIND(L_skip);
 248     }
 249 #endif // COMPILER2
 250 
 251     // store result depending on type
 252     // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 253     __ movptr(rdi, result);
 254     Label is_long, is_float, is_double, exit;
 255     __ movl(rsi, result_type);
 256     __ cmpl(rsi, T_LONG);
 257     __ jcc(Assembler::equal, is_long);
 258     __ cmpl(rsi, T_FLOAT);
 259     __ jcc(Assembler::equal, is_float);
 260     __ cmpl(rsi, T_DOUBLE);
 261     __ jcc(Assembler::equal, is_double);
 262 
 263     // handle T_INT case
 264     __ movl(Address(rdi, 0), rax);
 265     __ BIND(exit);
 266 
 267     // check that FPU stack is empty
 268     __ verify_FPU(0, "generate_call_stub");
 269 
 270     // pop parameters
 271     __ lea(rsp, rsp_after_call);
 272 
 273     // restore %mxcsr
 274     if (sse_save) {
 275       __ ldmxcsr(mxcsr_save);
 276     }
 277 
 278     // restore rdi, rsi and rbx,
 279     __ movptr(rbx, saved_rbx);
 280     __ movptr(rsi, saved_rsi);
 281     __ movptr(rdi, saved_rdi);
 282     __ addptr(rsp, 4*wordSize);
 283 
 284     // return
 285     __ pop(rbp);
 286     __ ret(0);
 287 
 288     // handle return types different from T_INT
 289     __ BIND(is_long);
 290     __ movl(Address(rdi, 0 * wordSize), rax);
 291     __ movl(Address(rdi, 1 * wordSize), rdx);
 292     __ jmp(exit);
 293 
 294     __ BIND(is_float);
 295     // interpreter uses xmm0 for return values
 296     if (UseSSE >= 1) {
 297       __ movflt(Address(rdi, 0), xmm0);
 298     } else {
 299       __ fstp_s(Address(rdi, 0));
 300     }
 301     __ jmp(exit);
 302 
 303     __ BIND(is_double);
 304     // interpreter uses xmm0 for return values
 305     if (UseSSE >= 2) {
 306       __ movdbl(Address(rdi, 0), xmm0);
 307     } else {
 308       __ fstp_d(Address(rdi, 0));
 309     }
 310     __ jmp(exit);
 311 
 312     return start;
 313   }
 314 
 315 
 316   //------------------------------------------------------------------------------------------------------------------------
 317   // Return point for a Java call if there's an exception thrown in Java code.
 318   // The exception is caught and transformed into a pending exception stored in
 319   // JavaThread that can be tested from within the VM.
 320   //
 321   // Note: Usually the parameters are removed by the callee. In case of an exception
 322   //       crossing an activation frame boundary, that is not the case if the callee
 323   //       is compiled code => need to setup the rsp.
 324   //
 325   // rax,: exception oop
 326 
 327   address generate_catch_exception() {
 328     StubCodeMark mark(this, "StubRoutines", "catch_exception");
 329     const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
 330     const Address thread        (rbp,  9 * wordSize); // same as in generate_call_stub()!
 331     address start = __ pc();
 332 
 333     // get thread directly
 334     __ movptr(rcx, thread);
 335 #ifdef ASSERT
 336     // verify that threads correspond
 337     { Label L;
 338       __ get_thread(rbx);
 339       __ cmpptr(rbx, rcx);
 340       __ jcc(Assembler::equal, L);
 341       __ stop("StubRoutines::catch_exception: threads must correspond");
 342       __ bind(L);
 343     }
 344 #endif
 345     // set pending exception
 346     __ verify_oop(rax);
 347     __ movptr(Address(rcx, Thread::pending_exception_offset()), rax          );
 348     __ lea(Address(rcx, Thread::exception_file_offset   ()),
 349            ExternalAddress((address)__FILE__));
 350     __ movl(Address(rcx, Thread::exception_line_offset   ()), __LINE__ );
 351     // complete return to VM
 352     assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
 353     __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 354 
 355     return start;
 356   }
 357 
 358 
 359   //------------------------------------------------------------------------------------------------------------------------
 360   // Continuation point for runtime calls returning with a pending exception.
 361   // The pending exception check happened in the runtime or native call stub.
 362   // The pending exception in Thread is converted into a Java-level exception.
 363   //
 364   // Contract with Java-level exception handlers:
 365   // rax: exception
 366   // rdx: throwing pc
 367   //
 368   // NOTE: At entry of this stub, exception-pc must be on stack !!
 369 
 370   address generate_forward_exception() {
 371     StubCodeMark mark(this, "StubRoutines", "forward exception");
 372     address start = __ pc();
 373     const Register thread = rcx;
 374 
 375     // other registers used in this stub
 376     const Register exception_oop = rax;
 377     const Register handler_addr  = rbx;
 378     const Register exception_pc  = rdx;
 379 
 380     // Upon entry, the sp points to the return address returning into Java
 381     // (interpreted or compiled) code; i.e., the return address becomes the
 382     // throwing pc.
 383     //
 384     // Arguments pushed before the runtime call are still on the stack but
 385     // the exception handler will reset the stack pointer -> ignore them.
 386     // A potential result in registers can be ignored as well.
 387 
 388 #ifdef ASSERT
 389     // make sure this code is only executed if there is a pending exception
 390     { Label L;
 391       __ get_thread(thread);
 392       __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 393       __ jcc(Assembler::notEqual, L);
 394       __ stop("StubRoutines::forward exception: no pending exception (1)");
 395       __ bind(L);
 396     }
 397 #endif
 398 
 399     // compute exception handler into rbx,
 400     __ get_thread(thread);
 401     __ movptr(exception_pc, Address(rsp, 0));
 402     BLOCK_COMMENT("call exception_handler_for_return_address");
 403     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
 404     __ mov(handler_addr, rax);
 405 
 406     // setup rax & rdx, remove return address & clear pending exception
 407     __ get_thread(thread);
 408     __ pop(exception_pc);
 409     __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
 410     __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 411 
 412 #ifdef ASSERT
 413     // make sure exception is set
 414     { Label L;
 415       __ testptr(exception_oop, exception_oop);
 416       __ jcc(Assembler::notEqual, L);
 417       __ stop("StubRoutines::forward exception: no pending exception (2)");
 418       __ bind(L);
 419     }
 420 #endif
 421 
 422     // Verify that there is really a valid exception in RAX.
 423     __ verify_oop(exception_oop);
 424 
 425     // continue at exception handler (return address removed)
 426     // rax: exception
 427     // rbx: exception handler
 428     // rdx: throwing pc
 429     __ jmp(handler_addr);
 430 
 431     return start;
 432   }
 433 
 434 
 435   //----------------------------------------------------------------------------------------------------
 436   // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
 437   //
 438   // xchg exists as far back as 8086, lock needed for MP only
 439   // Stack layout immediately after call:
 440   //
 441   // 0 [ret addr ] <--- rsp
 442   // 1 [  ex     ]
 443   // 2 [  dest   ]
 444   //
 445   // Result:   *dest <- ex, return (old *dest)
 446   //
 447   // Note: win32 does not currently use this code
 448 
 449   address generate_atomic_xchg() {
 450     StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
 451     address start = __ pc();
 452 
 453     __ push(rdx);
 454     Address exchange(rsp, 2 * wordSize);
 455     Address dest_addr(rsp, 3 * wordSize);
 456     __ movl(rax, exchange);
 457     __ movptr(rdx, dest_addr);
 458     __ xchgl(rax, Address(rdx, 0));
 459     __ pop(rdx);
 460     __ ret(0);
 461 
 462     return start;
 463   }
 464 
 465   //----------------------------------------------------------------------------------------------------
 466   // Support for void verify_mxcsr()
 467   //
 468   // This routine is used with -Xcheck:jni to verify that native
 469   // JNI code does not return to Java code without restoring the
 470   // MXCSR register to our expected state.
 471 
 472 
 473   address generate_verify_mxcsr() {
 474     StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 475     address start = __ pc();
 476 
 477     const Address mxcsr_save(rsp, 0);
 478 
 479     if (CheckJNICalls && UseSSE > 0 ) {
 480       Label ok_ret;
 481       ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
 482       __ push(rax);
 483       __ subptr(rsp, wordSize);      // allocate a temp location
 484       __ stmxcsr(mxcsr_save);
 485       __ movl(rax, mxcsr_save);
 486       __ andl(rax, MXCSR_MASK);
 487       __ cmp32(rax, mxcsr_std);
 488       __ jcc(Assembler::equal, ok_ret);
 489 
 490       __ warn("MXCSR changed by native JNI code.");
 491 
 492       __ ldmxcsr(mxcsr_std);
 493 
 494       __ bind(ok_ret);
 495       __ addptr(rsp, wordSize);
 496       __ pop(rax);
 497     }
 498 
 499     __ ret(0);
 500 
 501     return start;
 502   }
 503 
 504 
 505   //---------------------------------------------------------------------------
 506   // Support for void verify_fpu_cntrl_wrd()
 507   //
 508   // This routine is used with -Xcheck:jni to verify that native
 509   // JNI code does not return to Java code without restoring the
 510   // FP control word to our expected state.
 511 
 512   address generate_verify_fpu_cntrl_wrd() {
 513     StubCodeMark mark(this, "StubRoutines", "verify_spcw");
 514     address start = __ pc();
 515 
 516     const Address fpu_cntrl_wrd_save(rsp, 0);
 517 
 518     if (CheckJNICalls) {
 519       Label ok_ret;
 520       __ push(rax);
 521       __ subptr(rsp, wordSize);      // allocate a temp location
 522       __ fnstcw(fpu_cntrl_wrd_save);
 523       __ movl(rax, fpu_cntrl_wrd_save);
 524       __ andl(rax, FPU_CNTRL_WRD_MASK);
 525       ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
 526       __ cmp32(rax, fpu_std);
 527       __ jcc(Assembler::equal, ok_ret);
 528 
 529       __ warn("Floating point control word changed by native JNI code.");
 530 
 531       __ fldcw(fpu_std);
 532 
 533       __ bind(ok_ret);
 534       __ addptr(rsp, wordSize);
 535       __ pop(rax);
 536     }
 537 
 538     __ ret(0);
 539 
 540     return start;
 541   }
 542 
 543   //---------------------------------------------------------------------------
 544   // Wrapper for slow-case handling of double-to-integer conversion
 545   // d2i or f2i fast case failed either because it is nan or because
 546   // of under/overflow.
 547   // Input:  FPU TOS: float value
 548   // Output: rax, (rdx): integer (long) result
 549 
 550   address generate_d2i_wrapper(BasicType t, address fcn) {
 551     StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
 552     address start = __ pc();
 553 
 554   // Capture info about frame layout
 555   enum layout { FPUState_off         = 0,
 556                 rbp_off              = FPUStateSizeInWords,
 557                 rdi_off,
 558                 rsi_off,
 559                 rcx_off,
 560                 rbx_off,
 561                 saved_argument_off,
 562                 saved_argument_off2, // 2nd half of double
 563                 framesize
 564   };
 565 
 566   assert(FPUStateSizeInWords == 27, "update stack layout");
 567 
 568     // Save outgoing argument to stack across push_FPU_state()
 569     __ subptr(rsp, wordSize * 2);
 570     __ fstp_d(Address(rsp, 0));
 571 
 572     // Save CPU & FPU state
 573     __ push(rbx);
 574     __ push(rcx);
 575     __ push(rsi);
 576     __ push(rdi);
 577     __ push(rbp);
 578     __ push_FPU_state();
 579 
 580     // push_FPU_state() resets the FP top of stack
 581     // Load original double into FP top of stack
 582     __ fld_d(Address(rsp, saved_argument_off * wordSize));
 583     // Store double into stack as outgoing argument
 584     __ subptr(rsp, wordSize*2);
 585     __ fst_d(Address(rsp, 0));
 586 
 587     // Prepare FPU for doing math in C-land
 588     __ empty_FPU_stack();
 589     // Call the C code to massage the double.  Result in EAX
 590     if (t == T_INT)
 591       { BLOCK_COMMENT("SharedRuntime::d2i"); }
 592     else if (t == T_LONG)
 593       { BLOCK_COMMENT("SharedRuntime::d2l"); }
 594     __ call_VM_leaf( fcn, 2 );
 595 
 596     // Restore CPU & FPU state
 597     __ pop_FPU_state();
 598     __ pop(rbp);
 599     __ pop(rdi);
 600     __ pop(rsi);
 601     __ pop(rcx);
 602     __ pop(rbx);
 603     __ addptr(rsp, wordSize * 2);
 604 
 605     __ ret(0);
 606 
 607     return start;
 608   }
 609 
 610 
 611   //----------------------------------------------------------------------------------------------------
 612   // Non-destructive plausibility checks for oops
 613 
 614   address generate_verify_oop() {
 615     StubCodeMark mark(this, "StubRoutines", "verify_oop");
 616     address start = __ pc();
 617 
 618     // Incoming arguments on stack after saving rax,:
 619     //
 620     // [tos    ]: saved rdx
 621     // [tos + 1]: saved EFLAGS
 622     // [tos + 2]: return address
 623     // [tos + 3]: char* error message
 624     // [tos + 4]: oop   object to verify
 625     // [tos + 5]: saved rax, - saved by caller and bashed
 626 
 627     Label exit, error;
 628     __ pushf();
 629     __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
 630     __ push(rdx);                                // save rdx
 631     // make sure object is 'reasonable'
 632     __ movptr(rax, Address(rsp, 4 * wordSize));    // get object
 633     __ testptr(rax, rax);
 634     __ jcc(Assembler::zero, exit);               // if obj is NULL it is ok
 635 
 636     // Check if the oop is in the right area of memory
 637     const int oop_mask = Universe::verify_oop_mask();
 638     const int oop_bits = Universe::verify_oop_bits();
 639     __ mov(rdx, rax);
 640     __ andptr(rdx, oop_mask);
 641     __ cmpptr(rdx, oop_bits);
 642     __ jcc(Assembler::notZero, error);
 643 
 644     // make sure klass is 'reasonable', which is not zero.
 645     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
 646     __ testptr(rax, rax);
 647     __ jcc(Assembler::zero, error);              // if klass is NULL it is broken
 648 
 649     // return if everything seems ok
 650     __ bind(exit);
 651     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 652     __ pop(rdx);                                 // restore rdx
 653     __ popf();                                   // restore EFLAGS
 654     __ ret(3 * wordSize);                        // pop arguments
 655 
 656     // handle errors
 657     __ bind(error);
 658     __ movptr(rax, Address(rsp, 5 * wordSize));  // get saved rax, back
 659     __ pop(rdx);                                 // get saved rdx back
 660     __ popf();                                   // get saved EFLAGS off stack -- will be ignored
 661     __ pusha();                                  // push registers (eip = return address & msg are already pushed)
 662     BLOCK_COMMENT("call MacroAssembler::debug");
 663     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
 664     __ popa();
 665     __ ret(3 * wordSize);                        // pop arguments
 666     return start;
 667   }
 668 
 669   //
 670   //  Generate pre-barrier for array stores
 671   //
 672   //  Input:
 673   //     start   -  starting address
 674   //     count   -  element count
 675   void  gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
 676     assert_different_registers(start, count);
 677     BarrierSet* bs = Universe::heap()->barrier_set();
 678     switch (bs->kind()) {
 679       case BarrierSet::G1SATBCTLogging:
 680         // With G1, don't generate the call if we statically know that the target in uninitialized
 681         if (!uninitialized_target) {
 682            __ pusha();                      // push registers
 683            __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
 684                            start, count);
 685            __ popa();
 686          }
 687         break;
 688       case BarrierSet::CardTableForRS:
 689       case BarrierSet::CardTableExtension:
 690       case BarrierSet::ModRef:
 691         break;
 692       default      :
 693         ShouldNotReachHere();
 694 
 695     }
 696   }
 697 
 698 
 699   //
 700   // Generate a post-barrier for an array store
 701   //
 702   //     start    -  starting address
 703   //     count    -  element count
 704   //
 705   //  The two input registers are overwritten.
 706   //
 707   void  gen_write_ref_array_post_barrier(Register start, Register count) {
 708     BarrierSet* bs = Universe::heap()->barrier_set();
 709     assert_different_registers(start, count);
 710     switch (bs->kind()) {
 711       case BarrierSet::G1SATBCTLogging:
 712         {
 713           __ pusha();                      // push registers
 714           __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
 715                           start, count);
 716           __ popa();
 717         }
 718         break;
 719 
 720       case BarrierSet::CardTableForRS:
 721       case BarrierSet::CardTableExtension:
 722         {
 723           CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
 724           assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
 725 
 726           Label L_loop;
 727           const Register end = count;  // elements count; end == start+count-1
 728           assert_different_registers(start, end);
 729 
 730           __ lea(end,  Address(start, count, Address::times_ptr, -wordSize));
 731           __ shrptr(start, CardTableModRefBS::card_shift);
 732           __ shrptr(end,   CardTableModRefBS::card_shift);
 733           __ subptr(end, start); // end --> count
 734         __ BIND(L_loop);
 735           intptr_t disp = (intptr_t) ct->byte_map_base;
 736           Address cardtable(start, count, Address::times_1, disp);
 737           __ movb(cardtable, 0);
 738           __ decrement(count);
 739           __ jcc(Assembler::greaterEqual, L_loop);
 740         }
 741         break;
 742       case BarrierSet::ModRef:
 743         break;
 744       default      :
 745         ShouldNotReachHere();
 746 
 747     }
 748   }
 749 
 750 
 751   // Copy 64 bytes chunks
 752   //
 753   // Inputs:
 754   //   from        - source array address
 755   //   to_from     - destination array address - from
 756   //   qword_count - 8-bytes element count, negative
 757   //
 758   void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
 759     assert( UseSSE >= 2, "supported cpu only" );
 760     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 761     if (UseAVX > 2) {
 762       __ push(rbx);
 763       __ movl(rbx, 0xffff);
 764       __ kmovwl(k1, rbx);
 765       __ pop(rbx);
 766     }
 767     // Copy 64-byte chunks
 768     __ jmpb(L_copy_64_bytes);
 769     __ align(OptoLoopAlignment);
 770   __ BIND(L_copy_64_bytes_loop);
 771 
 772     if (UseUnalignedLoadStores) {
 773       if (UseAVX > 2) {
 774         __ evmovdqul(xmm0, Address(from, 0), Assembler::AVX_512bit);
 775         __ evmovdqul(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
 776       } else if (UseAVX == 2) {
 777         __ vmovdqu(xmm0, Address(from,  0));
 778         __ vmovdqu(Address(from, to_from, Address::times_1,  0), xmm0);
 779         __ vmovdqu(xmm1, Address(from, 32));
 780         __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
 781       } else {
 782         __ movdqu(xmm0, Address(from, 0));
 783         __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
 784         __ movdqu(xmm1, Address(from, 16));
 785         __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
 786         __ movdqu(xmm2, Address(from, 32));
 787         __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
 788         __ movdqu(xmm3, Address(from, 48));
 789         __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
 790       }
 791     } else {
 792       __ movq(xmm0, Address(from, 0));
 793       __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
 794       __ movq(xmm1, Address(from, 8));
 795       __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
 796       __ movq(xmm2, Address(from, 16));
 797       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 798       __ movq(xmm3, Address(from, 24));
 799       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 800       __ movq(xmm4, Address(from, 32));
 801       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 802       __ movq(xmm5, Address(from, 40));
 803       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 804       __ movq(xmm6, Address(from, 48));
 805       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 806       __ movq(xmm7, Address(from, 56));
 807       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 808     }
 809 
 810     __ addl(from, 64);
 811   __ BIND(L_copy_64_bytes);
 812     __ subl(qword_count, 8);
 813     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 814 
 815     if (UseUnalignedLoadStores && (UseAVX == 2)) {
 816       // clean upper bits of YMM registers
 817       __ vpxor(xmm0, xmm0);
 818       __ vpxor(xmm1, xmm1);
 819     }
 820     __ addl(qword_count, 8);
 821     __ jccb(Assembler::zero, L_exit);
 822     //
 823     // length is too short, just copy qwords
 824     //
 825   __ BIND(L_copy_8_bytes);
 826     __ movq(xmm0, Address(from, 0));
 827     __ movq(Address(from, to_from, Address::times_1), xmm0);
 828     __ addl(from, 8);
 829     __ decrement(qword_count);
 830     __ jcc(Assembler::greater, L_copy_8_bytes);
 831   __ BIND(L_exit);
 832   }
 833 
 834   // Copy 64 bytes chunks
 835   //
 836   // Inputs:
 837   //   from        - source array address
 838   //   to_from     - destination array address - from
 839   //   qword_count - 8-bytes element count, negative
 840   //
 841   void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
 842     assert( VM_Version::supports_mmx(), "supported cpu only" );
 843     Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
 844     // Copy 64-byte chunks
 845     __ jmpb(L_copy_64_bytes);
 846     __ align(OptoLoopAlignment);
 847   __ BIND(L_copy_64_bytes_loop);
 848     __ movq(mmx0, Address(from, 0));
 849     __ movq(mmx1, Address(from, 8));
 850     __ movq(mmx2, Address(from, 16));
 851     __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
 852     __ movq(mmx3, Address(from, 24));
 853     __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
 854     __ movq(mmx4, Address(from, 32));
 855     __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
 856     __ movq(mmx5, Address(from, 40));
 857     __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
 858     __ movq(mmx6, Address(from, 48));
 859     __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
 860     __ movq(mmx7, Address(from, 56));
 861     __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
 862     __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
 863     __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
 864     __ addptr(from, 64);
 865   __ BIND(L_copy_64_bytes);
 866     __ subl(qword_count, 8);
 867     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 868     __ addl(qword_count, 8);
 869     __ jccb(Assembler::zero, L_exit);
 870     //
 871     // length is too short, just copy qwords
 872     //
 873   __ BIND(L_copy_8_bytes);
 874     __ movq(mmx0, Address(from, 0));
 875     __ movq(Address(from, to_from, Address::times_1), mmx0);
 876     __ addptr(from, 8);
 877     __ decrement(qword_count);
 878     __ jcc(Assembler::greater, L_copy_8_bytes);
 879   __ BIND(L_exit);
 880     __ emms();
 881   }
 882 
 883   address generate_disjoint_copy(BasicType t, bool aligned,
 884                                  Address::ScaleFactor sf,
 885                                  address* entry, const char *name,
 886                                  bool dest_uninitialized = false) {
 887     __ align(CodeEntryAlignment);
 888     StubCodeMark mark(this, "StubRoutines", name);
 889     address start = __ pc();
 890 
 891     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
 892     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
 893 
 894     int shift = Address::times_ptr - sf;
 895 
 896     const Register from     = rsi;  // source array address
 897     const Register to       = rdi;  // destination array address
 898     const Register count    = rcx;  // elements count
 899     const Register to_from  = to;   // (to - from)
 900     const Register saved_to = rdx;  // saved destination array address
 901 
 902     __ enter(); // required for proper stackwalking of RuntimeStub frame
 903     __ push(rsi);
 904     __ push(rdi);
 905     __ movptr(from , Address(rsp, 12+ 4));
 906     __ movptr(to   , Address(rsp, 12+ 8));
 907     __ movl(count, Address(rsp, 12+ 12));
 908 
 909     if (entry != NULL) {
 910       *entry = __ pc(); // Entry point from conjoint arraycopy stub.
 911       BLOCK_COMMENT("Entry:");
 912     }
 913 
 914     if (t == T_OBJECT) {
 915       __ testl(count, count);
 916       __ jcc(Assembler::zero, L_0_count);
 917       gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
 918       __ mov(saved_to, to);          // save 'to'
 919     }
 920 
 921     __ subptr(to, from); // to --> to_from
 922     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
 923     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
 924     if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 925       // align source address at 4 bytes address boundary
 926       if (t == T_BYTE) {
 927         // One byte misalignment happens only for byte arrays
 928         __ testl(from, 1);
 929         __ jccb(Assembler::zero, L_skip_align1);
 930         __ movb(rax, Address(from, 0));
 931         __ movb(Address(from, to_from, Address::times_1, 0), rax);
 932         __ increment(from);
 933         __ decrement(count);
 934       __ BIND(L_skip_align1);
 935       }
 936       // Two bytes misalignment happens only for byte and short (char) arrays
 937       __ testl(from, 2);
 938       __ jccb(Assembler::zero, L_skip_align2);
 939       __ movw(rax, Address(from, 0));
 940       __ movw(Address(from, to_from, Address::times_1, 0), rax);
 941       __ addptr(from, 2);
 942       __ subl(count, 1<<(shift-1));
 943     __ BIND(L_skip_align2);
 944     }
 945     if (!VM_Version::supports_mmx()) {
 946       __ mov(rax, count);      // save 'count'
 947       __ shrl(count, shift); // bytes count
 948       __ addptr(to_from, from);// restore 'to'
 949       __ rep_mov();
 950       __ subptr(to_from, from);// restore 'to_from'
 951       __ mov(count, rax);      // restore 'count'
 952       __ jmpb(L_copy_2_bytes); // all dwords were copied
 953     } else {
 954       if (!UseUnalignedLoadStores) {
 955         // align to 8 bytes, we know we are 4 byte aligned to start
 956         __ testptr(from, 4);
 957         __ jccb(Assembler::zero, L_copy_64_bytes);
 958         __ movl(rax, Address(from, 0));
 959         __ movl(Address(from, to_from, Address::times_1, 0), rax);
 960         __ addptr(from, 4);
 961         __ subl(count, 1<<shift);
 962       }
 963     __ BIND(L_copy_64_bytes);
 964       __ mov(rax, count);
 965       __ shrl(rax, shift+1);  // 8 bytes chunk count
 966       //
 967       // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
 968       //
 969       if (UseXMMForArrayCopy) {
 970         xmm_copy_forward(from, to_from, rax);
 971       } else {
 972         mmx_copy_forward(from, to_from, rax);
 973       }
 974     }
 975     // copy tailing dword
 976   __ BIND(L_copy_4_bytes);
 977     __ testl(count, 1<<shift);
 978     __ jccb(Assembler::zero, L_copy_2_bytes);
 979     __ movl(rax, Address(from, 0));
 980     __ movl(Address(from, to_from, Address::times_1, 0), rax);
 981     if (t == T_BYTE || t == T_SHORT) {
 982       __ addptr(from, 4);
 983     __ BIND(L_copy_2_bytes);
 984       // copy tailing word
 985       __ testl(count, 1<<(shift-1));
 986       __ jccb(Assembler::zero, L_copy_byte);
 987       __ movw(rax, Address(from, 0));
 988       __ movw(Address(from, to_from, Address::times_1, 0), rax);
 989       if (t == T_BYTE) {
 990         __ addptr(from, 2);
 991       __ BIND(L_copy_byte);
 992         // copy tailing byte
 993         __ testl(count, 1);
 994         __ jccb(Assembler::zero, L_exit);
 995         __ movb(rax, Address(from, 0));
 996         __ movb(Address(from, to_from, Address::times_1, 0), rax);
 997       __ BIND(L_exit);
 998       } else {
 999       __ BIND(L_copy_byte);
1000       }
1001     } else {
1002     __ BIND(L_copy_2_bytes);
1003     }
1004 
1005     if (t == T_OBJECT) {
1006       __ movl(count, Address(rsp, 12+12)); // reread 'count'
1007       __ mov(to, saved_to); // restore 'to'
1008       gen_write_ref_array_post_barrier(to, count);
1009     __ BIND(L_0_count);
1010     }
1011     inc_copy_counter_np(t);
1012     __ pop(rdi);
1013     __ pop(rsi);
1014     __ leave(); // required for proper stackwalking of RuntimeStub frame
1015     __ xorptr(rax, rax); // return 0
1016     __ ret(0);
1017     return start;
1018   }
1019 
1020 
1021   address generate_fill(BasicType t, bool aligned, const char *name) {
1022     __ align(CodeEntryAlignment);
1023     StubCodeMark mark(this, "StubRoutines", name);
1024     address start = __ pc();
1025 
1026     BLOCK_COMMENT("Entry:");
1027 
1028     const Register to       = rdi;  // source array address
1029     const Register value    = rdx;  // value
1030     const Register count    = rsi;  // elements count
1031 
1032     __ enter(); // required for proper stackwalking of RuntimeStub frame
1033     __ push(rsi);
1034     __ push(rdi);
1035     __ movptr(to   , Address(rsp, 12+ 4));
1036     __ movl(value, Address(rsp, 12+ 8));
1037     __ movl(count, Address(rsp, 12+ 12));
1038 
1039     __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1040 
1041     __ pop(rdi);
1042     __ pop(rsi);
1043     __ leave(); // required for proper stackwalking of RuntimeStub frame
1044     __ ret(0);
1045     return start;
1046   }
1047 
1048   address generate_conjoint_copy(BasicType t, bool aligned,
1049                                  Address::ScaleFactor sf,
1050                                  address nooverlap_target,
1051                                  address* entry, const char *name,
1052                                  bool dest_uninitialized = false) {
1053     __ align(CodeEntryAlignment);
1054     StubCodeMark mark(this, "StubRoutines", name);
1055     address start = __ pc();
1056 
1057     Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1058     Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1059 
1060     int shift = Address::times_ptr - sf;
1061 
1062     const Register src   = rax;  // source array address
1063     const Register dst   = rdx;  // destination array address
1064     const Register from  = rsi;  // source array address
1065     const Register to    = rdi;  // destination array address
1066     const Register count = rcx;  // elements count
1067     const Register end   = rax;  // array end address
1068 
1069     __ enter(); // required for proper stackwalking of RuntimeStub frame
1070     __ push(rsi);
1071     __ push(rdi);
1072     __ movptr(src  , Address(rsp, 12+ 4));   // from
1073     __ movptr(dst  , Address(rsp, 12+ 8));   // to
1074     __ movl2ptr(count, Address(rsp, 12+12)); // count
1075 
1076     if (entry != NULL) {
1077       *entry = __ pc(); // Entry point from generic arraycopy stub.
1078       BLOCK_COMMENT("Entry:");
1079     }
1080 
1081     // nooverlap_target expects arguments in rsi and rdi.
1082     __ mov(from, src);
1083     __ mov(to  , dst);
1084 
1085     // arrays overlap test: dispatch to disjoint stub if necessary.
1086     RuntimeAddress nooverlap(nooverlap_target);
1087     __ cmpptr(dst, src);
1088     __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1089     __ jump_cc(Assembler::belowEqual, nooverlap);
1090     __ cmpptr(dst, end);
1091     __ jump_cc(Assembler::aboveEqual, nooverlap);
1092 
1093     if (t == T_OBJECT) {
1094       __ testl(count, count);
1095       __ jcc(Assembler::zero, L_0_count);
1096       gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1097     }
1098 
1099     // copy from high to low
1100     __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1101     __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1102     if (t == T_BYTE || t == T_SHORT) {
1103       // Align the end of destination array at 4 bytes address boundary
1104       __ lea(end, Address(dst, count, sf, 0));
1105       if (t == T_BYTE) {
1106         // One byte misalignment happens only for byte arrays
1107         __ testl(end, 1);
1108         __ jccb(Assembler::zero, L_skip_align1);
1109         __ decrement(count);
1110         __ movb(rdx, Address(from, count, sf, 0));
1111         __ movb(Address(to, count, sf, 0), rdx);
1112       __ BIND(L_skip_align1);
1113       }
1114       // Two bytes misalignment happens only for byte and short (char) arrays
1115       __ testl(end, 2);
1116       __ jccb(Assembler::zero, L_skip_align2);
1117       __ subptr(count, 1<<(shift-1));
1118       __ movw(rdx, Address(from, count, sf, 0));
1119       __ movw(Address(to, count, sf, 0), rdx);
1120     __ BIND(L_skip_align2);
1121       __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1122       __ jcc(Assembler::below, L_copy_4_bytes);
1123     }
1124 
1125     if (!VM_Version::supports_mmx()) {
1126       __ std();
1127       __ mov(rax, count); // Save 'count'
1128       __ mov(rdx, to);    // Save 'to'
1129       __ lea(rsi, Address(from, count, sf, -4));
1130       __ lea(rdi, Address(to  , count, sf, -4));
1131       __ shrptr(count, shift); // bytes count
1132       __ rep_mov();
1133       __ cld();
1134       __ mov(count, rax); // restore 'count'
1135       __ andl(count, (1<<shift)-1);      // mask the number of rest elements
1136       __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1137       __ mov(to, rdx);   // restore 'to'
1138       __ jmpb(L_copy_2_bytes); // all dword were copied
1139    } else {
1140       // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1141       __ testptr(end, 4);
1142       __ jccb(Assembler::zero, L_copy_8_bytes);
1143       __ subl(count, 1<<shift);
1144       __ movl(rdx, Address(from, count, sf, 0));
1145       __ movl(Address(to, count, sf, 0), rdx);
1146       __ jmpb(L_copy_8_bytes);
1147 
1148       __ align(OptoLoopAlignment);
1149       // Move 8 bytes
1150     __ BIND(L_copy_8_bytes_loop);
1151       if (UseXMMForArrayCopy) {
1152         __ movq(xmm0, Address(from, count, sf, 0));
1153         __ movq(Address(to, count, sf, 0), xmm0);
1154       } else {
1155         __ movq(mmx0, Address(from, count, sf, 0));
1156         __ movq(Address(to, count, sf, 0), mmx0);
1157       }
1158     __ BIND(L_copy_8_bytes);
1159       __ subl(count, 2<<shift);
1160       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1161       __ addl(count, 2<<shift);
1162       if (!UseXMMForArrayCopy) {
1163         __ emms();
1164       }
1165     }
1166   __ BIND(L_copy_4_bytes);
1167     // copy prefix qword
1168     __ testl(count, 1<<shift);
1169     __ jccb(Assembler::zero, L_copy_2_bytes);
1170     __ movl(rdx, Address(from, count, sf, -4));
1171     __ movl(Address(to, count, sf, -4), rdx);
1172 
1173     if (t == T_BYTE || t == T_SHORT) {
1174         __ subl(count, (1<<shift));
1175       __ BIND(L_copy_2_bytes);
1176         // copy prefix dword
1177         __ testl(count, 1<<(shift-1));
1178         __ jccb(Assembler::zero, L_copy_byte);
1179         __ movw(rdx, Address(from, count, sf, -2));
1180         __ movw(Address(to, count, sf, -2), rdx);
1181         if (t == T_BYTE) {
1182           __ subl(count, 1<<(shift-1));
1183         __ BIND(L_copy_byte);
1184           // copy prefix byte
1185           __ testl(count, 1);
1186           __ jccb(Assembler::zero, L_exit);
1187           __ movb(rdx, Address(from, 0));
1188           __ movb(Address(to, 0), rdx);
1189         __ BIND(L_exit);
1190         } else {
1191         __ BIND(L_copy_byte);
1192         }
1193     } else {
1194     __ BIND(L_copy_2_bytes);
1195     }
1196     if (t == T_OBJECT) {
1197       __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1198       gen_write_ref_array_post_barrier(to, count);
1199     __ BIND(L_0_count);
1200     }
1201     inc_copy_counter_np(t);
1202     __ pop(rdi);
1203     __ pop(rsi);
1204     __ leave(); // required for proper stackwalking of RuntimeStub frame
1205     __ xorptr(rax, rax); // return 0
1206     __ ret(0);
1207     return start;
1208   }
1209 
1210 
1211   address generate_disjoint_long_copy(address* entry, const char *name) {
1212     __ align(CodeEntryAlignment);
1213     StubCodeMark mark(this, "StubRoutines", name);
1214     address start = __ pc();
1215 
1216     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1217     const Register from       = rax;  // source array address
1218     const Register to         = rdx;  // destination array address
1219     const Register count      = rcx;  // elements count
1220     const Register to_from    = rdx;  // (to - from)
1221 
1222     __ enter(); // required for proper stackwalking of RuntimeStub frame
1223     __ movptr(from , Address(rsp, 8+0));       // from
1224     __ movptr(to   , Address(rsp, 8+4));       // to
1225     __ movl2ptr(count, Address(rsp, 8+8));     // count
1226 
1227     *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1228     BLOCK_COMMENT("Entry:");
1229 
1230     __ subptr(to, from); // to --> to_from
1231     if (VM_Version::supports_mmx()) {
1232       if (UseXMMForArrayCopy) {
1233         xmm_copy_forward(from, to_from, count);
1234       } else {
1235         mmx_copy_forward(from, to_from, count);
1236       }
1237     } else {
1238       __ jmpb(L_copy_8_bytes);
1239       __ align(OptoLoopAlignment);
1240     __ BIND(L_copy_8_bytes_loop);
1241       __ fild_d(Address(from, 0));
1242       __ fistp_d(Address(from, to_from, Address::times_1));
1243       __ addptr(from, 8);
1244     __ BIND(L_copy_8_bytes);
1245       __ decrement(count);
1246       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1247     }
1248     inc_copy_counter_np(T_LONG);
1249     __ leave(); // required for proper stackwalking of RuntimeStub frame
1250     __ xorptr(rax, rax); // return 0
1251     __ ret(0);
1252     return start;
1253   }
1254 
1255   address generate_conjoint_long_copy(address nooverlap_target,
1256                                       address* entry, const char *name) {
1257     __ align(CodeEntryAlignment);
1258     StubCodeMark mark(this, "StubRoutines", name);
1259     address start = __ pc();
1260 
1261     Label L_copy_8_bytes, L_copy_8_bytes_loop;
1262     const Register from       = rax;  // source array address
1263     const Register to         = rdx;  // destination array address
1264     const Register count      = rcx;  // elements count
1265     const Register end_from   = rax;  // source array end address
1266 
1267     __ enter(); // required for proper stackwalking of RuntimeStub frame
1268     __ movptr(from , Address(rsp, 8+0));       // from
1269     __ movptr(to   , Address(rsp, 8+4));       // to
1270     __ movl2ptr(count, Address(rsp, 8+8));     // count
1271 
1272     *entry = __ pc(); // Entry point from generic arraycopy stub.
1273     BLOCK_COMMENT("Entry:");
1274 
1275     // arrays overlap test
1276     __ cmpptr(to, from);
1277     RuntimeAddress nooverlap(nooverlap_target);
1278     __ jump_cc(Assembler::belowEqual, nooverlap);
1279     __ lea(end_from, Address(from, count, Address::times_8, 0));
1280     __ cmpptr(to, end_from);
1281     __ movptr(from, Address(rsp, 8));  // from
1282     __ jump_cc(Assembler::aboveEqual, nooverlap);
1283 
1284     __ jmpb(L_copy_8_bytes);
1285 
1286     __ align(OptoLoopAlignment);
1287   __ BIND(L_copy_8_bytes_loop);
1288     if (VM_Version::supports_mmx()) {
1289       if (UseXMMForArrayCopy) {
1290         __ movq(xmm0, Address(from, count, Address::times_8));
1291         __ movq(Address(to, count, Address::times_8), xmm0);
1292       } else {
1293         __ movq(mmx0, Address(from, count, Address::times_8));
1294         __ movq(Address(to, count, Address::times_8), mmx0);
1295       }
1296     } else {
1297       __ fild_d(Address(from, count, Address::times_8));
1298       __ fistp_d(Address(to, count, Address::times_8));
1299     }
1300   __ BIND(L_copy_8_bytes);
1301     __ decrement(count);
1302     __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1303 
1304     if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1305       __ emms();
1306     }
1307     inc_copy_counter_np(T_LONG);
1308     __ leave(); // required for proper stackwalking of RuntimeStub frame
1309     __ xorptr(rax, rax); // return 0
1310     __ ret(0);
1311     return start;
1312   }
1313 
1314 
1315   // Helper for generating a dynamic type check.
1316   // The sub_klass must be one of {rbx, rdx, rsi}.
1317   // The temp is killed.
1318   void generate_type_check(Register sub_klass,
1319                            Address& super_check_offset_addr,
1320                            Address& super_klass_addr,
1321                            Register temp,
1322                            Label* L_success, Label* L_failure) {
1323     BLOCK_COMMENT("type_check:");
1324 
1325     Label L_fallthrough;
1326 #define LOCAL_JCC(assembler_con, label_ptr)                             \
1327     if (label_ptr != NULL)  __ jcc(assembler_con, *(label_ptr));        \
1328     else                    __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1329 
1330     // The following is a strange variation of the fast path which requires
1331     // one less register, because needed values are on the argument stack.
1332     // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1333     //                                  L_success, L_failure, NULL);
1334     assert_different_registers(sub_klass, temp);
1335 
1336     int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1337 
1338     // if the pointers are equal, we are done (e.g., String[] elements)
1339     __ cmpptr(sub_klass, super_klass_addr);
1340     LOCAL_JCC(Assembler::equal, L_success);
1341 
1342     // check the supertype display:
1343     __ movl2ptr(temp, super_check_offset_addr);
1344     Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1345     __ movptr(temp, super_check_addr); // load displayed supertype
1346     __ cmpptr(temp, super_klass_addr); // test the super type
1347     LOCAL_JCC(Assembler::equal, L_success);
1348 
1349     // if it was a primary super, we can just fail immediately
1350     __ cmpl(super_check_offset_addr, sc_offset);
1351     LOCAL_JCC(Assembler::notEqual, L_failure);
1352 
1353     // The repne_scan instruction uses fixed registers, which will get spilled.
1354     // We happen to know this works best when super_klass is in rax.
1355     Register super_klass = temp;
1356     __ movptr(super_klass, super_klass_addr);
1357     __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1358                                      L_success, L_failure);
1359 
1360     __ bind(L_fallthrough);
1361 
1362     if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1363     if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1364 
1365 #undef LOCAL_JCC
1366   }
1367 
1368   //
1369   //  Generate checkcasting array copy stub
1370   //
1371   //  Input:
1372   //    4(rsp)   - source array address
1373   //    8(rsp)   - destination array address
1374   //   12(rsp)   - element count, can be zero
1375   //   16(rsp)   - size_t ckoff (super_check_offset)
1376   //   20(rsp)   - oop ckval (super_klass)
1377   //
1378   //  Output:
1379   //    rax, ==  0  -  success
1380   //    rax, == -1^K - failure, where K is partial transfer count
1381   //
1382   address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1383     __ align(CodeEntryAlignment);
1384     StubCodeMark mark(this, "StubRoutines", name);
1385     address start = __ pc();
1386 
1387     Label L_load_element, L_store_element, L_do_card_marks, L_done;
1388 
1389     // register use:
1390     //  rax, rdx, rcx -- loop control (end_from, end_to, count)
1391     //  rdi, rsi      -- element access (oop, klass)
1392     //  rbx,           -- temp
1393     const Register from       = rax;    // source array address
1394     const Register to         = rdx;    // destination array address
1395     const Register length     = rcx;    // elements count
1396     const Register elem       = rdi;    // each oop copied
1397     const Register elem_klass = rsi;    // each elem._klass (sub_klass)
1398     const Register temp       = rbx;    // lone remaining temp
1399 
1400     __ enter(); // required for proper stackwalking of RuntimeStub frame
1401 
1402     __ push(rsi);
1403     __ push(rdi);
1404     __ push(rbx);
1405 
1406     Address   from_arg(rsp, 16+ 4);     // from
1407     Address     to_arg(rsp, 16+ 8);     // to
1408     Address length_arg(rsp, 16+12);     // elements count
1409     Address  ckoff_arg(rsp, 16+16);     // super_check_offset
1410     Address  ckval_arg(rsp, 16+20);     // super_klass
1411 
1412     // Load up:
1413     __ movptr(from,     from_arg);
1414     __ movptr(to,         to_arg);
1415     __ movl2ptr(length, length_arg);
1416 
1417     if (entry != NULL) {
1418       *entry = __ pc(); // Entry point from generic arraycopy stub.
1419       BLOCK_COMMENT("Entry:");
1420     }
1421 
1422     //---------------------------------------------------------------
1423     // Assembler stub will be used for this call to arraycopy
1424     // if the two arrays are subtypes of Object[] but the
1425     // destination array type is not equal to or a supertype
1426     // of the source type.  Each element must be separately
1427     // checked.
1428 
1429     // Loop-invariant addresses.  They are exclusive end pointers.
1430     Address end_from_addr(from, length, Address::times_ptr, 0);
1431     Address   end_to_addr(to,   length, Address::times_ptr, 0);
1432 
1433     Register end_from = from;           // re-use
1434     Register end_to   = to;             // re-use
1435     Register count    = length;         // re-use
1436 
1437     // Loop-variant addresses.  They assume post-incremented count < 0.
1438     Address from_element_addr(end_from, count, Address::times_ptr, 0);
1439     Address   to_element_addr(end_to,   count, Address::times_ptr, 0);
1440     Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1441 
1442     // Copy from low to high addresses, indexed from the end of each array.
1443     gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1444     __ lea(end_from, end_from_addr);
1445     __ lea(end_to,   end_to_addr);
1446     assert(length == count, "");        // else fix next line:
1447     __ negptr(count);                   // negate and test the length
1448     __ jccb(Assembler::notZero, L_load_element);
1449 
1450     // Empty array:  Nothing to do.
1451     __ xorptr(rax, rax);                  // return 0 on (trivial) success
1452     __ jmp(L_done);
1453 
1454     // ======== begin loop ========
1455     // (Loop is rotated; its entry is L_load_element.)
1456     // Loop control:
1457     //   for (count = -count; count != 0; count++)
1458     // Base pointers src, dst are biased by 8*count,to last element.
1459     __ align(OptoLoopAlignment);
1460 
1461     __ BIND(L_store_element);
1462     __ movptr(to_element_addr, elem);     // store the oop
1463     __ increment(count);                // increment the count toward zero
1464     __ jccb(Assembler::zero, L_do_card_marks);
1465 
1466     // ======== loop entry is here ========
1467     __ BIND(L_load_element);
1468     __ movptr(elem, from_element_addr);   // load the oop
1469     __ testptr(elem, elem);
1470     __ jccb(Assembler::zero, L_store_element);
1471 
1472     // (Could do a trick here:  Remember last successful non-null
1473     // element stored and make a quick oop equality check on it.)
1474 
1475     __ movptr(elem_klass, elem_klass_addr); // query the object klass
1476     generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1477                         &L_store_element, NULL);
1478     // (On fall-through, we have failed the element type check.)
1479     // ======== end loop ========
1480 
1481     // It was a real error; we must depend on the caller to finish the job.
1482     // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1483     // Emit GC store barriers for the oops we have copied (length_arg + count),
1484     // and report their number to the caller.
1485     assert_different_registers(to, count, rax);
1486     Label L_post_barrier;
1487     __ addl(count, length_arg);         // transfers = (length - remaining)
1488     __ movl2ptr(rax, count);            // save the value
1489     __ notptr(rax);                     // report (-1^K) to caller (does not affect flags)
1490     __ jccb(Assembler::notZero, L_post_barrier);
1491     __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1492 
1493     // Come here on success only.
1494     __ BIND(L_do_card_marks);
1495     __ xorptr(rax, rax);                // return 0 on success
1496     __ movl2ptr(count, length_arg);
1497 
1498     __ BIND(L_post_barrier);
1499     __ movptr(to, to_arg);              // reload
1500     gen_write_ref_array_post_barrier(to, count);
1501 
1502     // Common exit point (success or failure).
1503     __ BIND(L_done);
1504     __ pop(rbx);
1505     __ pop(rdi);
1506     __ pop(rsi);
1507     inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1508     __ leave(); // required for proper stackwalking of RuntimeStub frame
1509     __ ret(0);
1510 
1511     return start;
1512   }
1513 
1514   //
1515   //  Generate 'unsafe' array copy stub
1516   //  Though just as safe as the other stubs, it takes an unscaled
1517   //  size_t argument instead of an element count.
1518   //
1519   //  Input:
1520   //    4(rsp)   - source array address
1521   //    8(rsp)   - destination array address
1522   //   12(rsp)   - byte count, can be zero
1523   //
1524   //  Output:
1525   //    rax, ==  0  -  success
1526   //    rax, == -1  -  need to call System.arraycopy
1527   //
1528   // Examines the alignment of the operands and dispatches
1529   // to a long, int, short, or byte copy loop.
1530   //
1531   address generate_unsafe_copy(const char *name,
1532                                address byte_copy_entry,
1533                                address short_copy_entry,
1534                                address int_copy_entry,
1535                                address long_copy_entry) {
1536 
1537     Label L_long_aligned, L_int_aligned, L_short_aligned;
1538 
1539     __ align(CodeEntryAlignment);
1540     StubCodeMark mark(this, "StubRoutines", name);
1541     address start = __ pc();
1542 
1543     const Register from       = rax;  // source array address
1544     const Register to         = rdx;  // destination array address
1545     const Register count      = rcx;  // elements count
1546 
1547     __ enter(); // required for proper stackwalking of RuntimeStub frame
1548     __ push(rsi);
1549     __ push(rdi);
1550     Address  from_arg(rsp, 12+ 4);      // from
1551     Address    to_arg(rsp, 12+ 8);      // to
1552     Address count_arg(rsp, 12+12);      // byte count
1553 
1554     // Load up:
1555     __ movptr(from ,  from_arg);
1556     __ movptr(to   ,    to_arg);
1557     __ movl2ptr(count, count_arg);
1558 
1559     // bump this on entry, not on exit:
1560     inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1561 
1562     const Register bits = rsi;
1563     __ mov(bits, from);
1564     __ orptr(bits, to);
1565     __ orptr(bits, count);
1566 
1567     __ testl(bits, BytesPerLong-1);
1568     __ jccb(Assembler::zero, L_long_aligned);
1569 
1570     __ testl(bits, BytesPerInt-1);
1571     __ jccb(Assembler::zero, L_int_aligned);
1572 
1573     __ testl(bits, BytesPerShort-1);
1574     __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1575 
1576     __ BIND(L_short_aligned);
1577     __ shrptr(count, LogBytesPerShort); // size => short_count
1578     __ movl(count_arg, count);          // update 'count'
1579     __ jump(RuntimeAddress(short_copy_entry));
1580 
1581     __ BIND(L_int_aligned);
1582     __ shrptr(count, LogBytesPerInt); // size => int_count
1583     __ movl(count_arg, count);          // update 'count'
1584     __ jump(RuntimeAddress(int_copy_entry));
1585 
1586     __ BIND(L_long_aligned);
1587     __ shrptr(count, LogBytesPerLong); // size => qword_count
1588     __ movl(count_arg, count);          // update 'count'
1589     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1590     __ pop(rsi);
1591     __ jump(RuntimeAddress(long_copy_entry));
1592 
1593     return start;
1594   }
1595 
1596 
1597   // Perform range checks on the proposed arraycopy.
1598   // Smashes src_pos and dst_pos.  (Uses them up for temps.)
1599   void arraycopy_range_checks(Register src,
1600                               Register src_pos,
1601                               Register dst,
1602                               Register dst_pos,
1603                               Address& length,
1604                               Label& L_failed) {
1605     BLOCK_COMMENT("arraycopy_range_checks:");
1606     const Register src_end = src_pos;   // source array end position
1607     const Register dst_end = dst_pos;   // destination array end position
1608     __ addl(src_end, length); // src_pos + length
1609     __ addl(dst_end, length); // dst_pos + length
1610 
1611     //  if (src_pos + length > arrayOop(src)->length() ) FAIL;
1612     __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1613     __ jcc(Assembler::above, L_failed);
1614 
1615     //  if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1616     __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1617     __ jcc(Assembler::above, L_failed);
1618 
1619     BLOCK_COMMENT("arraycopy_range_checks done");
1620   }
1621 
1622 
1623   //
1624   //  Generate generic array copy stubs
1625   //
1626   //  Input:
1627   //     4(rsp)    -  src oop
1628   //     8(rsp)    -  src_pos
1629   //    12(rsp)    -  dst oop
1630   //    16(rsp)    -  dst_pos
1631   //    20(rsp)    -  element count
1632   //
1633   //  Output:
1634   //    rax, ==  0  -  success
1635   //    rax, == -1^K - failure, where K is partial transfer count
1636   //
1637   address generate_generic_copy(const char *name,
1638                                 address entry_jbyte_arraycopy,
1639                                 address entry_jshort_arraycopy,
1640                                 address entry_jint_arraycopy,
1641                                 address entry_oop_arraycopy,
1642                                 address entry_jlong_arraycopy,
1643                                 address entry_checkcast_arraycopy) {
1644     Label L_failed, L_failed_0, L_objArray;
1645 
1646     { int modulus = CodeEntryAlignment;
1647       int target  = modulus - 5; // 5 = sizeof jmp(L_failed)
1648       int advance = target - (__ offset() % modulus);
1649       if (advance < 0)  advance += modulus;
1650       if (advance > 0)  __ nop(advance);
1651     }
1652     StubCodeMark mark(this, "StubRoutines", name);
1653 
1654     // Short-hop target to L_failed.  Makes for denser prologue code.
1655     __ BIND(L_failed_0);
1656     __ jmp(L_failed);
1657     assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1658 
1659     __ align(CodeEntryAlignment);
1660     address start = __ pc();
1661 
1662     __ enter(); // required for proper stackwalking of RuntimeStub frame
1663     __ push(rsi);
1664     __ push(rdi);
1665 
1666     // bump this on entry, not on exit:
1667     inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1668 
1669     // Input values
1670     Address SRC     (rsp, 12+ 4);
1671     Address SRC_POS (rsp, 12+ 8);
1672     Address DST     (rsp, 12+12);
1673     Address DST_POS (rsp, 12+16);
1674     Address LENGTH  (rsp, 12+20);
1675 
1676     //-----------------------------------------------------------------------
1677     // Assembler stub will be used for this call to arraycopy
1678     // if the following conditions are met:
1679     //
1680     // (1) src and dst must not be null.
1681     // (2) src_pos must not be negative.
1682     // (3) dst_pos must not be negative.
1683     // (4) length  must not be negative.
1684     // (5) src klass and dst klass should be the same and not NULL.
1685     // (6) src and dst should be arrays.
1686     // (7) src_pos + length must not exceed length of src.
1687     // (8) dst_pos + length must not exceed length of dst.
1688     //
1689 
1690     const Register src     = rax;       // source array oop
1691     const Register src_pos = rsi;
1692     const Register dst     = rdx;       // destination array oop
1693     const Register dst_pos = rdi;
1694     const Register length  = rcx;       // transfer count
1695 
1696     //  if (src == NULL) return -1;
1697     __ movptr(src, SRC);      // src oop
1698     __ testptr(src, src);
1699     __ jccb(Assembler::zero, L_failed_0);
1700 
1701     //  if (src_pos < 0) return -1;
1702     __ movl2ptr(src_pos, SRC_POS);  // src_pos
1703     __ testl(src_pos, src_pos);
1704     __ jccb(Assembler::negative, L_failed_0);
1705 
1706     //  if (dst == NULL) return -1;
1707     __ movptr(dst, DST);      // dst oop
1708     __ testptr(dst, dst);
1709     __ jccb(Assembler::zero, L_failed_0);
1710 
1711     //  if (dst_pos < 0) return -1;
1712     __ movl2ptr(dst_pos, DST_POS);  // dst_pos
1713     __ testl(dst_pos, dst_pos);
1714     __ jccb(Assembler::negative, L_failed_0);
1715 
1716     //  if (length < 0) return -1;
1717     __ movl2ptr(length, LENGTH);   // length
1718     __ testl(length, length);
1719     __ jccb(Assembler::negative, L_failed_0);
1720 
1721     //  if (src->klass() == NULL) return -1;
1722     Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1723     Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1724     const Register rcx_src_klass = rcx;    // array klass
1725     __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1726 
1727 #ifdef ASSERT
1728     //  assert(src->klass() != NULL);
1729     BLOCK_COMMENT("assert klasses not null");
1730     { Label L1, L2;
1731       __ testptr(rcx_src_klass, rcx_src_klass);
1732       __ jccb(Assembler::notZero, L2);   // it is broken if klass is NULL
1733       __ bind(L1);
1734       __ stop("broken null klass");
1735       __ bind(L2);
1736       __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1737       __ jccb(Assembler::equal, L1);      // this would be broken also
1738       BLOCK_COMMENT("assert done");
1739     }
1740 #endif //ASSERT
1741 
1742     // Load layout helper (32-bits)
1743     //
1744     //  |array_tag|     | header_size | element_type |     |log2_element_size|
1745     // 32        30    24            16              8     2                 0
1746     //
1747     //   array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1748     //
1749 
1750     int lh_offset = in_bytes(Klass::layout_helper_offset());
1751     Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1752 
1753     // Handle objArrays completely differently...
1754     jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1755     __ cmpl(src_klass_lh_addr, objArray_lh);
1756     __ jcc(Assembler::equal, L_objArray);
1757 
1758     //  if (src->klass() != dst->klass()) return -1;
1759     __ cmpptr(rcx_src_klass, dst_klass_addr);
1760     __ jccb(Assembler::notEqual, L_failed_0);
1761 
1762     const Register rcx_lh = rcx;  // layout helper
1763     assert(rcx_lh == rcx_src_klass, "known alias");
1764     __ movl(rcx_lh, src_klass_lh_addr);
1765 
1766     //  if (!src->is_Array()) return -1;
1767     __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1768     __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1769 
1770     // At this point, it is known to be a typeArray (array_tag 0x3).
1771 #ifdef ASSERT
1772     { Label L;
1773       __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1774       __ jcc(Assembler::greaterEqual, L); // signed cmp
1775       __ stop("must be a primitive array");
1776       __ bind(L);
1777     }
1778 #endif
1779 
1780     assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1781     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1782 
1783     // TypeArrayKlass
1784     //
1785     // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1786     // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1787     //
1788     const Register rsi_offset = rsi; // array offset
1789     const Register src_array  = src; // src array offset
1790     const Register dst_array  = dst; // dst array offset
1791     const Register rdi_elsize = rdi; // log2 element size
1792 
1793     __ mov(rsi_offset, rcx_lh);
1794     __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1795     __ andptr(rsi_offset, Klass::_lh_header_size_mask);   // array_offset
1796     __ addptr(src_array, rsi_offset);  // src array offset
1797     __ addptr(dst_array, rsi_offset);  // dst array offset
1798     __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1799 
1800     // next registers should be set before the jump to corresponding stub
1801     const Register from       = src; // source array address
1802     const Register to         = dst; // destination array address
1803     const Register count      = rcx; // elements count
1804     // some of them should be duplicated on stack
1805 #define FROM   Address(rsp, 12+ 4)
1806 #define TO     Address(rsp, 12+ 8)   // Not used now
1807 #define COUNT  Address(rsp, 12+12)   // Only for oop arraycopy
1808 
1809     BLOCK_COMMENT("scale indexes to element size");
1810     __ movl2ptr(rsi, SRC_POS);  // src_pos
1811     __ shlptr(rsi);             // src_pos << rcx (log2 elsize)
1812     assert(src_array == from, "");
1813     __ addptr(from, rsi);       // from = src_array + SRC_POS << log2 elsize
1814     __ movl2ptr(rdi, DST_POS);  // dst_pos
1815     __ shlptr(rdi);             // dst_pos << rcx (log2 elsize)
1816     assert(dst_array == to, "");
1817     __ addptr(to,  rdi);        // to   = dst_array + DST_POS << log2 elsize
1818     __ movptr(FROM, from);      // src_addr
1819     __ mov(rdi_elsize, rcx_lh); // log2 elsize
1820     __ movl2ptr(count, LENGTH); // elements count
1821 
1822     BLOCK_COMMENT("choose copy loop based on element size");
1823     __ cmpl(rdi_elsize, 0);
1824 
1825     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1826     __ cmpl(rdi_elsize, LogBytesPerShort);
1827     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1828     __ cmpl(rdi_elsize, LogBytesPerInt);
1829     __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1830 #ifdef ASSERT
1831     __ cmpl(rdi_elsize, LogBytesPerLong);
1832     __ jccb(Assembler::notEqual, L_failed);
1833 #endif
1834     __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1835     __ pop(rsi);
1836     __ jump(RuntimeAddress(entry_jlong_arraycopy));
1837 
1838   __ BIND(L_failed);
1839     __ xorptr(rax, rax);
1840     __ notptr(rax); // return -1
1841     __ pop(rdi);
1842     __ pop(rsi);
1843     __ leave(); // required for proper stackwalking of RuntimeStub frame
1844     __ ret(0);
1845 
1846     // ObjArrayKlass
1847   __ BIND(L_objArray);
1848     // live at this point:  rcx_src_klass, src[_pos], dst[_pos]
1849 
1850     Label L_plain_copy, L_checkcast_copy;
1851     //  test array classes for subtyping
1852     __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1853     __ jccb(Assembler::notEqual, L_checkcast_copy);
1854 
1855     // Identically typed arrays can be copied without element-wise checks.
1856     assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1857     arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1858 
1859   __ BIND(L_plain_copy);
1860     __ movl2ptr(count, LENGTH); // elements count
1861     __ movl2ptr(src_pos, SRC_POS);  // reload src_pos
1862     __ lea(from, Address(src, src_pos, Address::times_ptr,
1863                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1864     __ movl2ptr(dst_pos, DST_POS);  // reload dst_pos
1865     __ lea(to,   Address(dst, dst_pos, Address::times_ptr,
1866                  arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1867     __ movptr(FROM,  from);   // src_addr
1868     __ movptr(TO,    to);     // dst_addr
1869     __ movl(COUNT, count);  // count
1870     __ jump(RuntimeAddress(entry_oop_arraycopy));
1871 
1872   __ BIND(L_checkcast_copy);
1873     // live at this point:  rcx_src_klass, dst[_pos], src[_pos]
1874     {
1875       // Handy offsets:
1876       int  ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1877       int sco_offset = in_bytes(Klass::super_check_offset_offset());
1878 
1879       Register rsi_dst_klass = rsi;
1880       Register rdi_temp      = rdi;
1881       assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1882       assert(rdi_temp      == dst_pos, "expected alias w/ dst_pos");
1883       Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1884 
1885       // Before looking at dst.length, make sure dst is also an objArray.
1886       __ movptr(rsi_dst_klass, dst_klass_addr);
1887       __ cmpl(dst_klass_lh_addr, objArray_lh);
1888       __ jccb(Assembler::notEqual, L_failed);
1889 
1890       // It is safe to examine both src.length and dst.length.
1891       __ movl2ptr(src_pos, SRC_POS);        // reload rsi
1892       arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1893       // (Now src_pos and dst_pos are killed, but not src and dst.)
1894 
1895       // We'll need this temp (don't forget to pop it after the type check).
1896       __ push(rbx);
1897       Register rbx_src_klass = rbx;
1898 
1899       __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1900       __ movptr(rsi_dst_klass, dst_klass_addr);
1901       Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1902       Label L_fail_array_check;
1903       generate_type_check(rbx_src_klass,
1904                           super_check_offset_addr, dst_klass_addr,
1905                           rdi_temp, NULL, &L_fail_array_check);
1906       // (On fall-through, we have passed the array type check.)
1907       __ pop(rbx);
1908       __ jmp(L_plain_copy);
1909 
1910       __ BIND(L_fail_array_check);
1911       // Reshuffle arguments so we can call checkcast_arraycopy:
1912 
1913       // match initial saves for checkcast_arraycopy
1914       // push(rsi);    // already done; see above
1915       // push(rdi);    // already done; see above
1916       // push(rbx);    // already done; see above
1917 
1918       // Marshal outgoing arguments now, freeing registers.
1919       Address   from_arg(rsp, 16+ 4);   // from
1920       Address     to_arg(rsp, 16+ 8);   // to
1921       Address length_arg(rsp, 16+12);   // elements count
1922       Address  ckoff_arg(rsp, 16+16);   // super_check_offset
1923       Address  ckval_arg(rsp, 16+20);   // super_klass
1924 
1925       Address SRC_POS_arg(rsp, 16+ 8);
1926       Address DST_POS_arg(rsp, 16+16);
1927       Address  LENGTH_arg(rsp, 16+20);
1928       // push rbx, changed the incoming offsets (why not just use rbp,??)
1929       // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1930 
1931       __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1932       __ movl2ptr(length, LENGTH_arg);    // reload elements count
1933       __ movl2ptr(src_pos, SRC_POS_arg);  // reload src_pos
1934       __ movl2ptr(dst_pos, DST_POS_arg);  // reload dst_pos
1935 
1936       __ movptr(ckval_arg, rbx);          // destination element type
1937       __ movl(rbx, Address(rbx, sco_offset));
1938       __ movl(ckoff_arg, rbx);          // corresponding class check offset
1939 
1940       __ movl(length_arg, length);      // outgoing length argument
1941 
1942       __ lea(from, Address(src, src_pos, Address::times_ptr,
1943                             arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1944       __ movptr(from_arg, from);
1945 
1946       __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1947                           arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1948       __ movptr(to_arg, to);
1949       __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1950     }
1951 
1952     return start;
1953   }
1954 
1955   void generate_arraycopy_stubs() {
1956     address entry;
1957     address entry_jbyte_arraycopy;
1958     address entry_jshort_arraycopy;
1959     address entry_jint_arraycopy;
1960     address entry_oop_arraycopy;
1961     address entry_jlong_arraycopy;
1962     address entry_checkcast_arraycopy;
1963 
1964     StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1965         generate_disjoint_copy(T_BYTE,  true, Address::times_1, &entry,
1966                                "arrayof_jbyte_disjoint_arraycopy");
1967     StubRoutines::_arrayof_jbyte_arraycopy =
1968         generate_conjoint_copy(T_BYTE,  true, Address::times_1,  entry,
1969                                NULL, "arrayof_jbyte_arraycopy");
1970     StubRoutines::_jbyte_disjoint_arraycopy =
1971         generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1972                                "jbyte_disjoint_arraycopy");
1973     StubRoutines::_jbyte_arraycopy =
1974         generate_conjoint_copy(T_BYTE, false, Address::times_1,  entry,
1975                                &entry_jbyte_arraycopy, "jbyte_arraycopy");
1976 
1977     StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1978         generate_disjoint_copy(T_SHORT,  true, Address::times_2, &entry,
1979                                "arrayof_jshort_disjoint_arraycopy");
1980     StubRoutines::_arrayof_jshort_arraycopy =
1981         generate_conjoint_copy(T_SHORT,  true, Address::times_2,  entry,
1982                                NULL, "arrayof_jshort_arraycopy");
1983     StubRoutines::_jshort_disjoint_arraycopy =
1984         generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1985                                "jshort_disjoint_arraycopy");
1986     StubRoutines::_jshort_arraycopy =
1987         generate_conjoint_copy(T_SHORT, false, Address::times_2,  entry,
1988                                &entry_jshort_arraycopy, "jshort_arraycopy");
1989 
1990     // Next arrays are always aligned on 4 bytes at least.
1991     StubRoutines::_jint_disjoint_arraycopy =
1992         generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1993                                "jint_disjoint_arraycopy");
1994     StubRoutines::_jint_arraycopy =
1995         generate_conjoint_copy(T_INT, true, Address::times_4,  entry,
1996                                &entry_jint_arraycopy, "jint_arraycopy");
1997 
1998     StubRoutines::_oop_disjoint_arraycopy =
1999         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2000                                "oop_disjoint_arraycopy");
2001     StubRoutines::_oop_arraycopy =
2002         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2003                                &entry_oop_arraycopy, "oop_arraycopy");
2004 
2005     StubRoutines::_oop_disjoint_arraycopy_uninit =
2006         generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2007                                "oop_disjoint_arraycopy_uninit",
2008                                /*dest_uninitialized*/true);
2009     StubRoutines::_oop_arraycopy_uninit =
2010         generate_conjoint_copy(T_OBJECT, true, Address::times_ptr,  entry,
2011                                NULL, "oop_arraycopy_uninit",
2012                                /*dest_uninitialized*/true);
2013 
2014     StubRoutines::_jlong_disjoint_arraycopy =
2015         generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2016     StubRoutines::_jlong_arraycopy =
2017         generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2018                                     "jlong_arraycopy");
2019 
2020     StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2021     StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2022     StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2023     StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2024     StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2025     StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2026 
2027     StubRoutines::_arrayof_jint_disjoint_arraycopy       = StubRoutines::_jint_disjoint_arraycopy;
2028     StubRoutines::_arrayof_oop_disjoint_arraycopy        = StubRoutines::_oop_disjoint_arraycopy;
2029     StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2030     StubRoutines::_arrayof_jlong_disjoint_arraycopy      = StubRoutines::_jlong_disjoint_arraycopy;
2031 
2032     StubRoutines::_arrayof_jint_arraycopy       = StubRoutines::_jint_arraycopy;
2033     StubRoutines::_arrayof_oop_arraycopy        = StubRoutines::_oop_arraycopy;
2034     StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2035     StubRoutines::_arrayof_jlong_arraycopy      = StubRoutines::_jlong_arraycopy;
2036 
2037     StubRoutines::_checkcast_arraycopy =
2038         generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2039     StubRoutines::_checkcast_arraycopy_uninit =
2040         generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2041 
2042     StubRoutines::_unsafe_arraycopy =
2043         generate_unsafe_copy("unsafe_arraycopy",
2044                                entry_jbyte_arraycopy,
2045                                entry_jshort_arraycopy,
2046                                entry_jint_arraycopy,
2047                                entry_jlong_arraycopy);
2048 
2049     StubRoutines::_generic_arraycopy =
2050         generate_generic_copy("generic_arraycopy",
2051                                entry_jbyte_arraycopy,
2052                                entry_jshort_arraycopy,
2053                                entry_jint_arraycopy,
2054                                entry_oop_arraycopy,
2055                                entry_jlong_arraycopy,
2056                                entry_checkcast_arraycopy);
2057   }
2058 
2059   // AES intrinsic stubs
2060   enum {AESBlockSize = 16};
2061 
2062   address generate_key_shuffle_mask() {
2063     __ align(16);
2064     StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2065     address start = __ pc();
2066     __ emit_data(0x00010203, relocInfo::none, 0 );
2067     __ emit_data(0x04050607, relocInfo::none, 0 );
2068     __ emit_data(0x08090a0b, relocInfo::none, 0 );
2069     __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2070     return start;
2071   }
2072 
2073   address generate_counter_shuffle_mask() {
2074     __ align(16);
2075     StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2076     address start = __ pc();
2077     __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2078     __ emit_data(0x08090a0b, relocInfo::none, 0);
2079     __ emit_data(0x04050607, relocInfo::none, 0);
2080     __ emit_data(0x00010203, relocInfo::none, 0);
2081     return start;
2082   }
2083 
2084   // Utility routine for loading a 128-bit key word in little endian format
2085   // can optionally specify that the shuffle mask is already in an xmmregister
2086   void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2087     __ movdqu(xmmdst, Address(key, offset));
2088     if (xmm_shuf_mask != NULL) {
2089       __ pshufb(xmmdst, xmm_shuf_mask);
2090     } else {
2091       __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2092     }
2093   }
2094 
2095   // aesenc using specified key+offset
2096   // can optionally specify that the shuffle mask is already in an xmmregister
2097   void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2098     load_key(xmmtmp, key, offset, xmm_shuf_mask);
2099     __ aesenc(xmmdst, xmmtmp);
2100   }
2101 
2102   // aesdec using specified key+offset
2103   // can optionally specify that the shuffle mask is already in an xmmregister
2104   void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2105     load_key(xmmtmp, key, offset, xmm_shuf_mask);
2106     __ aesdec(xmmdst, xmmtmp);
2107   }
2108 
2109   // Utility routine for increase 128bit counter (iv in CTR mode)
2110   //  XMM_128bit,  D3, D2, D1, D0
2111   void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2112     __ pextrd(reg, xmmdst, 0x0);
2113     __ addl(reg, inc_delta);
2114     __ pinsrd(xmmdst, reg, 0x0);
2115     __ jcc(Assembler::carryClear, next_block); // jump if no carry
2116 
2117     __ pextrd(reg, xmmdst, 0x01); // Carry-> D1
2118     __ addl(reg, 0x01);
2119     __ pinsrd(xmmdst, reg, 0x01);
2120     __ jcc(Assembler::carryClear, next_block); // jump if no carry
2121 
2122     __ pextrd(reg, xmmdst, 0x02); // Carry-> D2
2123     __ addl(reg, 0x01);
2124     __ pinsrd(xmmdst, reg, 0x02);
2125     __ jcc(Assembler::carryClear, next_block); // jump if no carry
2126 
2127     __ pextrd(reg, xmmdst, 0x03); // Carry -> D3
2128     __ addl(reg, 0x01);
2129     __ pinsrd(xmmdst, reg, 0x03);
2130 
2131     __ BIND(next_block);          // next instruction
2132   }
2133 
2134 
2135   // Arguments:
2136   //
2137   // Inputs:
2138   //   c_rarg0   - source byte array address
2139   //   c_rarg1   - destination byte array address
2140   //   c_rarg2   - K (key) in little endian int array
2141   //
2142   address generate_aescrypt_encryptBlock() {
2143     assert(UseAES, "need AES instructions and misaligned SSE support");
2144     __ align(CodeEntryAlignment);
2145     StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2146     Label L_doLast;
2147     address start = __ pc();
2148 
2149     const Register from        = rdx;      // source array address
2150     const Register to          = rdx;      // destination array address
2151     const Register key         = rcx;      // key array address
2152     const Register keylen      = rax;
2153     const Address  from_param(rbp, 8+0);
2154     const Address  to_param  (rbp, 8+4);
2155     const Address  key_param (rbp, 8+8);
2156 
2157     const XMMRegister xmm_result = xmm0;
2158     const XMMRegister xmm_key_shuf_mask = xmm1;
2159     const XMMRegister xmm_temp1  = xmm2;
2160     const XMMRegister xmm_temp2  = xmm3;
2161     const XMMRegister xmm_temp3  = xmm4;
2162     const XMMRegister xmm_temp4  = xmm5;
2163 
2164     __ enter();   // required for proper stackwalking of RuntimeStub frame
2165 
2166     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2167     // context for the registers used, where all instructions below are using 128-bit mode
2168     // On EVEX without VL and BW, these instructions will all be AVX.
2169     if (VM_Version::supports_avx512vlbw()) {
2170       __ movl(rdx, 0xffff);
2171       __ kmovdl(k1, rdx);
2172     }
2173 
2174     __ movptr(from, from_param);
2175     __ movptr(key, key_param);
2176 
2177     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2178     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2179 
2180     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2181     __ movdqu(xmm_result, Address(from, 0));  // get 16 bytes of input
2182     __ movptr(to, to_param);
2183 
2184     // For encryption, the java expanded key ordering is just what we need
2185 
2186     load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2187     __ pxor(xmm_result, xmm_temp1);
2188 
2189     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2190     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2191     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2192     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2193 
2194     __ aesenc(xmm_result, xmm_temp1);
2195     __ aesenc(xmm_result, xmm_temp2);
2196     __ aesenc(xmm_result, xmm_temp3);
2197     __ aesenc(xmm_result, xmm_temp4);
2198 
2199     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2200     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2201     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2202     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2203 
2204     __ aesenc(xmm_result, xmm_temp1);
2205     __ aesenc(xmm_result, xmm_temp2);
2206     __ aesenc(xmm_result, xmm_temp3);
2207     __ aesenc(xmm_result, xmm_temp4);
2208 
2209     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2210     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2211 
2212     __ cmpl(keylen, 44);
2213     __ jccb(Assembler::equal, L_doLast);
2214 
2215     __ aesenc(xmm_result, xmm_temp1);
2216     __ aesenc(xmm_result, xmm_temp2);
2217 
2218     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2219     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2220 
2221     __ cmpl(keylen, 52);
2222     __ jccb(Assembler::equal, L_doLast);
2223 
2224     __ aesenc(xmm_result, xmm_temp1);
2225     __ aesenc(xmm_result, xmm_temp2);
2226 
2227     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2228     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2229 
2230     __ BIND(L_doLast);
2231     __ aesenc(xmm_result, xmm_temp1);
2232     __ aesenclast(xmm_result, xmm_temp2);
2233     __ movdqu(Address(to, 0), xmm_result);        // store the result
2234     __ xorptr(rax, rax); // return 0
2235     __ leave(); // required for proper stackwalking of RuntimeStub frame
2236     __ ret(0);
2237 
2238     return start;
2239   }
2240 
2241 
2242   // Arguments:
2243   //
2244   // Inputs:
2245   //   c_rarg0   - source byte array address
2246   //   c_rarg1   - destination byte array address
2247   //   c_rarg2   - K (key) in little endian int array
2248   //
2249   address generate_aescrypt_decryptBlock() {
2250     assert(UseAES, "need AES instructions and misaligned SSE support");
2251     __ align(CodeEntryAlignment);
2252     StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2253     Label L_doLast;
2254     address start = __ pc();
2255 
2256     const Register from        = rdx;      // source array address
2257     const Register to          = rdx;      // destination array address
2258     const Register key         = rcx;      // key array address
2259     const Register keylen      = rax;
2260     const Address  from_param(rbp, 8+0);
2261     const Address  to_param  (rbp, 8+4);
2262     const Address  key_param (rbp, 8+8);
2263 
2264     const XMMRegister xmm_result = xmm0;
2265     const XMMRegister xmm_key_shuf_mask = xmm1;
2266     const XMMRegister xmm_temp1  = xmm2;
2267     const XMMRegister xmm_temp2  = xmm3;
2268     const XMMRegister xmm_temp3  = xmm4;
2269     const XMMRegister xmm_temp4  = xmm5;
2270 
2271     __ enter(); // required for proper stackwalking of RuntimeStub frame
2272 
2273     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2274     // context for the registers used, where all instructions below are using 128-bit mode
2275     // On EVEX without VL and BW, these instructions will all be AVX.
2276     if (VM_Version::supports_avx512vlbw()) {
2277       __ movl(rdx, 0xffff);
2278       __ kmovdl(k1, rdx);
2279     }
2280 
2281     __ movptr(from, from_param);
2282     __ movptr(key, key_param);
2283 
2284     // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2285     __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2286 
2287     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2288     __ movdqu(xmm_result, Address(from, 0));
2289     __ movptr(to, to_param);
2290 
2291     // for decryption java expanded key ordering is rotated one position from what we want
2292     // so we start from 0x10 here and hit 0x00 last
2293     // we don't know if the key is aligned, hence not using load-execute form
2294     load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2295     load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2296     load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2297     load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2298 
2299     __ pxor  (xmm_result, xmm_temp1);
2300     __ aesdec(xmm_result, xmm_temp2);
2301     __ aesdec(xmm_result, xmm_temp3);
2302     __ aesdec(xmm_result, xmm_temp4);
2303 
2304     load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2305     load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2306     load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2307     load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2308 
2309     __ aesdec(xmm_result, xmm_temp1);
2310     __ aesdec(xmm_result, xmm_temp2);
2311     __ aesdec(xmm_result, xmm_temp3);
2312     __ aesdec(xmm_result, xmm_temp4);
2313 
2314     load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2315     load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2316     load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2317 
2318     __ cmpl(keylen, 44);
2319     __ jccb(Assembler::equal, L_doLast);
2320 
2321     __ aesdec(xmm_result, xmm_temp1);
2322     __ aesdec(xmm_result, xmm_temp2);
2323 
2324     load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2325     load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2326 
2327     __ cmpl(keylen, 52);
2328     __ jccb(Assembler::equal, L_doLast);
2329 
2330     __ aesdec(xmm_result, xmm_temp1);
2331     __ aesdec(xmm_result, xmm_temp2);
2332 
2333     load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2334     load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2335 
2336     __ BIND(L_doLast);
2337     __ aesdec(xmm_result, xmm_temp1);
2338     __ aesdec(xmm_result, xmm_temp2);
2339 
2340     // for decryption the aesdeclast operation is always on key+0x00
2341     __ aesdeclast(xmm_result, xmm_temp3);
2342     __ movdqu(Address(to, 0), xmm_result);  // store the result
2343     __ xorptr(rax, rax); // return 0
2344     __ leave(); // required for proper stackwalking of RuntimeStub frame
2345     __ ret(0);
2346 
2347     return start;
2348   }
2349 
2350   void handleSOERegisters(bool saving) {
2351     const int saveFrameSizeInBytes = 4 * wordSize;
2352     const Address saved_rbx     (rbp, -3 * wordSize);
2353     const Address saved_rsi     (rbp, -2 * wordSize);
2354     const Address saved_rdi     (rbp, -1 * wordSize);
2355 
2356     if (saving) {
2357       __ subptr(rsp, saveFrameSizeInBytes);
2358       __ movptr(saved_rsi, rsi);
2359       __ movptr(saved_rdi, rdi);
2360       __ movptr(saved_rbx, rbx);
2361     } else {
2362       // restoring
2363       __ movptr(rsi, saved_rsi);
2364       __ movptr(rdi, saved_rdi);
2365       __ movptr(rbx, saved_rbx);
2366     }
2367   }
2368 
2369   // Arguments:
2370   //
2371   // Inputs:
2372   //   c_rarg0   - source byte array address
2373   //   c_rarg1   - destination byte array address
2374   //   c_rarg2   - K (key) in little endian int array
2375   //   c_rarg3   - r vector byte array address
2376   //   c_rarg4   - input length
2377   //
2378   // Output:
2379   //   rax       - input length
2380   //
2381   address generate_cipherBlockChaining_encryptAESCrypt() {
2382     assert(UseAES, "need AES instructions and misaligned SSE support");
2383     __ align(CodeEntryAlignment);
2384     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2385     address start = __ pc();
2386 
2387     Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2388     const Register from        = rsi;      // source array address
2389     const Register to          = rdx;      // destination array address
2390     const Register key         = rcx;      // key array address
2391     const Register rvec        = rdi;      // r byte array initialized from initvector array address
2392                                            // and left with the results of the last encryption block
2393     const Register len_reg     = rbx;      // src len (must be multiple of blocksize 16)
2394     const Register pos         = rax;
2395 
2396     // xmm register assignments for the loops below
2397     const XMMRegister xmm_result = xmm0;
2398     const XMMRegister xmm_temp   = xmm1;
2399     // first 6 keys preloaded into xmm2-xmm7
2400     const int XMM_REG_NUM_KEY_FIRST = 2;
2401     const int XMM_REG_NUM_KEY_LAST  = 7;
2402     const XMMRegister xmm_key0   = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2403 
2404     __ enter(); // required for proper stackwalking of RuntimeStub frame
2405     handleSOERegisters(true /*saving*/);
2406 
2407     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2408     // context for the registers used, where all instructions below are using 128-bit mode
2409     // On EVEX without VL and BW, these instructions will all be AVX.
2410     if (VM_Version::supports_avx512vlbw()) {
2411       __ movl(rdx, 0xffff);
2412       __ kmovdl(k1, rdx);
2413     }
2414 
2415     // load registers from incoming parameters
2416     const Address  from_param(rbp, 8+0);
2417     const Address  to_param  (rbp, 8+4);
2418     const Address  key_param (rbp, 8+8);
2419     const Address  rvec_param (rbp, 8+12);
2420     const Address  len_param  (rbp, 8+16);
2421     __ movptr(from , from_param);
2422     __ movptr(to   , to_param);
2423     __ movptr(key  , key_param);
2424     __ movptr(rvec , rvec_param);
2425     __ movptr(len_reg , len_param);
2426 
2427     const XMMRegister xmm_key_shuf_mask = xmm_temp;  // used temporarily to swap key bytes up front
2428     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2429     // load up xmm regs 2 thru 7 with keys 0-5
2430     for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2431       load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2432       offset += 0x10;
2433     }
2434 
2435     __ movdqu(xmm_result, Address(rvec, 0x00));   // initialize xmm_result with r vec
2436 
2437     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2438     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2439     __ cmpl(rax, 44);
2440     __ jcc(Assembler::notEqual, L_key_192_256);
2441 
2442     // 128 bit code follows here
2443     __ movl(pos, 0);
2444     __ align(OptoLoopAlignment);
2445     __ BIND(L_loopTop_128);
2446     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2447     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2448 
2449     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2450     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2451       __ aesenc(xmm_result, as_XMMRegister(rnum));
2452     }
2453     for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2454       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2455     }
2456     load_key(xmm_temp, key, 0xa0);
2457     __ aesenclast(xmm_result, xmm_temp);
2458 
2459     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);     // store into the next 16 bytes of output
2460     // no need to store r to memory until we exit
2461     __ addptr(pos, AESBlockSize);
2462     __ subptr(len_reg, AESBlockSize);
2463     __ jcc(Assembler::notEqual, L_loopTop_128);
2464 
2465     __ BIND(L_exit);
2466     __ movdqu(Address(rvec, 0), xmm_result);     // final value of r stored in rvec of CipherBlockChaining object
2467 
2468     handleSOERegisters(false /*restoring*/);
2469     __ movptr(rax, len_param); // return length
2470     __ leave();                                  // required for proper stackwalking of RuntimeStub frame
2471     __ ret(0);
2472 
2473     __ BIND(L_key_192_256);
2474     // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2475     __ cmpl(rax, 52);
2476     __ jcc(Assembler::notEqual, L_key_256);
2477 
2478     // 192-bit code follows here (could be changed to use more xmm registers)
2479     __ movl(pos, 0);
2480     __ align(OptoLoopAlignment);
2481     __ BIND(L_loopTop_192);
2482     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2483     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2484 
2485     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2486     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2487       __ aesenc(xmm_result, as_XMMRegister(rnum));
2488     }
2489     for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2490       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2491     }
2492     load_key(xmm_temp, key, 0xc0);
2493     __ aesenclast(xmm_result, xmm_temp);
2494 
2495     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);   // store into the next 16 bytes of output
2496     // no need to store r to memory until we exit
2497     __ addptr(pos, AESBlockSize);
2498     __ subptr(len_reg, AESBlockSize);
2499     __ jcc(Assembler::notEqual, L_loopTop_192);
2500     __ jmp(L_exit);
2501 
2502     __ BIND(L_key_256);
2503     // 256-bit code follows here (could be changed to use more xmm registers)
2504     __ movl(pos, 0);
2505     __ align(OptoLoopAlignment);
2506     __ BIND(L_loopTop_256);
2507     __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0));   // get next 16 bytes of input
2508     __ pxor  (xmm_result, xmm_temp);                                // xor with the current r vector
2509 
2510     __ pxor  (xmm_result, xmm_key0);                                // do the aes rounds
2511     for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum  <= XMM_REG_NUM_KEY_LAST; rnum++) {
2512       __ aesenc(xmm_result, as_XMMRegister(rnum));
2513     }
2514     for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2515       aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2516     }
2517     load_key(xmm_temp, key, 0xe0);
2518     __ aesenclast(xmm_result, xmm_temp);
2519 
2520     __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result);   // store into the next 16 bytes of output
2521     // no need to store r to memory until we exit
2522     __ addptr(pos, AESBlockSize);
2523     __ subptr(len_reg, AESBlockSize);
2524     __ jcc(Assembler::notEqual, L_loopTop_256);
2525     __ jmp(L_exit);
2526 
2527     return start;
2528   }
2529 
2530 
2531   // CBC AES Decryption.
2532   // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2533   //
2534   // Arguments:
2535   //
2536   // Inputs:
2537   //   c_rarg0   - source byte array address
2538   //   c_rarg1   - destination byte array address
2539   //   c_rarg2   - K (key) in little endian int array
2540   //   c_rarg3   - r vector byte array address
2541   //   c_rarg4   - input length
2542   //
2543   // Output:
2544   //   rax       - input length
2545   //
2546 
2547   address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
2548     assert(UseAES, "need AES instructions and misaligned SSE support");
2549     __ align(CodeEntryAlignment);
2550     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2551     address start = __ pc();
2552 
2553     const Register from        = rsi;      // source array address
2554     const Register to          = rdx;      // destination array address
2555     const Register key         = rcx;      // key array address
2556     const Register rvec        = rdi;      // r byte array initialized from initvector array address
2557                                            // and left with the results of the last encryption block
2558     const Register len_reg     = rbx;      // src len (must be multiple of blocksize 16)
2559     const Register pos         = rax;
2560 
2561     const int PARALLEL_FACTOR = 4;
2562     const int ROUNDS[3] = { 10, 12, 14 }; //aes rounds for key128, key192, key256
2563 
2564     Label L_exit;
2565     Label L_singleBlock_loopTop[3]; //128, 192, 256
2566     Label L_multiBlock_loopTop[3]; //128, 192, 256
2567 
2568     const XMMRegister xmm_prev_block_cipher = xmm0; // holds cipher of previous block
2569     const XMMRegister xmm_key_shuf_mask = xmm1;
2570 
2571     const XMMRegister xmm_key_tmp0 = xmm2;
2572     const XMMRegister xmm_key_tmp1 = xmm3;
2573 
2574     // registers holding the six results in the parallelized loop
2575     const XMMRegister xmm_result0 = xmm4;
2576     const XMMRegister xmm_result1 = xmm5;
2577     const XMMRegister xmm_result2 = xmm6;
2578     const XMMRegister xmm_result3 = xmm7;
2579 
2580     __ enter(); // required for proper stackwalking of RuntimeStub frame
2581     handleSOERegisters(true /*saving*/);
2582 
2583     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2584     // context for the registers used, where all instructions below are using 128-bit mode
2585     // On EVEX without VL and BW, these instructions will all be AVX.
2586     if (VM_Version::supports_avx512vlbw()) {
2587       __ movl(rdx, 0xffff);
2588       __ kmovdl(k1, rdx);
2589     }
2590 
2591     // load registers from incoming parameters
2592     const Address  from_param(rbp, 8+0);
2593     const Address  to_param  (rbp, 8+4);
2594     const Address  key_param (rbp, 8+8);
2595     const Address  rvec_param (rbp, 8+12);
2596     const Address  len_param  (rbp, 8+16);
2597 
2598     __ movptr(from , from_param);
2599     __ movptr(to   , to_param);
2600     __ movptr(key  , key_param);
2601     __ movptr(rvec , rvec_param);
2602     __ movptr(len_reg , len_param);
2603 
2604     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2605     __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
2606 
2607     __ xorptr(pos, pos);
2608 
2609     // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2610     // rvec is reused
2611     __ movl(rvec, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2612     __ cmpl(rvec, 52);
2613     __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
2614     __ cmpl(rvec, 60);
2615     __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
2616 
2617 #define DoFour(opc, src_reg)           \
2618   __ opc(xmm_result0, src_reg);         \
2619   __ opc(xmm_result1, src_reg);         \
2620   __ opc(xmm_result2, src_reg);         \
2621   __ opc(xmm_result3, src_reg);         \
2622 
2623     for (int k = 0; k < 3; ++k) {
2624       __ align(OptoLoopAlignment);
2625       __ BIND(L_multiBlock_loopTop[k]);
2626       __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
2627       __ jcc(Assembler::less, L_singleBlock_loopTop[k]);
2628 
2629       __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
2630       __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2631       __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2632       __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2633 
2634       // the java expanded key ordering is rotated one position from what we want
2635       // so we start from 0x10 here and hit 0x00 last
2636       load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2637       DoFour(pxor, xmm_key_tmp0); //xor with first key
2638       // do the aes dec rounds
2639       for (int rnum = 1; rnum <= ROUNDS[k];) {
2640         //load two keys at a time
2641         //k1->0x20, ..., k9->0xa0, k10->0x00
2642         load_key(xmm_key_tmp1, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2643         load_key(xmm_key_tmp0, key, ((rnum + 2) % (ROUNDS[k] + 1)) * 0x10, xmm_key_shuf_mask); // hit 0x00 last!
2644         DoFour(aesdec, xmm_key_tmp1);
2645         rnum++;
2646         if (rnum != ROUNDS[k]) {
2647           DoFour(aesdec, xmm_key_tmp0);
2648         }
2649         else {
2650           DoFour(aesdeclast, xmm_key_tmp0);
2651         }
2652         rnum++;
2653       }
2654 
2655       // for each result, xor with the r vector of previous cipher block
2656       __ pxor(xmm_result0, xmm_prev_block_cipher);
2657       __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2658       __ pxor(xmm_result1, xmm_prev_block_cipher);
2659       __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2660       __ pxor(xmm_result2, xmm_prev_block_cipher);
2661       __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2662       __ pxor(xmm_result3, xmm_prev_block_cipher);
2663       __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
2664 
2665             // store 4 results into the next 64 bytes of output
2666        __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2667        __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2668        __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2669        __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2670 
2671        __ addptr(pos, 4 * AESBlockSize);
2672        __ subptr(len_reg, 4 * AESBlockSize);
2673        __ jmp(L_multiBlock_loopTop[k]);
2674 
2675        //singleBlock starts here
2676        __ align(OptoLoopAlignment);
2677        __ BIND(L_singleBlock_loopTop[k]);
2678        __ cmpptr(len_reg, 0); // any blocks left?
2679        __ jcc(Assembler::equal, L_exit);
2680        __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2681        __ movdqa(xmm_result1, xmm_result0);
2682 
2683        load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2684        __ pxor(xmm_result0, xmm_key_tmp0);
2685        // do the aes dec rounds
2686        for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
2687          // the java expanded key ordering is rotated one position from what we want
2688          load_key(xmm_key_tmp0, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2689          __ aesdec(xmm_result0, xmm_key_tmp0);
2690        }
2691        load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
2692        __ aesdeclast(xmm_result0, xmm_key_tmp0);
2693        __ pxor(xmm_result0, xmm_prev_block_cipher); // xor with the current r vector
2694        __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result0); // store into the next 16 bytes of output
2695        // no need to store r to memory until we exit
2696        __ movdqa(xmm_prev_block_cipher, xmm_result1); // set up next r vector with cipher input from this block
2697 
2698        __ addptr(pos, AESBlockSize);
2699        __ subptr(len_reg, AESBlockSize);
2700        __ jmp(L_singleBlock_loopTop[k]);
2701     }//for 128/192/256
2702 
2703     __ BIND(L_exit);
2704     __ movptr(rvec, rvec_param);                        // restore this since reused earlier
2705     __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
2706     handleSOERegisters(false /*restoring*/);
2707     __ movptr(rax, len_param);                          // return length
2708     __ leave();                                         // required for proper stackwalking of RuntimeStub frame
2709     __ ret(0);
2710 
2711     return start;
2712   }
2713 
2714   // CTR AES crypt.
2715   // In 32-bit stub, parallelize 4 blocks at a time
2716   // Arguments:
2717   //
2718   // Inputs:
2719   //   c_rarg0   - source byte array address
2720   //   c_rarg1   - destination byte array address
2721   //   c_rarg2   - K (key) in little endian int array
2722   //   c_rarg3   - counter vector byte array address
2723   //   c_rarg4   - input length
2724   //
2725   // Output:
2726   //   rax       - input length
2727   //
2728   address generate_counterMode_AESCrypt_Parallel() {
2729     assert(UseAES, "need AES instructions and misaligned SSE support");
2730     __ align(CodeEntryAlignment);
2731     StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
2732     address start = __ pc();
2733     const Register from        = rsi;      // source array address
2734     const Register to          = rdx;      // destination array address
2735     const Register key         = rcx;      // key array address
2736     const Register counter     = rdi;      // counter byte array initialized from initvector array address
2737                                            // and updated with the incremented counter in the end
2738     const Register len_reg     = rbx;
2739     const Register pos         = rax;
2740 
2741     __ enter(); // required for proper stackwalking of RuntimeStub frame
2742     handleSOERegisters(true /*saving*/); // save rbx, rsi, rdi
2743 
2744     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2745     // context for the registers used, where all instructions below are using 128-bit mode
2746     // On EVEX without VL and BW, these instructions will all be AVX.
2747     if (VM_Version::supports_avx512vlbw()) {
2748       __ movl(rdx, 0xffff);
2749       __ kmovdl(k1, rdx);
2750     }
2751 
2752     // load registers from incoming parameters
2753     const Address  from_param(rbp, 8+0);
2754     const Address  to_param  (rbp, 8+4);
2755     const Address  key_param (rbp, 8+8);
2756     const Address  rvec_param (rbp, 8+12);
2757     const Address  len_param  (rbp, 8+16);
2758     const Address  saved_counter_param(rbp, 8 + 20);
2759     const Address  used_addr_param(rbp, 8 + 24);
2760 
2761     __ movptr(from , from_param);
2762     __ movptr(to   , to_param);
2763     __ movptr(len_reg , len_param);
2764 
2765     // Use the partially used encrpyted counter from last invocation
2766     Label L_exit_preLoop, L_preLoop_start;
2767 
2768     // Use the registers 'counter' and 'key' here in this preloop
2769     // to hold of last 2 params 'used' and 'saved_encCounter_start'
2770     Register used = counter;
2771     Register saved_encCounter_start = key;
2772     Register used_addr = saved_encCounter_start;
2773 
2774     __ movptr(used_addr, used_addr_param);
2775     __ movptr(used, Address(used_addr, 0));
2776     __ movptr(saved_encCounter_start, saved_counter_param);
2777 
2778     __ BIND(L_preLoop_start);
2779     __ cmpptr(used, 16);
2780     __ jcc(Assembler::aboveEqual, L_exit_preLoop);
2781     __ cmpptr(len_reg, 0);
2782     __ jcc(Assembler::lessEqual, L_exit_preLoop);
2783     __ movb(rax, Address(saved_encCounter_start, used));
2784     __ xorb(rax, Address(from, 0));
2785     __ movb(Address(to, 0), rax);
2786     __ addptr(from, 1);
2787     __ addptr(to, 1);
2788     __ addptr(used, 1);
2789     __ subptr(len_reg, 1);
2790 
2791     __ jmp(L_preLoop_start);
2792 
2793     __ BIND(L_exit_preLoop);
2794     __ movptr(used_addr, used_addr_param);
2795     __ movptr(used_addr, used_addr_param);
2796     __ movl(Address(used_addr, 0), used);
2797 
2798     // load the parameters 'key' and 'counter'
2799     __ movptr(key, key_param);
2800     __ movptr(counter, rvec_param);
2801 
2802     // xmm register assignments for the loops below
2803     const XMMRegister xmm_curr_counter      = xmm0;
2804     const XMMRegister xmm_counter_shuf_mask = xmm1;  // need to be reloaded
2805     const XMMRegister xmm_key_shuf_mask     = xmm2;  // need to be reloaded
2806     const XMMRegister xmm_key               = xmm3;
2807     const XMMRegister xmm_result0           = xmm4;
2808     const XMMRegister xmm_result1           = xmm5;
2809     const XMMRegister xmm_result2           = xmm6;
2810     const XMMRegister xmm_result3           = xmm7;
2811     const XMMRegister xmm_from0             = xmm1;   //reuse XMM register
2812     const XMMRegister xmm_from1             = xmm2;
2813     const XMMRegister xmm_from2             = xmm3;
2814     const XMMRegister xmm_from3             = xmm4;
2815 
2816     //for key_128, key_192, key_256
2817     const int rounds[3] = {10, 12, 14};
2818     Label L_singleBlockLoopTop[3];
2819     Label L_multiBlock_loopTop[3];
2820     Label L_key192_top, L_key256_top;
2821     Label L_incCounter[3][4]; // 3: different key length,  4: 4 blocks at a time
2822     Label L_incCounter_single[3]; //for single block, key128, key192, key256
2823     Label L_processTail_insr[3], L_processTail_4_insr[3], L_processTail_2_insr[3], L_processTail_1_insr[3], L_processTail_exit_insr[3];
2824     Label L_processTail_extr[3], L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
2825 
2826     Label L_exit;
2827     const int PARALLEL_FACTOR = 4;  //because of the limited register number
2828 
2829     // initialize counter with initial counter
2830     __ movdqu(xmm_curr_counter, Address(counter, 0x00));
2831     __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2832     __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled for increase
2833 
2834     // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
2835     __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2836     __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2837     __ cmpl(rax, 52);
2838     __ jcc(Assembler::equal, L_key192_top);
2839     __ cmpl(rax, 60);
2840     __ jcc(Assembler::equal, L_key256_top);
2841 
2842     //key128 begins here
2843     __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2844 
2845 #define CTR_DoFour(opc, src_reg)               \
2846     __ opc(xmm_result0, src_reg);              \
2847     __ opc(xmm_result1, src_reg);              \
2848     __ opc(xmm_result2, src_reg);              \
2849     __ opc(xmm_result3, src_reg);
2850 
2851     // k == 0 :  generate code for key_128
2852     // k == 1 :  generate code for key_192
2853     // k == 2 :  generate code for key_256
2854     for (int k = 0; k < 3; ++k) {
2855       //multi blocks starts here
2856       __ align(OptoLoopAlignment);
2857       __ BIND(L_multiBlock_loopTop[k]);
2858       __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
2859       __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
2860 
2861       __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2862       __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2863 
2864       //load, then increase counters
2865       CTR_DoFour(movdqa, xmm_curr_counter);
2866       __ push(rbx);
2867       inc_counter(rbx, xmm_result1, 0x01, L_incCounter[k][0]);
2868       inc_counter(rbx, xmm_result2, 0x02, L_incCounter[k][1]);
2869       inc_counter(rbx, xmm_result3, 0x03, L_incCounter[k][2]);
2870       inc_counter(rbx, xmm_curr_counter, 0x04, L_incCounter[k][3]);
2871       __ pop (rbx);
2872 
2873       load_key(xmm_key, key, 0x00, xmm_key_shuf_mask); // load Round 0 key. interleaving for better performance
2874 
2875       CTR_DoFour(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
2876       CTR_DoFour(pxor, xmm_key);   //PXOR with Round 0 key
2877 
2878       for (int i = 1; i < rounds[k]; ++i) {
2879         load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2880         CTR_DoFour(aesenc, xmm_key);
2881       }
2882       load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2883       CTR_DoFour(aesenclast, xmm_key);
2884 
2885       // get next PARALLEL_FACTOR blocks into xmm_from registers
2886       __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2887       __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2888       __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2889 
2890       // PXOR with input text
2891       __ pxor(xmm_result0, xmm_from0); //result0 is xmm4
2892       __ pxor(xmm_result1, xmm_from1);
2893       __ pxor(xmm_result2, xmm_from2);
2894 
2895       // store PARALLEL_FACTOR results into the next 64 bytes of output
2896       __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2897       __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2898       __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2899 
2900       // do it here after xmm_result0 is saved, because xmm_from3 reuse the same register of xmm_result0.
2901       __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2902       __ pxor(xmm_result3, xmm_from3);
2903       __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2904 
2905       __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
2906       __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
2907       __ jmp(L_multiBlock_loopTop[k]);
2908 
2909       // singleBlock starts here
2910       __ align(OptoLoopAlignment);
2911       __ BIND(L_singleBlockLoopTop[k]);
2912       __ cmpptr(len_reg, 0);
2913       __ jcc(Assembler::equal, L_exit);
2914       __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2915       __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2916       __ movdqa(xmm_result0, xmm_curr_counter);
2917       load_key(xmm_key, key, 0x00, xmm_key_shuf_mask);
2918       __ push(rbx);//rbx is used for increasing counter
2919       inc_counter(rbx, xmm_curr_counter, 0x01, L_incCounter_single[k]);
2920       __ pop (rbx);
2921       __ pshufb(xmm_result0, xmm_counter_shuf_mask);
2922       __ pxor(xmm_result0, xmm_key);
2923       for (int i = 1; i < rounds[k]; i++) {
2924         load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2925         __ aesenc(xmm_result0, xmm_key);
2926       }
2927       load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2928       __ aesenclast(xmm_result0, xmm_key);
2929       __ cmpptr(len_reg, AESBlockSize);
2930       __ jcc(Assembler::less, L_processTail_insr[k]);
2931         __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2932         __ pxor(xmm_result0, xmm_from0);
2933         __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2934         __ addptr(pos, AESBlockSize);
2935         __ subptr(len_reg, AESBlockSize);
2936         __ jmp(L_singleBlockLoopTop[k]);
2937 
2938       __ BIND(L_processTail_insr[k]);                                               // Process the tail part of the input array
2939         __ addptr(pos, len_reg);                                                    // 1. Insert bytes from src array into xmm_from0 register
2940         __ testptr(len_reg, 8);
2941         __ jcc(Assembler::zero, L_processTail_4_insr[k]);
2942           __ subptr(pos,8);
2943           __ pinsrd(xmm_from0, Address(from, pos), 0);
2944           __ pinsrd(xmm_from0, Address(from, pos, Address::times_1, 4), 1);
2945         __ BIND(L_processTail_4_insr[k]);
2946         __ testptr(len_reg, 4);
2947         __ jcc(Assembler::zero, L_processTail_2_insr[k]);
2948           __ subptr(pos,4);
2949           __ pslldq(xmm_from0, 4);
2950           __ pinsrd(xmm_from0, Address(from, pos), 0);
2951         __ BIND(L_processTail_2_insr[k]);
2952         __ testptr(len_reg, 2);
2953         __ jcc(Assembler::zero, L_processTail_1_insr[k]);
2954           __ subptr(pos, 2);
2955           __ pslldq(xmm_from0, 2);
2956           __ pinsrw(xmm_from0, Address(from, pos), 0);
2957         __ BIND(L_processTail_1_insr[k]);
2958         __ testptr(len_reg, 1);
2959         __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
2960           __ subptr(pos, 1);
2961           __ pslldq(xmm_from0, 1);
2962           __ pinsrb(xmm_from0, Address(from, pos), 0);
2963         __ BIND(L_processTail_exit_insr[k]);
2964 
2965         __ movptr(saved_encCounter_start, saved_counter_param);
2966         __ movdqu(Address(saved_encCounter_start, 0), xmm_result0);               // 2. Perform pxor of the encrypted counter and plaintext Bytes.
2967         __ pxor(xmm_result0, xmm_from0);                                          //    Also the encrypted counter is saved for next invocation.
2968 
2969         __ testptr(len_reg, 8);
2970         __ jcc(Assembler::zero, L_processTail_4_extr[k]);                        // 3. Extract bytes from xmm_result0 into the dest. array
2971           __ pextrd(Address(to, pos), xmm_result0, 0);
2972           __ pextrd(Address(to, pos, Address::times_1, 4), xmm_result0, 1);
2973           __ psrldq(xmm_result0, 8);
2974           __ addptr(pos, 8);
2975         __ BIND(L_processTail_4_extr[k]);
2976         __ testptr(len_reg, 4);
2977         __ jcc(Assembler::zero, L_processTail_2_extr[k]);
2978           __ pextrd(Address(to, pos), xmm_result0, 0);
2979           __ psrldq(xmm_result0, 4);
2980           __ addptr(pos, 4);
2981         __ BIND(L_processTail_2_extr[k]);
2982         __ testptr(len_reg, 2);
2983         __ jcc(Assembler::zero, L_processTail_1_extr[k]);
2984           __ pextrb(Address(to, pos), xmm_result0, 0);
2985           __ pextrb(Address(to, pos, Address::times_1, 1), xmm_result0, 1);
2986           __ psrldq(xmm_result0, 2);
2987           __ addptr(pos, 2);
2988         __ BIND(L_processTail_1_extr[k]);
2989         __ testptr(len_reg, 1);
2990         __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
2991           __ pextrb(Address(to, pos), xmm_result0, 0);
2992 
2993         __ BIND(L_processTail_exit_extr[k]);
2994         __ movptr(used_addr, used_addr_param);
2995         __ movl(Address(used_addr, 0), len_reg);
2996         __ jmp(L_exit);
2997     }
2998 
2999     __ BIND(L_exit);
3000     __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
3001     __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
3002     __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
3003     handleSOERegisters(false /*restoring*/);
3004     __ movptr(rax, len_param); // return length
3005     __ leave();                // required for proper stackwalking of RuntimeStub frame
3006     __ ret(0);
3007 
3008     __ BIND (L_key192_top);
3009     __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
3010     __ jmp(L_multiBlock_loopTop[1]); //key192
3011 
3012     __ BIND (L_key256_top);
3013     __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
3014     __ jmp(L_multiBlock_loopTop[2]); //key192
3015 
3016     return start;
3017   }
3018 
3019   address generate_upper_word_mask() {
3020     __ align(64);
3021     StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3022     address start = __ pc();
3023     __ emit_data(0x00000000, relocInfo::none, 0);
3024     __ emit_data(0x00000000, relocInfo::none, 0);
3025     __ emit_data(0x00000000, relocInfo::none, 0);
3026     __ emit_data(0xFFFFFFFF, relocInfo::none, 0);
3027     return start;
3028   }
3029 
3030   address generate_shuffle_byte_flip_mask() {
3031     __ align(64);
3032     StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3033     address start = __ pc();
3034     __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3035     __ emit_data(0x08090a0b, relocInfo::none, 0);
3036     __ emit_data(0x04050607, relocInfo::none, 0);
3037     __ emit_data(0x00010203, relocInfo::none, 0);
3038     return start;
3039   }
3040 
3041   // ofs and limit are use for multi-block byte array.
3042   // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3043   address generate_sha1_implCompress(bool multi_block, const char *name) {
3044     __ align(CodeEntryAlignment);
3045     StubCodeMark mark(this, "StubRoutines", name);
3046     address start = __ pc();
3047 
3048     Register buf   = rax;
3049     Register state = rdx;
3050     Register ofs   = rcx;
3051     Register limit = rdi;
3052 
3053     const Address  buf_param(rbp, 8 + 0);
3054     const Address  state_param(rbp, 8 + 4);
3055     const Address  ofs_param(rbp, 8 + 8);
3056     const Address  limit_param(rbp, 8 + 12);
3057 
3058     const XMMRegister abcd = xmm0;
3059     const XMMRegister e0 = xmm1;
3060     const XMMRegister e1 = xmm2;
3061     const XMMRegister msg0 = xmm3;
3062 
3063     const XMMRegister msg1 = xmm4;
3064     const XMMRegister msg2 = xmm5;
3065     const XMMRegister msg3 = xmm6;
3066     const XMMRegister shuf_mask = xmm7;
3067 
3068     __ enter();
3069     __ subptr(rsp, 8 * wordSize);
3070     if (multi_block) {
3071       __ push(limit);
3072     }
3073     __ movptr(buf, buf_param);
3074     __ movptr(state, state_param);
3075     if (multi_block) {
3076       __ movptr(ofs, ofs_param);
3077       __ movptr(limit, limit_param);
3078     }
3079 
3080     __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3081       buf, state, ofs, limit, rsp, multi_block);
3082 
3083     if (multi_block) {
3084       __ pop(limit);
3085     }
3086     __ addptr(rsp, 8 * wordSize);
3087     __ leave();
3088     __ ret(0);
3089     return start;
3090   }
3091 
3092   address generate_pshuffle_byte_flip_mask() {
3093     __ align(64);
3094     StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3095     address start = __ pc();
3096     __ emit_data(0x00010203, relocInfo::none, 0);
3097     __ emit_data(0x04050607, relocInfo::none, 0);
3098     __ emit_data(0x08090a0b, relocInfo::none, 0);
3099     __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3100     return start;
3101   }
3102 
3103   // ofs and limit are use for multi-block byte array.
3104   // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3105  address generate_sha256_implCompress(bool multi_block, const char *name) {
3106     __ align(CodeEntryAlignment);
3107     StubCodeMark mark(this, "StubRoutines", name);
3108     address start = __ pc();
3109 
3110     Register buf = rbx;
3111     Register state = rsi;
3112     Register ofs = rdx;
3113     Register limit = rcx;
3114 
3115     const Address  buf_param(rbp, 8 + 0);
3116     const Address  state_param(rbp, 8 + 4);
3117     const Address  ofs_param(rbp, 8 + 8);
3118     const Address  limit_param(rbp, 8 + 12);
3119 
3120     const XMMRegister msg = xmm0;
3121     const XMMRegister state0 = xmm1;
3122     const XMMRegister state1 = xmm2;
3123     const XMMRegister msgtmp0 = xmm3;
3124 
3125     const XMMRegister msgtmp1 = xmm4;
3126     const XMMRegister msgtmp2 = xmm5;
3127     const XMMRegister msgtmp3 = xmm6;
3128     const XMMRegister msgtmp4 = xmm7;
3129 
3130     __ enter();
3131     __ subptr(rsp, 8 * wordSize);
3132     handleSOERegisters(true /*saving*/);
3133     __ movptr(buf, buf_param);
3134     __ movptr(state, state_param);
3135     if (multi_block) {
3136      __ movptr(ofs, ofs_param);
3137      __ movptr(limit, limit_param);
3138     }
3139 
3140     __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3141       buf, state, ofs, limit, rsp, multi_block);
3142 
3143     handleSOERegisters(false);
3144     __ addptr(rsp, 8 * wordSize);
3145     __ leave();
3146     __ ret(0);
3147     return start;
3148   }
3149 
3150   // byte swap x86 long
3151   address generate_ghash_long_swap_mask() {
3152     __ align(CodeEntryAlignment);
3153     StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3154     address start = __ pc();
3155     __ emit_data(0x0b0a0908, relocInfo::none, 0);
3156     __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
3157     __ emit_data(0x03020100, relocInfo::none, 0);
3158     __ emit_data(0x07060504, relocInfo::none, 0);
3159 
3160   return start;
3161   }
3162 
3163   // byte swap x86 byte array
3164   address generate_ghash_byte_swap_mask() {
3165     __ align(CodeEntryAlignment);
3166     StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3167     address start = __ pc();
3168     __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3169     __ emit_data(0x08090a0b, relocInfo::none, 0);
3170     __ emit_data(0x04050607, relocInfo::none, 0);
3171     __ emit_data(0x00010203, relocInfo::none, 0);
3172   return start;
3173   }
3174 
3175   /* Single and multi-block ghash operations */
3176   address generate_ghash_processBlocks() {
3177     assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
3178     __ align(CodeEntryAlignment);
3179     Label L_ghash_loop, L_exit;
3180     StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3181     address start = __ pc();
3182 
3183     const Register state        = rdi;
3184     const Register subkeyH      = rsi;
3185     const Register data         = rdx;
3186     const Register blocks       = rcx;
3187 
3188     const Address  state_param(rbp, 8+0);
3189     const Address  subkeyH_param(rbp, 8+4);
3190     const Address  data_param(rbp, 8+8);
3191     const Address  blocks_param(rbp, 8+12);
3192 
3193     const XMMRegister xmm_temp0 = xmm0;
3194     const XMMRegister xmm_temp1 = xmm1;
3195     const XMMRegister xmm_temp2 = xmm2;
3196     const XMMRegister xmm_temp3 = xmm3;
3197     const XMMRegister xmm_temp4 = xmm4;
3198     const XMMRegister xmm_temp5 = xmm5;
3199     const XMMRegister xmm_temp6 = xmm6;
3200     const XMMRegister xmm_temp7 = xmm7;
3201 
3202     __ enter();
3203     handleSOERegisters(true);  // Save registers
3204 
3205     // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3206     // context for the registers used, where all instructions below are using 128-bit mode
3207     // On EVEX without VL and BW, these instructions will all be AVX.
3208     if (VM_Version::supports_avx512vlbw()) {
3209       __ movl(rdx, 0xffff);
3210       __ kmovdl(k1, rdx);
3211     }
3212 
3213     __ movptr(state, state_param);
3214     __ movptr(subkeyH, subkeyH_param);
3215     __ movptr(data, data_param);
3216     __ movptr(blocks, blocks_param);
3217 
3218     __ movdqu(xmm_temp0, Address(state, 0));
3219     __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3220 
3221     __ movdqu(xmm_temp1, Address(subkeyH, 0));
3222     __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3223 
3224     __ BIND(L_ghash_loop);
3225     __ movdqu(xmm_temp2, Address(data, 0));
3226     __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3227 
3228     __ pxor(xmm_temp0, xmm_temp2);
3229 
3230     //
3231     // Multiply with the hash key
3232     //
3233     __ movdqu(xmm_temp3, xmm_temp0);
3234     __ pclmulqdq(xmm_temp3, xmm_temp1, 0);      // xmm3 holds a0*b0
3235     __ movdqu(xmm_temp4, xmm_temp0);
3236     __ pclmulqdq(xmm_temp4, xmm_temp1, 16);     // xmm4 holds a0*b1
3237 
3238     __ movdqu(xmm_temp5, xmm_temp0);
3239     __ pclmulqdq(xmm_temp5, xmm_temp1, 1);      // xmm5 holds a1*b0
3240     __ movdqu(xmm_temp6, xmm_temp0);
3241     __ pclmulqdq(xmm_temp6, xmm_temp1, 17);     // xmm6 holds a1*b1
3242 
3243     __ pxor(xmm_temp4, xmm_temp5);      // xmm4 holds a0*b1 + a1*b0
3244 
3245     __ movdqu(xmm_temp5, xmm_temp4);    // move the contents of xmm4 to xmm5
3246     __ psrldq(xmm_temp4, 8);    // shift by xmm4 64 bits to the right
3247     __ pslldq(xmm_temp5, 8);    // shift by xmm5 64 bits to the left
3248     __ pxor(xmm_temp3, xmm_temp5);
3249     __ pxor(xmm_temp6, xmm_temp4);      // Register pair <xmm6:xmm3> holds the result
3250                                         // of the carry-less multiplication of
3251                                         // xmm0 by xmm1.
3252 
3253     // We shift the result of the multiplication by one bit position
3254     // to the left to cope for the fact that the bits are reversed.
3255     __ movdqu(xmm_temp7, xmm_temp3);
3256     __ movdqu(xmm_temp4, xmm_temp6);
3257     __ pslld (xmm_temp3, 1);
3258     __ pslld(xmm_temp6, 1);
3259     __ psrld(xmm_temp7, 31);
3260     __ psrld(xmm_temp4, 31);
3261     __ movdqu(xmm_temp5, xmm_temp7);
3262     __ pslldq(xmm_temp4, 4);
3263     __ pslldq(xmm_temp7, 4);
3264     __ psrldq(xmm_temp5, 12);
3265     __ por(xmm_temp3, xmm_temp7);
3266     __ por(xmm_temp6, xmm_temp4);
3267     __ por(xmm_temp6, xmm_temp5);
3268 
3269     //
3270     // First phase of the reduction
3271     //
3272     // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
3273     // independently.
3274     __ movdqu(xmm_temp7, xmm_temp3);
3275     __ movdqu(xmm_temp4, xmm_temp3);
3276     __ movdqu(xmm_temp5, xmm_temp3);
3277     __ pslld(xmm_temp7, 31);    // packed right shift shifting << 31
3278     __ pslld(xmm_temp4, 30);    // packed right shift shifting << 30
3279     __ pslld(xmm_temp5, 25);    // packed right shift shifting << 25
3280     __ pxor(xmm_temp7, xmm_temp4);      // xor the shifted versions
3281     __ pxor(xmm_temp7, xmm_temp5);
3282     __ movdqu(xmm_temp4, xmm_temp7);
3283     __ pslldq(xmm_temp7, 12);
3284     __ psrldq(xmm_temp4, 4);
3285     __ pxor(xmm_temp3, xmm_temp7);      // first phase of the reduction complete
3286 
3287     //
3288     // Second phase of the reduction
3289     //
3290     // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
3291     // shift operations.
3292     __ movdqu(xmm_temp2, xmm_temp3);
3293     __ movdqu(xmm_temp7, xmm_temp3);
3294     __ movdqu(xmm_temp5, xmm_temp3);
3295     __ psrld(xmm_temp2, 1);     // packed left shifting >> 1
3296     __ psrld(xmm_temp7, 2);     // packed left shifting >> 2
3297     __ psrld(xmm_temp5, 7);     // packed left shifting >> 7
3298     __ pxor(xmm_temp2, xmm_temp7);      // xor the shifted versions
3299     __ pxor(xmm_temp2, xmm_temp5);
3300     __ pxor(xmm_temp2, xmm_temp4);
3301     __ pxor(xmm_temp3, xmm_temp2);
3302     __ pxor(xmm_temp6, xmm_temp3);      // the result is in xmm6
3303 
3304     __ decrement(blocks);
3305     __ jcc(Assembler::zero, L_exit);
3306     __ movdqu(xmm_temp0, xmm_temp6);
3307     __ addptr(data, 16);
3308     __ jmp(L_ghash_loop);
3309 
3310     __ BIND(L_exit);
3311        // Byte swap 16-byte result
3312     __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3313     __ movdqu(Address(state, 0), xmm_temp6);   // store the result
3314 
3315     handleSOERegisters(false);  // restore registers
3316     __ leave();
3317     __ ret(0);
3318     return start;
3319   }
3320 
3321   /**
3322    *  Arguments:
3323    *
3324    * Inputs:
3325    *   rsp(4)   - int crc
3326    *   rsp(8)   - byte* buf
3327    *   rsp(12)  - int length
3328    *
3329    * Ouput:
3330    *       rax   - int crc result
3331    */
3332   address generate_updateBytesCRC32() {
3333     assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3334 
3335     __ align(CodeEntryAlignment);
3336     StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3337 
3338     address start = __ pc();
3339 
3340     const Register crc   = rdx;  // crc
3341     const Register buf   = rsi;  // source java byte array address
3342     const Register len   = rcx;  // length
3343     const Register table = rdi;  // crc_table address (reuse register)
3344     const Register tmp   = rbx;
3345     assert_different_registers(crc, buf, len, table, tmp, rax);
3346 
3347     BLOCK_COMMENT("Entry:");
3348     __ enter(); // required for proper stackwalking of RuntimeStub frame
3349     __ push(rsi);
3350     __ push(rdi);
3351     __ push(rbx);
3352 
3353     Address crc_arg(rbp, 8 + 0);
3354     Address buf_arg(rbp, 8 + 4);
3355     Address len_arg(rbp, 8 + 8);
3356 
3357     // Load up:
3358     __ movl(crc,   crc_arg);
3359     __ movptr(buf, buf_arg);
3360     __ movl(len,   len_arg);
3361 
3362     __ kernel_crc32(crc, buf, len, table, tmp);
3363 
3364     __ movl(rax, crc);
3365     __ pop(rbx);
3366     __ pop(rdi);
3367     __ pop(rsi);
3368     __ leave(); // required for proper stackwalking of RuntimeStub frame
3369     __ ret(0);
3370 
3371     return start;
3372   }
3373 
3374   /**
3375   *  Arguments:
3376   *
3377   * Inputs:
3378   *   rsp(4)   - int crc
3379   *   rsp(8)   - byte* buf
3380   *   rsp(12)  - int length
3381   *   rsp(16)  - table_start - optional (present only when doing a library_calll,
3382   *              not used by x86 algorithm)
3383   *
3384   * Ouput:
3385   *       rax  - int crc result
3386   */
3387   address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3388     assert(UseCRC32CIntrinsics, "need SSE4_2");
3389     __ align(CodeEntryAlignment);
3390     StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3391     address start = __ pc();
3392     const Register crc = rax;  // crc
3393     const Register buf = rcx;  // source java byte array address
3394     const Register len = rdx;  // length
3395     const Register d = rbx;
3396     const Register g = rsi;
3397     const Register h = rdi;
3398     const Register empty = 0; // will never be used, in order not
3399                               // to change a signature for crc32c_IPL_Alg2_Alt2
3400                               // between 64/32 I'm just keeping it here
3401     assert_different_registers(crc, buf, len, d, g, h);
3402 
3403     BLOCK_COMMENT("Entry:");
3404     __ enter(); // required for proper stackwalking of RuntimeStub frame
3405     Address crc_arg(rsp, 4 + 4 + 0); // ESP+4 +
3406                                      // we need to add additional 4 because __ enter
3407                                      // have just pushed ebp on a stack
3408     Address buf_arg(rsp, 4 + 4 + 4);
3409     Address len_arg(rsp, 4 + 4 + 8);
3410       // Load up:
3411       __ movl(crc, crc_arg);
3412       __ movl(buf, buf_arg);
3413       __ movl(len, len_arg);
3414       __ push(d);
3415       __ push(g);
3416       __ push(h);
3417       __ crc32c_ipl_alg2_alt2(crc, buf, len,
3418                               d, g, h,
3419                               empty, empty, empty,
3420                               xmm0, xmm1, xmm2,
3421                               is_pclmulqdq_supported);
3422       __ pop(h);
3423       __ pop(g);
3424       __ pop(d);
3425     __ leave(); // required for proper stackwalking of RuntimeStub frame
3426     __ ret(0);
3427 
3428     return start;
3429   }
3430 
3431  address generate_libmExp() {
3432     address start = __ pc();
3433 
3434     const XMMRegister x0  = xmm0;
3435     const XMMRegister x1  = xmm1;
3436     const XMMRegister x2  = xmm2;
3437     const XMMRegister x3  = xmm3;
3438 
3439     const XMMRegister x4  = xmm4;
3440     const XMMRegister x5  = xmm5;
3441     const XMMRegister x6  = xmm6;
3442     const XMMRegister x7  = xmm7;
3443 
3444     const Register tmp   = rbx;
3445 
3446     BLOCK_COMMENT("Entry:");
3447     __ enter(); // required for proper stackwalking of RuntimeStub frame
3448     __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3449     __ leave(); // required for proper stackwalking of RuntimeStub frame
3450     __ ret(0);
3451 
3452     return start;
3453 
3454   }
3455 
3456  address generate_libmLog() {
3457    address start = __ pc();
3458 
3459    const XMMRegister x0 = xmm0;
3460    const XMMRegister x1 = xmm1;
3461    const XMMRegister x2 = xmm2;
3462    const XMMRegister x3 = xmm3;
3463 
3464    const XMMRegister x4 = xmm4;
3465    const XMMRegister x5 = xmm5;
3466    const XMMRegister x6 = xmm6;
3467    const XMMRegister x7 = xmm7;
3468 
3469    const Register tmp = rbx;
3470 
3471    BLOCK_COMMENT("Entry:");
3472    __ enter(); // required for proper stackwalking of RuntimeStub frame
3473    __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3474    __ leave(); // required for proper stackwalking of RuntimeStub frame
3475    __ ret(0);
3476 
3477    return start;
3478 
3479  }
3480 
3481  address generate_libmLog10() {
3482    address start = __ pc();
3483 
3484    const XMMRegister x0 = xmm0;
3485    const XMMRegister x1 = xmm1;
3486    const XMMRegister x2 = xmm2;
3487    const XMMRegister x3 = xmm3;
3488 
3489    const XMMRegister x4 = xmm4;
3490    const XMMRegister x5 = xmm5;
3491    const XMMRegister x6 = xmm6;
3492    const XMMRegister x7 = xmm7;
3493 
3494    const Register tmp = rbx;
3495 
3496    BLOCK_COMMENT("Entry:");
3497    __ enter(); // required for proper stackwalking of RuntimeStub frame
3498    __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3499    __ leave(); // required for proper stackwalking of RuntimeStub frame
3500    __ ret(0);
3501 
3502    return start;
3503 
3504  }
3505 
3506  address generate_libmPow() {
3507    address start = __ pc();
3508 
3509    const XMMRegister x0 = xmm0;
3510    const XMMRegister x1 = xmm1;
3511    const XMMRegister x2 = xmm2;
3512    const XMMRegister x3 = xmm3;
3513 
3514    const XMMRegister x4 = xmm4;
3515    const XMMRegister x5 = xmm5;
3516    const XMMRegister x6 = xmm6;
3517    const XMMRegister x7 = xmm7;
3518 
3519    const Register tmp = rbx;
3520 
3521    BLOCK_COMMENT("Entry:");
3522    __ enter(); // required for proper stackwalking of RuntimeStub frame
3523    __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3524    __ leave(); // required for proper stackwalking of RuntimeStub frame
3525    __ ret(0);
3526 
3527    return start;
3528 
3529  }
3530 
3531  address generate_libm_reduce_pi04l() {
3532    address start = __ pc();
3533 
3534    BLOCK_COMMENT("Entry:");
3535    __ libm_reduce_pi04l(rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3536 
3537    return start;
3538 
3539  }
3540 
3541  address generate_libm_sin_cos_huge() {
3542    address start = __ pc();
3543 
3544    const XMMRegister x0 = xmm0;
3545    const XMMRegister x1 = xmm1;
3546 
3547    BLOCK_COMMENT("Entry:");
3548    __ libm_sincos_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3549 
3550    return start;
3551 
3552  }
3553 
3554  address generate_libmSin() {
3555    address start = __ pc();
3556 
3557    const XMMRegister x0 = xmm0;
3558    const XMMRegister x1 = xmm1;
3559    const XMMRegister x2 = xmm2;
3560    const XMMRegister x3 = xmm3;
3561 
3562    const XMMRegister x4 = xmm4;
3563    const XMMRegister x5 = xmm5;
3564    const XMMRegister x6 = xmm6;
3565    const XMMRegister x7 = xmm7;
3566 
3567    BLOCK_COMMENT("Entry:");
3568    __ enter(); // required for proper stackwalking of RuntimeStub frame
3569    __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rdx);
3570    __ leave(); // required for proper stackwalking of RuntimeStub frame
3571    __ ret(0);
3572 
3573    return start;
3574 
3575  }
3576 
3577  address generate_libmCos() {
3578    address start = __ pc();
3579 
3580    const XMMRegister x0 = xmm0;
3581    const XMMRegister x1 = xmm1;
3582    const XMMRegister x2 = xmm2;
3583    const XMMRegister x3 = xmm3;
3584 
3585    const XMMRegister x4 = xmm4;
3586    const XMMRegister x5 = xmm5;
3587    const XMMRegister x6 = xmm6;
3588    const XMMRegister x7 = xmm7;
3589 
3590    const Register tmp = rbx;
3591 
3592    BLOCK_COMMENT("Entry:");
3593    __ enter(); // required for proper stackwalking of RuntimeStub frame
3594    __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3595    __ leave(); // required for proper stackwalking of RuntimeStub frame
3596    __ ret(0);
3597 
3598    return start;
3599 
3600  }
3601 
3602  address generate_libm_tan_cot_huge() {
3603    address start = __ pc();
3604 
3605    const XMMRegister x0 = xmm0;
3606    const XMMRegister x1 = xmm1;
3607 
3608    BLOCK_COMMENT("Entry:");
3609    __ libm_tancot_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3610 
3611    return start;
3612 
3613  }
3614 
3615  address generate_libmTan() {
3616    address start = __ pc();
3617 
3618    const XMMRegister x0 = xmm0;
3619    const XMMRegister x1 = xmm1;
3620    const XMMRegister x2 = xmm2;
3621    const XMMRegister x3 = xmm3;
3622 
3623    const XMMRegister x4 = xmm4;
3624    const XMMRegister x5 = xmm5;
3625    const XMMRegister x6 = xmm6;
3626    const XMMRegister x7 = xmm7;
3627 
3628    const Register tmp = rbx;
3629 
3630    BLOCK_COMMENT("Entry:");
3631    __ enter(); // required for proper stackwalking of RuntimeStub frame
3632    __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3633    __ leave(); // required for proper stackwalking of RuntimeStub frame
3634    __ ret(0);
3635 
3636    return start;
3637 
3638  }
3639 
3640   // Safefetch stubs.
3641   void generate_safefetch(const char* name, int size, address* entry,
3642                           address* fault_pc, address* continuation_pc) {
3643     // safefetch signatures:
3644     //   int      SafeFetch32(int*      adr, int      errValue);
3645     //   intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3646 
3647     StubCodeMark mark(this, "StubRoutines", name);
3648 
3649     // Entry point, pc or function descriptor.
3650     *entry = __ pc();
3651 
3652     __ movl(rax, Address(rsp, 0x8));
3653     __ movl(rcx, Address(rsp, 0x4));
3654     // Load *adr into eax, may fault.
3655     *fault_pc = __ pc();
3656     switch (size) {
3657       case 4:
3658         // int32_t
3659         __ movl(rax, Address(rcx, 0));
3660         break;
3661       case 8:
3662         // int64_t
3663         Unimplemented();
3664         break;
3665       default:
3666         ShouldNotReachHere();
3667     }
3668 
3669     // Return errValue or *adr.
3670     *continuation_pc = __ pc();
3671     __ ret(0);
3672   }
3673 
3674  public:
3675   // Information about frame layout at time of blocking runtime call.
3676   // Note that we only have to preserve callee-saved registers since
3677   // the compilers are responsible for supplying a continuation point
3678   // if they expect all registers to be preserved.
3679   enum layout {
3680     thread_off,    // last_java_sp
3681     arg1_off,
3682     arg2_off,
3683     rbp_off,       // callee saved register
3684     ret_pc,
3685     framesize
3686   };
3687 
3688  private:
3689 
3690 #undef  __
3691 #define __ masm->
3692 
3693   //------------------------------------------------------------------------------------------------------------------------
3694   // Continuation point for throwing of implicit exceptions that are not handled in
3695   // the current activation. Fabricates an exception oop and initiates normal
3696   // exception dispatching in this frame.
3697   //
3698   // Previously the compiler (c2) allowed for callee save registers on Java calls.
3699   // This is no longer true after adapter frames were removed but could possibly
3700   // be brought back in the future if the interpreter code was reworked and it
3701   // was deemed worthwhile. The comment below was left to describe what must
3702   // happen here if callee saves were resurrected. As it stands now this stub
3703   // could actually be a vanilla BufferBlob and have now oopMap at all.
3704   // Since it doesn't make much difference we've chosen to leave it the
3705   // way it was in the callee save days and keep the comment.
3706 
3707   // If we need to preserve callee-saved values we need a callee-saved oop map and
3708   // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3709   // If the compiler needs all registers to be preserved between the fault
3710   // point and the exception handler then it must assume responsibility for that in
3711   // AbstractCompiler::continuation_for_implicit_null_exception or
3712   // continuation_for_implicit_division_by_zero_exception. All other implicit
3713   // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3714   // either at call sites or otherwise assume that stack unwinding will be initiated,
3715   // so caller saved registers were assumed volatile in the compiler.
3716   address generate_throw_exception(const char* name, address runtime_entry,
3717                                    Register arg1 = noreg, Register arg2 = noreg) {
3718 
3719     int insts_size = 256;
3720     int locs_size  = 32;
3721 
3722     CodeBuffer code(name, insts_size, locs_size);
3723     OopMapSet* oop_maps  = new OopMapSet();
3724     MacroAssembler* masm = new MacroAssembler(&code);
3725 
3726     address start = __ pc();
3727 
3728     // This is an inlined and slightly modified version of call_VM
3729     // which has the ability to fetch the return PC out of
3730     // thread-local storage and also sets up last_Java_sp slightly
3731     // differently than the real call_VM
3732     Register java_thread = rbx;
3733     __ get_thread(java_thread);
3734 
3735     __ enter(); // required for proper stackwalking of RuntimeStub frame
3736 
3737     // pc and rbp, already pushed
3738     __ subptr(rsp, (framesize-2) * wordSize); // prolog
3739 
3740     // Frame is now completed as far as size and linkage.
3741 
3742     int frame_complete = __ pc() - start;
3743 
3744     // push java thread (becomes first argument of C function)
3745     __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3746     if (arg1 != noreg) {
3747       __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3748     }
3749     if (arg2 != noreg) {
3750       assert(arg1 != noreg, "missing reg arg");
3751       __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3752     }
3753 
3754     // Set up last_Java_sp and last_Java_fp
3755     __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3756 
3757     // Call runtime
3758     BLOCK_COMMENT("call runtime_entry");
3759     __ call(RuntimeAddress(runtime_entry));
3760     // Generate oop map
3761     OopMap* map =  new OopMap(framesize, 0);
3762     oop_maps->add_gc_map(__ pc() - start, map);
3763 
3764     // restore the thread (cannot use the pushed argument since arguments
3765     // may be overwritten by C code generated by an optimizing compiler);
3766     // however can use the register value directly if it is callee saved.
3767     __ get_thread(java_thread);
3768 
3769     __ reset_last_Java_frame(java_thread, true);
3770 
3771     __ leave(); // required for proper stackwalking of RuntimeStub frame
3772 
3773     // check for pending exceptions
3774 #ifdef ASSERT
3775     Label L;
3776     __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3777     __ jcc(Assembler::notEqual, L);
3778     __ should_not_reach_here();
3779     __ bind(L);
3780 #endif /* ASSERT */
3781     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3782 
3783 
3784     RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3785     return stub->entry_point();
3786   }
3787 
3788 
3789   void create_control_words() {
3790     // Round to nearest, 53-bit mode, exceptions masked
3791     StubRoutines::_fpu_cntrl_wrd_std   = 0x027F;
3792     // Round to zero, 53-bit mode, exception mased
3793     StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3794     // Round to nearest, 24-bit mode, exceptions masked
3795     StubRoutines::_fpu_cntrl_wrd_24    = 0x007F;
3796     // Round to nearest, 64-bit mode, exceptions masked
3797     StubRoutines::_fpu_cntrl_wrd_64    = 0x037F;
3798     // Round to nearest, 64-bit mode, exceptions masked
3799     StubRoutines::_mxcsr_std           = 0x1F80;
3800     // Note: the following two constants are 80-bit values
3801     //       layout is critical for correct loading by FPU.
3802     // Bias for strict fp multiply/divide
3803     StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3804     StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3805     StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3806     // Un-Bias for strict fp multiply/divide
3807     StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3808     StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3809     StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3810   }
3811 
3812   //---------------------------------------------------------------------------
3813   // Initialization
3814 
3815   void generate_initial() {
3816     // Generates all stubs and initializes the entry points
3817 
3818     //------------------------------------------------------------------------------------------------------------------------
3819     // entry points that exist in all platforms
3820     // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3821     //       the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3822     StubRoutines::_forward_exception_entry      = generate_forward_exception();
3823 
3824     StubRoutines::_call_stub_entry              =
3825       generate_call_stub(StubRoutines::_call_stub_return_address);
3826     // is referenced by megamorphic call
3827     StubRoutines::_catch_exception_entry        = generate_catch_exception();
3828 
3829     // These are currently used by Solaris/Intel
3830     StubRoutines::_atomic_xchg_entry            = generate_atomic_xchg();
3831 
3832     // platform dependent
3833     create_control_words();
3834 
3835     StubRoutines::x86::_verify_mxcsr_entry                 = generate_verify_mxcsr();
3836     StubRoutines::x86::_verify_fpu_cntrl_wrd_entry         = generate_verify_fpu_cntrl_wrd();
3837     StubRoutines::_d2i_wrapper                              = generate_d2i_wrapper(T_INT,
3838                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3839     StubRoutines::_d2l_wrapper                              = generate_d2i_wrapper(T_LONG,
3840                                                                                    CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3841 
3842     // Build this early so it's available for the interpreter
3843     StubRoutines::_throw_StackOverflowError_entry          = generate_throw_exception("StackOverflowError throw_exception",
3844                                                                                       CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3845     StubRoutines::_throw_delayed_StackOverflowError_entry  = generate_throw_exception("delayed StackOverflowError throw_exception",
3846                                                                                       CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError));
3847 
3848     if (UseCRC32Intrinsics) {
3849       // set table address before stub generation which use it
3850       StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3851       StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3852     }
3853 
3854     if (UseCRC32CIntrinsics) {
3855       bool supports_clmul = VM_Version::supports_clmul();
3856       StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3857       StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3858       StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3859     }
3860     if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
3861       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3862           vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3863           vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3864         StubRoutines::x86::_L_2il0floatpacket_0_adr = (address)StubRoutines::x86::_L_2il0floatpacket_0;
3865         StubRoutines::x86::_Pi4Inv_adr = (address)StubRoutines::x86::_Pi4Inv;
3866         StubRoutines::x86::_Pi4x3_adr = (address)StubRoutines::x86::_Pi4x3;
3867         StubRoutines::x86::_Pi4x4_adr = (address)StubRoutines::x86::_Pi4x4;
3868         StubRoutines::x86::_ones_adr = (address)StubRoutines::x86::_ones;
3869       }
3870       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3871         StubRoutines::_dexp = generate_libmExp();
3872       }
3873       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3874         StubRoutines::_dlog = generate_libmLog();
3875       }
3876       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3877         StubRoutines::_dlog10 = generate_libmLog10();
3878       }
3879       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3880         StubRoutines::_dpow = generate_libmPow();
3881       }
3882       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3883         vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3884         vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3885         StubRoutines::_dlibm_reduce_pi04l = generate_libm_reduce_pi04l();
3886       }
3887       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3888         vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3889         StubRoutines::_dlibm_sin_cos_huge = generate_libm_sin_cos_huge();
3890       }
3891       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3892         StubRoutines::_dsin = generate_libmSin();
3893       }
3894       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3895         StubRoutines::_dcos = generate_libmCos();
3896       }
3897       if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3898         StubRoutines::_dlibm_tan_cot_huge = generate_libm_tan_cot_huge();
3899         StubRoutines::_dtan = generate_libmTan();
3900       }
3901     }
3902   }
3903 
3904   void generate_all() {
3905     // Generates all stubs and initializes the entry points
3906 
3907     // These entry points require SharedInfo::stack0 to be set up in non-core builds
3908     // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3909     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3910     StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3911     StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3912 
3913     //------------------------------------------------------------------------------------------------------------------------
3914     // entry points that are platform specific
3915 
3916     // support for verify_oop (must happen after universe_init)
3917     StubRoutines::_verify_oop_subroutine_entry     = generate_verify_oop();
3918 
3919     // arraycopy stubs used by compilers
3920     generate_arraycopy_stubs();
3921 
3922     // don't bother generating these AES intrinsic stubs unless global flag is set
3923     if (UseAESIntrinsics) {
3924       StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask();  // might be needed by the others
3925 
3926       StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3927       StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3928       StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3929       StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3930     }
3931 
3932     if (UseAESCTRIntrinsics) {
3933       StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
3934       StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
3935     }
3936 
3937     if (UseSHA1Intrinsics) {
3938       StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
3939       StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
3940       StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
3941       StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
3942     }
3943     if (UseSHA256Intrinsics) {
3944       StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
3945       StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
3946       StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
3947       StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
3948     }
3949 
3950     // Generate GHASH intrinsics code
3951     if (UseGHASHIntrinsics) {
3952       StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
3953       StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
3954       StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3955     }
3956 
3957     // Safefetch stubs.
3958     generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3959                                                    &StubRoutines::_safefetch32_fault_pc,
3960                                                    &StubRoutines::_safefetch32_continuation_pc);
3961     StubRoutines::_safefetchN_entry           = StubRoutines::_safefetch32_entry;
3962     StubRoutines::_safefetchN_fault_pc        = StubRoutines::_safefetch32_fault_pc;
3963     StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3964   }
3965 
3966 
3967  public:
3968   StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3969     if (all) {
3970       generate_all();
3971     } else {
3972       generate_initial();
3973     }
3974   }
3975 }; // end class declaration
3976 
3977 
3978 void StubGenerator_generate(CodeBuffer* code, bool all) {
3979   StubGenerator g(code, all);
3980 }