1 /*
   2  * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/assembler.hpp"
  27 #include "asm/macroAssembler.hpp"
  28 #include "classfile/vmIntrinsics.hpp"
  29 #include "compiler/oopMap.hpp"
  30 #include "gc/shared/barrierSet.hpp"
  31 #include "gc/shared/barrierSetAssembler.hpp"
  32 #include "gc/shared/barrierSetNMethod.hpp"
  33 #include "gc/shared/gc_globals.hpp"
  34 #include "memory/universe.hpp"
  35 #include "prims/jvmtiExport.hpp"
  36 #include "runtime/arguments.hpp"
  37 #include "runtime/javaThread.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "utilities/macros.hpp"
  41 #include "vmreg_x86.inline.hpp"
  42 #include "stubGenerator_x86_64.hpp"
  43 #ifdef COMPILER2
  44 #include "opto/runtime.hpp"
  45 #include "opto/c2_globals.hpp"
  46 #endif
  47 #if INCLUDE_JVMCI
  48 #include "jvmci/jvmci_globals.hpp"
  49 #endif
  50 #if INCLUDE_JFR
  51 #include "jfr/support/jfrIntrinsics.hpp"
  52 #endif
  53 
  54 // For a more detailed description of the stub routine structure
  55 // see the comment in stubRoutines.hpp
  56 
  57 #define __ _masm->
  58 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
  59 
  60 #ifdef PRODUCT
  61 #define BLOCK_COMMENT(str) /* nothing */
  62 #else
  63 #define BLOCK_COMMENT(str) __ block_comment(str)
  64 #endif // PRODUCT
  65 
  66 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  67 
  68 //
  69 // Linux Arguments:
  70 //    c_rarg0:   call wrapper address                   address
  71 //    c_rarg1:   result                                 address
  72 //    c_rarg2:   result type                            BasicType
  73 //    c_rarg3:   method                                 Method*
  74 //    c_rarg4:   (interpreter) entry point              address
  75 //    c_rarg5:   parameters                             intptr_t*
  76 //    16(rbp): parameter size (in words)              int
  77 //    24(rbp): thread                                 Thread*
  78 //
  79 //     [ return_from_Java     ] <--- rsp
  80 //     [ argument word n      ]
  81 //      ...
  82 // -12 [ argument word 1      ]
  83 // -11 [ saved r15            ] <--- rsp_after_call
  84 // -10 [ saved r14            ]
  85 //  -9 [ saved r13            ]
  86 //  -8 [ saved r12            ]
  87 //  -7 [ saved rbx            ]
  88 //  -6 [ call wrapper         ]
  89 //  -5 [ result               ]
  90 //  -4 [ result type          ]
  91 //  -3 [ method               ]
  92 //  -2 [ entry point          ]
  93 //  -1 [ parameters           ]
  94 //   0 [ saved rbp            ] <--- rbp
  95 //   1 [ return address       ]
  96 //   2 [ parameter size       ]
  97 //   3 [ thread               ]
  98 //
  99 // Windows Arguments:
 100 //    c_rarg0:   call wrapper address                   address
 101 //    c_rarg1:   result                                 address
 102 //    c_rarg2:   result type                            BasicType
 103 //    c_rarg3:   method                                 Method*
 104 //    48(rbp): (interpreter) entry point              address
 105 //    56(rbp): parameters                             intptr_t*
 106 //    64(rbp): parameter size (in words)              int
 107 //    72(rbp): thread                                 Thread*
 108 //
 109 //     [ return_from_Java     ] <--- rsp
 110 //     [ argument word n      ]
 111 //      ...
 112 // -60 [ argument word 1      ]
 113 // -59 [ saved xmm31          ] <--- rsp after_call
 114 //     [ saved xmm16-xmm30    ] (EVEX enabled, else the space is blank)
 115 // -27 [ saved xmm15          ]
 116 //     [ saved xmm7-xmm14     ]
 117 //  -9 [ saved xmm6           ] (each xmm register takes 2 slots)
 118 //  -7 [ saved r15            ]
 119 //  -6 [ saved r14            ]
 120 //  -5 [ saved r13            ]
 121 //  -4 [ saved r12            ]
 122 //  -3 [ saved rdi            ]
 123 //  -2 [ saved rsi            ]
 124 //  -1 [ saved rbx            ]
 125 //   0 [ saved rbp            ] <--- rbp
 126 //   1 [ return address       ]
 127 //   2 [ call wrapper         ]
 128 //   3 [ result               ]
 129 //   4 [ result type          ]
 130 //   5 [ method               ]
 131 //   6 [ entry point          ]
 132 //   7 [ parameters           ]
 133 //   8 [ parameter size       ]
 134 //   9 [ thread               ]
 135 //
 136 //    Windows reserves the callers stack space for arguments 1-4.
 137 //    We spill c_rarg0-c_rarg3 to this space.
 138 
 139 // Call stub stack layout word offsets from rbp
 140 #ifdef _WIN64
 141 enum call_stub_layout {
 142   xmm_save_first     = 6,  // save from xmm6
 143   xmm_save_last      = 31, // to xmm31
 144   xmm_save_base      = -9,
 145   rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
 146   r15_off            = -7,
 147   r14_off            = -6,
 148   r13_off            = -5,
 149   r12_off            = -4,
 150   rdi_off            = -3,
 151   rsi_off            = -2,
 152   rbx_off            = -1,
 153   rbp_off            =  0,
 154   retaddr_off        =  1,
 155   call_wrapper_off   =  2,
 156   result_off         =  3,
 157   result_type_off    =  4,
 158   method_off         =  5,
 159   entry_point_off    =  6,
 160   parameters_off     =  7,
 161   parameter_size_off =  8,
 162   thread_off         =  9
 163 };
 164 
 165 static Address xmm_save(int reg) {
 166   assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
 167   return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
 168 }
 169 #else // !_WIN64
 170 enum call_stub_layout {
 171   rsp_after_call_off = -12,
 172   mxcsr_off          = rsp_after_call_off,
 173   r15_off            = -11,
 174   r14_off            = -10,
 175   r13_off            = -9,
 176   r12_off            = -8,
 177   rbx_off            = -7,
 178   call_wrapper_off   = -6,
 179   result_off         = -5,
 180   result_type_off    = -4,
 181   method_off         = -3,
 182   entry_point_off    = -2,
 183   parameters_off     = -1,
 184   rbp_off            =  0,
 185   retaddr_off        =  1,
 186   parameter_size_off =  2,
 187   thread_off         =  3
 188 };
 189 #endif // _WIN64
 190 
 191 address StubGenerator::generate_call_stub(address& return_address) {
 192 
 193   assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 194          (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 195          "adjust this code");
 196   StubCodeMark mark(this, "StubRoutines", "call_stub");
 197   address start = __ pc();
 198 
 199   // same as in generate_catch_exception()!
 200   const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 201 
 202   const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 203   const Address result        (rbp, result_off         * wordSize);
 204   const Address result_type   (rbp, result_type_off    * wordSize);
 205   const Address method        (rbp, method_off         * wordSize);
 206   const Address entry_point   (rbp, entry_point_off    * wordSize);
 207   const Address parameters    (rbp, parameters_off     * wordSize);
 208   const Address parameter_size(rbp, parameter_size_off * wordSize);
 209 
 210   // same as in generate_catch_exception()!
 211   const Address thread        (rbp, thread_off         * wordSize);
 212 
 213   const Address r15_save(rbp, r15_off * wordSize);
 214   const Address r14_save(rbp, r14_off * wordSize);
 215   const Address r13_save(rbp, r13_off * wordSize);
 216   const Address r12_save(rbp, r12_off * wordSize);
 217   const Address rbx_save(rbp, rbx_off * wordSize);
 218 
 219   // stub code
 220   __ enter();
 221   __ subptr(rsp, -rsp_after_call_off * wordSize);
 222 
 223   // save register parameters
 224 #ifndef _WIN64
 225   __ movptr(parameters,   c_rarg5); // parameters
 226   __ movptr(entry_point,  c_rarg4); // entry_point
 227 #endif
 228 
 229   __ movptr(method,       c_rarg3); // method
 230   __ movl(result_type,  c_rarg2);   // result type
 231   __ movptr(result,       c_rarg1); // result
 232   __ movptr(call_wrapper, c_rarg0); // call wrapper
 233 
 234   // save regs belonging to calling function
 235   __ movptr(rbx_save, rbx);
 236   __ movptr(r12_save, r12);
 237   __ movptr(r13_save, r13);
 238   __ movptr(r14_save, r14);
 239   __ movptr(r15_save, r15);
 240 
 241 #ifdef _WIN64
 242   int last_reg = 15;
 243   if (UseAVX > 2) {
 244     last_reg = 31;
 245   }
 246   if (VM_Version::supports_evex()) {
 247     for (int i = xmm_save_first; i <= last_reg; i++) {
 248       __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
 249     }
 250   } else {
 251     for (int i = xmm_save_first; i <= last_reg; i++) {
 252       __ movdqu(xmm_save(i), as_XMMRegister(i));
 253     }
 254   }
 255 
 256   const Address rdi_save(rbp, rdi_off * wordSize);
 257   const Address rsi_save(rbp, rsi_off * wordSize);
 258 
 259   __ movptr(rsi_save, rsi);
 260   __ movptr(rdi_save, rdi);
 261 #else
 262   const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 263   {
 264     Label skip_ldmx;
 265     __ stmxcsr(mxcsr_save);
 266     __ movl(rax, mxcsr_save);
 267     __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
 268     ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 269     __ cmp32(rax, mxcsr_std, rscratch1);
 270     __ jcc(Assembler::equal, skip_ldmx);
 271     __ ldmxcsr(mxcsr_std, rscratch1);
 272     __ bind(skip_ldmx);
 273   }
 274 #endif
 275 
 276   // Load up thread register
 277   __ movptr(r15_thread, thread);
 278   __ reinit_heapbase();
 279 
 280 #ifdef ASSERT
 281   // make sure we have no pending exceptions
 282   {
 283     Label L;
 284     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 285     __ jcc(Assembler::equal, L);
 286     __ stop("StubRoutines::call_stub: entered with pending exception");
 287     __ bind(L);
 288   }
 289 #endif
 290 
 291   // pass parameters if any
 292   BLOCK_COMMENT("pass parameters if any");
 293   Label parameters_done;
 294   __ movl(c_rarg3, parameter_size);
 295   __ testl(c_rarg3, c_rarg3);
 296   __ jcc(Assembler::zero, parameters_done);
 297 
 298   Label loop;
 299   __ movptr(c_rarg2, parameters);       // parameter pointer
 300   __ movl(c_rarg1, c_rarg3);            // parameter counter is in c_rarg1
 301   __ BIND(loop);
 302   __ movptr(rax, Address(c_rarg2, 0));// get parameter
 303   __ addptr(c_rarg2, wordSize);       // advance to next parameter
 304   __ decrementl(c_rarg1);             // decrement counter
 305   __ push(rax);                       // pass parameter
 306   __ jcc(Assembler::notZero, loop);
 307 
 308   // call Java function
 309   __ BIND(parameters_done);
 310   __ movptr(rbx, method);             // get Method*
 311   __ movptr(c_rarg1, entry_point);    // get entry_point
 312   __ mov(r13, rsp);                   // set sender sp
 313   BLOCK_COMMENT("call Java function");
 314   __ call(c_rarg1);
 315 
 316   BLOCK_COMMENT("call_stub_return_address:");
 317   return_address = __ pc();
 318 
 319   // store result depending on type (everything that is not
 320   // T_OBJECT, T_PRIMITIVE_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 321   __ movptr(r13, result);
 322   Label is_long, is_float, is_double, check_prim, exit;
 323   __ movl(rbx, result_type);
 324   __ cmpl(rbx, T_OBJECT);
 325   __ jcc(Assembler::equal, check_prim);
 326   __ cmpl(rbx, T_PRIMITIVE_OBJECT);
 327   __ jcc(Assembler::equal, check_prim);
 328   __ cmpl(rbx, T_LONG);
 329   __ jcc(Assembler::equal, is_long);
 330   __ cmpl(rbx, T_FLOAT);
 331   __ jcc(Assembler::equal, is_float);
 332   __ cmpl(rbx, T_DOUBLE);
 333   __ jcc(Assembler::equal, is_double);
 334 
 335   // handle T_INT case
 336   __ movl(Address(r13, 0), rax);
 337 
 338   __ BIND(exit);
 339 
 340   // pop parameters
 341   __ lea(rsp, rsp_after_call);
 342 
 343 #ifdef ASSERT
 344   // verify that threads correspond
 345   {
 346    Label L1, L2, L3;
 347     __ cmpptr(r15_thread, thread);
 348     __ jcc(Assembler::equal, L1);
 349     __ stop("StubRoutines::call_stub: r15_thread is corrupted");
 350     __ bind(L1);
 351     __ get_thread(rbx);
 352     __ cmpptr(r15_thread, thread);
 353     __ jcc(Assembler::equal, L2);
 354     __ stop("StubRoutines::call_stub: r15_thread is modified by call");
 355     __ bind(L2);
 356     __ cmpptr(r15_thread, rbx);
 357     __ jcc(Assembler::equal, L3);
 358     __ stop("StubRoutines::call_stub: threads must correspond");
 359     __ bind(L3);
 360   }
 361 #endif
 362 
 363   __ pop_cont_fastpath();
 364 
 365   // restore regs belonging to calling function
 366 #ifdef _WIN64
 367   // emit the restores for xmm regs
 368   if (VM_Version::supports_evex()) {
 369     for (int i = xmm_save_first; i <= last_reg; i++) {
 370       __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
 371     }
 372   } else {
 373     for (int i = xmm_save_first; i <= last_reg; i++) {
 374       __ movdqu(as_XMMRegister(i), xmm_save(i));
 375     }
 376   }
 377 #endif
 378   __ movptr(r15, r15_save);
 379   __ movptr(r14, r14_save);
 380   __ movptr(r13, r13_save);
 381   __ movptr(r12, r12_save);
 382   __ movptr(rbx, rbx_save);
 383 
 384 #ifdef _WIN64
 385   __ movptr(rdi, rdi_save);
 386   __ movptr(rsi, rsi_save);
 387 #else
 388   __ ldmxcsr(mxcsr_save);
 389 #endif
 390 
 391   // restore rsp
 392   __ addptr(rsp, -rsp_after_call_off * wordSize);
 393 
 394   // return
 395   __ vzeroupper();
 396   __ pop(rbp);
 397   __ ret(0);
 398 
 399   // handle return types different from T_INT
 400   __ BIND(check_prim);
 401   if (InlineTypeReturnedAsFields) {
 402     // Check for scalarized return value
 403     __ testptr(rax, 1);
 404     __ jcc(Assembler::zero, is_long);
 405     // Load pack handler address
 406     __ andptr(rax, -2);
 407     __ movptr(rax, Address(rax, InstanceKlass::adr_inlineklass_fixed_block_offset()));
 408     __ movptr(rbx, Address(rax, InlineKlass::pack_handler_jobject_offset()));
 409     // Call pack handler to initialize the buffer
 410     __ call(rbx);
 411     __ jmp(exit);
 412   }
 413   __ BIND(is_long);
 414   __ movq(Address(r13, 0), rax);
 415   __ jmp(exit);
 416 
 417   __ BIND(is_float);
 418   __ movflt(Address(r13, 0), xmm0);
 419   __ jmp(exit);
 420 
 421   __ BIND(is_double);
 422   __ movdbl(Address(r13, 0), xmm0);
 423   __ jmp(exit);
 424 
 425   return start;
 426 }
 427 
 428 // Return point for a Java call if there's an exception thrown in
 429 // Java code.  The exception is caught and transformed into a
 430 // pending exception stored in JavaThread that can be tested from
 431 // within the VM.
 432 //
 433 // Note: Usually the parameters are removed by the callee. In case
 434 // of an exception crossing an activation frame boundary, that is
 435 // not the case if the callee is compiled code => need to setup the
 436 // rsp.
 437 //
 438 // rax: exception oop
 439 
 440 address StubGenerator::generate_catch_exception() {
 441   StubCodeMark mark(this, "StubRoutines", "catch_exception");
 442   address start = __ pc();
 443 
 444   // same as in generate_call_stub():
 445   const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 446   const Address thread        (rbp, thread_off         * wordSize);
 447 
 448 #ifdef ASSERT
 449   // verify that threads correspond
 450   {
 451     Label L1, L2, L3;
 452     __ cmpptr(r15_thread, thread);
 453     __ jcc(Assembler::equal, L1);
 454     __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
 455     __ bind(L1);
 456     __ get_thread(rbx);
 457     __ cmpptr(r15_thread, thread);
 458     __ jcc(Assembler::equal, L2);
 459     __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
 460     __ bind(L2);
 461     __ cmpptr(r15_thread, rbx);
 462     __ jcc(Assembler::equal, L3);
 463     __ stop("StubRoutines::catch_exception: threads must correspond");
 464     __ bind(L3);
 465   }
 466 #endif
 467 
 468   // set pending exception
 469   __ verify_oop(rax);
 470 
 471   __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
 472   __ lea(rscratch1, ExternalAddress((address)__FILE__));
 473   __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 474   __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 475 
 476   // complete return to VM
 477   assert(StubRoutines::_call_stub_return_address != nullptr,
 478          "_call_stub_return_address must have been generated before");
 479   __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 480 
 481   return start;
 482 }
 483 
 484 // Continuation point for runtime calls returning with a pending
 485 // exception.  The pending exception check happened in the runtime
 486 // or native call stub.  The pending exception in Thread is
 487 // converted into a Java-level exception.
 488 //
 489 // Contract with Java-level exception handlers:
 490 // rax: exception
 491 // rdx: throwing pc
 492 //
 493 // NOTE: At entry of this stub, exception-pc must be on stack !!
 494 
 495 address StubGenerator::generate_forward_exception() {
 496   StubCodeMark mark(this, "StubRoutines", "forward exception");
 497   address start = __ pc();
 498 
 499   // Upon entry, the sp points to the return address returning into
 500   // Java (interpreted or compiled) code; i.e., the return address
 501   // becomes the throwing pc.
 502   //
 503   // Arguments pushed before the runtime call are still on the stack
 504   // but the exception handler will reset the stack pointer ->
 505   // ignore them.  A potential result in registers can be ignored as
 506   // well.
 507 
 508 #ifdef ASSERT
 509   // make sure this code is only executed if there is a pending exception
 510   {
 511     Label L;
 512     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 513     __ jcc(Assembler::notEqual, L);
 514     __ stop("StubRoutines::forward exception: no pending exception (1)");
 515     __ bind(L);
 516   }
 517 #endif
 518 
 519   // compute exception handler into rbx
 520   __ movptr(c_rarg0, Address(rsp, 0));
 521   BLOCK_COMMENT("call exception_handler_for_return_address");
 522   __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 523                        SharedRuntime::exception_handler_for_return_address),
 524                   r15_thread, c_rarg0);
 525   __ mov(rbx, rax);
 526 
 527   // setup rax & rdx, remove return address & clear pending exception
 528   __ pop(rdx);
 529   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 530   __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 531 
 532 #ifdef ASSERT
 533   // make sure exception is set
 534   {
 535     Label L;
 536     __ testptr(rax, rax);
 537     __ jcc(Assembler::notEqual, L);
 538     __ stop("StubRoutines::forward exception: no pending exception (2)");
 539     __ bind(L);
 540   }
 541 #endif
 542 
 543   // continue at exception handler (return address removed)
 544   // rax: exception
 545   // rbx: exception handler
 546   // rdx: throwing pc
 547   __ verify_oop(rax);
 548   __ jmp(rbx);
 549 
 550   return start;
 551 }
 552 
 553 // Support for intptr_t OrderAccess::fence()
 554 //
 555 // Arguments :
 556 //
 557 // Result:
 558 address StubGenerator::generate_orderaccess_fence() {
 559   StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 560   address start = __ pc();
 561 
 562   __ membar(Assembler::StoreLoad);
 563   __ ret(0);
 564 
 565   return start;
 566 }
 567 
 568 
 569 // Support for intptr_t get_previous_sp()
 570 //
 571 // This routine is used to find the previous stack pointer for the
 572 // caller.
 573 address StubGenerator::generate_get_previous_sp() {
 574   StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
 575   address start = __ pc();
 576 
 577   __ movptr(rax, rsp);
 578   __ addptr(rax, 8); // return address is at the top of the stack.
 579   __ ret(0);
 580 
 581   return start;
 582 }
 583 
 584 //----------------------------------------------------------------------------------------------------
 585 // Support for void verify_mxcsr()
 586 //
 587 // This routine is used with -Xcheck:jni to verify that native
 588 // JNI code does not return to Java code without restoring the
 589 // MXCSR register to our expected state.
 590 
 591 address StubGenerator::generate_verify_mxcsr() {
 592   StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 593   address start = __ pc();
 594 
 595   const Address mxcsr_save(rsp, 0);
 596 
 597   if (CheckJNICalls) {
 598     Label ok_ret;
 599     ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 600     __ push(rax);
 601     __ subptr(rsp, wordSize);      // allocate a temp location
 602     __ stmxcsr(mxcsr_save);
 603     __ movl(rax, mxcsr_save);
 604     __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
 605     __ cmp32(rax, mxcsr_std, rscratch1);
 606     __ jcc(Assembler::equal, ok_ret);
 607 
 608     __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 609 
 610     __ ldmxcsr(mxcsr_std, rscratch1);
 611 
 612     __ bind(ok_ret);
 613     __ addptr(rsp, wordSize);
 614     __ pop(rax);
 615   }
 616 
 617   __ ret(0);
 618 
 619   return start;
 620 }
 621 
 622 address StubGenerator::generate_f2i_fixup() {
 623   StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 624   Address inout(rsp, 5 * wordSize); // return address + 4 saves
 625 
 626   address start = __ pc();
 627 
 628   Label L;
 629 
 630   __ push(rax);
 631   __ push(c_rarg3);
 632   __ push(c_rarg2);
 633   __ push(c_rarg1);
 634 
 635   __ movl(rax, 0x7f800000);
 636   __ xorl(c_rarg3, c_rarg3);
 637   __ movl(c_rarg2, inout);
 638   __ movl(c_rarg1, c_rarg2);
 639   __ andl(c_rarg1, 0x7fffffff);
 640   __ cmpl(rax, c_rarg1); // NaN? -> 0
 641   __ jcc(Assembler::negative, L);
 642   __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 643   __ movl(c_rarg3, 0x80000000);
 644   __ movl(rax, 0x7fffffff);
 645   __ cmovl(Assembler::positive, c_rarg3, rax);
 646 
 647   __ bind(L);
 648   __ movptr(inout, c_rarg3);
 649 
 650   __ pop(c_rarg1);
 651   __ pop(c_rarg2);
 652   __ pop(c_rarg3);
 653   __ pop(rax);
 654 
 655   __ ret(0);
 656 
 657   return start;
 658 }
 659 
 660 address StubGenerator::generate_f2l_fixup() {
 661   StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 662   Address inout(rsp, 5 * wordSize); // return address + 4 saves
 663   address start = __ pc();
 664 
 665   Label L;
 666 
 667   __ push(rax);
 668   __ push(c_rarg3);
 669   __ push(c_rarg2);
 670   __ push(c_rarg1);
 671 
 672   __ movl(rax, 0x7f800000);
 673   __ xorl(c_rarg3, c_rarg3);
 674   __ movl(c_rarg2, inout);
 675   __ movl(c_rarg1, c_rarg2);
 676   __ andl(c_rarg1, 0x7fffffff);
 677   __ cmpl(rax, c_rarg1); // NaN? -> 0
 678   __ jcc(Assembler::negative, L);
 679   __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 680   __ mov64(c_rarg3, 0x8000000000000000);
 681   __ mov64(rax, 0x7fffffffffffffff);
 682   __ cmov(Assembler::positive, c_rarg3, rax);
 683 
 684   __ bind(L);
 685   __ movptr(inout, c_rarg3);
 686 
 687   __ pop(c_rarg1);
 688   __ pop(c_rarg2);
 689   __ pop(c_rarg3);
 690   __ pop(rax);
 691 
 692   __ ret(0);
 693 
 694   return start;
 695 }
 696 
 697 address StubGenerator::generate_d2i_fixup() {
 698   StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 699   Address inout(rsp, 6 * wordSize); // return address + 5 saves
 700 
 701   address start = __ pc();
 702 
 703   Label L;
 704 
 705   __ push(rax);
 706   __ push(c_rarg3);
 707   __ push(c_rarg2);
 708   __ push(c_rarg1);
 709   __ push(c_rarg0);
 710 
 711   __ movl(rax, 0x7ff00000);
 712   __ movq(c_rarg2, inout);
 713   __ movl(c_rarg3, c_rarg2);
 714   __ mov(c_rarg1, c_rarg2);
 715   __ mov(c_rarg0, c_rarg2);
 716   __ negl(c_rarg3);
 717   __ shrptr(c_rarg1, 0x20);
 718   __ orl(c_rarg3, c_rarg2);
 719   __ andl(c_rarg1, 0x7fffffff);
 720   __ xorl(c_rarg2, c_rarg2);
 721   __ shrl(c_rarg3, 0x1f);
 722   __ orl(c_rarg1, c_rarg3);
 723   __ cmpl(rax, c_rarg1);
 724   __ jcc(Assembler::negative, L); // NaN -> 0
 725   __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 726   __ movl(c_rarg2, 0x80000000);
 727   __ movl(rax, 0x7fffffff);
 728   __ cmov(Assembler::positive, c_rarg2, rax);
 729 
 730   __ bind(L);
 731   __ movptr(inout, c_rarg2);
 732 
 733   __ pop(c_rarg0);
 734   __ pop(c_rarg1);
 735   __ pop(c_rarg2);
 736   __ pop(c_rarg3);
 737   __ pop(rax);
 738 
 739   __ ret(0);
 740 
 741   return start;
 742 }
 743 
 744 address StubGenerator::generate_d2l_fixup() {
 745   StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 746   Address inout(rsp, 6 * wordSize); // return address + 5 saves
 747 
 748   address start = __ pc();
 749 
 750   Label L;
 751 
 752   __ push(rax);
 753   __ push(c_rarg3);
 754   __ push(c_rarg2);
 755   __ push(c_rarg1);
 756   __ push(c_rarg0);
 757 
 758   __ movl(rax, 0x7ff00000);
 759   __ movq(c_rarg2, inout);
 760   __ movl(c_rarg3, c_rarg2);
 761   __ mov(c_rarg1, c_rarg2);
 762   __ mov(c_rarg0, c_rarg2);
 763   __ negl(c_rarg3);
 764   __ shrptr(c_rarg1, 0x20);
 765   __ orl(c_rarg3, c_rarg2);
 766   __ andl(c_rarg1, 0x7fffffff);
 767   __ xorl(c_rarg2, c_rarg2);
 768   __ shrl(c_rarg3, 0x1f);
 769   __ orl(c_rarg1, c_rarg3);
 770   __ cmpl(rax, c_rarg1);
 771   __ jcc(Assembler::negative, L); // NaN -> 0
 772   __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 773   __ mov64(c_rarg2, 0x8000000000000000);
 774   __ mov64(rax, 0x7fffffffffffffff);
 775   __ cmovq(Assembler::positive, c_rarg2, rax);
 776 
 777   __ bind(L);
 778   __ movq(inout, c_rarg2);
 779 
 780   __ pop(c_rarg0);
 781   __ pop(c_rarg1);
 782   __ pop(c_rarg2);
 783   __ pop(c_rarg3);
 784   __ pop(rax);
 785 
 786   __ ret(0);
 787 
 788   return start;
 789 }
 790 
 791 address StubGenerator::generate_count_leading_zeros_lut(const char *stub_name) {
 792   __ align64();
 793   StubCodeMark mark(this, "StubRoutines", stub_name);
 794   address start = __ pc();
 795 
 796   __ emit_data64(0x0101010102020304, relocInfo::none);
 797   __ emit_data64(0x0000000000000000, relocInfo::none);
 798   __ emit_data64(0x0101010102020304, relocInfo::none);
 799   __ emit_data64(0x0000000000000000, relocInfo::none);
 800   __ emit_data64(0x0101010102020304, relocInfo::none);
 801   __ emit_data64(0x0000000000000000, relocInfo::none);
 802   __ emit_data64(0x0101010102020304, relocInfo::none);
 803   __ emit_data64(0x0000000000000000, relocInfo::none);
 804 
 805   return start;
 806 }
 807 
 808 address StubGenerator::generate_popcount_avx_lut(const char *stub_name) {
 809   __ align64();
 810   StubCodeMark mark(this, "StubRoutines", stub_name);
 811   address start = __ pc();
 812 
 813   __ emit_data64(0x0302020102010100, relocInfo::none);
 814   __ emit_data64(0x0403030203020201, relocInfo::none);
 815   __ emit_data64(0x0302020102010100, relocInfo::none);
 816   __ emit_data64(0x0403030203020201, relocInfo::none);
 817   __ emit_data64(0x0302020102010100, relocInfo::none);
 818   __ emit_data64(0x0403030203020201, relocInfo::none);
 819   __ emit_data64(0x0302020102010100, relocInfo::none);
 820   __ emit_data64(0x0403030203020201, relocInfo::none);
 821 
 822   return start;
 823 }
 824 
 825 address StubGenerator::generate_iota_indices(const char *stub_name) {
 826   __ align(CodeEntryAlignment);
 827   StubCodeMark mark(this, "StubRoutines", stub_name);
 828   address start = __ pc();
 829   // B
 830   __ emit_data64(0x0706050403020100, relocInfo::none);
 831   __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
 832   __ emit_data64(0x1716151413121110, relocInfo::none);
 833   __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
 834   __ emit_data64(0x2726252423222120, relocInfo::none);
 835   __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
 836   __ emit_data64(0x3736353433323130, relocInfo::none);
 837   __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
 838   // W
 839   __ emit_data64(0x0003000200010000, relocInfo::none);
 840   __ emit_data64(0x0007000600050004, relocInfo::none);
 841   __ emit_data64(0x000B000A00090008, relocInfo::none);
 842   __ emit_data64(0x000F000E000D000C, relocInfo::none);
 843   __ emit_data64(0x0013001200110010, relocInfo::none);
 844   __ emit_data64(0x0017001600150014, relocInfo::none);
 845   __ emit_data64(0x001B001A00190018, relocInfo::none);
 846   __ emit_data64(0x001F001E001D001C, relocInfo::none);
 847   // D
 848   __ emit_data64(0x0000000100000000, relocInfo::none);
 849   __ emit_data64(0x0000000300000002, relocInfo::none);
 850   __ emit_data64(0x0000000500000004, relocInfo::none);
 851   __ emit_data64(0x0000000700000006, relocInfo::none);
 852   __ emit_data64(0x0000000900000008, relocInfo::none);
 853   __ emit_data64(0x0000000B0000000A, relocInfo::none);
 854   __ emit_data64(0x0000000D0000000C, relocInfo::none);
 855   __ emit_data64(0x0000000F0000000E, relocInfo::none);
 856   // Q
 857   __ emit_data64(0x0000000000000000, relocInfo::none);
 858   __ emit_data64(0x0000000000000001, relocInfo::none);
 859   __ emit_data64(0x0000000000000002, relocInfo::none);
 860   __ emit_data64(0x0000000000000003, relocInfo::none);
 861   __ emit_data64(0x0000000000000004, relocInfo::none);
 862   __ emit_data64(0x0000000000000005, relocInfo::none);
 863   __ emit_data64(0x0000000000000006, relocInfo::none);
 864   __ emit_data64(0x0000000000000007, relocInfo::none);
 865   // D - FP
 866   __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
 867   __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
 868   __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
 869   __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
 870   __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
 871   __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
 872   __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
 873   __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
 874   // Q - FP
 875   __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
 876   __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
 877   __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
 878   __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
 879   __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
 880   __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
 881   __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
 882   __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
 883   return start;
 884 }
 885 
 886 address StubGenerator::generate_vector_reverse_bit_lut(const char *stub_name) {
 887   __ align(CodeEntryAlignment);
 888   StubCodeMark mark(this, "StubRoutines", stub_name);
 889   address start = __ pc();
 890 
 891   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 892   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 893   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 894   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 895   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 896   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 897   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 898   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 899 
 900   return start;
 901 }
 902 
 903 address StubGenerator::generate_vector_reverse_byte_perm_mask_long(const char *stub_name) {
 904   __ align(CodeEntryAlignment);
 905   StubCodeMark mark(this, "StubRoutines", stub_name);
 906   address start = __ pc();
 907 
 908   __ emit_data64(0x0001020304050607, relocInfo::none);
 909   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 910   __ emit_data64(0x0001020304050607, relocInfo::none);
 911   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 912   __ emit_data64(0x0001020304050607, relocInfo::none);
 913   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 914   __ emit_data64(0x0001020304050607, relocInfo::none);
 915   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 916 
 917   return start;
 918 }
 919 
 920 address StubGenerator::generate_vector_reverse_byte_perm_mask_int(const char *stub_name) {
 921   __ align(CodeEntryAlignment);
 922   StubCodeMark mark(this, "StubRoutines", stub_name);
 923   address start = __ pc();
 924 
 925   __ emit_data64(0x0405060700010203, relocInfo::none);
 926   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 927   __ emit_data64(0x0405060700010203, relocInfo::none);
 928   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 929   __ emit_data64(0x0405060700010203, relocInfo::none);
 930   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 931   __ emit_data64(0x0405060700010203, relocInfo::none);
 932   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 933 
 934   return start;
 935 }
 936 
 937 address StubGenerator::generate_vector_reverse_byte_perm_mask_short(const char *stub_name) {
 938   __ align(CodeEntryAlignment);
 939   StubCodeMark mark(this, "StubRoutines", stub_name);
 940   address start = __ pc();
 941 
 942   __ emit_data64(0x0607040502030001, relocInfo::none);
 943   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 944   __ emit_data64(0x0607040502030001, relocInfo::none);
 945   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 946   __ emit_data64(0x0607040502030001, relocInfo::none);
 947   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 948   __ emit_data64(0x0607040502030001, relocInfo::none);
 949   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 950 
 951   return start;
 952 }
 953 
 954 address StubGenerator::generate_vector_byte_shuffle_mask(const char *stub_name) {
 955   __ align(CodeEntryAlignment);
 956   StubCodeMark mark(this, "StubRoutines", stub_name);
 957   address start = __ pc();
 958 
 959   __ emit_data64(0x7070707070707070, relocInfo::none);
 960   __ emit_data64(0x7070707070707070, relocInfo::none);
 961   __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
 962   __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
 963 
 964   return start;
 965 }
 966 
 967 address StubGenerator::generate_fp_mask(const char *stub_name, int64_t mask) {
 968   __ align(CodeEntryAlignment);
 969   StubCodeMark mark(this, "StubRoutines", stub_name);
 970   address start = __ pc();
 971 
 972   __ emit_data64( mask, relocInfo::none );
 973   __ emit_data64( mask, relocInfo::none );
 974 
 975   return start;
 976 }
 977 
 978 address StubGenerator::generate_vector_mask(const char *stub_name, int64_t mask) {
 979   __ align(CodeEntryAlignment);
 980   StubCodeMark mark(this, "StubRoutines", stub_name);
 981   address start = __ pc();
 982 
 983   __ emit_data64(mask, relocInfo::none);
 984   __ emit_data64(mask, relocInfo::none);
 985   __ emit_data64(mask, relocInfo::none);
 986   __ emit_data64(mask, relocInfo::none);
 987   __ emit_data64(mask, relocInfo::none);
 988   __ emit_data64(mask, relocInfo::none);
 989   __ emit_data64(mask, relocInfo::none);
 990   __ emit_data64(mask, relocInfo::none);
 991 
 992   return start;
 993 }
 994 
 995 address StubGenerator::generate_vector_byte_perm_mask(const char *stub_name) {
 996   __ align(CodeEntryAlignment);
 997   StubCodeMark mark(this, "StubRoutines", stub_name);
 998   address start = __ pc();
 999 
1000   __ emit_data64(0x0000000000000001, relocInfo::none);
1001   __ emit_data64(0x0000000000000003, relocInfo::none);
1002   __ emit_data64(0x0000000000000005, relocInfo::none);
1003   __ emit_data64(0x0000000000000007, relocInfo::none);
1004   __ emit_data64(0x0000000000000000, relocInfo::none);
1005   __ emit_data64(0x0000000000000002, relocInfo::none);
1006   __ emit_data64(0x0000000000000004, relocInfo::none);
1007   __ emit_data64(0x0000000000000006, relocInfo::none);
1008 
1009   return start;
1010 }
1011 
1012 address StubGenerator::generate_vector_fp_mask(const char *stub_name, int64_t mask) {
1013   __ align(CodeEntryAlignment);
1014   StubCodeMark mark(this, "StubRoutines", stub_name);
1015   address start = __ pc();
1016 
1017   __ emit_data64(mask, relocInfo::none);
1018   __ emit_data64(mask, relocInfo::none);
1019   __ emit_data64(mask, relocInfo::none);
1020   __ emit_data64(mask, relocInfo::none);
1021   __ emit_data64(mask, relocInfo::none);
1022   __ emit_data64(mask, relocInfo::none);
1023   __ emit_data64(mask, relocInfo::none);
1024   __ emit_data64(mask, relocInfo::none);
1025 
1026   return start;
1027 }
1028 
1029 address StubGenerator::generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
1030                                    int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1031                                    int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1032                                    int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1033                                    int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1034   __ align(CodeEntryAlignment);
1035   StubCodeMark mark(this, "StubRoutines", stub_name);
1036   address start = __ pc();
1037 
1038   assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1039   __ emit_data(val0, relocInfo::none, 0);
1040   __ emit_data(val1, relocInfo::none, 0);
1041   __ emit_data(val2, relocInfo::none, 0);
1042   __ emit_data(val3, relocInfo::none, 0);
1043   if (len >= Assembler::AVX_256bit) {
1044     __ emit_data(val4, relocInfo::none, 0);
1045     __ emit_data(val5, relocInfo::none, 0);
1046     __ emit_data(val6, relocInfo::none, 0);
1047     __ emit_data(val7, relocInfo::none, 0);
1048     if (len >= Assembler::AVX_512bit) {
1049       __ emit_data(val8, relocInfo::none, 0);
1050       __ emit_data(val9, relocInfo::none, 0);
1051       __ emit_data(val10, relocInfo::none, 0);
1052       __ emit_data(val11, relocInfo::none, 0);
1053       __ emit_data(val12, relocInfo::none, 0);
1054       __ emit_data(val13, relocInfo::none, 0);
1055       __ emit_data(val14, relocInfo::none, 0);
1056       __ emit_data(val15, relocInfo::none, 0);
1057     }
1058   }
1059   return start;
1060 }
1061 
1062 // Non-destructive plausibility checks for oops
1063 //
1064 // Arguments:
1065 //    all args on stack!
1066 //
1067 // Stack after saving c_rarg3:
1068 //    [tos + 0]: saved c_rarg3
1069 //    [tos + 1]: saved c_rarg2
1070 //    [tos + 2]: saved r12 (several TemplateTable methods use it)
1071 //    [tos + 3]: saved flags
1072 //    [tos + 4]: return address
1073 //  * [tos + 5]: error message (char*)
1074 //  * [tos + 6]: object to verify (oop)
1075 //  * [tos + 7]: saved rax - saved by caller and bashed
1076 //  * [tos + 8]: saved r10 (rscratch1) - saved by caller
1077 //  * = popped on exit
1078 address StubGenerator::generate_verify_oop() {
1079   StubCodeMark mark(this, "StubRoutines", "verify_oop");
1080   address start = __ pc();
1081 
1082   Label exit, error;
1083 
1084   __ pushf();
1085   __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1086 
1087   __ push(r12);
1088 
1089   // save c_rarg2 and c_rarg3
1090   __ push(c_rarg2);
1091   __ push(c_rarg3);
1092 
1093   enum {
1094     // After previous pushes.
1095     oop_to_verify = 6 * wordSize,
1096     saved_rax     = 7 * wordSize,
1097     saved_r10     = 8 * wordSize,
1098 
1099     // Before the call to MacroAssembler::debug(), see below.
1100     return_addr   = 16 * wordSize,
1101     error_msg     = 17 * wordSize
1102   };
1103 
1104   // get object
1105   __ movptr(rax, Address(rsp, oop_to_verify));
1106 
1107   // make sure object is 'reasonable'
1108   __ testptr(rax, rax);
1109   __ jcc(Assembler::zero, exit); // if obj is null it is OK
1110 
1111    BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1112    bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1113 
1114   // return if everything seems ok
1115   __ bind(exit);
1116   __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1117   __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1118   __ pop(c_rarg3);                             // restore c_rarg3
1119   __ pop(c_rarg2);                             // restore c_rarg2
1120   __ pop(r12);                                 // restore r12
1121   __ popf();                                   // restore flags
1122   __ ret(4 * wordSize);                        // pop caller saved stuff
1123 
1124   // handle errors
1125   __ bind(error);
1126   __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1127   __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1128   __ pop(c_rarg3);                             // get saved c_rarg3 back
1129   __ pop(c_rarg2);                             // get saved c_rarg2 back
1130   __ pop(r12);                                 // get saved r12 back
1131   __ popf();                                   // get saved flags off stack --
1132                                                // will be ignored
1133 
1134   __ pusha();                                  // push registers
1135                                                // (rip is already
1136                                                // already pushed)
1137   // debug(char* msg, int64_t pc, int64_t regs[])
1138   // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1139   // pushed all the registers, so now the stack looks like:
1140   //     [tos +  0] 16 saved registers
1141   //     [tos + 16] return address
1142   //   * [tos + 17] error message (char*)
1143   //   * [tos + 18] object to verify (oop)
1144   //   * [tos + 19] saved rax - saved by caller and bashed
1145   //   * [tos + 20] saved r10 (rscratch1) - saved by caller
1146   //   * = popped on exit
1147 
1148   __ movptr(c_rarg0, Address(rsp, error_msg));    // pass address of error message
1149   __ movptr(c_rarg1, Address(rsp, return_addr));  // pass return address
1150   __ movq(c_rarg2, rsp);                          // pass address of regs on stack
1151   __ mov(r12, rsp);                               // remember rsp
1152   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1153   __ andptr(rsp, -16);                            // align stack as required by ABI
1154   BLOCK_COMMENT("call MacroAssembler::debug");
1155   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1156   __ hlt();
1157 
1158   return start;
1159 }
1160 
1161 
1162 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1163 //
1164 // Outputs:
1165 //    rdi - rcx
1166 //    rsi - rdx
1167 //    rdx - r8
1168 //    rcx - r9
1169 //
1170 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1171 // are non-volatile.  r9 and r10 should not be used by the caller.
1172 //
1173 void StubGenerator::setup_arg_regs(int nargs) {
1174   const Register saved_rdi = r9;
1175   const Register saved_rsi = r10;
1176   assert(nargs == 3 || nargs == 4, "else fix");
1177 #ifdef _WIN64
1178   assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1179          "unexpected argument registers");
1180   if (nargs == 4) {
1181     __ mov(rax, r9);  // r9 is also saved_rdi
1182   }
1183   __ movptr(saved_rdi, rdi);
1184   __ movptr(saved_rsi, rsi);
1185   __ mov(rdi, rcx); // c_rarg0
1186   __ mov(rsi, rdx); // c_rarg1
1187   __ mov(rdx, r8);  // c_rarg2
1188   if (nargs == 4) {
1189     __ mov(rcx, rax); // c_rarg3 (via rax)
1190   }
1191 #else
1192   assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1193          "unexpected argument registers");
1194 #endif
1195   DEBUG_ONLY(_regs_in_thread = false;)
1196 }
1197 
1198 
1199 void StubGenerator::restore_arg_regs() {
1200   assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1201   const Register saved_rdi = r9;
1202   const Register saved_rsi = r10;
1203 #ifdef _WIN64
1204   __ movptr(rdi, saved_rdi);
1205   __ movptr(rsi, saved_rsi);
1206 #endif
1207 }
1208 
1209 
1210 // This is used in places where r10 is a scratch register, and can
1211 // be adapted if r9 is needed also.
1212 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1213   const Register saved_r15 = r9;
1214   assert(nargs == 3 || nargs == 4, "else fix");
1215 #ifdef _WIN64
1216   if (nargs == 4) {
1217     __ mov(rax, r9);       // r9 is also saved_r15
1218   }
1219   __ mov(saved_r15, r15);  // r15 is callee saved and needs to be restored
1220   __ get_thread(r15_thread);
1221   assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1222          "unexpected argument registers");
1223   __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1224   __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1225 
1226   __ mov(rdi, rcx); // c_rarg0
1227   __ mov(rsi, rdx); // c_rarg1
1228   __ mov(rdx, r8);  // c_rarg2
1229   if (nargs == 4) {
1230     __ mov(rcx, rax); // c_rarg3 (via rax)
1231   }
1232 #else
1233   assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1234          "unexpected argument registers");
1235 #endif
1236   DEBUG_ONLY(_regs_in_thread = true;)
1237 }
1238 
1239 
1240 void StubGenerator::restore_arg_regs_using_thread() {
1241   assert(_regs_in_thread, "wrong call to restore_arg_regs");
1242   const Register saved_r15 = r9;
1243 #ifdef _WIN64
1244   __ get_thread(r15_thread);
1245   __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1246   __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1247   __ mov(r15, saved_r15);  // r15 is callee saved and needs to be restored
1248 #endif
1249 }
1250 
1251 
1252 void StubGenerator::setup_argument_regs(BasicType type) {
1253   if (type == T_BYTE || type == T_SHORT) {
1254     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1255                       // r9 and r10 may be used to save non-volatile registers
1256   } else {
1257     setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1258                                    // r9 is used to save r15_thread
1259   }
1260 }
1261 
1262 
1263 void StubGenerator::restore_argument_regs(BasicType type) {
1264   if (type == T_BYTE || type == T_SHORT) {
1265     restore_arg_regs();
1266   } else {
1267     restore_arg_regs_using_thread();
1268   }
1269 }
1270 
1271 address StubGenerator::generate_data_cache_writeback() {
1272   const Register src        = c_rarg0;  // source address
1273 
1274   __ align(CodeEntryAlignment);
1275 
1276   StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
1277 
1278   address start = __ pc();
1279 
1280   __ enter();
1281   __ cache_wb(Address(src, 0));
1282   __ leave();
1283   __ ret(0);
1284 
1285   return start;
1286 }
1287 
1288 address StubGenerator::generate_data_cache_writeback_sync() {
1289   const Register is_pre    = c_rarg0;  // pre or post sync
1290 
1291   __ align(CodeEntryAlignment);
1292 
1293   StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
1294 
1295   // pre wbsync is a no-op
1296   // post wbsync translates to an sfence
1297 
1298   Label skip;
1299   address start = __ pc();
1300 
1301   __ enter();
1302   __ cmpl(is_pre, 0);
1303   __ jcc(Assembler::notEqual, skip);
1304   __ cache_wbsync(false);
1305   __ bind(skip);
1306   __ leave();
1307   __ ret(0);
1308 
1309   return start;
1310 }
1311 
1312 // ofs and limit are use for multi-block byte array.
1313 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1314 address StubGenerator::generate_md5_implCompress(bool multi_block, const char *name) {
1315   __ align(CodeEntryAlignment);
1316   StubCodeMark mark(this, "StubRoutines", name);
1317   address start = __ pc();
1318 
1319   const Register buf_param = r15;
1320   const Address state_param(rsp, 0 * wordSize);
1321   const Address ofs_param  (rsp, 1 * wordSize    );
1322   const Address limit_param(rsp, 1 * wordSize + 4);
1323 
1324   __ enter();
1325   __ push(rbx);
1326   __ push(rdi);
1327   __ push(rsi);
1328   __ push(r15);
1329   __ subptr(rsp, 2 * wordSize);
1330 
1331   __ movptr(buf_param, c_rarg0);
1332   __ movptr(state_param, c_rarg1);
1333   if (multi_block) {
1334     __ movl(ofs_param, c_rarg2);
1335     __ movl(limit_param, c_rarg3);
1336   }
1337   __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1338 
1339   __ addptr(rsp, 2 * wordSize);
1340   __ pop(r15);
1341   __ pop(rsi);
1342   __ pop(rdi);
1343   __ pop(rbx);
1344   __ leave();
1345   __ ret(0);
1346 
1347   return start;
1348 }
1349 
1350 address StubGenerator::generate_upper_word_mask() {
1351   __ align64();
1352   StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
1353   address start = __ pc();
1354 
1355   __ emit_data64(0x0000000000000000, relocInfo::none);
1356   __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1357 
1358   return start;
1359 }
1360 
1361 address StubGenerator::generate_shuffle_byte_flip_mask() {
1362   __ align64();
1363   StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
1364   address start = __ pc();
1365 
1366   __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1367   __ emit_data64(0x0001020304050607, relocInfo::none);
1368 
1369   return start;
1370 }
1371 
1372 // ofs and limit are use for multi-block byte array.
1373 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1374 address StubGenerator::generate_sha1_implCompress(bool multi_block, const char *name) {
1375   __ align(CodeEntryAlignment);
1376   StubCodeMark mark(this, "StubRoutines", name);
1377   address start = __ pc();
1378 
1379   Register buf = c_rarg0;
1380   Register state = c_rarg1;
1381   Register ofs = c_rarg2;
1382   Register limit = c_rarg3;
1383 
1384   const XMMRegister abcd = xmm0;
1385   const XMMRegister e0 = xmm1;
1386   const XMMRegister e1 = xmm2;
1387   const XMMRegister msg0 = xmm3;
1388 
1389   const XMMRegister msg1 = xmm4;
1390   const XMMRegister msg2 = xmm5;
1391   const XMMRegister msg3 = xmm6;
1392   const XMMRegister shuf_mask = xmm7;
1393 
1394   __ enter();
1395 
1396   __ subptr(rsp, 4 * wordSize);
1397 
1398   __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1399     buf, state, ofs, limit, rsp, multi_block);
1400 
1401   __ addptr(rsp, 4 * wordSize);
1402 
1403   __ leave();
1404   __ ret(0);
1405 
1406   return start;
1407 }
1408 
1409 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1410   __ align64();
1411   StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
1412   address start = __ pc();
1413 
1414   __ emit_data64(0x0405060700010203, relocInfo::none);
1415   __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1416 
1417   if (VM_Version::supports_avx2()) {
1418     __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1419     __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1420     // _SHUF_00BA
1421     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1422     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1423     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1424     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1425     // _SHUF_DC00
1426     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1427     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1428     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1429     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1430   }
1431 
1432   return start;
1433 }
1434 
1435 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1436 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1437   __ align32();
1438   StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
1439   address start = __ pc();
1440 
1441   if (VM_Version::supports_avx2()) {
1442     __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1443     __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1444     __ emit_data64(0x1011121314151617, relocInfo::none);
1445     __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1446     __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1447     __ emit_data64(0x0000000000000000, relocInfo::none);
1448     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1449     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1450   }
1451 
1452   return start;
1453 }
1454 
1455 // ofs and limit are use for multi-block byte array.
1456 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1457 address StubGenerator::generate_sha256_implCompress(bool multi_block, const char *name) {
1458   assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1459   __ align(CodeEntryAlignment);
1460   StubCodeMark mark(this, "StubRoutines", name);
1461   address start = __ pc();
1462 
1463   Register buf = c_rarg0;
1464   Register state = c_rarg1;
1465   Register ofs = c_rarg2;
1466   Register limit = c_rarg3;
1467 
1468   const XMMRegister msg = xmm0;
1469   const XMMRegister state0 = xmm1;
1470   const XMMRegister state1 = xmm2;
1471   const XMMRegister msgtmp0 = xmm3;
1472 
1473   const XMMRegister msgtmp1 = xmm4;
1474   const XMMRegister msgtmp2 = xmm5;
1475   const XMMRegister msgtmp3 = xmm6;
1476   const XMMRegister msgtmp4 = xmm7;
1477 
1478   const XMMRegister shuf_mask = xmm8;
1479 
1480   __ enter();
1481 
1482   __ subptr(rsp, 4 * wordSize);
1483 
1484   if (VM_Version::supports_sha()) {
1485     __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1486       buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1487   } else if (VM_Version::supports_avx2()) {
1488     __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1489       buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1490   }
1491   __ addptr(rsp, 4 * wordSize);
1492   __ vzeroupper();
1493   __ leave();
1494   __ ret(0);
1495 
1496   return start;
1497 }
1498 
1499 address StubGenerator::generate_sha512_implCompress(bool multi_block, const char *name) {
1500   assert(VM_Version::supports_avx2(), "");
1501   assert(VM_Version::supports_bmi2(), "");
1502   __ align(CodeEntryAlignment);
1503   StubCodeMark mark(this, "StubRoutines", name);
1504   address start = __ pc();
1505 
1506   Register buf = c_rarg0;
1507   Register state = c_rarg1;
1508   Register ofs = c_rarg2;
1509   Register limit = c_rarg3;
1510 
1511   const XMMRegister msg = xmm0;
1512   const XMMRegister state0 = xmm1;
1513   const XMMRegister state1 = xmm2;
1514   const XMMRegister msgtmp0 = xmm3;
1515   const XMMRegister msgtmp1 = xmm4;
1516   const XMMRegister msgtmp2 = xmm5;
1517   const XMMRegister msgtmp3 = xmm6;
1518   const XMMRegister msgtmp4 = xmm7;
1519 
1520   const XMMRegister shuf_mask = xmm8;
1521 
1522   __ enter();
1523 
1524   __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1525   buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1526 
1527   __ vzeroupper();
1528   __ leave();
1529   __ ret(0);
1530 
1531   return start;
1532 }
1533 
1534 address StubGenerator::base64_shuffle_addr() {
1535   __ align64();
1536   StubCodeMark mark(this, "StubRoutines", "shuffle_base64");
1537   address start = __ pc();
1538 
1539   assert(((unsigned long long)start & 0x3f) == 0,
1540          "Alignment problem (0x%08llx)", (unsigned long long)start);
1541   __ emit_data64(0x0405030401020001, relocInfo::none);
1542   __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1543   __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1544   __ emit_data64(0x1617151613141213, relocInfo::none);
1545   __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1546   __ emit_data64(0x222321221f201e1f, relocInfo::none);
1547   __ emit_data64(0x2829272825262425, relocInfo::none);
1548   __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1549 
1550   return start;
1551 }
1552 
1553 address StubGenerator::base64_avx2_shuffle_addr() {
1554   __ align32();
1555   StubCodeMark mark(this, "StubRoutines", "avx2_shuffle_base64");
1556   address start = __ pc();
1557 
1558   __ emit_data64(0x0809070805060405, relocInfo::none);
1559   __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1560   __ emit_data64(0x0405030401020001, relocInfo::none);
1561   __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1562 
1563   return start;
1564 }
1565 
1566 address StubGenerator::base64_avx2_input_mask_addr() {
1567   __ align32();
1568   StubCodeMark mark(this, "StubRoutines", "avx2_input_mask_base64");
1569   address start = __ pc();
1570 
1571   __ emit_data64(0x8000000000000000, relocInfo::none);
1572   __ emit_data64(0x8000000080000000, relocInfo::none);
1573   __ emit_data64(0x8000000080000000, relocInfo::none);
1574   __ emit_data64(0x8000000080000000, relocInfo::none);
1575 
1576   return start;
1577 }
1578 
1579 address StubGenerator::base64_avx2_lut_addr() {
1580   __ align32();
1581   StubCodeMark mark(this, "StubRoutines", "avx2_lut_base64");
1582   address start = __ pc();
1583 
1584   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1585   __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1586   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1587   __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1588 
1589   // URL LUT
1590   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1591   __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1592   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1593   __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1594 
1595   return start;
1596 }
1597 
1598 address StubGenerator::base64_encoding_table_addr() {
1599   __ align64();
1600   StubCodeMark mark(this, "StubRoutines", "encoding_table_base64");
1601   address start = __ pc();
1602 
1603   assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1604   __ emit_data64(0x4847464544434241, relocInfo::none);
1605   __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1606   __ emit_data64(0x5857565554535251, relocInfo::none);
1607   __ emit_data64(0x6665646362615a59, relocInfo::none);
1608   __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1609   __ emit_data64(0x767574737271706f, relocInfo::none);
1610   __ emit_data64(0x333231307a797877, relocInfo::none);
1611   __ emit_data64(0x2f2b393837363534, relocInfo::none);
1612 
1613   // URL table
1614   __ emit_data64(0x4847464544434241, relocInfo::none);
1615   __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1616   __ emit_data64(0x5857565554535251, relocInfo::none);
1617   __ emit_data64(0x6665646362615a59, relocInfo::none);
1618   __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1619   __ emit_data64(0x767574737271706f, relocInfo::none);
1620   __ emit_data64(0x333231307a797877, relocInfo::none);
1621   __ emit_data64(0x5f2d393837363534, relocInfo::none);
1622 
1623   return start;
1624 }
1625 
1626 // Code for generating Base64 encoding.
1627 // Intrinsic function prototype in Base64.java:
1628 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1629 // boolean isURL) {
1630 address StubGenerator::generate_base64_encodeBlock()
1631 {
1632   __ align(CodeEntryAlignment);
1633   StubCodeMark mark(this, "StubRoutines", "implEncode");
1634   address start = __ pc();
1635 
1636   __ enter();
1637 
1638   // Save callee-saved registers before using them
1639   __ push(r12);
1640   __ push(r13);
1641   __ push(r14);
1642   __ push(r15);
1643 
1644   // arguments
1645   const Register source = c_rarg0;       // Source Array
1646   const Register start_offset = c_rarg1; // start offset
1647   const Register end_offset = c_rarg2;   // end offset
1648   const Register dest = c_rarg3;   // destination array
1649 
1650 #ifndef _WIN64
1651   const Register dp = c_rarg4;    // Position for writing to dest array
1652   const Register isURL = c_rarg5; // Base64 or URL character set
1653 #else
1654   const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1655   const Address isURL_mem(rbp, 7 * wordSize);
1656   const Register isURL = r10; // pick the volatile windows register
1657   const Register dp = r12;
1658   __ movl(dp, dp_mem);
1659   __ movl(isURL, isURL_mem);
1660 #endif
1661 
1662   const Register length = r14;
1663   const Register encode_table = r13;
1664   Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1665 
1666   // calculate length from offsets
1667   __ movl(length, end_offset);
1668   __ subl(length, start_offset);
1669   __ jcc(Assembler::lessEqual, L_exit);
1670 
1671   // Code for 512-bit VBMI encoding.  Encodes 48 input bytes into 64
1672   // output bytes. We read 64 input bytes and ignore the last 16, so be
1673   // sure not to read past the end of the input buffer.
1674   if (VM_Version::supports_avx512_vbmi()) {
1675     __ cmpl(length, 64); // Do not overrun input buffer.
1676     __ jcc(Assembler::below, L_not512);
1677 
1678     __ shll(isURL, 6); // index into decode table based on isURL
1679     __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1680     __ addptr(encode_table, isURL);
1681     __ shrl(isURL, 6); // restore isURL
1682 
1683     __ mov64(rax, 0x3036242a1016040aull); // Shifts
1684     __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1685     __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1686     __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1687 
1688     __ align32();
1689     __ BIND(L_vbmiLoop);
1690 
1691     __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1692     __ subl(length, 48);
1693 
1694     // Put the input bytes into the proper lanes for writing, then
1695     // encode them.
1696     __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1697     __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1698 
1699     // Write to destination
1700     __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1701 
1702     __ addptr(dest, 64);
1703     __ addptr(source, 48);
1704     __ cmpl(length, 64);
1705     __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1706 
1707     __ vzeroupper();
1708   }
1709 
1710   __ BIND(L_not512);
1711   if (VM_Version::supports_avx2()) {
1712     /*
1713     ** This AVX2 encoder is based off the paper at:
1714     **      https://dl.acm.org/doi/10.1145/3132709
1715     **
1716     ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1717     ** output bytes.
1718     **
1719     */
1720     // Lengths under 32 bytes are done with scalar routine
1721     __ cmpl(length, 31);
1722     __ jcc(Assembler::belowEqual, L_process3);
1723 
1724     // Set up supporting constant table data
1725     __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1726     // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1727     __ movl(rax, 0x0fc0fc00);
1728     __ movdl(xmm8, rax);
1729     __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1730     __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1731 
1732     // Multiplication constant for "shifting" right by 6 and 10
1733     // bits
1734     __ movl(rax, 0x04000040);
1735 
1736     __ subl(length, 24);
1737     __ movdl(xmm7, rax);
1738     __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1739 
1740     // For the first load, we mask off reading of the first 4
1741     // bytes into the register. This is so we can get 4 3-byte
1742     // chunks into each lane of the register, avoiding having to
1743     // handle end conditions.  We then shuffle these bytes into a
1744     // specific order so that manipulation is easier.
1745     //
1746     // The initial read loads the XMM register like this:
1747     //
1748     // Lower 128-bit lane:
1749     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1750     // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1751     // | C2 | D0 | D1 | D2 |
1752     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1753     //
1754     // Upper 128-bit lane:
1755     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1756     // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1757     // | XX | XX | XX | XX |
1758     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1759     //
1760     // Where A0 is the first input byte, B0 is the fourth, etc.
1761     // The alphabetical significance denotes the 3 bytes to be
1762     // consumed and encoded into 4 bytes.
1763     //
1764     // We then shuffle the register so each 32-bit word contains
1765     // the sequence:
1766     //    A1 A0 A2 A1, B1, B0, B2, B1, etc.
1767     // Each of these byte sequences are then manipulated into 4
1768     // 6-bit values ready for encoding.
1769     //
1770     // If we focus on one set of 3-byte chunks, changing the
1771     // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1772     // shuffle such that each 24-bit chunk contains:
1773     //
1774     // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1775     // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1776     // Explain this step.
1777     // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1778     // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1779     //
1780     // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1781     // a5..a0) and shift them using a vector multiplication
1782     // operation (vpmulhuw) which effectively shifts c right by 6
1783     // bits and a right by 10 bits.  We similarly mask bits 10-15
1784     // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1785     // bits respectively.  This is done using vpmullw.  We end up
1786     // with 4 6-bit values, thus splitting the 3 input bytes,
1787     // ready for encoding:
1788     //    0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1789     //
1790     // For translation, we recognize that there are 5 distinct
1791     // ranges of legal Base64 characters as below:
1792     //
1793     //   +-------------+-------------+------------+
1794     //   | 6-bit value | ASCII range |   offset   |
1795     //   +-------------+-------------+------------+
1796     //   |    0..25    |    A..Z     |     65     |
1797     //   |   26..51    |    a..z     |     71     |
1798     //   |   52..61    |    0..9     |     -4     |
1799     //   |     62      |   + or -    | -19 or -17 |
1800     //   |     63      |   / or _    | -16 or 32  |
1801     //   +-------------+-------------+------------+
1802     //
1803     // We note that vpshufb does a parallel lookup in a
1804     // destination register using the lower 4 bits of bytes from a
1805     // source register.  If we use a saturated subtraction and
1806     // subtract 51 from each 6-bit value, bytes from [0,51]
1807     // saturate to 0, and [52,63] map to a range of [1,12].  We
1808     // distinguish the [0,25] and [26,51] ranges by assigning a
1809     // value of 13 for all 6-bit values less than 26.  We end up
1810     // with:
1811     //
1812     //   +-------------+-------------+------------+
1813     //   | 6-bit value |   Reduced   |   offset   |
1814     //   +-------------+-------------+------------+
1815     //   |    0..25    |     13      |     65     |
1816     //   |   26..51    |      0      |     71     |
1817     //   |   52..61    |    0..9     |     -4     |
1818     //   |     62      |     11      | -19 or -17 |
1819     //   |     63      |     12      | -16 or 32  |
1820     //   +-------------+-------------+------------+
1821     //
1822     // We then use a final vpshufb to add the appropriate offset,
1823     // translating the bytes.
1824     //
1825     // Load input bytes - only 28 bytes.  Mask the first load to
1826     // not load into the full register.
1827     __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1828 
1829     // Move 3-byte chunks of input (12 bytes) into 16 bytes,
1830     // ordering by:
1831     //   1, 0, 2, 1; 4, 3, 5, 4; etc.  This groups 6-bit chunks
1832     //   for easy masking
1833     __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1834 
1835     __ addl(start_offset, 24);
1836 
1837     // Load masking register for first and third (and multiples)
1838     // 6-bit values.
1839     __ movl(rax, 0x003f03f0);
1840     __ movdl(xmm6, rax);
1841     __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
1842     // Multiplication constant for "shifting" left by 4 and 8 bits
1843     __ movl(rax, 0x01000010);
1844     __ movdl(xmm5, rax);
1845     __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
1846 
1847     // Isolate 6-bit chunks of interest
1848     __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1849 
1850     // Load constants for encoding
1851     __ movl(rax, 0x19191919);
1852     __ movdl(xmm3, rax);
1853     __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
1854     __ movl(rax, 0x33333333);
1855     __ movdl(xmm4, rax);
1856     __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
1857 
1858     // Shift output bytes 0 and 2 into proper lanes
1859     __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
1860 
1861     // Mask and shift output bytes 1 and 3 into proper lanes and
1862     // combine
1863     __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1864     __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1865     __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
1866 
1867     // Find out which are 0..25.  This indicates which input
1868     // values fall in the range of 'A'-'Z', which require an
1869     // additional offset (see comments above)
1870     __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
1871     __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1872     __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
1873 
1874     // Load the proper lookup table
1875     __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
1876     __ movl(r15, isURL);
1877     __ shll(r15, 5);
1878     __ vmovdqu(xmm2, Address(r11, r15));
1879 
1880     // Shuffle the offsets based on the range calculation done
1881     // above. This allows us to add the correct offset to the
1882     // 6-bit value corresponding to the range documented above.
1883     __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1884     __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1885 
1886     // Store the encoded bytes
1887     __ vmovdqu(Address(dest, dp), xmm0);
1888     __ addl(dp, 32);
1889 
1890     __ cmpl(length, 31);
1891     __ jcc(Assembler::belowEqual, L_process3);
1892 
1893     __ align32();
1894     __ BIND(L_32byteLoop);
1895 
1896     // Get next 32 bytes
1897     __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
1898 
1899     __ subl(length, 24);
1900     __ addl(start_offset, 24);
1901 
1902     // This logic is identical to the above, with only constant
1903     // register loads removed.  Shuffle the input, mask off 6-bit
1904     // chunks, shift them into place, then add the offset to
1905     // encode.
1906     __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1907 
1908     __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1909     __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
1910     __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1911     __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1912     __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
1913     __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
1914     __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1915     __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
1916     __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1917     __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1918 
1919     // Store the encoded bytes
1920     __ vmovdqu(Address(dest, dp), xmm0);
1921     __ addl(dp, 32);
1922 
1923     __ cmpl(length, 31);
1924     __ jcc(Assembler::above, L_32byteLoop);
1925 
1926     __ BIND(L_process3);
1927     __ vzeroupper();
1928   } else {
1929     __ BIND(L_process3);
1930   }
1931 
1932   __ cmpl(length, 3);
1933   __ jcc(Assembler::below, L_exit);
1934 
1935   // Load the encoding table based on isURL
1936   __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1937   __ movl(r15, isURL);
1938   __ shll(r15, 6);
1939   __ addptr(r11, r15);
1940 
1941   __ BIND(L_processdata);
1942 
1943   // Load 3 bytes
1944   __ load_unsigned_byte(r15, Address(source, start_offset));
1945   __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
1946   __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
1947 
1948   // Build a 32-bit word with bytes 1, 2, 0, 1
1949   __ movl(rax, r10);
1950   __ shll(r10, 24);
1951   __ orl(rax, r10);
1952 
1953   __ subl(length, 3);
1954 
1955   __ shll(r15, 8);
1956   __ shll(r13, 16);
1957   __ orl(rax, r15);
1958 
1959   __ addl(start_offset, 3);
1960 
1961   __ orl(rax, r13);
1962   // At this point, rax contains | byte1 | byte2 | byte0 | byte1
1963   // r13 has byte2 << 16 - need low-order 6 bits to translate.
1964   // This translated byte is the fourth output byte.
1965   __ shrl(r13, 16);
1966   __ andl(r13, 0x3f);
1967 
1968   // The high-order 6 bits of r15 (byte0) is translated.
1969   // The translated byte is the first output byte.
1970   __ shrl(r15, 10);
1971 
1972   __ load_unsigned_byte(r13, Address(r11, r13));
1973   __ load_unsigned_byte(r15, Address(r11, r15));
1974 
1975   __ movb(Address(dest, dp, Address::times_1, 3), r13);
1976 
1977   // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
1978   // This translated byte is the second output byte.
1979   __ shrl(rax, 4);
1980   __ movl(r10, rax);
1981   __ andl(rax, 0x3f);
1982 
1983   __ movb(Address(dest, dp, Address::times_1, 0), r15);
1984 
1985   __ load_unsigned_byte(rax, Address(r11, rax));
1986 
1987   // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
1988   // This translated byte is the third output byte.
1989   __ shrl(r10, 18);
1990   __ andl(r10, 0x3f);
1991 
1992   __ load_unsigned_byte(r10, Address(r11, r10));
1993 
1994   __ movb(Address(dest, dp, Address::times_1, 1), rax);
1995   __ movb(Address(dest, dp, Address::times_1, 2), r10);
1996 
1997   __ addl(dp, 4);
1998   __ cmpl(length, 3);
1999   __ jcc(Assembler::aboveEqual, L_processdata);
2000 
2001   __ BIND(L_exit);
2002   __ pop(r15);
2003   __ pop(r14);
2004   __ pop(r13);
2005   __ pop(r12);
2006   __ leave();
2007   __ ret(0);
2008 
2009   return start;
2010 }
2011 
2012 // base64 AVX512vbmi tables
2013 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2014   __ align64();
2015   StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64");
2016   address start = __ pc();
2017 
2018   assert(((unsigned long long)start & 0x3f) == 0,
2019          "Alignment problem (0x%08llx)", (unsigned long long)start);
2020   __ emit_data64(0x8080808080808080, relocInfo::none);
2021   __ emit_data64(0x8080808080808080, relocInfo::none);
2022   __ emit_data64(0x8080808080808080, relocInfo::none);
2023   __ emit_data64(0x8080808080808080, relocInfo::none);
2024   __ emit_data64(0x8080808080808080, relocInfo::none);
2025   __ emit_data64(0x3f8080803e808080, relocInfo::none);
2026   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2027   __ emit_data64(0x8080808080803d3c, relocInfo::none);
2028 
2029   return start;
2030 }
2031 
2032 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2033   __ align64();
2034   StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64");
2035   address start = __ pc();
2036 
2037   assert(((unsigned long long)start & 0x3f) == 0,
2038          "Alignment problem (0x%08llx)", (unsigned long long)start);
2039   __ emit_data64(0x0605040302010080, relocInfo::none);
2040   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2041   __ emit_data64(0x161514131211100f, relocInfo::none);
2042   __ emit_data64(0x8080808080191817, relocInfo::none);
2043   __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2044   __ emit_data64(0x2827262524232221, relocInfo::none);
2045   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2046   __ emit_data64(0x8080808080333231, relocInfo::none);
2047 
2048   return start;
2049 }
2050 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2051   __ align64();
2052   StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64url");
2053   address start = __ pc();
2054 
2055   assert(((unsigned long long)start & 0x3f) == 0,
2056          "Alignment problem (0x%08llx)", (unsigned long long)start);
2057   __ emit_data64(0x8080808080808080, relocInfo::none);
2058   __ emit_data64(0x8080808080808080, relocInfo::none);
2059   __ emit_data64(0x8080808080808080, relocInfo::none);
2060   __ emit_data64(0x8080808080808080, relocInfo::none);
2061   __ emit_data64(0x8080808080808080, relocInfo::none);
2062   __ emit_data64(0x80803e8080808080, relocInfo::none);
2063   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2064   __ emit_data64(0x8080808080803d3c, relocInfo::none);
2065 
2066   return start;
2067 }
2068 
2069 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2070   __ align64();
2071   StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64url");
2072   address start = __ pc();
2073 
2074   assert(((unsigned long long)start & 0x3f) == 0,
2075          "Alignment problem (0x%08llx)", (unsigned long long)start);
2076   __ emit_data64(0x0605040302010080, relocInfo::none);
2077   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2078   __ emit_data64(0x161514131211100f, relocInfo::none);
2079   __ emit_data64(0x3f80808080191817, relocInfo::none);
2080   __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2081   __ emit_data64(0x2827262524232221, relocInfo::none);
2082   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2083   __ emit_data64(0x8080808080333231, relocInfo::none);
2084 
2085   return start;
2086 }
2087 
2088 address StubGenerator::base64_vbmi_pack_vec_addr() {
2089   __ align64();
2090   StubCodeMark mark(this, "StubRoutines", "pack_vec_base64");
2091   address start = __ pc();
2092 
2093   assert(((unsigned long long)start & 0x3f) == 0,
2094          "Alignment problem (0x%08llx)", (unsigned long long)start);
2095   __ emit_data64(0x090a040506000102, relocInfo::none);
2096   __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2097   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2098   __ emit_data64(0x292a242526202122, relocInfo::none);
2099   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2100   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2101   __ emit_data64(0x0000000000000000, relocInfo::none);
2102   __ emit_data64(0x0000000000000000, relocInfo::none);
2103 
2104   return start;
2105 }
2106 
2107 address StubGenerator::base64_vbmi_join_0_1_addr() {
2108   __ align64();
2109   StubCodeMark mark(this, "StubRoutines", "join_0_1_base64");
2110   address start = __ pc();
2111 
2112   assert(((unsigned long long)start & 0x3f) == 0,
2113          "Alignment problem (0x%08llx)", (unsigned long long)start);
2114   __ emit_data64(0x090a040506000102, relocInfo::none);
2115   __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2116   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2117   __ emit_data64(0x292a242526202122, relocInfo::none);
2118   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2119   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2120   __ emit_data64(0x494a444546404142, relocInfo::none);
2121   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2122 
2123   return start;
2124 }
2125 
2126 address StubGenerator::base64_vbmi_join_1_2_addr() {
2127   __ align64();
2128   StubCodeMark mark(this, "StubRoutines", "join_1_2_base64");
2129   address start = __ pc();
2130 
2131   assert(((unsigned long long)start & 0x3f) == 0,
2132          "Alignment problem (0x%08llx)", (unsigned long long)start);
2133   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2134   __ emit_data64(0x292a242526202122, relocInfo::none);
2135   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2136   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2137   __ emit_data64(0x494a444546404142, relocInfo::none);
2138   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2139   __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2140   __ emit_data64(0x696a646566606162, relocInfo::none);
2141 
2142   return start;
2143 }
2144 
2145 address StubGenerator::base64_vbmi_join_2_3_addr() {
2146   __ align64();
2147   StubCodeMark mark(this, "StubRoutines", "join_2_3_base64");
2148   address start = __ pc();
2149 
2150   assert(((unsigned long long)start & 0x3f) == 0,
2151          "Alignment problem (0x%08llx)", (unsigned long long)start);
2152   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2153   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2154   __ emit_data64(0x494a444546404142, relocInfo::none);
2155   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2156   __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2157   __ emit_data64(0x696a646566606162, relocInfo::none);
2158   __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2159   __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2160 
2161   return start;
2162 }
2163 
2164 address StubGenerator::base64_AVX2_decode_tables_addr() {
2165   __ align64();
2166   StubCodeMark mark(this, "StubRoutines", "AVX2_tables_base64");
2167   address start = __ pc();
2168 
2169   assert(((unsigned long long)start & 0x3f) == 0,
2170          "Alignment problem (0x%08llx)", (unsigned long long)start);
2171   __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2172   __ emit_data(0x5f5f5f5f, relocInfo::none, 0);  // for URL
2173 
2174   __ emit_data(0xffffffff, relocInfo::none, 0);
2175   __ emit_data(0xfcfcfcfc, relocInfo::none, 0);  // for URL
2176 
2177   // Permute table
2178   __ emit_data64(0x0000000100000000, relocInfo::none);
2179   __ emit_data64(0x0000000400000002, relocInfo::none);
2180   __ emit_data64(0x0000000600000005, relocInfo::none);
2181   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2182 
2183   // Shuffle table
2184   __ emit_data64(0x090a040506000102, relocInfo::none);
2185   __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2186   __ emit_data64(0x090a040506000102, relocInfo::none);
2187   __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2188 
2189   // merge table
2190   __ emit_data(0x01400140, relocInfo::none, 0);
2191 
2192   // merge multiplier
2193   __ emit_data(0x00011000, relocInfo::none, 0);
2194 
2195   return start;
2196 }
2197 
2198 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2199   __ align64();
2200   StubCodeMark mark(this, "StubRoutines", "AVX2_tables_URL_base64");
2201   address start = __ pc();
2202 
2203   assert(((unsigned long long)start & 0x3f) == 0,
2204          "Alignment problem (0x%08llx)", (unsigned long long)start);
2205   // lut_lo
2206   __ emit_data64(0x1111111111111115, relocInfo::none);
2207   __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2208   __ emit_data64(0x1111111111111115, relocInfo::none);
2209   __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2210 
2211   // lut_roll
2212   __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2213   __ emit_data64(0x0000000000000000, relocInfo::none);
2214   __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2215   __ emit_data64(0x0000000000000000, relocInfo::none);
2216 
2217   // lut_lo URL
2218   __ emit_data64(0x1111111111111115, relocInfo::none);
2219   __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2220   __ emit_data64(0x1111111111111115, relocInfo::none);
2221   __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2222 
2223   // lut_roll URL
2224   __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2225   __ emit_data64(0x0000000000000000, relocInfo::none);
2226   __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2227   __ emit_data64(0x0000000000000000, relocInfo::none);
2228 
2229   // lut_hi
2230   __ emit_data64(0x0804080402011010, relocInfo::none);
2231   __ emit_data64(0x1010101010101010, relocInfo::none);
2232   __ emit_data64(0x0804080402011010, relocInfo::none);
2233   __ emit_data64(0x1010101010101010, relocInfo::none);
2234 
2235   return start;
2236 }
2237 
2238 address StubGenerator::base64_decoding_table_addr() {
2239   StubCodeMark mark(this, "StubRoutines", "decoding_table_base64");
2240   address start = __ pc();
2241 
2242   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2243   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2244   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2245   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2246   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2247   __ emit_data64(0x3fffffff3effffff, relocInfo::none);
2248   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2249   __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2250   __ emit_data64(0x06050403020100ff, relocInfo::none);
2251   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2252   __ emit_data64(0x161514131211100f, relocInfo::none);
2253   __ emit_data64(0xffffffffff191817, relocInfo::none);
2254   __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2255   __ emit_data64(0x2827262524232221, relocInfo::none);
2256   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2257   __ emit_data64(0xffffffffff333231, relocInfo::none);
2258   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2259   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2260   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2261   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2262   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2263   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2264   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2265   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2266   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2267   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2268   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2269   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2270   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2271   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2272   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2273   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2274 
2275   // URL table
2276   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2277   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2278   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2279   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2280   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2281   __ emit_data64(0xffff3effffffffff, relocInfo::none);
2282   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2283   __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2284   __ emit_data64(0x06050403020100ff, relocInfo::none);
2285   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2286   __ emit_data64(0x161514131211100f, relocInfo::none);
2287   __ emit_data64(0x3fffffffff191817, relocInfo::none);
2288   __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2289   __ emit_data64(0x2827262524232221, relocInfo::none);
2290   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2291   __ emit_data64(0xffffffffff333231, relocInfo::none);
2292   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2293   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2294   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2295   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2296   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2297   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2298   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2299   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2300   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2301   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2302   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2303   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2304   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2305   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2306   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2307   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2308 
2309   return start;
2310 }
2311 
2312 
2313 // Code for generating Base64 decoding.
2314 //
2315 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2316 //
2317 // Intrinsic function prototype in Base64.java:
2318 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2319 address StubGenerator::generate_base64_decodeBlock() {
2320   __ align(CodeEntryAlignment);
2321   StubCodeMark mark(this, "StubRoutines", "implDecode");
2322   address start = __ pc();
2323 
2324   __ enter();
2325 
2326   // Save callee-saved registers before using them
2327   __ push(r12);
2328   __ push(r13);
2329   __ push(r14);
2330   __ push(r15);
2331   __ push(rbx);
2332 
2333   // arguments
2334   const Register source = c_rarg0; // Source Array
2335   const Register start_offset = c_rarg1; // start offset
2336   const Register end_offset = c_rarg2; // end offset
2337   const Register dest = c_rarg3; // destination array
2338   const Register isMIME = rbx;
2339 
2340 #ifndef _WIN64
2341   const Register dp = c_rarg4;  // Position for writing to dest array
2342   const Register isURL = c_rarg5;// Base64 or URL character set
2343   __ movl(isMIME, Address(rbp, 2 * wordSize));
2344 #else
2345   const Address  dp_mem(rbp, 6 * wordSize);  // length is on stack on Win64
2346   const Address isURL_mem(rbp, 7 * wordSize);
2347   const Register isURL = r10;      // pick the volatile windows register
2348   const Register dp = r12;
2349   __ movl(dp, dp_mem);
2350   __ movl(isURL, isURL_mem);
2351   __ movl(isMIME, Address(rbp, 8 * wordSize));
2352 #endif
2353 
2354   const XMMRegister lookup_lo = xmm5;
2355   const XMMRegister lookup_hi = xmm6;
2356   const XMMRegister errorvec = xmm7;
2357   const XMMRegister pack16_op = xmm9;
2358   const XMMRegister pack32_op = xmm8;
2359   const XMMRegister input0 = xmm3;
2360   const XMMRegister input1 = xmm20;
2361   const XMMRegister input2 = xmm21;
2362   const XMMRegister input3 = xmm19;
2363   const XMMRegister join01 = xmm12;
2364   const XMMRegister join12 = xmm11;
2365   const XMMRegister join23 = xmm10;
2366   const XMMRegister translated0 = xmm2;
2367   const XMMRegister translated1 = xmm1;
2368   const XMMRegister translated2 = xmm0;
2369   const XMMRegister translated3 = xmm4;
2370 
2371   const XMMRegister merged0 = xmm2;
2372   const XMMRegister merged1 = xmm1;
2373   const XMMRegister merged2 = xmm0;
2374   const XMMRegister merged3 = xmm4;
2375   const XMMRegister merge_ab_bc0 = xmm2;
2376   const XMMRegister merge_ab_bc1 = xmm1;
2377   const XMMRegister merge_ab_bc2 = xmm0;
2378   const XMMRegister merge_ab_bc3 = xmm4;
2379 
2380   const XMMRegister pack24bits = xmm4;
2381 
2382   const Register length = r14;
2383   const Register output_size = r13;
2384   const Register output_mask = r15;
2385   const KRegister input_mask = k1;
2386 
2387   const XMMRegister input_initial_valid_b64 = xmm0;
2388   const XMMRegister tmp = xmm10;
2389   const XMMRegister mask = xmm0;
2390   const XMMRegister invalid_b64 = xmm1;
2391 
2392   Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2393   Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2394   Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2395 
2396   // calculate length from offsets
2397   __ movl(length, end_offset);
2398   __ subl(length, start_offset);
2399   __ push(dest);          // Save for return value calc
2400 
2401   // If AVX512 VBMI not supported, just compile non-AVX code
2402   if(VM_Version::supports_avx512_vbmi() &&
2403      VM_Version::supports_avx512bw()) {
2404     __ cmpl(length, 31);     // 32-bytes is break-even for AVX-512
2405     __ jcc(Assembler::lessEqual, L_lastChunk);
2406 
2407     __ cmpl(isMIME, 0);
2408     __ jcc(Assembler::notEqual, L_lastChunk);
2409 
2410     // Load lookup tables based on isURL
2411     __ cmpl(isURL, 0);
2412     __ jcc(Assembler::notZero, L_loadURL);
2413 
2414     __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2415     __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2416 
2417     __ BIND(L_continue);
2418 
2419     __ movl(r15, 0x01400140);
2420     __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2421 
2422     __ movl(r15, 0x00011000);
2423     __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2424 
2425     __ cmpl(length, 0xff);
2426     __ jcc(Assembler::lessEqual, L_process64);
2427 
2428     // load masks required for decoding data
2429     __ BIND(L_processdata);
2430     __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2431     __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2432     __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2433 
2434     __ align32();
2435     __ BIND(L_process256);
2436     // Grab input data
2437     __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2438     __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2439     __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2440     __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2441 
2442     // Copy the low part of the lookup table into the destination of the permutation
2443     __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2444     __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2445     __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2446     __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2447 
2448     // Translate the base64 input into "decoded" bytes
2449     __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2450     __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2451     __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2452     __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2453 
2454     // OR all of the translations together to check for errors (high-order bit of byte set)
2455     __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2456 
2457     __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2458     __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2459     __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2460 
2461     // Check if there was an error - if so, try 64-byte chunks
2462     __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2463     __ kortestql(k3, k3);
2464     __ jcc(Assembler::notZero, L_process64);
2465 
2466     // The merging and shuffling happens here
2467     // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2468     // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2469     // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2470     __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2471     __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2472     __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2473     __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2474 
2475     // Now do the same with packed 16-bit values.
2476     // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2477     // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2478     // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2479     __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2480     __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2481     __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2482     __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2483 
2484     // The join vectors specify which byte from which vector goes into the outputs
2485     // One of every 4 bytes in the extended vector is zero, so we pack them into their
2486     // final positions in the register for storing (256 bytes in, 192 bytes out)
2487     __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2488     __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2489     __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2490 
2491     // Store result
2492     __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2493     __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2494     __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2495 
2496     __ addptr(source, 0x100);
2497     __ addptr(dest, 0xc0);
2498     __ subl(length, 0x100);
2499     __ cmpl(length, 64 * 4);
2500     __ jcc(Assembler::greaterEqual, L_process256);
2501 
2502     // At this point, we've decoded 64 * 4 * n bytes.
2503     // The remaining length will be <= 64 * 4 - 1.
2504     // UNLESS there was an error decoding the first 256-byte chunk.  In this
2505     // case, the length will be arbitrarily long.
2506     //
2507     // Note that this will be the path for MIME-encoded strings.
2508 
2509     __ BIND(L_process64);
2510 
2511     __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2512 
2513     __ cmpl(length, 63);
2514     __ jcc(Assembler::lessEqual, L_finalBit);
2515 
2516     __ mov64(rax, 0x0000ffffffffffff);
2517     __ kmovql(k2, rax);
2518 
2519     __ align32();
2520     __ BIND(L_process64Loop);
2521 
2522     // Handle first 64-byte block
2523 
2524     __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2525     __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2526     __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2527 
2528     __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2529 
2530     // Check for error and bomb out before updating dest
2531     __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2532     __ kortestql(k3, k3);
2533     __ jcc(Assembler::notZero, L_exit);
2534 
2535     // Pack output register, selecting correct byte ordering
2536     __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2537     __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2538     __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2539 
2540     __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2541 
2542     __ subl(length, 64);
2543     __ addptr(source, 64);
2544     __ addptr(dest, 48);
2545 
2546     __ cmpl(length, 64);
2547     __ jcc(Assembler::greaterEqual, L_process64Loop);
2548 
2549     __ cmpl(length, 0);
2550     __ jcc(Assembler::lessEqual, L_exit);
2551 
2552     __ BIND(L_finalBit);
2553     // Now have 1 to 63 bytes left to decode
2554 
2555     // I was going to let Java take care of the final fragment
2556     // however it will repeatedly call this routine for every 4 bytes
2557     // of input data, so handle the rest here.
2558     __ movq(rax, -1);
2559     __ bzhiq(rax, rax, length);    // Input mask in rax
2560 
2561     __ movl(output_size, length);
2562     __ shrl(output_size, 2);   // Find (len / 4) * 3 (output length)
2563     __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2564     // output_size in r13
2565 
2566     // Strip pad characters, if any, and adjust length and mask
2567     __ cmpb(Address(source, length, Address::times_1, -1), '=');
2568     __ jcc(Assembler::equal, L_padding);
2569 
2570     __ BIND(L_donePadding);
2571 
2572     // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2573     __ kmovql(input_mask, rax);
2574     __ movq(output_mask, -1);
2575     __ bzhiq(output_mask, output_mask, output_size);
2576 
2577     // Load initial input with all valid base64 characters.  Will be used
2578     // in merging source bytes to avoid masking when determining if an error occurred.
2579     __ movl(rax, 0x61616161);
2580     __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2581 
2582     // A register containing all invalid base64 decoded values
2583     __ movl(rax, 0x80808080);
2584     __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2585 
2586     // input_mask is in k1
2587     // output_size is in r13
2588     // output_mask is in r15
2589     // zmm0 - free
2590     // zmm1 - 0x00011000
2591     // zmm2 - 0x01400140
2592     // zmm3 - errorvec
2593     // zmm4 - pack vector
2594     // zmm5 - lookup_lo
2595     // zmm6 - lookup_hi
2596     // zmm7 - errorvec
2597     // zmm8 - 0x61616161
2598     // zmm9 - 0x80808080
2599 
2600     // Load only the bytes from source, merging into our "fully-valid" register
2601     __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2602 
2603     // Decode all bytes within our merged input
2604     __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2605     __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2606     __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2607 
2608     // Check for error.  Compare (decoded | initial) to all invalid.
2609     // If any bytes have their high-order bit set, then we have an error.
2610     __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2611     __ kortestql(k2, k2);
2612 
2613     // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2614     __ jcc(Assembler::notZero, L_bruteForce);
2615 
2616     // Shuffle output bytes
2617     __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2618     __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2619 
2620     __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2621     __ kmovql(k1, output_mask);
2622     __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2623 
2624     __ addptr(dest, output_size);
2625 
2626     __ BIND(L_exit);
2627     __ vzeroupper();
2628     __ pop(rax);             // Get original dest value
2629     __ subptr(dest, rax);      // Number of bytes converted
2630     __ movptr(rax, dest);
2631     __ pop(rbx);
2632     __ pop(r15);
2633     __ pop(r14);
2634     __ pop(r13);
2635     __ pop(r12);
2636     __ leave();
2637     __ ret(0);
2638 
2639     __ BIND(L_loadURL);
2640     __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2641     __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2642     __ jmp(L_continue);
2643 
2644     __ BIND(L_padding);
2645     __ decrementq(output_size, 1);
2646     __ shrq(rax, 1);
2647 
2648     __ cmpb(Address(source, length, Address::times_1, -2), '=');
2649     __ jcc(Assembler::notEqual, L_donePadding);
2650 
2651     __ decrementq(output_size, 1);
2652     __ shrq(rax, 1);
2653     __ jmp(L_donePadding);
2654 
2655     __ align32();
2656     __ BIND(L_bruteForce);
2657   }   // End of if(avx512_vbmi)
2658 
2659   if (VM_Version::supports_avx2()) {
2660     Label L_tailProc, L_topLoop, L_enterLoop;
2661 
2662     __ cmpl(isMIME, 0);
2663     __ jcc(Assembler::notEqual, L_lastChunk);
2664 
2665     // Check for buffer too small (for algorithm)
2666     __ subl(length, 0x2c);
2667     __ jcc(Assembler::less, L_tailProc);
2668 
2669     __ shll(isURL, 2);
2670 
2671     // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2672     // Encoding and Decoding using AVX2 Instructions".  URL modifications added.
2673 
2674     // Set up constants
2675     __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2676     __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit);  // 2F or 5F
2677     __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit);  // -1 or -4
2678     __ vmovdqu(xmm12, Address(r13, 0x10));  // permute
2679     __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2680     __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit);  // merge
2681     __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit);  // merge mult
2682 
2683     __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2684     __ shll(isURL, 4);
2685     __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00));  // lut_lo
2686     __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2687     __ shrl(isURL, 6);  // restore isURL
2688     __ vmovdqu(xmm9, Address(r13, 0x80));  // lut_hi
2689     __ jmp(L_enterLoop);
2690 
2691     __ align32();
2692     __ bind(L_topLoop);
2693     // Add in the offset value (roll) to get 6-bit out values
2694     __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2695     // Merge and permute the output bits into appropriate output byte lanes
2696     __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2697     __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2698     __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2699     __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2700     // Store the output bytes
2701     __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2702     __ addptr(source, 0x20);
2703     __ addptr(dest, 0x18);
2704     __ subl(length, 0x20);
2705     __ jcc(Assembler::less, L_tailProc);
2706 
2707     __ bind(L_enterLoop);
2708 
2709     // Load in encoded string (32 bytes)
2710     __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2711     // Extract the high nibble for indexing into the lut tables.  High 4 bits are don't care.
2712     __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2713     __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2714     // Extract the low nibble. 5F/2F will isolate the low-order 4 bits.  High 4 bits are don't care.
2715     __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2716     // Check for special-case (0x2F or 0x5F (URL))
2717     __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2718     // Get the bitset based on the low nibble.  vpshufb uses low-order 4 bits only.
2719     __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2720     // Get the bit value of the high nibble
2721     __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2722     // Make sure 2F / 5F shows as valid
2723     __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2724     // Make adjustment for roll index.  For non-URL, this is a no-op,
2725     // for URL, this adjusts by -4.  This is to properly index the
2726     // roll value for 2F / 5F.
2727     __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2728     // If the and of the two is non-zero, we have an invalid input character
2729     __ vptest(xmm3, xmm5);
2730     // Extract the "roll" value - value to add to the input to get 6-bit out value
2731     __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2732     __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2733     __ jcc(Assembler::equal, L_topLoop);  // Fall through on error
2734 
2735     __ bind(L_tailProc);
2736 
2737     __ addl(length, 0x2c);
2738 
2739     __ vzeroupper();
2740   }
2741 
2742   // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2743 
2744   // Register state (Linux):
2745   // r12-15 - saved on stack
2746   // rdi - src
2747   // rsi - sp
2748   // rdx - sl
2749   // rcx - dst
2750   // r8 - dp
2751   // r9 - isURL
2752 
2753   // Register state (Windows):
2754   // r12-15 - saved on stack
2755   // rcx - src
2756   // rdx - sp
2757   // r8 - sl
2758   // r9 - dst
2759   // r12 - dp
2760   // r10 - isURL
2761 
2762   // Registers (common):
2763   // length (r14) - bytes in src
2764 
2765   const Register decode_table = r11;
2766   const Register out_byte_count = rbx;
2767   const Register byte1 = r13;
2768   const Register byte2 = r15;
2769   const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2770   const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2771 
2772   __ bind(L_lastChunk);
2773 
2774   __ shrl(length, 2);    // Multiple of 4 bytes only - length is # 4-byte chunks
2775   __ cmpl(length, 0);
2776   __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2777 
2778   __ shll(isURL, 8);    // index into decode table based on isURL
2779   __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2780   __ addptr(decode_table, isURL);
2781 
2782   __ jmp(L_bottomLoop);
2783 
2784   __ align32();
2785   __ BIND(L_forceLoop);
2786   __ shll(byte1, 18);
2787   __ shll(byte2, 12);
2788   __ shll(byte3, 6);
2789   __ orl(byte1, byte2);
2790   __ orl(byte1, byte3);
2791   __ orl(byte1, byte4);
2792 
2793   __ addptr(source, 4);
2794 
2795   __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2796   __ shrl(byte1, 8);
2797   __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2798   __ shrl(byte1, 8);
2799   __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2800 
2801   __ addptr(dest, 3);
2802   __ decrementl(length, 1);
2803   __ jcc(Assembler::zero, L_exit_no_vzero);
2804 
2805   __ BIND(L_bottomLoop);
2806   __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2807   __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2808   __ load_signed_byte(byte1, Address(decode_table, byte1));
2809   __ load_signed_byte(byte2, Address(decode_table, byte2));
2810   __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2811   __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2812   __ load_signed_byte(byte3, Address(decode_table, byte3));
2813   __ load_signed_byte(byte4, Address(decode_table, byte4));
2814 
2815   __ mov(rax, byte1);
2816   __ orl(rax, byte2);
2817   __ orl(rax, byte3);
2818   __ orl(rax, byte4);
2819   __ jcc(Assembler::positive, L_forceLoop);
2820 
2821   __ BIND(L_exit_no_vzero);
2822   __ pop(rax);             // Get original dest value
2823   __ subptr(dest, rax);      // Number of bytes converted
2824   __ movptr(rax, dest);
2825   __ pop(rbx);
2826   __ pop(r15);
2827   __ pop(r14);
2828   __ pop(r13);
2829   __ pop(r12);
2830   __ leave();
2831   __ ret(0);
2832 
2833   return start;
2834 }
2835 
2836 
2837 /**
2838  *  Arguments:
2839  *
2840  * Inputs:
2841  *   c_rarg0   - int crc
2842  *   c_rarg1   - byte* buf
2843  *   c_rarg2   - int length
2844  *
2845  * Output:
2846  *       rax   - int crc result
2847  */
2848 address StubGenerator::generate_updateBytesCRC32() {
2849   assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2850 
2851   __ align(CodeEntryAlignment);
2852   StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2853 
2854   address start = __ pc();
2855 
2856   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
2857   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
2858   // rscratch1: r10
2859   const Register crc   = c_rarg0;  // crc
2860   const Register buf   = c_rarg1;  // source java byte array address
2861   const Register len   = c_rarg2;  // length
2862   const Register table = c_rarg3;  // crc_table address (reuse register)
2863   const Register tmp1   = r11;
2864   const Register tmp2   = r10;
2865   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
2866 
2867   BLOCK_COMMENT("Entry:");
2868   __ enter(); // required for proper stackwalking of RuntimeStub frame
2869 
2870   if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2871       VM_Version::supports_avx512bw() &&
2872       VM_Version::supports_avx512vl()) {
2873       // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
2874       // However, the constant table for CRC32-C assumes the original crc value.  Account for this
2875       // difference before calling and after returning.
2876     __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
2877     __ notl(crc);
2878     __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
2879     __ notl(crc);
2880   } else {
2881     __ kernel_crc32(crc, buf, len, table, tmp1);
2882   }
2883 
2884   __ movl(rax, crc);
2885   __ vzeroupper();
2886   __ leave(); // required for proper stackwalking of RuntimeStub frame
2887   __ ret(0);
2888 
2889   return start;
2890 }
2891 
2892 /**
2893 *  Arguments:
2894 *
2895 * Inputs:
2896 *   c_rarg0   - int crc
2897 *   c_rarg1   - byte* buf
2898 *   c_rarg2   - long length
2899 *   c_rarg3   - table_start - optional (present only when doing a library_call,
2900 *              not used by x86 algorithm)
2901 *
2902 * Output:
2903 *       rax   - int crc result
2904 */
2905 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
2906   assert(UseCRC32CIntrinsics, "need SSE4_2");
2907   __ align(CodeEntryAlignment);
2908   StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
2909   address start = __ pc();
2910 
2911   //reg.arg        int#0        int#1        int#2        int#3        int#4        int#5        float regs
2912   //Windows        RCX          RDX          R8           R9           none         none         XMM0..XMM3
2913   //Lin / Sol      RDI          RSI          RDX          RCX          R8           R9           XMM0..XMM7
2914   const Register crc = c_rarg0;  // crc
2915   const Register buf = c_rarg1;  // source java byte array address
2916   const Register len = c_rarg2;  // length
2917   const Register a = rax;
2918   const Register j = r9;
2919   const Register k = r10;
2920   const Register l = r11;
2921 #ifdef _WIN64
2922   const Register y = rdi;
2923   const Register z = rsi;
2924 #else
2925   const Register y = rcx;
2926   const Register z = r8;
2927 #endif
2928   assert_different_registers(crc, buf, len, a, j, k, l, y, z);
2929 
2930   BLOCK_COMMENT("Entry:");
2931   __ enter(); // required for proper stackwalking of RuntimeStub frame
2932   Label L_continue;
2933 
2934   if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2935       VM_Version::supports_avx512bw() &&
2936       VM_Version::supports_avx512vl()) {
2937     Label L_doSmall;
2938 
2939     __ cmpl(len, 384);
2940     __ jcc(Assembler::lessEqual, L_doSmall);
2941 
2942     __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
2943     __ kernel_crc32_avx512(crc, buf, len, j, l, k);
2944 
2945     __ jmp(L_continue);
2946 
2947     __ bind(L_doSmall);
2948   }
2949 #ifdef _WIN64
2950   __ push(y);
2951   __ push(z);
2952 #endif
2953   __ crc32c_ipl_alg2_alt2(crc, buf, len,
2954                           a, j, k,
2955                           l, y, z,
2956                           c_farg0, c_farg1, c_farg2,
2957                           is_pclmulqdq_supported);
2958 #ifdef _WIN64
2959   __ pop(z);
2960   __ pop(y);
2961 #endif
2962 
2963   __ bind(L_continue);
2964   __ movl(rax, crc);
2965   __ vzeroupper();
2966   __ leave(); // required for proper stackwalking of RuntimeStub frame
2967   __ ret(0);
2968 
2969   return start;
2970 }
2971 
2972 
2973 /**
2974  *  Arguments:
2975  *
2976  *  Input:
2977  *    c_rarg0   - x address
2978  *    c_rarg1   - x length
2979  *    c_rarg2   - y address
2980  *    c_rarg3   - y length
2981  * not Win64
2982  *    c_rarg4   - z address
2983  *    c_rarg5   - z length
2984  * Win64
2985  *    rsp+40    - z address
2986  *    rsp+48    - z length
2987  */
2988 address StubGenerator::generate_multiplyToLen() {
2989   __ align(CodeEntryAlignment);
2990   StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
2991   address start = __ pc();
2992 
2993   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
2994   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
2995   const Register x     = rdi;
2996   const Register xlen  = rax;
2997   const Register y     = rsi;
2998   const Register ylen  = rcx;
2999   const Register z     = r8;
3000   const Register zlen  = r11;
3001 
3002   // Next registers will be saved on stack in multiply_to_len().
3003   const Register tmp1  = r12;
3004   const Register tmp2  = r13;
3005   const Register tmp3  = r14;
3006   const Register tmp4  = r15;
3007   const Register tmp5  = rbx;
3008 
3009   BLOCK_COMMENT("Entry:");
3010   __ enter(); // required for proper stackwalking of RuntimeStub frame
3011 
3012 #ifndef _WIN64
3013   __ movptr(zlen, r9); // Save r9 in r11 - zlen
3014 #endif
3015   setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3016                      // ylen => rcx, z => r8, zlen => r11
3017                      // r9 and r10 may be used to save non-volatile registers
3018 #ifdef _WIN64
3019   // last 2 arguments (#4, #5) are on stack on Win64
3020   __ movptr(z, Address(rsp, 6 * wordSize));
3021   __ movptr(zlen, Address(rsp, 7 * wordSize));
3022 #endif
3023 
3024   __ movptr(xlen, rsi);
3025   __ movptr(y,    rdx);
3026   __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
3027 
3028   restore_arg_regs();
3029 
3030   __ leave(); // required for proper stackwalking of RuntimeStub frame
3031   __ ret(0);
3032 
3033   return start;
3034 }
3035 
3036 /**
3037 *  Arguments:
3038 *
3039 *  Input:
3040 *    c_rarg0   - obja     address
3041 *    c_rarg1   - objb     address
3042 *    c_rarg3   - length   length
3043 *    c_rarg4   - scale    log2_array_indxscale
3044 *
3045 *  Output:
3046 *        rax   - int >= mismatched index, < 0 bitwise complement of tail
3047 */
3048 address StubGenerator::generate_vectorizedMismatch() {
3049   __ align(CodeEntryAlignment);
3050   StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
3051   address start = __ pc();
3052 
3053   BLOCK_COMMENT("Entry:");
3054   __ enter();
3055 
3056 #ifdef _WIN64  // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3057   const Register scale = c_rarg0;  //rcx, will exchange with r9
3058   const Register objb = c_rarg1;   //rdx
3059   const Register length = c_rarg2; //r8
3060   const Register obja = c_rarg3;   //r9
3061   __ xchgq(obja, scale);  //now obja and scale contains the correct contents
3062 
3063   const Register tmp1 = r10;
3064   const Register tmp2 = r11;
3065 #endif
3066 #ifndef _WIN64 // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3067   const Register obja = c_rarg0;   //U:rdi
3068   const Register objb = c_rarg1;   //U:rsi
3069   const Register length = c_rarg2; //U:rdx
3070   const Register scale = c_rarg3;  //U:rcx
3071   const Register tmp1 = r8;
3072   const Register tmp2 = r9;
3073 #endif
3074   const Register result = rax; //return value
3075   const XMMRegister vec0 = xmm0;
3076   const XMMRegister vec1 = xmm1;
3077   const XMMRegister vec2 = xmm2;
3078 
3079   __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3080 
3081   __ vzeroupper();
3082   __ leave();
3083   __ ret(0);
3084 
3085   return start;
3086 }
3087 
3088 /**
3089  *  Arguments:
3090  *
3091 //  Input:
3092 //    c_rarg0   - x address
3093 //    c_rarg1   - x length
3094 //    c_rarg2   - z address
3095 //    c_rarg3   - z length
3096  *
3097  */
3098 address StubGenerator::generate_squareToLen() {
3099 
3100   __ align(CodeEntryAlignment);
3101   StubCodeMark mark(this, "StubRoutines", "squareToLen");
3102   address start = __ pc();
3103 
3104   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3105   // Unix:  rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3106   const Register x      = rdi;
3107   const Register len    = rsi;
3108   const Register z      = r8;
3109   const Register zlen   = rcx;
3110 
3111  const Register tmp1      = r12;
3112  const Register tmp2      = r13;
3113  const Register tmp3      = r14;
3114  const Register tmp4      = r15;
3115  const Register tmp5      = rbx;
3116 
3117   BLOCK_COMMENT("Entry:");
3118   __ enter(); // required for proper stackwalking of RuntimeStub frame
3119 
3120   setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3121                      // zlen => rcx
3122                      // r9 and r10 may be used to save non-volatile registers
3123   __ movptr(r8, rdx);
3124   __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3125 
3126   restore_arg_regs();
3127 
3128   __ leave(); // required for proper stackwalking of RuntimeStub frame
3129   __ ret(0);
3130 
3131   return start;
3132 }
3133 
3134 address StubGenerator::generate_method_entry_barrier() {
3135   __ align(CodeEntryAlignment);
3136   StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3137   address start = __ pc();
3138 
3139   Label deoptimize_label;
3140 
3141   __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3142 
3143   BLOCK_COMMENT("Entry:");
3144   __ enter(); // save rbp
3145 
3146   // save c_rarg0, because we want to use that value.
3147   // We could do without it but then we depend on the number of slots used by pusha
3148   __ push(c_rarg0);
3149 
3150   __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3151 
3152   __ pusha();
3153 
3154   // The method may have floats as arguments, and we must spill them before calling
3155   // the VM runtime.
3156   assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3157   const int xmm_size = wordSize * 2;
3158   const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3159   __ subptr(rsp, xmm_spill_size);
3160   __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3161   __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3162   __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3163   __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3164   __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3165   __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3166   __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3167   __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3168 
3169   __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3170 
3171   __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3172   __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3173   __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3174   __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3175   __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3176   __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3177   __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3178   __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3179   __ addptr(rsp, xmm_spill_size);
3180 
3181   __ cmpl(rax, 1); // 1 means deoptimize
3182   __ jcc(Assembler::equal, deoptimize_label);
3183 
3184   __ popa();
3185   __ pop(c_rarg0);
3186 
3187   __ leave();
3188 
3189   __ addptr(rsp, 1 * wordSize); // cookie
3190   __ ret(0);
3191 
3192 
3193   __ BIND(deoptimize_label);
3194 
3195   __ popa();
3196   __ pop(c_rarg0);
3197 
3198   __ leave();
3199 
3200   // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3201   // here while still having a correct stack is valuable
3202   __ testptr(rsp, Address(rsp, 0));
3203 
3204   __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3205   __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3206 
3207   return start;
3208 }
3209 
3210  /**
3211  *  Arguments:
3212  *
3213  *  Input:
3214  *    c_rarg0   - out address
3215  *    c_rarg1   - in address
3216  *    c_rarg2   - offset
3217  *    c_rarg3   - len
3218  * not Win64
3219  *    c_rarg4   - k
3220  * Win64
3221  *    rsp+40    - k
3222  */
3223 address StubGenerator::generate_mulAdd() {
3224   __ align(CodeEntryAlignment);
3225   StubCodeMark mark(this, "StubRoutines", "mulAdd");
3226   address start = __ pc();
3227 
3228   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3229   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3230   const Register out     = rdi;
3231   const Register in      = rsi;
3232   const Register offset  = r11;
3233   const Register len     = rcx;
3234   const Register k       = r8;
3235 
3236   // Next registers will be saved on stack in mul_add().
3237   const Register tmp1  = r12;
3238   const Register tmp2  = r13;
3239   const Register tmp3  = r14;
3240   const Register tmp4  = r15;
3241   const Register tmp5  = rbx;
3242 
3243   BLOCK_COMMENT("Entry:");
3244   __ enter(); // required for proper stackwalking of RuntimeStub frame
3245 
3246   setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3247                      // len => rcx, k => r8
3248                      // r9 and r10 may be used to save non-volatile registers
3249 #ifdef _WIN64
3250   // last argument is on stack on Win64
3251   __ movl(k, Address(rsp, 6 * wordSize));
3252 #endif
3253   __ movptr(r11, rdx);  // move offset in rdx to offset(r11)
3254   __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3255 
3256   restore_arg_regs();
3257 
3258   __ leave(); // required for proper stackwalking of RuntimeStub frame
3259   __ ret(0);
3260 
3261   return start;
3262 }
3263 
3264 address StubGenerator::generate_bigIntegerRightShift() {
3265   __ align(CodeEntryAlignment);
3266   StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
3267   address start = __ pc();
3268 
3269   Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3270   // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3271   const Register newArr = rdi;
3272   const Register oldArr = rsi;
3273   const Register newIdx = rdx;
3274   const Register shiftCount = rcx;  // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3275   const Register totalNumIter = r8;
3276 
3277   // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3278   // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3279   const Register tmp1 = r11;                    // Caller save.
3280   const Register tmp2 = rax;                    // Caller save.
3281   const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9);   // Windows: Callee save. Linux: Caller save.
3282   const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10);  // Windows: Callee save. Linux: Caller save.
3283   const Register tmp5 = r14;                    // Callee save.
3284   const Register tmp6 = r15;
3285 
3286   const XMMRegister x0 = xmm0;
3287   const XMMRegister x1 = xmm1;
3288   const XMMRegister x2 = xmm2;
3289 
3290   BLOCK_COMMENT("Entry:");
3291   __ enter(); // required for proper stackwalking of RuntimeStub frame
3292 
3293 #ifdef _WIN64
3294   setup_arg_regs(4);
3295   // For windows, since last argument is on stack, we need to move it to the appropriate register.
3296   __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3297   // Save callee save registers.
3298   __ push(tmp3);
3299   __ push(tmp4);
3300 #endif
3301   __ push(tmp5);
3302 
3303   // Rename temps used throughout the code.
3304   const Register idx = tmp1;
3305   const Register nIdx = tmp2;
3306 
3307   __ xorl(idx, idx);
3308 
3309   // Start right shift from end of the array.
3310   // For example, if #iteration = 4 and newIdx = 1
3311   // then dest[4] = src[4] >> shiftCount  | src[3] <<< (shiftCount - 32)
3312   // if #iteration = 4 and newIdx = 0
3313   // then dest[3] = src[4] >> shiftCount  | src[3] <<< (shiftCount - 32)
3314   __ movl(idx, totalNumIter);
3315   __ movl(nIdx, idx);
3316   __ addl(nIdx, newIdx);
3317 
3318   // If vectorization is enabled, check if the number of iterations is at least 64
3319   // If not, then go to ShifTwo processing 2 iterations
3320   if (VM_Version::supports_avx512_vbmi2()) {
3321     __ cmpptr(totalNumIter, (AVX3Threshold/64));
3322     __ jcc(Assembler::less, ShiftTwo);
3323 
3324     if (AVX3Threshold < 16 * 64) {
3325       __ cmpl(totalNumIter, 16);
3326       __ jcc(Assembler::less, ShiftTwo);
3327     }
3328     __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3329     __ subl(idx, 16);
3330     __ subl(nIdx, 16);
3331     __ BIND(Shift512Loop);
3332     __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
3333     __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3334     __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
3335     __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
3336     __ subl(nIdx, 16);
3337     __ subl(idx, 16);
3338     __ jcc(Assembler::greaterEqual, Shift512Loop);
3339     __ addl(idx, 16);
3340     __ addl(nIdx, 16);
3341   }
3342   __ BIND(ShiftTwo);
3343   __ cmpl(idx, 2);
3344   __ jcc(Assembler::less, ShiftOne);
3345   __ subl(idx, 2);
3346   __ subl(nIdx, 2);
3347   __ BIND(ShiftTwoLoop);
3348   __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
3349   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3350   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3351   __ shrdl(tmp5, tmp4);
3352   __ shrdl(tmp4, tmp3);
3353   __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
3354   __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3355   __ subl(nIdx, 2);
3356   __ subl(idx, 2);
3357   __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3358   __ addl(idx, 2);
3359   __ addl(nIdx, 2);
3360 
3361   // Do the last iteration
3362   __ BIND(ShiftOne);
3363   __ cmpl(idx, 1);
3364   __ jcc(Assembler::less, Exit);
3365   __ subl(idx, 1);
3366   __ subl(nIdx, 1);
3367   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3368   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3369   __ shrdl(tmp4, tmp3);
3370   __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3371   __ BIND(Exit);
3372   __ vzeroupper();
3373   // Restore callee save registers.
3374   __ pop(tmp5);
3375 #ifdef _WIN64
3376   __ pop(tmp4);
3377   __ pop(tmp3);
3378   restore_arg_regs();
3379 #endif
3380   __ leave(); // required for proper stackwalking of RuntimeStub frame
3381   __ ret(0);
3382 
3383   return start;
3384 }
3385 
3386  /**
3387  *  Arguments:
3388  *
3389  *  Input:
3390  *    c_rarg0   - newArr address
3391  *    c_rarg1   - oldArr address
3392  *    c_rarg2   - newIdx
3393  *    c_rarg3   - shiftCount
3394  * not Win64
3395  *    c_rarg4   - numIter
3396  * Win64
3397  *    rsp40    - numIter
3398  */
3399 address StubGenerator::generate_bigIntegerLeftShift() {
3400   __ align(CodeEntryAlignment);
3401   StubCodeMark mark(this,  "StubRoutines", "bigIntegerLeftShiftWorker");
3402   address start = __ pc();
3403 
3404   Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3405   // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3406   const Register newArr = rdi;
3407   const Register oldArr = rsi;
3408   const Register newIdx = rdx;
3409   const Register shiftCount = rcx;  // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3410   const Register totalNumIter = r8;
3411   // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3412   // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3413   const Register tmp1 = r11;                    // Caller save.
3414   const Register tmp2 = rax;                    // Caller save.
3415   const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9);   // Windows: Callee save. Linux: Caller save.
3416   const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10);  // Windows: Callee save. Linux: Caller save.
3417   const Register tmp5 = r14;                    // Callee save.
3418 
3419   const XMMRegister x0 = xmm0;
3420   const XMMRegister x1 = xmm1;
3421   const XMMRegister x2 = xmm2;
3422   BLOCK_COMMENT("Entry:");
3423   __ enter(); // required for proper stackwalking of RuntimeStub frame
3424 
3425 #ifdef _WIN64
3426   setup_arg_regs(4);
3427   // For windows, since last argument is on stack, we need to move it to the appropriate register.
3428   __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3429   // Save callee save registers.
3430   __ push(tmp3);
3431   __ push(tmp4);
3432 #endif
3433   __ push(tmp5);
3434 
3435   // Rename temps used throughout the code
3436   const Register idx = tmp1;
3437   const Register numIterTmp = tmp2;
3438 
3439   // Start idx from zero.
3440   __ xorl(idx, idx);
3441   // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
3442   __ lea(newArr, Address(newArr, newIdx, Address::times_4));
3443   __ movl(numIterTmp, totalNumIter);
3444 
3445   // If vectorization is enabled, check if the number of iterations is at least 64
3446   // If not, then go to ShiftTwo shifting two numbers at a time
3447   if (VM_Version::supports_avx512_vbmi2()) {
3448     __ cmpl(totalNumIter, (AVX3Threshold/64));
3449     __ jcc(Assembler::less, ShiftTwo);
3450 
3451     if (AVX3Threshold < 16 * 64) {
3452       __ cmpl(totalNumIter, 16);
3453       __ jcc(Assembler::less, ShiftTwo);
3454     }
3455     __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3456     __ subl(numIterTmp, 16);
3457     __ BIND(Shift512Loop);
3458     __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3459     __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
3460     __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
3461     __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
3462     __ addl(idx, 16);
3463     __ subl(numIterTmp, 16);
3464     __ jcc(Assembler::greaterEqual, Shift512Loop);
3465     __ addl(numIterTmp, 16);
3466   }
3467   __ BIND(ShiftTwo);
3468   __ cmpl(totalNumIter, 1);
3469   __ jcc(Assembler::less, Exit);
3470   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3471   __ subl(numIterTmp, 2);
3472   __ jcc(Assembler::less, ShiftOne);
3473 
3474   __ BIND(ShiftTwoLoop);
3475   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3476   __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
3477   __ shldl(tmp3, tmp4);
3478   __ shldl(tmp4, tmp5);
3479   __ movl(Address(newArr, idx, Address::times_4), tmp3);
3480   __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
3481   __ movl(tmp3, tmp5);
3482   __ addl(idx, 2);
3483   __ subl(numIterTmp, 2);
3484   __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3485 
3486   // Do the last iteration
3487   __ BIND(ShiftOne);
3488   __ addl(numIterTmp, 2);
3489   __ cmpl(numIterTmp, 1);
3490   __ jcc(Assembler::less, Exit);
3491   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3492   __ shldl(tmp3, tmp4);
3493   __ movl(Address(newArr, idx, Address::times_4), tmp3);
3494 
3495   __ BIND(Exit);
3496   __ vzeroupper();
3497   // Restore callee save registers.
3498   __ pop(tmp5);
3499 #ifdef _WIN64
3500   __ pop(tmp4);
3501   __ pop(tmp3);
3502   restore_arg_regs();
3503 #endif
3504   __ leave(); // required for proper stackwalking of RuntimeStub frame
3505   __ ret(0);
3506 
3507   return start;
3508 }
3509 
3510 void StubGenerator::generate_libm_stubs() {
3511   if (UseLibmIntrinsic && InlineIntrinsics) {
3512     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3513       StubRoutines::_dsin = generate_libmSin(); // from stubGenerator_x86_64_sin.cpp
3514     }
3515     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3516       StubRoutines::_dcos = generate_libmCos(); // from stubGenerator_x86_64_cos.cpp
3517     }
3518     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3519       StubRoutines::_dtan = generate_libmTan(); // from stubGenerator_x86_64_tan.cpp
3520     }
3521     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3522       StubRoutines::_dexp = generate_libmExp(); // from stubGenerator_x86_64_exp.cpp
3523     }
3524     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3525       StubRoutines::_dpow = generate_libmPow(); // from stubGenerator_x86_64_pow.cpp
3526     }
3527     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3528       StubRoutines::_dlog = generate_libmLog(); // from stubGenerator_x86_64_log.cpp
3529     }
3530     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3531       StubRoutines::_dlog10 = generate_libmLog10(); // from stubGenerator_x86_64_log.cpp
3532     }
3533   }
3534 }
3535 
3536 /**
3537 *  Arguments:
3538 *
3539 *  Input:
3540 *    c_rarg0   - float16  jshort
3541 *
3542 *  Output:
3543 *       xmm0   - float
3544 */
3545 address StubGenerator::generate_float16ToFloat() {
3546   StubCodeMark mark(this, "StubRoutines", "float16ToFloat");
3547 
3548   address start = __ pc();
3549 
3550   BLOCK_COMMENT("Entry:");
3551   // No need for RuntimeStub frame since it is called only during JIT compilation
3552 
3553   // Load value into xmm0 and convert
3554   __ flt16_to_flt(xmm0, c_rarg0);
3555 
3556   __ ret(0);
3557 
3558   return start;
3559 }
3560 
3561 /**
3562 *  Arguments:
3563 *
3564 *  Input:
3565 *       xmm0   - float
3566 *
3567 *  Output:
3568 *        rax   - float16  jshort
3569 */
3570 address StubGenerator::generate_floatToFloat16() {
3571   StubCodeMark mark(this, "StubRoutines", "floatToFloat16");
3572 
3573   address start = __ pc();
3574 
3575   BLOCK_COMMENT("Entry:");
3576   // No need for RuntimeStub frame since it is called only during JIT compilation
3577 
3578   // Convert and put result into rax
3579   __ flt_to_flt16(rax, xmm0, xmm1);
3580 
3581   __ ret(0);
3582 
3583   return start;
3584 }
3585 
3586 address StubGenerator::generate_cont_thaw(const char* label, Continuation::thaw_kind kind) {
3587   if (!Continuations::enabled()) return nullptr;
3588 
3589   bool return_barrier = Continuation::is_thaw_return_barrier(kind);
3590   bool return_barrier_exception = Continuation::is_thaw_return_barrier_exception(kind);
3591 
3592   StubCodeMark mark(this, "StubRoutines", label);
3593   address start = __ pc();
3594 
3595   // TODO: Handle Valhalla return types. May require generating different return barriers.
3596 
3597   if (!return_barrier) {
3598     // Pop return address. If we don't do this, we get a drift,
3599     // where the bottom-most frozen frame continuously grows.
3600     __ pop(c_rarg3);
3601   } else {
3602     __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3603   }
3604 
3605 #ifdef ASSERT
3606   {
3607     Label L_good_sp;
3608     __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3609     __ jcc(Assembler::equal, L_good_sp);
3610     __ stop("Incorrect rsp at thaw entry");
3611     __ BIND(L_good_sp);
3612   }
3613 #endif // ASSERT
3614 
3615   if (return_barrier) {
3616     // Preserve possible return value from a method returning to the return barrier.
3617     __ push(rax);
3618     __ push_d(xmm0);
3619   }
3620 
3621   __ movptr(c_rarg0, r15_thread);
3622   __ movptr(c_rarg1, (return_barrier ? 1 : 0));
3623   __ call_VM_leaf(CAST_FROM_FN_PTR(address, Continuation::prepare_thaw), 2);
3624   __ movptr(rbx, rax);
3625 
3626   if (return_barrier) {
3627     // Restore return value from a method returning to the return barrier.
3628     // No safepoint in the call to thaw, so even an oop return value should be OK.
3629     __ pop_d(xmm0);
3630     __ pop(rax);
3631   }
3632 
3633 #ifdef ASSERT
3634   {
3635     Label L_good_sp;
3636     __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3637     __ jcc(Assembler::equal, L_good_sp);
3638     __ stop("Incorrect rsp after prepare thaw");
3639     __ BIND(L_good_sp);
3640   }
3641 #endif // ASSERT
3642 
3643   // rbx contains the size of the frames to thaw, 0 if overflow or no more frames
3644   Label L_thaw_success;
3645   __ testptr(rbx, rbx);
3646   __ jccb(Assembler::notZero, L_thaw_success);
3647   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
3648   __ bind(L_thaw_success);
3649 
3650   // Make room for the thawed frames and align the stack.
3651   __ subptr(rsp, rbx);
3652   __ andptr(rsp, -StackAlignmentInBytes);
3653 
3654   if (return_barrier) {
3655     // Preserve possible return value from a method returning to the return barrier. (Again.)
3656     __ push(rax);
3657     __ push_d(xmm0);
3658   }
3659 
3660   // If we want, we can templatize thaw by kind, and have three different entries.
3661   __ movptr(c_rarg0, r15_thread);
3662   __ movptr(c_rarg1, kind);
3663   __ call_VM_leaf(Continuation::thaw_entry(), 2);
3664   __ movptr(rbx, rax);
3665 
3666   if (return_barrier) {
3667     // Restore return value from a method returning to the return barrier. (Again.)
3668     // No safepoint in the call to thaw, so even an oop return value should be OK.
3669     __ pop_d(xmm0);
3670     __ pop(rax);
3671   } else {
3672     // Return 0 (success) from doYield.
3673     __ xorptr(rax, rax);
3674   }
3675 
3676   // After thawing, rbx is the SP of the yielding frame.
3677   // Move there, and then to saved RBP slot.
3678   __ movptr(rsp, rbx);
3679   __ subptr(rsp, 2*wordSize);
3680 
3681   if (return_barrier_exception) {
3682     __ movptr(c_rarg0, r15_thread);
3683     __ movptr(c_rarg1, Address(rsp, wordSize)); // return address
3684 
3685     // rax still holds the original exception oop, save it before the call
3686     __ push(rax);
3687 
3688     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 2);
3689     __ movptr(rbx, rax);
3690 
3691     // Continue at exception handler:
3692     //   rax: exception oop
3693     //   rbx: exception handler
3694     //   rdx: exception pc
3695     __ pop(rax);
3696     __ verify_oop(rax);
3697     __ pop(rbp); // pop out RBP here too
3698     __ pop(rdx);
3699     __ jmp(rbx);
3700   } else {
3701     // We are "returning" into the topmost thawed frame; see Thaw::push_return_frame
3702     __ pop(rbp);
3703     __ ret(0);
3704   }
3705 
3706   return start;
3707 }
3708 
3709 address StubGenerator::generate_cont_thaw() {
3710   return generate_cont_thaw("Cont thaw", Continuation::thaw_top);
3711 }
3712 
3713 // TODO: will probably need multiple return barriers depending on return type
3714 
3715 address StubGenerator::generate_cont_returnBarrier() {
3716   return generate_cont_thaw("Cont thaw return barrier", Continuation::thaw_return_barrier);
3717 }
3718 
3719 address StubGenerator::generate_cont_returnBarrier_exception() {
3720   return generate_cont_thaw("Cont thaw return barrier exception", Continuation::thaw_return_barrier_exception);
3721 }
3722 
3723 #if INCLUDE_JFR
3724 
3725 // For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
3726 // It returns a jobject handle to the event writer.
3727 // The handle is dereferenced and the return value is the event writer oop.
3728 RuntimeStub* StubGenerator::generate_jfr_write_checkpoint() {
3729   enum layout {
3730     rbp_off,
3731     rbpH_off,
3732     return_off,
3733     return_off2,
3734     framesize // inclusive of return address
3735   };
3736 
3737   CodeBuffer code("jfr_write_checkpoint", 1024, 64);
3738   MacroAssembler* _masm = new MacroAssembler(&code);
3739   address start = __ pc();
3740 
3741   __ enter();
3742   address the_pc = __ pc();
3743 
3744   int frame_complete = the_pc - start;
3745 
3746   __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3747   __ movptr(c_rarg0, r15_thread);
3748   __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
3749   __ reset_last_Java_frame(true);
3750 
3751   // rax is jobject handle result, unpack and process it through a barrier.
3752   __ resolve_global_jobject(rax, r15_thread, c_rarg0);
3753 
3754   __ leave();
3755   __ ret(0);
3756 
3757   OopMapSet* oop_maps = new OopMapSet();
3758   OopMap* map = new OopMap(framesize, 1);
3759   oop_maps->add_gc_map(frame_complete, map);
3760 
3761   RuntimeStub* stub =
3762     RuntimeStub::new_runtime_stub(code.name(),
3763                                   &code,
3764                                   frame_complete,
3765                                   (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3766                                   oop_maps,
3767                                   false);
3768   return stub;
3769 }
3770 
3771 #endif // INCLUDE_JFR
3772 
3773 // Continuation point for throwing of implicit exceptions that are
3774 // not handled in the current activation. Fabricates an exception
3775 // oop and initiates normal exception dispatching in this
3776 // frame. Since we need to preserve callee-saved values (currently
3777 // only for C2, but done for C1 as well) we need a callee-saved oop
3778 // map and therefore have to make these stubs into RuntimeStubs
3779 // rather than BufferBlobs.  If the compiler needs all registers to
3780 // be preserved between the fault point and the exception handler
3781 // then it must assume responsibility for that in
3782 // AbstractCompiler::continuation_for_implicit_null_exception or
3783 // continuation_for_implicit_division_by_zero_exception. All other
3784 // implicit exceptions (e.g., NullPointerException or
3785 // AbstractMethodError on entry) are either at call sites or
3786 // otherwise assume that stack unwinding will be initiated, so
3787 // caller saved registers were assumed volatile in the compiler.
3788 address StubGenerator::generate_throw_exception(const char* name,
3789                                                 address runtime_entry,
3790                                                 Register arg1,
3791                                                 Register arg2) {
3792   // Information about frame layout at time of blocking runtime call.
3793   // Note that we only have to preserve callee-saved registers since
3794   // the compilers are responsible for supplying a continuation point
3795   // if they expect all registers to be preserved.
3796   enum layout {
3797     rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3798     rbp_off2,
3799     return_off,
3800     return_off2,
3801     framesize // inclusive of return address
3802   };
3803 
3804   int insts_size = 512;
3805   int locs_size  = 64;
3806 
3807   CodeBuffer code(name, insts_size, locs_size);
3808   OopMapSet* oop_maps  = new OopMapSet();
3809   MacroAssembler* _masm = new MacroAssembler(&code);
3810 
3811   address start = __ pc();
3812 
3813   // This is an inlined and slightly modified version of call_VM
3814   // which has the ability to fetch the return PC out of
3815   // thread-local storage and also sets up last_Java_sp slightly
3816   // differently than the real call_VM
3817 
3818   __ enter(); // required for proper stackwalking of RuntimeStub frame
3819 
3820   assert(is_even(framesize/2), "sp not 16-byte aligned");
3821 
3822   // return address and rbp are already in place
3823   __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3824 
3825   int frame_complete = __ pc() - start;
3826 
3827   // Set up last_Java_sp and last_Java_fp
3828   address the_pc = __ pc();
3829   __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3830   __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
3831 
3832   // Call runtime
3833   if (arg1 != noreg) {
3834     assert(arg2 != c_rarg1, "clobbered");
3835     __ movptr(c_rarg1, arg1);
3836   }
3837   if (arg2 != noreg) {
3838     __ movptr(c_rarg2, arg2);
3839   }
3840   __ movptr(c_rarg0, r15_thread);
3841   BLOCK_COMMENT("call runtime_entry");
3842   __ call(RuntimeAddress(runtime_entry));
3843 
3844   // Generate oop map
3845   OopMap* map = new OopMap(framesize, 0);
3846 
3847   oop_maps->add_gc_map(the_pc - start, map);
3848 
3849   __ reset_last_Java_frame(true);
3850 
3851   __ leave(); // required for proper stackwalking of RuntimeStub frame
3852 
3853   // check for pending exceptions
3854 #ifdef ASSERT
3855   Label L;
3856   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
3857   __ jcc(Assembler::notEqual, L);
3858   __ should_not_reach_here();
3859   __ bind(L);
3860 #endif // ASSERT
3861   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3862 
3863 
3864   // codeBlob framesize is in words (not VMRegImpl::slot_size)
3865   RuntimeStub* stub =
3866     RuntimeStub::new_runtime_stub(name,
3867                                   &code,
3868                                   frame_complete,
3869                                   (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3870                                   oop_maps, false);
3871   return stub->entry_point();
3872 }
3873 
3874 void StubGenerator::create_control_words() {
3875   // Round to nearest, 64-bit mode, exceptions masked
3876   StubRoutines::x86::_mxcsr_std = 0x1F80;
3877 }
3878 
3879 // Initialization
3880 void StubGenerator::generate_initial_stubs() {
3881   // Generates all stubs and initializes the entry points
3882 
3883   // This platform-specific settings are needed by generate_call_stub()
3884   create_control_words();
3885 
3886   // Initialize table for unsafe copy memeory check.
3887   if (UnsafeCopyMemory::_table == nullptr) {
3888     UnsafeCopyMemory::create_table(16);
3889   }
3890 
3891   // entry points that exist in all platforms Note: This is code
3892   // that could be shared among different platforms - however the
3893   // benefit seems to be smaller than the disadvantage of having a
3894   // much more complicated generator structure. See also comment in
3895   // stubRoutines.hpp.
3896 
3897   StubRoutines::_forward_exception_entry = generate_forward_exception();
3898 
3899   // Generate these first because they are called from other stubs
3900   if (InlineTypeReturnedAsFields) {
3901     StubRoutines::_load_inline_type_fields_in_regs =
3902       generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_inline_type_fields_in_regs),
3903                                  "load_inline_type_fields_in_regs", false);
3904     StubRoutines::_store_inline_type_fields_to_buf =
3905       generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_inline_type_fields_to_buf),
3906                                  "store_inline_type_fields_to_buf", true);
3907   }
3908 
3909   StubRoutines::_call_stub_entry =
3910     generate_call_stub(StubRoutines::_call_stub_return_address);
3911 
3912   // is referenced by megamorphic call
3913   StubRoutines::_catch_exception_entry = generate_catch_exception();
3914 
3915   // atomic calls
3916   StubRoutines::_fence_entry                = generate_orderaccess_fence();
3917 
3918   // platform dependent
3919   StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
3920 
3921   StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
3922 
3923   StubRoutines::x86::_f2i_fixup             = generate_f2i_fixup();
3924   StubRoutines::x86::_f2l_fixup             = generate_f2l_fixup();
3925   StubRoutines::x86::_d2i_fixup             = generate_d2i_fixup();
3926   StubRoutines::x86::_d2l_fixup             = generate_d2l_fixup();
3927 
3928   StubRoutines::x86::_float_sign_mask       = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
3929   StubRoutines::x86::_float_sign_flip       = generate_fp_mask("float_sign_flip",  0x8000000080000000);
3930   StubRoutines::x86::_double_sign_mask      = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3931   StubRoutines::x86::_double_sign_flip      = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3932 
3933   // Build this early so it's available for the interpreter.
3934   StubRoutines::_throw_StackOverflowError_entry =
3935     generate_throw_exception("StackOverflowError throw_exception",
3936                              CAST_FROM_FN_PTR(address,
3937                                               SharedRuntime::
3938                                               throw_StackOverflowError));
3939   StubRoutines::_throw_delayed_StackOverflowError_entry =
3940     generate_throw_exception("delayed StackOverflowError throw_exception",
3941                              CAST_FROM_FN_PTR(address,
3942                                               SharedRuntime::
3943                                               throw_delayed_StackOverflowError));
3944   if (UseCRC32Intrinsics) {
3945     // set table address before stub generation which use it
3946     StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3947     StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3948   }
3949 
3950   if (UseCRC32CIntrinsics) {
3951     bool supports_clmul = VM_Version::supports_clmul();
3952     StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3953     StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3954     StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3955   }
3956 
3957   if (VM_Version::supports_float16()) {
3958     // For results consistency both intrinsics should be enabled.
3959     // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
3960     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
3961         vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
3962       StubRoutines::_hf2f = generate_float16ToFloat();
3963       StubRoutines::_f2hf = generate_floatToFloat16();
3964     }
3965   }
3966 
3967   generate_libm_stubs();
3968 }
3969 
3970 // Call here from the interpreter or compiled code to either load
3971 // multiple returned values from the inline type instance being
3972 // returned to registers or to store returned values to a newly
3973 // allocated inline type instance.
3974 // Register is a class, but it would be assigned numerical value.
3975 // "0" is assigned for xmm0. Thus we need to ignore -Wnonnull.
3976 PRAGMA_DIAG_PUSH
3977 PRAGMA_NONNULL_IGNORED
3978 address StubGenerator::generate_return_value_stub(address destination, const char* name, bool has_res) {
3979   // We need to save all registers the calling convention may use so
3980   // the runtime calls read or update those registers. This needs to
3981   // be in sync with SharedRuntime::java_return_convention().
3982   enum layout {
3983     pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
3984     rax_off, rax_off_2,
3985     j_rarg5_off, j_rarg5_2,
3986     j_rarg4_off, j_rarg4_2,
3987     j_rarg3_off, j_rarg3_2,
3988     j_rarg2_off, j_rarg2_2,
3989     j_rarg1_off, j_rarg1_2,
3990     j_rarg0_off, j_rarg0_2,
3991     j_farg0_off, j_farg0_2,
3992     j_farg1_off, j_farg1_2,
3993     j_farg2_off, j_farg2_2,
3994     j_farg3_off, j_farg3_2,
3995     j_farg4_off, j_farg4_2,
3996     j_farg5_off, j_farg5_2,
3997     j_farg6_off, j_farg6_2,
3998     j_farg7_off, j_farg7_2,
3999     rbp_off, rbp_off_2,
4000     return_off, return_off_2,
4001 
4002     framesize
4003   };
4004 
4005   CodeBuffer buffer(name, 1000, 512);
4006   MacroAssembler* _masm = new MacroAssembler(&buffer);
4007 
4008   int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
4009   assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
4010   int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
4011   int frame_size_in_words = frame_size_in_bytes / wordSize;
4012 
4013   OopMapSet *oop_maps = new OopMapSet();
4014   OopMap* map = new OopMap(frame_size_in_slots, 0);
4015 
4016   map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
4017   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
4018   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
4019   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
4020   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
4021   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
4022   map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
4023   map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
4024   map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
4025   map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
4026   map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
4027   map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
4028   map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
4029   map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
4030   map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
4031 
4032   int start = __ offset();
4033 
4034   __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
4035 
4036   __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
4037   __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
4038   __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
4039   __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
4040   __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
4041   __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
4042   __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
4043   __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
4044   __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
4045 
4046   __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
4047   __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
4048   __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
4049   __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
4050   __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
4051   __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
4052   __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
4053 
4054   int frame_complete = __ offset();
4055 
4056   __ set_last_Java_frame(noreg, noreg, NULL, rscratch1);
4057 
4058   __ mov(c_rarg0, r15_thread);
4059   __ mov(c_rarg1, rax);
4060 
4061   __ call(RuntimeAddress(destination));
4062 
4063   // Set an oopmap for the call site.
4064 
4065   oop_maps->add_gc_map( __ offset() - start, map);
4066 
4067   // clear last_Java_sp
4068   __ reset_last_Java_frame(false);
4069 
4070   __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
4071   __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
4072   __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
4073   __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
4074   __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
4075   __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
4076   __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
4077   __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
4078   __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
4079 
4080   __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
4081   __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
4082   __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
4083   __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
4084   __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
4085   __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
4086   __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
4087 
4088   __ addptr(rsp, frame_size_in_bytes-8);
4089 
4090   // check for pending exceptions
4091   Label pending;
4092   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
4093   __ jcc(Assembler::notEqual, pending);
4094 
4095   if (has_res) {
4096     __ get_vm_result(rax, r15_thread);
4097   }
4098 
4099   __ ret(0);
4100 
4101   __ bind(pending);
4102 
4103   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
4104   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
4105 
4106   // -------------
4107   // make sure all code is generated
4108   _masm->flush();
4109 
4110   RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false);
4111   return stub->entry_point();
4112 }
4113 
4114 void StubGenerator::generate_continuation_stubs() {
4115   // Continuation stubs:
4116   StubRoutines::_cont_thaw          = generate_cont_thaw();
4117   StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
4118   StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
4119 
4120   JFR_ONLY(StubRoutines::_jfr_write_checkpoint_stub = generate_jfr_write_checkpoint();)
4121   JFR_ONLY(StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();)
4122 }
4123 
4124 void StubGenerator::generate_final_stubs() {
4125   // Generates the rest of stubs and initializes the entry points
4126 
4127   // These entry points require SharedInfo::stack0 to be set up in
4128   // non-core builds and need to be relocatable, so they each
4129   // fabricate a RuntimeStub internally.
4130   StubRoutines::_throw_AbstractMethodError_entry =
4131     generate_throw_exception("AbstractMethodError throw_exception",
4132                              CAST_FROM_FN_PTR(address,
4133                                               SharedRuntime::
4134                                               throw_AbstractMethodError));
4135 
4136   StubRoutines::_throw_IncompatibleClassChangeError_entry =
4137     generate_throw_exception("IncompatibleClassChangeError throw_exception",
4138                              CAST_FROM_FN_PTR(address,
4139                                               SharedRuntime::
4140                                               throw_IncompatibleClassChangeError));
4141 
4142   StubRoutines::_throw_NullPointerException_at_call_entry =
4143     generate_throw_exception("NullPointerException at call throw_exception",
4144                              CAST_FROM_FN_PTR(address,
4145                                               SharedRuntime::
4146                                               throw_NullPointerException_at_call));
4147 
4148   // support for verify_oop (must happen after universe_init)
4149   if (VerifyOops) {
4150     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4151   }
4152 
4153   // data cache line writeback
4154   StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4155   StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4156 
4157   // arraycopy stubs used by compilers
4158   generate_arraycopy_stubs();
4159 
4160   BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4161   if (bs_nm != nullptr) {
4162     StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
4163   }
4164 
4165   if (UseVectorizedMismatchIntrinsic) {
4166     StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4167   }
4168 }
4169 
4170 void StubGenerator::generate_compiler_stubs() {
4171 #if COMPILER2_OR_JVMCI
4172 
4173   // Entry points that are C2 compiler specific.
4174 
4175   StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4176   StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
4177   StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4178   StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
4179   StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFFFFFFFFFF);
4180   StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask("vector_int_mask_cmp_bits", 0x0000000100000001);
4181   StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
4182   StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
4183   StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff000000ff);
4184   StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff0000ffff);
4185   StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
4186                                                                       0xFFFFFFFF, 0, 0, 0);
4187   StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
4188                                                                       0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4189   StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x0302010003020100);
4190   StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask("vector_byte_shuffle_mask");
4191   StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x0100010001000100);
4192   StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask("vector_long_shuffle_mask", 0x0000000100000000);
4193   StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
4194   StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
4195   StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut("count_leading_zeros_lut");
4196   StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut("reverse_bit_lut");
4197   StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long("perm_mask_long");
4198   StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int("perm_mask_int");
4199   StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short("perm_mask_short");
4200 
4201   if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4202     // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4203     StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut("popcount_lut");
4204   }
4205 
4206   generate_aes_stubs();
4207 
4208   generate_ghash_stubs();
4209 
4210   generate_chacha_stubs();
4211 
4212   if (UseAdler32Intrinsics) {
4213      StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4214   }
4215 
4216   if (UsePoly1305Intrinsics) {
4217     StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4218   }
4219 
4220   if (UseMD5Intrinsics) {
4221     StubRoutines::_md5_implCompress = generate_md5_implCompress(false, "md5_implCompress");
4222     StubRoutines::_md5_implCompressMB = generate_md5_implCompress(true, "md5_implCompressMB");
4223   }
4224 
4225   if (UseSHA1Intrinsics) {
4226     StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4227     StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4228     StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
4229     StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
4230   }
4231 
4232   if (UseSHA256Intrinsics) {
4233     StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4234     char* dst = (char*)StubRoutines::x86::_k256_W;
4235     char* src = (char*)StubRoutines::x86::_k256;
4236     for (int ii = 0; ii < 16; ++ii) {
4237       memcpy(dst + 32 * ii,      src + 16 * ii, 16);
4238       memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4239     }
4240     StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4241     StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4242     StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4243     StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4244   }
4245 
4246   if (UseSHA512Intrinsics) {
4247     StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4248     StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4249     StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
4250     StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
4251   }
4252 
4253   if (UseBASE64Intrinsics) {
4254     if(VM_Version::supports_avx2()) {
4255       StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4256       StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4257       StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4258       StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4259       StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4260     }
4261     StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4262     if (VM_Version::supports_avx512_vbmi()) {
4263       StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4264       StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4265       StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4266       StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4267       StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4268       StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4269       StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4270       StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4271       StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4272     }
4273     StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4274     StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4275     StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4276   }
4277 
4278 #ifdef COMPILER2
4279   if (UseMultiplyToLenIntrinsic) {
4280     StubRoutines::_multiplyToLen = generate_multiplyToLen();
4281   }
4282   if (UseSquareToLenIntrinsic) {
4283     StubRoutines::_squareToLen = generate_squareToLen();
4284   }
4285   if (UseMulAddIntrinsic) {
4286     StubRoutines::_mulAdd = generate_mulAdd();
4287   }
4288   if (VM_Version::supports_avx512_vbmi2()) {
4289     StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4290     StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4291   }
4292   if (UseMontgomeryMultiplyIntrinsic) {
4293     StubRoutines::_montgomeryMultiply
4294       = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4295   }
4296   if (UseMontgomerySquareIntrinsic) {
4297     StubRoutines::_montgomerySquare
4298       = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4299   }
4300 
4301   // Get svml stub routine addresses
4302   void *libjsvml = nullptr;
4303   char ebuf[1024];
4304   char dll_name[JVM_MAXPATHLEN];
4305   if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4306     libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4307   }
4308   if (libjsvml != nullptr) {
4309     // SVML method naming convention
4310     //   All the methods are named as __jsvml_op<T><N>_ha_<VV>
4311     //   Where:
4312     //      ha stands for high accuracy
4313     //      <T> is optional to indicate float/double
4314     //              Set to f for vector float operation
4315     //              Omitted for vector double operation
4316     //      <N> is the number of elements in the vector
4317     //              1, 2, 4, 8, 16
4318     //              e.g. 128 bit float vector has 4 float elements
4319     //      <VV> indicates the avx/sse level:
4320     //              z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4321     //      e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4322     //           __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4323 
4324     log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4325     if (UseAVX > 2) {
4326       for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4327         int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4328         if ((!VM_Version::supports_avx512dq()) &&
4329             (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4330           continue;
4331         }
4332         snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::svmlname[op]);
4333         StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4334 
4335         snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::svmlname[op]);
4336         StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4337       }
4338     }
4339     const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4340     for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4341       int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4342       if (vop == VectorSupport::VECTOR_OP_POW) {
4343         continue;
4344       }
4345       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4346       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4347 
4348       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4349       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4350 
4351       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4352       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4353 
4354       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4355       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4356 
4357       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4358       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4359 
4360       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4361       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4362     }
4363   }
4364 #endif // COMPILER2
4365 #endif // COMPILER2_OR_JVMCI
4366 }
4367 
4368 StubGenerator::StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
4369     DEBUG_ONLY( _regs_in_thread = false; )
4370     switch(kind) {
4371     case Initial_stubs:
4372       generate_initial_stubs();
4373       break;
4374      case Continuation_stubs:
4375       generate_continuation_stubs();
4376       break;
4377     case Compiler_stubs:
4378       generate_compiler_stubs();
4379       break;
4380     case Final_stubs:
4381       generate_final_stubs();
4382       break;
4383     default:
4384       fatal("unexpected stubs kind: %d", kind);
4385       break;
4386     };
4387 }
4388 
4389 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
4390   StubGenerator g(code, kind);
4391 }
4392 
4393 #undef __