1 /*
   2  * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "classfile/vmIntrinsics.hpp"
  28 #include "compiler/oopMap.hpp"
  29 #include "gc/shared/barrierSet.hpp"
  30 #include "gc/shared/barrierSetAssembler.hpp"
  31 #include "gc/shared/barrierSetNMethod.hpp"
  32 #include "gc/shared/gc_globals.hpp"
  33 #include "memory/universe.hpp"
  34 #include "prims/jvmtiExport.hpp"
  35 #include "prims/upcallLinker.hpp"
  36 #include "runtime/arguments.hpp"
  37 #include "runtime/javaThread.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "stubGenerator_x86_64.hpp"
  41 #ifdef COMPILER2
  42 #include "opto/runtime.hpp"
  43 #include "opto/c2_globals.hpp"
  44 #endif
  45 #if INCLUDE_JVMCI
  46 #include "jvmci/jvmci_globals.hpp"
  47 #endif
  48 #if INCLUDE_JFR
  49 #include "jfr/support/jfrIntrinsics.hpp"
  50 #endif
  51 
  52 // For a more detailed description of the stub routine structure
  53 // see the comment in stubRoutines.hpp
  54 
  55 #define __ _masm->
  56 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
  57 
  58 #ifdef PRODUCT
  59 #define BLOCK_COMMENT(str) /* nothing */
  60 #else
  61 #define BLOCK_COMMENT(str) __ block_comment(str)
  62 #endif // PRODUCT
  63 
  64 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  65 
  66 //
  67 // Linux Arguments:
  68 //    c_rarg0:   call wrapper address                   address
  69 //    c_rarg1:   result                                 address
  70 //    c_rarg2:   result type                            BasicType
  71 //    c_rarg3:   method                                 Method*
  72 //    c_rarg4:   (interpreter) entry point              address
  73 //    c_rarg5:   parameters                             intptr_t*
  74 //    16(rbp): parameter size (in words)              int
  75 //    24(rbp): thread                                 Thread*
  76 //
  77 //     [ return_from_Java     ] <--- rsp
  78 //     [ argument word n      ]
  79 //      ...
  80 // -12 [ argument word 1      ]
  81 // -11 [ saved r15            ] <--- rsp_after_call
  82 // -10 [ saved r14            ]
  83 //  -9 [ saved r13            ]
  84 //  -8 [ saved r12            ]
  85 //  -7 [ saved rbx            ]
  86 //  -6 [ call wrapper         ]
  87 //  -5 [ result               ]
  88 //  -4 [ result type          ]
  89 //  -3 [ method               ]
  90 //  -2 [ entry point          ]
  91 //  -1 [ parameters           ]
  92 //   0 [ saved rbp            ] <--- rbp
  93 //   1 [ return address       ]
  94 //   2 [ parameter size       ]
  95 //   3 [ thread               ]
  96 //
  97 // Windows Arguments:
  98 //    c_rarg0:   call wrapper address                   address
  99 //    c_rarg1:   result                                 address
 100 //    c_rarg2:   result type                            BasicType
 101 //    c_rarg3:   method                                 Method*
 102 //    48(rbp): (interpreter) entry point              address
 103 //    56(rbp): parameters                             intptr_t*
 104 //    64(rbp): parameter size (in words)              int
 105 //    72(rbp): thread                                 Thread*
 106 //
 107 //     [ return_from_Java     ] <--- rsp
 108 //     [ argument word n      ]
 109 //      ...
 110 // -28 [ argument word 1      ]
 111 // -27 [ saved xmm15          ] <--- rsp after_call
 112 //     [ saved xmm7-xmm14     ]
 113 //  -9 [ saved xmm6           ] (each xmm register takes 2 slots)
 114 //  -7 [ saved r15            ]
 115 //  -6 [ saved r14            ]
 116 //  -5 [ saved r13            ]
 117 //  -4 [ saved r12            ]
 118 //  -3 [ saved rdi            ]
 119 //  -2 [ saved rsi            ]
 120 //  -1 [ saved rbx            ]
 121 //   0 [ saved rbp            ] <--- rbp
 122 //   1 [ return address       ]
 123 //   2 [ call wrapper         ]
 124 //   3 [ result               ]
 125 //   4 [ result type          ]
 126 //   5 [ method               ]
 127 //   6 [ entry point          ]
 128 //   7 [ parameters           ]
 129 //   8 [ parameter size       ]
 130 //   9 [ thread               ]
 131 //
 132 //    Windows reserves the callers stack space for arguments 1-4.
 133 //    We spill c_rarg0-c_rarg3 to this space.
 134 
 135 // Call stub stack layout word offsets from rbp
 136 #ifdef _WIN64
 137 enum call_stub_layout {
 138   xmm_save_first     = 6,  // save from xmm6
 139   xmm_save_last      = 15, // to xmm15
 140   xmm_save_base      = -9,
 141   rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
 142   r15_off            = -7,
 143   r14_off            = -6,
 144   r13_off            = -5,
 145   r12_off            = -4,
 146   rdi_off            = -3,
 147   rsi_off            = -2,
 148   rbx_off            = -1,
 149   rbp_off            =  0,
 150   retaddr_off        =  1,
 151   call_wrapper_off   =  2,
 152   result_off         =  3,
 153   result_type_off    =  4,
 154   method_off         =  5,
 155   entry_point_off    =  6,
 156   parameters_off     =  7,
 157   parameter_size_off =  8,
 158   thread_off         =  9
 159 };
 160 
 161 static Address xmm_save(int reg) {
 162   assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
 163   return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
 164 }
 165 #else // !_WIN64
 166 enum call_stub_layout {
 167   rsp_after_call_off = -12,
 168   mxcsr_off          = rsp_after_call_off,
 169   r15_off            = -11,
 170   r14_off            = -10,
 171   r13_off            = -9,
 172   r12_off            = -8,
 173   rbx_off            = -7,
 174   call_wrapper_off   = -6,
 175   result_off         = -5,
 176   result_type_off    = -4,
 177   method_off         = -3,
 178   entry_point_off    = -2,
 179   parameters_off     = -1,
 180   rbp_off            =  0,
 181   retaddr_off        =  1,
 182   parameter_size_off =  2,
 183   thread_off         =  3
 184 };
 185 #endif // _WIN64
 186 
 187 address StubGenerator::generate_call_stub(address& return_address) {
 188 
 189   assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
 190          (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
 191          "adjust this code");
 192   StubCodeMark mark(this, "StubRoutines", "call_stub");
 193   address start = __ pc();
 194 
 195   // same as in generate_catch_exception()!
 196   const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 197 
 198   const Address call_wrapper  (rbp, call_wrapper_off   * wordSize);
 199   const Address result        (rbp, result_off         * wordSize);
 200   const Address result_type   (rbp, result_type_off    * wordSize);
 201   const Address method        (rbp, method_off         * wordSize);
 202   const Address entry_point   (rbp, entry_point_off    * wordSize);
 203   const Address parameters    (rbp, parameters_off     * wordSize);
 204   const Address parameter_size(rbp, parameter_size_off * wordSize);
 205 
 206   // same as in generate_catch_exception()!
 207   const Address thread        (rbp, thread_off         * wordSize);
 208 
 209   const Address r15_save(rbp, r15_off * wordSize);
 210   const Address r14_save(rbp, r14_off * wordSize);
 211   const Address r13_save(rbp, r13_off * wordSize);
 212   const Address r12_save(rbp, r12_off * wordSize);
 213   const Address rbx_save(rbp, rbx_off * wordSize);
 214 
 215   // stub code
 216   __ enter();
 217   __ subptr(rsp, -rsp_after_call_off * wordSize);
 218 
 219   // save register parameters
 220 #ifndef _WIN64
 221   __ movptr(parameters,   c_rarg5); // parameters
 222   __ movptr(entry_point,  c_rarg4); // entry_point
 223 #endif
 224 
 225   __ movptr(method,       c_rarg3); // method
 226   __ movl(result_type,  c_rarg2);   // result type
 227   __ movptr(result,       c_rarg1); // result
 228   __ movptr(call_wrapper, c_rarg0); // call wrapper
 229 
 230   // save regs belonging to calling function
 231   __ movptr(rbx_save, rbx);
 232   __ movptr(r12_save, r12);
 233   __ movptr(r13_save, r13);
 234   __ movptr(r14_save, r14);
 235   __ movptr(r15_save, r15);
 236 
 237 #ifdef _WIN64
 238   int last_reg = 15;
 239   for (int i = xmm_save_first; i <= last_reg; i++) {
 240     __ movdqu(xmm_save(i), as_XMMRegister(i));
 241   }
 242 
 243   const Address rdi_save(rbp, rdi_off * wordSize);
 244   const Address rsi_save(rbp, rsi_off * wordSize);
 245 
 246   __ movptr(rsi_save, rsi);
 247   __ movptr(rdi_save, rdi);
 248 #else
 249   const Address mxcsr_save(rbp, mxcsr_off * wordSize);
 250   {
 251     Label skip_ldmx;
 252     __ stmxcsr(mxcsr_save);
 253     __ movl(rax, mxcsr_save);
 254     __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
 255     ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 256     __ cmp32(rax, mxcsr_std, rscratch1);
 257     __ jcc(Assembler::equal, skip_ldmx);
 258     __ ldmxcsr(mxcsr_std, rscratch1);
 259     __ bind(skip_ldmx);
 260   }
 261 #endif
 262 
 263   // Load up thread register
 264   __ movptr(r15_thread, thread);
 265   __ reinit_heapbase();
 266 
 267 #ifdef ASSERT
 268   // make sure we have no pending exceptions
 269   {
 270     Label L;
 271     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 272     __ jcc(Assembler::equal, L);
 273     __ stop("StubRoutines::call_stub: entered with pending exception");
 274     __ bind(L);
 275   }
 276 #endif
 277 
 278   // pass parameters if any
 279   BLOCK_COMMENT("pass parameters if any");
 280   Label parameters_done;
 281   __ movl(c_rarg3, parameter_size);
 282   __ testl(c_rarg3, c_rarg3);
 283   __ jcc(Assembler::zero, parameters_done);
 284 
 285   Label loop;
 286   __ movptr(c_rarg2, parameters);       // parameter pointer
 287   __ movl(c_rarg1, c_rarg3);            // parameter counter is in c_rarg1
 288   __ BIND(loop);
 289   __ movptr(rax, Address(c_rarg2, 0));// get parameter
 290   __ addptr(c_rarg2, wordSize);       // advance to next parameter
 291   __ decrementl(c_rarg1);             // decrement counter
 292   __ push(rax);                       // pass parameter
 293   __ jcc(Assembler::notZero, loop);
 294 
 295   // call Java function
 296   __ BIND(parameters_done);
 297   __ movptr(rbx, method);             // get Method*
 298   __ movptr(c_rarg1, entry_point);    // get entry_point
 299   __ mov(r13, rsp);                   // set sender sp
 300   BLOCK_COMMENT("call Java function");
 301   __ call(c_rarg1);
 302 
 303   BLOCK_COMMENT("call_stub_return_address:");
 304   return_address = __ pc();
 305 
 306   // store result depending on type (everything that is not
 307   // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
 308   __ movptr(c_rarg0, result);
 309   Label is_long, is_float, is_double, exit;
 310   __ movl(c_rarg1, result_type);
 311   __ cmpl(c_rarg1, T_OBJECT);
 312   __ jcc(Assembler::equal, is_long);
 313   __ cmpl(c_rarg1, T_LONG);
 314   __ jcc(Assembler::equal, is_long);
 315   __ cmpl(c_rarg1, T_FLOAT);
 316   __ jcc(Assembler::equal, is_float);
 317   __ cmpl(c_rarg1, T_DOUBLE);
 318   __ jcc(Assembler::equal, is_double);
 319 #ifdef ASSERT
 320   // make sure the type is INT
 321   {
 322     Label L;
 323     __ cmpl(c_rarg1, T_INT);
 324     __ jcc(Assembler::equal, L);
 325     __ stop("StubRoutines::call_stub: unexpected result type");
 326     __ bind(L);
 327   }
 328 #endif
 329 
 330   // handle T_INT case
 331   __ movl(Address(c_rarg0, 0), rax);
 332 
 333   __ BIND(exit);
 334 
 335   // pop parameters
 336   __ lea(rsp, rsp_after_call);
 337 
 338 #ifdef ASSERT
 339   // verify that threads correspond
 340   {
 341    Label L1, L2, L3;
 342     __ cmpptr(r15_thread, thread);
 343     __ jcc(Assembler::equal, L1);
 344     __ stop("StubRoutines::call_stub: r15_thread is corrupted");
 345     __ bind(L1);
 346     __ get_thread(rbx);
 347     __ cmpptr(r15_thread, thread);
 348     __ jcc(Assembler::equal, L2);
 349     __ stop("StubRoutines::call_stub: r15_thread is modified by call");
 350     __ bind(L2);
 351     __ cmpptr(r15_thread, rbx);
 352     __ jcc(Assembler::equal, L3);
 353     __ stop("StubRoutines::call_stub: threads must correspond");
 354     __ bind(L3);
 355   }
 356 #endif
 357 
 358   __ pop_cont_fastpath();
 359 
 360   // restore regs belonging to calling function
 361 #ifdef _WIN64
 362   // emit the restores for xmm regs
 363   for (int i = xmm_save_first; i <= last_reg; i++) {
 364     __ movdqu(as_XMMRegister(i), xmm_save(i));
 365   }
 366 #endif
 367   __ movptr(r15, r15_save);
 368   __ movptr(r14, r14_save);
 369   __ movptr(r13, r13_save);
 370   __ movptr(r12, r12_save);
 371   __ movptr(rbx, rbx_save);
 372 
 373 #ifdef _WIN64
 374   __ movptr(rdi, rdi_save);
 375   __ movptr(rsi, rsi_save);
 376 #else
 377   __ ldmxcsr(mxcsr_save);
 378 #endif
 379 
 380   // restore rsp
 381   __ addptr(rsp, -rsp_after_call_off * wordSize);
 382 
 383   // return
 384   __ vzeroupper();
 385   __ pop(rbp);
 386   __ ret(0);
 387 
 388   // handle return types different from T_INT
 389   __ BIND(is_long);
 390   __ movq(Address(c_rarg0, 0), rax);
 391   __ jmp(exit);
 392 
 393   __ BIND(is_float);
 394   __ movflt(Address(c_rarg0, 0), xmm0);
 395   __ jmp(exit);
 396 
 397   __ BIND(is_double);
 398   __ movdbl(Address(c_rarg0, 0), xmm0);
 399   __ jmp(exit);
 400 
 401   return start;
 402 }
 403 
 404 // Return point for a Java call if there's an exception thrown in
 405 // Java code.  The exception is caught and transformed into a
 406 // pending exception stored in JavaThread that can be tested from
 407 // within the VM.
 408 //
 409 // Note: Usually the parameters are removed by the callee. In case
 410 // of an exception crossing an activation frame boundary, that is
 411 // not the case if the callee is compiled code => need to setup the
 412 // rsp.
 413 //
 414 // rax: exception oop
 415 
 416 address StubGenerator::generate_catch_exception() {
 417   StubCodeMark mark(this, "StubRoutines", "catch_exception");
 418   address start = __ pc();
 419 
 420   // same as in generate_call_stub():
 421   const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
 422   const Address thread        (rbp, thread_off         * wordSize);
 423 
 424 #ifdef ASSERT
 425   // verify that threads correspond
 426   {
 427     Label L1, L2, L3;
 428     __ cmpptr(r15_thread, thread);
 429     __ jcc(Assembler::equal, L1);
 430     __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
 431     __ bind(L1);
 432     __ get_thread(rbx);
 433     __ cmpptr(r15_thread, thread);
 434     __ jcc(Assembler::equal, L2);
 435     __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
 436     __ bind(L2);
 437     __ cmpptr(r15_thread, rbx);
 438     __ jcc(Assembler::equal, L3);
 439     __ stop("StubRoutines::catch_exception: threads must correspond");
 440     __ bind(L3);
 441   }
 442 #endif
 443 
 444   // set pending exception
 445   __ verify_oop(rax);
 446 
 447   __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
 448   __ lea(rscratch1, ExternalAddress((address)__FILE__));
 449   __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
 450   __ movl(Address(r15_thread, Thread::exception_line_offset()), (int)  __LINE__);
 451 
 452   // complete return to VM
 453   assert(StubRoutines::_call_stub_return_address != nullptr,
 454          "_call_stub_return_address must have been generated before");
 455   __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
 456 
 457   return start;
 458 }
 459 
 460 // Continuation point for runtime calls returning with a pending
 461 // exception.  The pending exception check happened in the runtime
 462 // or native call stub.  The pending exception in Thread is
 463 // converted into a Java-level exception.
 464 //
 465 // Contract with Java-level exception handlers:
 466 // rax: exception
 467 // rdx: throwing pc
 468 //
 469 // NOTE: At entry of this stub, exception-pc must be on stack !!
 470 
 471 address StubGenerator::generate_forward_exception() {
 472   StubCodeMark mark(this, "StubRoutines", "forward exception");
 473   address start = __ pc();
 474 
 475   // Upon entry, the sp points to the return address returning into
 476   // Java (interpreted or compiled) code; i.e., the return address
 477   // becomes the throwing pc.
 478   //
 479   // Arguments pushed before the runtime call are still on the stack
 480   // but the exception handler will reset the stack pointer ->
 481   // ignore them.  A potential result in registers can be ignored as
 482   // well.
 483 
 484 #ifdef ASSERT
 485   // make sure this code is only executed if there is a pending exception
 486   {
 487     Label L;
 488     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 489     __ jcc(Assembler::notEqual, L);
 490     __ stop("StubRoutines::forward exception: no pending exception (1)");
 491     __ bind(L);
 492   }
 493 #endif
 494 
 495   // compute exception handler into rbx
 496   __ movptr(c_rarg0, Address(rsp, 0));
 497   BLOCK_COMMENT("call exception_handler_for_return_address");
 498   __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 499                        SharedRuntime::exception_handler_for_return_address),
 500                   r15_thread, c_rarg0);
 501   __ mov(rbx, rax);
 502 
 503   // setup rax & rdx, remove return address & clear pending exception
 504   __ pop(rdx);
 505   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 506   __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 507 
 508 #ifdef ASSERT
 509   // make sure exception is set
 510   {
 511     Label L;
 512     __ testptr(rax, rax);
 513     __ jcc(Assembler::notEqual, L);
 514     __ stop("StubRoutines::forward exception: no pending exception (2)");
 515     __ bind(L);
 516   }
 517 #endif
 518 
 519   // continue at exception handler (return address removed)
 520   // rax: exception
 521   // rbx: exception handler
 522   // rdx: throwing pc
 523   __ verify_oop(rax);
 524   __ jmp(rbx);
 525 
 526   return start;
 527 }
 528 
 529 // Support for intptr_t OrderAccess::fence()
 530 //
 531 // Arguments :
 532 //
 533 // Result:
 534 address StubGenerator::generate_orderaccess_fence() {
 535   StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
 536   address start = __ pc();
 537 
 538   __ membar(Assembler::StoreLoad);
 539   __ ret(0);
 540 
 541   return start;
 542 }
 543 
 544 
 545 // Support for intptr_t get_previous_sp()
 546 //
 547 // This routine is used to find the previous stack pointer for the
 548 // caller.
 549 address StubGenerator::generate_get_previous_sp() {
 550   StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
 551   address start = __ pc();
 552 
 553   __ movptr(rax, rsp);
 554   __ addptr(rax, 8); // return address is at the top of the stack.
 555   __ ret(0);
 556 
 557   return start;
 558 }
 559 
 560 //----------------------------------------------------------------------------------------------------
 561 // Support for void verify_mxcsr()
 562 //
 563 // This routine is used with -Xcheck:jni to verify that native
 564 // JNI code does not return to Java code without restoring the
 565 // MXCSR register to our expected state.
 566 
 567 address StubGenerator::generate_verify_mxcsr() {
 568   StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
 569   address start = __ pc();
 570 
 571   const Address mxcsr_save(rsp, 0);
 572 
 573   if (CheckJNICalls) {
 574     Label ok_ret;
 575     ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 576     __ push(rax);
 577     __ subptr(rsp, wordSize);      // allocate a temp location
 578     __ stmxcsr(mxcsr_save);
 579     __ movl(rax, mxcsr_save);
 580     __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
 581     __ cmp32(rax, mxcsr_std, rscratch1);
 582     __ jcc(Assembler::equal, ok_ret);
 583 
 584     __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
 585 
 586     __ ldmxcsr(mxcsr_std, rscratch1);
 587 
 588     __ bind(ok_ret);
 589     __ addptr(rsp, wordSize);
 590     __ pop(rax);
 591   }
 592 
 593   __ ret(0);
 594 
 595   return start;
 596 }
 597 
 598 address StubGenerator::generate_f2i_fixup() {
 599   StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
 600   Address inout(rsp, 5 * wordSize); // return address + 4 saves
 601 
 602   address start = __ pc();
 603 
 604   Label L;
 605 
 606   __ push(rax);
 607   __ push(c_rarg3);
 608   __ push(c_rarg2);
 609   __ push(c_rarg1);
 610 
 611   __ movl(rax, 0x7f800000);
 612   __ xorl(c_rarg3, c_rarg3);
 613   __ movl(c_rarg2, inout);
 614   __ movl(c_rarg1, c_rarg2);
 615   __ andl(c_rarg1, 0x7fffffff);
 616   __ cmpl(rax, c_rarg1); // NaN? -> 0
 617   __ jcc(Assembler::negative, L);
 618   __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
 619   __ movl(c_rarg3, 0x80000000);
 620   __ movl(rax, 0x7fffffff);
 621   __ cmovl(Assembler::positive, c_rarg3, rax);
 622 
 623   __ bind(L);
 624   __ movptr(inout, c_rarg3);
 625 
 626   __ pop(c_rarg1);
 627   __ pop(c_rarg2);
 628   __ pop(c_rarg3);
 629   __ pop(rax);
 630 
 631   __ ret(0);
 632 
 633   return start;
 634 }
 635 
 636 address StubGenerator::generate_f2l_fixup() {
 637   StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
 638   Address inout(rsp, 5 * wordSize); // return address + 4 saves
 639   address start = __ pc();
 640 
 641   Label L;
 642 
 643   __ push(rax);
 644   __ push(c_rarg3);
 645   __ push(c_rarg2);
 646   __ push(c_rarg1);
 647 
 648   __ movl(rax, 0x7f800000);
 649   __ xorl(c_rarg3, c_rarg3);
 650   __ movl(c_rarg2, inout);
 651   __ movl(c_rarg1, c_rarg2);
 652   __ andl(c_rarg1, 0x7fffffff);
 653   __ cmpl(rax, c_rarg1); // NaN? -> 0
 654   __ jcc(Assembler::negative, L);
 655   __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
 656   __ mov64(c_rarg3, 0x8000000000000000);
 657   __ mov64(rax, 0x7fffffffffffffff);
 658   __ cmov(Assembler::positive, c_rarg3, rax);
 659 
 660   __ bind(L);
 661   __ movptr(inout, c_rarg3);
 662 
 663   __ pop(c_rarg1);
 664   __ pop(c_rarg2);
 665   __ pop(c_rarg3);
 666   __ pop(rax);
 667 
 668   __ ret(0);
 669 
 670   return start;
 671 }
 672 
 673 address StubGenerator::generate_d2i_fixup() {
 674   StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
 675   Address inout(rsp, 6 * wordSize); // return address + 5 saves
 676 
 677   address start = __ pc();
 678 
 679   Label L;
 680 
 681   __ push(rax);
 682   __ push(c_rarg3);
 683   __ push(c_rarg2);
 684   __ push(c_rarg1);
 685   __ push(c_rarg0);
 686 
 687   __ movl(rax, 0x7ff00000);
 688   __ movq(c_rarg2, inout);
 689   __ movl(c_rarg3, c_rarg2);
 690   __ mov(c_rarg1, c_rarg2);
 691   __ mov(c_rarg0, c_rarg2);
 692   __ negl(c_rarg3);
 693   __ shrptr(c_rarg1, 0x20);
 694   __ orl(c_rarg3, c_rarg2);
 695   __ andl(c_rarg1, 0x7fffffff);
 696   __ xorl(c_rarg2, c_rarg2);
 697   __ shrl(c_rarg3, 0x1f);
 698   __ orl(c_rarg1, c_rarg3);
 699   __ cmpl(rax, c_rarg1);
 700   __ jcc(Assembler::negative, L); // NaN -> 0
 701   __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
 702   __ movl(c_rarg2, 0x80000000);
 703   __ movl(rax, 0x7fffffff);
 704   __ cmov(Assembler::positive, c_rarg2, rax);
 705 
 706   __ bind(L);
 707   __ movptr(inout, c_rarg2);
 708 
 709   __ pop(c_rarg0);
 710   __ pop(c_rarg1);
 711   __ pop(c_rarg2);
 712   __ pop(c_rarg3);
 713   __ pop(rax);
 714 
 715   __ ret(0);
 716 
 717   return start;
 718 }
 719 
 720 address StubGenerator::generate_d2l_fixup() {
 721   StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
 722   Address inout(rsp, 6 * wordSize); // return address + 5 saves
 723 
 724   address start = __ pc();
 725 
 726   Label L;
 727 
 728   __ push(rax);
 729   __ push(c_rarg3);
 730   __ push(c_rarg2);
 731   __ push(c_rarg1);
 732   __ push(c_rarg0);
 733 
 734   __ movl(rax, 0x7ff00000);
 735   __ movq(c_rarg2, inout);
 736   __ movl(c_rarg3, c_rarg2);
 737   __ mov(c_rarg1, c_rarg2);
 738   __ mov(c_rarg0, c_rarg2);
 739   __ negl(c_rarg3);
 740   __ shrptr(c_rarg1, 0x20);
 741   __ orl(c_rarg3, c_rarg2);
 742   __ andl(c_rarg1, 0x7fffffff);
 743   __ xorl(c_rarg2, c_rarg2);
 744   __ shrl(c_rarg3, 0x1f);
 745   __ orl(c_rarg1, c_rarg3);
 746   __ cmpl(rax, c_rarg1);
 747   __ jcc(Assembler::negative, L); // NaN -> 0
 748   __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
 749   __ mov64(c_rarg2, 0x8000000000000000);
 750   __ mov64(rax, 0x7fffffffffffffff);
 751   __ cmovq(Assembler::positive, c_rarg2, rax);
 752 
 753   __ bind(L);
 754   __ movq(inout, c_rarg2);
 755 
 756   __ pop(c_rarg0);
 757   __ pop(c_rarg1);
 758   __ pop(c_rarg2);
 759   __ pop(c_rarg3);
 760   __ pop(rax);
 761 
 762   __ ret(0);
 763 
 764   return start;
 765 }
 766 
 767 address StubGenerator::generate_count_leading_zeros_lut(const char *stub_name) {
 768   __ align64();
 769   StubCodeMark mark(this, "StubRoutines", stub_name);
 770   address start = __ pc();
 771 
 772   __ emit_data64(0x0101010102020304, relocInfo::none);
 773   __ emit_data64(0x0000000000000000, relocInfo::none);
 774   __ emit_data64(0x0101010102020304, relocInfo::none);
 775   __ emit_data64(0x0000000000000000, relocInfo::none);
 776   __ emit_data64(0x0101010102020304, relocInfo::none);
 777   __ emit_data64(0x0000000000000000, relocInfo::none);
 778   __ emit_data64(0x0101010102020304, relocInfo::none);
 779   __ emit_data64(0x0000000000000000, relocInfo::none);
 780 
 781   return start;
 782 }
 783 
 784 address StubGenerator::generate_popcount_avx_lut(const char *stub_name) {
 785   __ align64();
 786   StubCodeMark mark(this, "StubRoutines", stub_name);
 787   address start = __ pc();
 788 
 789   __ emit_data64(0x0302020102010100, relocInfo::none);
 790   __ emit_data64(0x0403030203020201, relocInfo::none);
 791   __ emit_data64(0x0302020102010100, relocInfo::none);
 792   __ emit_data64(0x0403030203020201, relocInfo::none);
 793   __ emit_data64(0x0302020102010100, relocInfo::none);
 794   __ emit_data64(0x0403030203020201, relocInfo::none);
 795   __ emit_data64(0x0302020102010100, relocInfo::none);
 796   __ emit_data64(0x0403030203020201, relocInfo::none);
 797 
 798   return start;
 799 }
 800 
 801 address StubGenerator::generate_iota_indices(const char *stub_name) {
 802   __ align(CodeEntryAlignment);
 803   StubCodeMark mark(this, "StubRoutines", stub_name);
 804   address start = __ pc();
 805   // B
 806   __ emit_data64(0x0706050403020100, relocInfo::none);
 807   __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
 808   __ emit_data64(0x1716151413121110, relocInfo::none);
 809   __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
 810   __ emit_data64(0x2726252423222120, relocInfo::none);
 811   __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
 812   __ emit_data64(0x3736353433323130, relocInfo::none);
 813   __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
 814   // W
 815   __ emit_data64(0x0003000200010000, relocInfo::none);
 816   __ emit_data64(0x0007000600050004, relocInfo::none);
 817   __ emit_data64(0x000B000A00090008, relocInfo::none);
 818   __ emit_data64(0x000F000E000D000C, relocInfo::none);
 819   __ emit_data64(0x0013001200110010, relocInfo::none);
 820   __ emit_data64(0x0017001600150014, relocInfo::none);
 821   __ emit_data64(0x001B001A00190018, relocInfo::none);
 822   __ emit_data64(0x001F001E001D001C, relocInfo::none);
 823   // D
 824   __ emit_data64(0x0000000100000000, relocInfo::none);
 825   __ emit_data64(0x0000000300000002, relocInfo::none);
 826   __ emit_data64(0x0000000500000004, relocInfo::none);
 827   __ emit_data64(0x0000000700000006, relocInfo::none);
 828   __ emit_data64(0x0000000900000008, relocInfo::none);
 829   __ emit_data64(0x0000000B0000000A, relocInfo::none);
 830   __ emit_data64(0x0000000D0000000C, relocInfo::none);
 831   __ emit_data64(0x0000000F0000000E, relocInfo::none);
 832   // Q
 833   __ emit_data64(0x0000000000000000, relocInfo::none);
 834   __ emit_data64(0x0000000000000001, relocInfo::none);
 835   __ emit_data64(0x0000000000000002, relocInfo::none);
 836   __ emit_data64(0x0000000000000003, relocInfo::none);
 837   __ emit_data64(0x0000000000000004, relocInfo::none);
 838   __ emit_data64(0x0000000000000005, relocInfo::none);
 839   __ emit_data64(0x0000000000000006, relocInfo::none);
 840   __ emit_data64(0x0000000000000007, relocInfo::none);
 841   // D - FP
 842   __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
 843   __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
 844   __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
 845   __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
 846   __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
 847   __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
 848   __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
 849   __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
 850   // Q - FP
 851   __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
 852   __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
 853   __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
 854   __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
 855   __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
 856   __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
 857   __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
 858   __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
 859   return start;
 860 }
 861 
 862 address StubGenerator::generate_vector_reverse_bit_lut(const char *stub_name) {
 863   __ align(CodeEntryAlignment);
 864   StubCodeMark mark(this, "StubRoutines", stub_name);
 865   address start = __ pc();
 866 
 867   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 868   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 869   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 870   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 871   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 872   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 873   __ emit_data64(0x0E060A020C040800, relocInfo::none);
 874   __ emit_data64(0x0F070B030D050901, relocInfo::none);
 875 
 876   return start;
 877 }
 878 
 879 address StubGenerator::generate_vector_reverse_byte_perm_mask_long(const char *stub_name) {
 880   __ align(CodeEntryAlignment);
 881   StubCodeMark mark(this, "StubRoutines", stub_name);
 882   address start = __ pc();
 883 
 884   __ emit_data64(0x0001020304050607, relocInfo::none);
 885   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 886   __ emit_data64(0x0001020304050607, relocInfo::none);
 887   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 888   __ emit_data64(0x0001020304050607, relocInfo::none);
 889   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 890   __ emit_data64(0x0001020304050607, relocInfo::none);
 891   __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
 892 
 893   return start;
 894 }
 895 
 896 address StubGenerator::generate_vector_reverse_byte_perm_mask_int(const char *stub_name) {
 897   __ align(CodeEntryAlignment);
 898   StubCodeMark mark(this, "StubRoutines", stub_name);
 899   address start = __ pc();
 900 
 901   __ emit_data64(0x0405060700010203, relocInfo::none);
 902   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 903   __ emit_data64(0x0405060700010203, relocInfo::none);
 904   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 905   __ emit_data64(0x0405060700010203, relocInfo::none);
 906   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 907   __ emit_data64(0x0405060700010203, relocInfo::none);
 908   __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
 909 
 910   return start;
 911 }
 912 
 913 address StubGenerator::generate_vector_reverse_byte_perm_mask_short(const char *stub_name) {
 914   __ align(CodeEntryAlignment);
 915   StubCodeMark mark(this, "StubRoutines", stub_name);
 916   address start = __ pc();
 917 
 918   __ emit_data64(0x0607040502030001, relocInfo::none);
 919   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 920   __ emit_data64(0x0607040502030001, relocInfo::none);
 921   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 922   __ emit_data64(0x0607040502030001, relocInfo::none);
 923   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 924   __ emit_data64(0x0607040502030001, relocInfo::none);
 925   __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
 926 
 927   return start;
 928 }
 929 
 930 address StubGenerator::generate_vector_byte_shuffle_mask(const char *stub_name) {
 931   __ align(CodeEntryAlignment);
 932   StubCodeMark mark(this, "StubRoutines", stub_name);
 933   address start = __ pc();
 934 
 935   __ emit_data64(0x7070707070707070, relocInfo::none);
 936   __ emit_data64(0x7070707070707070, relocInfo::none);
 937   __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
 938   __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
 939 
 940   return start;
 941 }
 942 
 943 address StubGenerator::generate_fp_mask(const char *stub_name, int64_t mask) {
 944   __ align(CodeEntryAlignment);
 945   StubCodeMark mark(this, "StubRoutines", stub_name);
 946   address start = __ pc();
 947 
 948   __ emit_data64( mask, relocInfo::none );
 949   __ emit_data64( mask, relocInfo::none );
 950 
 951   return start;
 952 }
 953 
 954 address StubGenerator::generate_compress_perm_table(const char *stub_name, int32_t esize) {
 955   __ align(CodeEntryAlignment);
 956   StubCodeMark mark(this, "StubRoutines", stub_name);
 957   address start = __ pc();
 958   if (esize == 32) {
 959     // Loop to generate 256 x 8 int compression permute index table. A row is
 960     // accessed using 8 bit index computed using vector mask. An entry in
 961     // a row holds either a valid permute index corresponding to set bit position
 962     // or a -1 (default) value.
 963     for (int mask = 0; mask < 256; mask++) {
 964       int ctr = 0;
 965       for (int j = 0; j < 8; j++) {
 966         if (mask & (1 << j)) {
 967           __ emit_data(j, relocInfo::none);
 968           ctr++;
 969         }
 970       }
 971       for (; ctr < 8; ctr++) {
 972         __ emit_data(-1, relocInfo::none);
 973       }
 974     }
 975   } else {
 976     assert(esize == 64, "");
 977     // Loop to generate 16 x 4 long compression permute index table. A row is
 978     // accessed using 4 bit index computed using vector mask. An entry in
 979     // a row holds either a valid permute index pair for a quadword corresponding
 980     // to set bit position or a -1 (default) value.
 981     for (int mask = 0; mask < 16; mask++) {
 982       int ctr = 0;
 983       for (int j = 0; j < 4; j++) {
 984         if (mask & (1 << j)) {
 985           __ emit_data(2 * j, relocInfo::none);
 986           __ emit_data(2 * j + 1, relocInfo::none);
 987           ctr++;
 988         }
 989       }
 990       for (; ctr < 4; ctr++) {
 991         __ emit_data64(-1L, relocInfo::none);
 992       }
 993     }
 994   }
 995   return start;
 996 }
 997 
 998 address StubGenerator::generate_expand_perm_table(const char *stub_name, int32_t esize) {
 999   __ align(CodeEntryAlignment);
1000   StubCodeMark mark(this, "StubRoutines", stub_name);
1001   address start = __ pc();
1002   if (esize == 32) {
1003     // Loop to generate 256 x 8 int expand permute index table. A row is accessed
1004     // using 8 bit index computed using vector mask. An entry in a row holds either
1005     // a valid permute index (starting from least significant lane) placed at poisition
1006     // corresponding to set bit position or a -1 (default) value.
1007     for (int mask = 0; mask < 256; mask++) {
1008       int ctr = 0;
1009       for (int j = 0; j < 8; j++) {
1010         if (mask & (1 << j)) {
1011           __ emit_data(ctr++, relocInfo::none);
1012         } else {
1013           __ emit_data(-1, relocInfo::none);
1014         }
1015       }
1016     }
1017   } else {
1018     assert(esize == 64, "");
1019     // Loop to generate 16 x 4 long expand permute index table. A row is accessed
1020     // using 4 bit index computed using vector mask. An entry in a row holds either
1021     // a valid doubleword permute index pair representing a quadword index (starting
1022     // from least significant lane) placed at poisition corresponding to set bit
1023     // position or a -1 (default) value.
1024     for (int mask = 0; mask < 16; mask++) {
1025       int ctr = 0;
1026       for (int j = 0; j < 4; j++) {
1027         if (mask & (1 << j)) {
1028           __ emit_data(2 * ctr, relocInfo::none);
1029           __ emit_data(2 * ctr + 1, relocInfo::none);
1030           ctr++;
1031         } else {
1032           __ emit_data64(-1L, relocInfo::none);
1033         }
1034       }
1035     }
1036   }
1037   return start;
1038 }
1039 
1040 address StubGenerator::generate_vector_mask(const char *stub_name, int64_t mask) {
1041   __ align(CodeEntryAlignment);
1042   StubCodeMark mark(this, "StubRoutines", stub_name);
1043   address start = __ pc();
1044 
1045   __ emit_data64(mask, relocInfo::none);
1046   __ emit_data64(mask, relocInfo::none);
1047   __ emit_data64(mask, relocInfo::none);
1048   __ emit_data64(mask, relocInfo::none);
1049   __ emit_data64(mask, relocInfo::none);
1050   __ emit_data64(mask, relocInfo::none);
1051   __ emit_data64(mask, relocInfo::none);
1052   __ emit_data64(mask, relocInfo::none);
1053 
1054   return start;
1055 }
1056 
1057 address StubGenerator::generate_vector_byte_perm_mask(const char *stub_name) {
1058   __ align(CodeEntryAlignment);
1059   StubCodeMark mark(this, "StubRoutines", stub_name);
1060   address start = __ pc();
1061 
1062   __ emit_data64(0x0000000000000001, relocInfo::none);
1063   __ emit_data64(0x0000000000000003, relocInfo::none);
1064   __ emit_data64(0x0000000000000005, relocInfo::none);
1065   __ emit_data64(0x0000000000000007, relocInfo::none);
1066   __ emit_data64(0x0000000000000000, relocInfo::none);
1067   __ emit_data64(0x0000000000000002, relocInfo::none);
1068   __ emit_data64(0x0000000000000004, relocInfo::none);
1069   __ emit_data64(0x0000000000000006, relocInfo::none);
1070 
1071   return start;
1072 }
1073 
1074 address StubGenerator::generate_vector_fp_mask(const char *stub_name, int64_t mask) {
1075   __ align(CodeEntryAlignment);
1076   StubCodeMark mark(this, "StubRoutines", stub_name);
1077   address start = __ pc();
1078 
1079   __ emit_data64(mask, relocInfo::none);
1080   __ emit_data64(mask, relocInfo::none);
1081   __ emit_data64(mask, relocInfo::none);
1082   __ emit_data64(mask, relocInfo::none);
1083   __ emit_data64(mask, relocInfo::none);
1084   __ emit_data64(mask, relocInfo::none);
1085   __ emit_data64(mask, relocInfo::none);
1086   __ emit_data64(mask, relocInfo::none);
1087 
1088   return start;
1089 }
1090 
1091 address StubGenerator::generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
1092                                    int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1093                                    int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1094                                    int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1095                                    int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1096   __ align(CodeEntryAlignment);
1097   StubCodeMark mark(this, "StubRoutines", stub_name);
1098   address start = __ pc();
1099 
1100   assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1101   __ emit_data(val0, relocInfo::none, 0);
1102   __ emit_data(val1, relocInfo::none, 0);
1103   __ emit_data(val2, relocInfo::none, 0);
1104   __ emit_data(val3, relocInfo::none, 0);
1105   if (len >= Assembler::AVX_256bit) {
1106     __ emit_data(val4, relocInfo::none, 0);
1107     __ emit_data(val5, relocInfo::none, 0);
1108     __ emit_data(val6, relocInfo::none, 0);
1109     __ emit_data(val7, relocInfo::none, 0);
1110     if (len >= Assembler::AVX_512bit) {
1111       __ emit_data(val8, relocInfo::none, 0);
1112       __ emit_data(val9, relocInfo::none, 0);
1113       __ emit_data(val10, relocInfo::none, 0);
1114       __ emit_data(val11, relocInfo::none, 0);
1115       __ emit_data(val12, relocInfo::none, 0);
1116       __ emit_data(val13, relocInfo::none, 0);
1117       __ emit_data(val14, relocInfo::none, 0);
1118       __ emit_data(val15, relocInfo::none, 0);
1119     }
1120   }
1121   return start;
1122 }
1123 
1124 // Non-destructive plausibility checks for oops
1125 //
1126 // Arguments:
1127 //    all args on stack!
1128 //
1129 // Stack after saving c_rarg3:
1130 //    [tos + 0]: saved c_rarg3
1131 //    [tos + 1]: saved c_rarg2
1132 //    [tos + 2]: saved r12 (several TemplateTable methods use it)
1133 //    [tos + 3]: saved flags
1134 //    [tos + 4]: return address
1135 //  * [tos + 5]: error message (char*)
1136 //  * [tos + 6]: object to verify (oop)
1137 //  * [tos + 7]: saved rax - saved by caller and bashed
1138 //  * [tos + 8]: saved r10 (rscratch1) - saved by caller
1139 //  * = popped on exit
1140 address StubGenerator::generate_verify_oop() {
1141   StubCodeMark mark(this, "StubRoutines", "verify_oop");
1142   address start = __ pc();
1143 
1144   Label exit, error;
1145 
1146   __ pushf();
1147   __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1148 
1149   __ push(r12);
1150 
1151   // save c_rarg2 and c_rarg3
1152   __ push(c_rarg2);
1153   __ push(c_rarg3);
1154 
1155   enum {
1156     // After previous pushes.
1157     oop_to_verify = 6 * wordSize,
1158     saved_rax     = 7 * wordSize,
1159     saved_r10     = 8 * wordSize,
1160 
1161     // Before the call to MacroAssembler::debug(), see below.
1162     return_addr   = 16 * wordSize,
1163     error_msg     = 17 * wordSize
1164   };
1165 
1166   // get object
1167   __ movptr(rax, Address(rsp, oop_to_verify));
1168 
1169   // make sure object is 'reasonable'
1170   __ testptr(rax, rax);
1171   __ jcc(Assembler::zero, exit); // if obj is null it is OK
1172 
1173    BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1174    bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1175 
1176   // return if everything seems ok
1177   __ bind(exit);
1178   __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1179   __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1180   __ pop(c_rarg3);                             // restore c_rarg3
1181   __ pop(c_rarg2);                             // restore c_rarg2
1182   __ pop(r12);                                 // restore r12
1183   __ popf();                                   // restore flags
1184   __ ret(4 * wordSize);                        // pop caller saved stuff
1185 
1186   // handle errors
1187   __ bind(error);
1188   __ movptr(rax, Address(rsp, saved_rax));     // get saved rax back
1189   __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1190   __ pop(c_rarg3);                             // get saved c_rarg3 back
1191   __ pop(c_rarg2);                             // get saved c_rarg2 back
1192   __ pop(r12);                                 // get saved r12 back
1193   __ popf();                                   // get saved flags off stack --
1194                                                // will be ignored
1195 
1196   __ pusha();                                  // push registers
1197                                                // (rip is already
1198                                                // already pushed)
1199   // debug(char* msg, int64_t pc, int64_t regs[])
1200   // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1201   // pushed all the registers, so now the stack looks like:
1202   //     [tos +  0] 16 saved registers
1203   //     [tos + 16] return address
1204   //   * [tos + 17] error message (char*)
1205   //   * [tos + 18] object to verify (oop)
1206   //   * [tos + 19] saved rax - saved by caller and bashed
1207   //   * [tos + 20] saved r10 (rscratch1) - saved by caller
1208   //   * = popped on exit
1209 
1210   __ movptr(c_rarg0, Address(rsp, error_msg));    // pass address of error message
1211   __ movptr(c_rarg1, Address(rsp, return_addr));  // pass return address
1212   __ movq(c_rarg2, rsp);                          // pass address of regs on stack
1213   __ mov(r12, rsp);                               // remember rsp
1214   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1215   __ andptr(rsp, -16);                            // align stack as required by ABI
1216   BLOCK_COMMENT("call MacroAssembler::debug");
1217   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1218   __ hlt();
1219 
1220   return start;
1221 }
1222 
1223 
1224 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1225 //
1226 // Outputs:
1227 //    rdi - rcx
1228 //    rsi - rdx
1229 //    rdx - r8
1230 //    rcx - r9
1231 //
1232 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1233 // are non-volatile.  r9 and r10 should not be used by the caller.
1234 //
1235 void StubGenerator::setup_arg_regs(int nargs) {
1236   const Register saved_rdi = r9;
1237   const Register saved_rsi = r10;
1238   assert(nargs == 3 || nargs == 4, "else fix");
1239 #ifdef _WIN64
1240   assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1241          "unexpected argument registers");
1242   if (nargs == 4) {
1243     __ mov(rax, r9);  // r9 is also saved_rdi
1244   }
1245   __ movptr(saved_rdi, rdi);
1246   __ movptr(saved_rsi, rsi);
1247   __ mov(rdi, rcx); // c_rarg0
1248   __ mov(rsi, rdx); // c_rarg1
1249   __ mov(rdx, r8);  // c_rarg2
1250   if (nargs == 4) {
1251     __ mov(rcx, rax); // c_rarg3 (via rax)
1252   }
1253 #else
1254   assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1255          "unexpected argument registers");
1256 #endif
1257   DEBUG_ONLY(_regs_in_thread = false;)
1258 }
1259 
1260 
1261 void StubGenerator::restore_arg_regs() {
1262   assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1263   const Register saved_rdi = r9;
1264   const Register saved_rsi = r10;
1265 #ifdef _WIN64
1266   __ movptr(rdi, saved_rdi);
1267   __ movptr(rsi, saved_rsi);
1268 #endif
1269 }
1270 
1271 
1272 // This is used in places where r10 is a scratch register, and can
1273 // be adapted if r9 is needed also.
1274 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1275   const Register saved_r15 = r9;
1276   assert(nargs == 3 || nargs == 4, "else fix");
1277 #ifdef _WIN64
1278   if (nargs == 4) {
1279     __ mov(rax, r9);       // r9 is also saved_r15
1280   }
1281   __ mov(saved_r15, r15);  // r15 is callee saved and needs to be restored
1282   __ get_thread(r15_thread);
1283   assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1284          "unexpected argument registers");
1285   __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1286   __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1287 
1288   __ mov(rdi, rcx); // c_rarg0
1289   __ mov(rsi, rdx); // c_rarg1
1290   __ mov(rdx, r8);  // c_rarg2
1291   if (nargs == 4) {
1292     __ mov(rcx, rax); // c_rarg3 (via rax)
1293   }
1294 #else
1295   assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1296          "unexpected argument registers");
1297 #endif
1298   DEBUG_ONLY(_regs_in_thread = true;)
1299 }
1300 
1301 
1302 void StubGenerator::restore_arg_regs_using_thread() {
1303   assert(_regs_in_thread, "wrong call to restore_arg_regs");
1304   const Register saved_r15 = r9;
1305 #ifdef _WIN64
1306   __ get_thread(r15_thread);
1307   __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1308   __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1309   __ mov(r15, saved_r15);  // r15 is callee saved and needs to be restored
1310 #endif
1311 }
1312 
1313 
1314 void StubGenerator::setup_argument_regs(BasicType type) {
1315   if (type == T_BYTE || type == T_SHORT) {
1316     setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1317                       // r9 and r10 may be used to save non-volatile registers
1318   } else {
1319     setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1320                                    // r9 is used to save r15_thread
1321   }
1322 }
1323 
1324 
1325 void StubGenerator::restore_argument_regs(BasicType type) {
1326   if (type == T_BYTE || type == T_SHORT) {
1327     restore_arg_regs();
1328   } else {
1329     restore_arg_regs_using_thread();
1330   }
1331 }
1332 
1333 address StubGenerator::generate_data_cache_writeback() {
1334   const Register src        = c_rarg0;  // source address
1335 
1336   __ align(CodeEntryAlignment);
1337 
1338   StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
1339 
1340   address start = __ pc();
1341 
1342   __ enter();
1343   __ cache_wb(Address(src, 0));
1344   __ leave();
1345   __ ret(0);
1346 
1347   return start;
1348 }
1349 
1350 address StubGenerator::generate_data_cache_writeback_sync() {
1351   const Register is_pre    = c_rarg0;  // pre or post sync
1352 
1353   __ align(CodeEntryAlignment);
1354 
1355   StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
1356 
1357   // pre wbsync is a no-op
1358   // post wbsync translates to an sfence
1359 
1360   Label skip;
1361   address start = __ pc();
1362 
1363   __ enter();
1364   __ cmpl(is_pre, 0);
1365   __ jcc(Assembler::notEqual, skip);
1366   __ cache_wbsync(false);
1367   __ bind(skip);
1368   __ leave();
1369   __ ret(0);
1370 
1371   return start;
1372 }
1373 
1374 // ofs and limit are use for multi-block byte array.
1375 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1376 address StubGenerator::generate_md5_implCompress(bool multi_block, const char *name) {
1377   __ align(CodeEntryAlignment);
1378   StubCodeMark mark(this, "StubRoutines", name);
1379   address start = __ pc();
1380 
1381   const Register buf_param = r15;
1382   const Address state_param(rsp, 0 * wordSize);
1383   const Address ofs_param  (rsp, 1 * wordSize    );
1384   const Address limit_param(rsp, 1 * wordSize + 4);
1385 
1386   __ enter();
1387   __ push(rbx);
1388   __ push(rdi);
1389   __ push(rsi);
1390   __ push(r15);
1391   __ subptr(rsp, 2 * wordSize);
1392 
1393   __ movptr(buf_param, c_rarg0);
1394   __ movptr(state_param, c_rarg1);
1395   if (multi_block) {
1396     __ movl(ofs_param, c_rarg2);
1397     __ movl(limit_param, c_rarg3);
1398   }
1399   __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1400 
1401   __ addptr(rsp, 2 * wordSize);
1402   __ pop(r15);
1403   __ pop(rsi);
1404   __ pop(rdi);
1405   __ pop(rbx);
1406   __ leave();
1407   __ ret(0);
1408 
1409   return start;
1410 }
1411 
1412 address StubGenerator::generate_upper_word_mask() {
1413   __ align64();
1414   StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
1415   address start = __ pc();
1416 
1417   __ emit_data64(0x0000000000000000, relocInfo::none);
1418   __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1419 
1420   return start;
1421 }
1422 
1423 address StubGenerator::generate_shuffle_byte_flip_mask() {
1424   __ align64();
1425   StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
1426   address start = __ pc();
1427 
1428   __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1429   __ emit_data64(0x0001020304050607, relocInfo::none);
1430 
1431   return start;
1432 }
1433 
1434 // ofs and limit are use for multi-block byte array.
1435 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1436 address StubGenerator::generate_sha1_implCompress(bool multi_block, const char *name) {
1437   __ align(CodeEntryAlignment);
1438   StubCodeMark mark(this, "StubRoutines", name);
1439   address start = __ pc();
1440 
1441   Register buf = c_rarg0;
1442   Register state = c_rarg1;
1443   Register ofs = c_rarg2;
1444   Register limit = c_rarg3;
1445 
1446   const XMMRegister abcd = xmm0;
1447   const XMMRegister e0 = xmm1;
1448   const XMMRegister e1 = xmm2;
1449   const XMMRegister msg0 = xmm3;
1450 
1451   const XMMRegister msg1 = xmm4;
1452   const XMMRegister msg2 = xmm5;
1453   const XMMRegister msg3 = xmm6;
1454   const XMMRegister shuf_mask = xmm7;
1455 
1456   __ enter();
1457 
1458   __ subptr(rsp, 4 * wordSize);
1459 
1460   __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1461     buf, state, ofs, limit, rsp, multi_block);
1462 
1463   __ addptr(rsp, 4 * wordSize);
1464 
1465   __ leave();
1466   __ ret(0);
1467 
1468   return start;
1469 }
1470 
1471 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1472   __ align64();
1473   StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
1474   address start = __ pc();
1475 
1476   __ emit_data64(0x0405060700010203, relocInfo::none);
1477   __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1478 
1479   if (VM_Version::supports_avx2()) {
1480     __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1481     __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1482     // _SHUF_00BA
1483     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1484     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1485     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1486     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1487     // _SHUF_DC00
1488     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1489     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1490     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1491     __ emit_data64(0x0b0a090803020100, relocInfo::none);
1492   }
1493 
1494   return start;
1495 }
1496 
1497 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1498 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1499   __ align32();
1500   StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
1501   address start = __ pc();
1502 
1503   if (VM_Version::supports_avx2()) {
1504     __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1505     __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1506     __ emit_data64(0x1011121314151617, relocInfo::none);
1507     __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1508     __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1509     __ emit_data64(0x0000000000000000, relocInfo::none);
1510     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1511     __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1512   }
1513 
1514   return start;
1515 }
1516 
1517 // ofs and limit are use for multi-block byte array.
1518 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1519 address StubGenerator::generate_sha256_implCompress(bool multi_block, const char *name) {
1520   assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1521   __ align(CodeEntryAlignment);
1522   StubCodeMark mark(this, "StubRoutines", name);
1523   address start = __ pc();
1524 
1525   Register buf = c_rarg0;
1526   Register state = c_rarg1;
1527   Register ofs = c_rarg2;
1528   Register limit = c_rarg3;
1529 
1530   const XMMRegister msg = xmm0;
1531   const XMMRegister state0 = xmm1;
1532   const XMMRegister state1 = xmm2;
1533   const XMMRegister msgtmp0 = xmm3;
1534 
1535   const XMMRegister msgtmp1 = xmm4;
1536   const XMMRegister msgtmp2 = xmm5;
1537   const XMMRegister msgtmp3 = xmm6;
1538   const XMMRegister msgtmp4 = xmm7;
1539 
1540   const XMMRegister shuf_mask = xmm8;
1541 
1542   __ enter();
1543 
1544   __ subptr(rsp, 4 * wordSize);
1545 
1546   if (VM_Version::supports_sha()) {
1547     __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1548       buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1549   } else if (VM_Version::supports_avx2()) {
1550     __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1551       buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1552   }
1553   __ addptr(rsp, 4 * wordSize);
1554   __ vzeroupper();
1555   __ leave();
1556   __ ret(0);
1557 
1558   return start;
1559 }
1560 
1561 address StubGenerator::generate_sha512_implCompress(bool multi_block, const char *name) {
1562   assert(VM_Version::supports_avx2(), "");
1563   assert(VM_Version::supports_bmi2(), "");
1564   __ align(CodeEntryAlignment);
1565   StubCodeMark mark(this, "StubRoutines", name);
1566   address start = __ pc();
1567 
1568   Register buf = c_rarg0;
1569   Register state = c_rarg1;
1570   Register ofs = c_rarg2;
1571   Register limit = c_rarg3;
1572 
1573   const XMMRegister msg = xmm0;
1574   const XMMRegister state0 = xmm1;
1575   const XMMRegister state1 = xmm2;
1576   const XMMRegister msgtmp0 = xmm3;
1577   const XMMRegister msgtmp1 = xmm4;
1578   const XMMRegister msgtmp2 = xmm5;
1579   const XMMRegister msgtmp3 = xmm6;
1580   const XMMRegister msgtmp4 = xmm7;
1581 
1582   const XMMRegister shuf_mask = xmm8;
1583 
1584   __ enter();
1585 
1586   __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1587   buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1588 
1589   __ vzeroupper();
1590   __ leave();
1591   __ ret(0);
1592 
1593   return start;
1594 }
1595 
1596 address StubGenerator::base64_shuffle_addr() {
1597   __ align64();
1598   StubCodeMark mark(this, "StubRoutines", "shuffle_base64");
1599   address start = __ pc();
1600 
1601   assert(((unsigned long long)start & 0x3f) == 0,
1602          "Alignment problem (0x%08llx)", (unsigned long long)start);
1603   __ emit_data64(0x0405030401020001, relocInfo::none);
1604   __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1605   __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1606   __ emit_data64(0x1617151613141213, relocInfo::none);
1607   __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1608   __ emit_data64(0x222321221f201e1f, relocInfo::none);
1609   __ emit_data64(0x2829272825262425, relocInfo::none);
1610   __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1611 
1612   return start;
1613 }
1614 
1615 address StubGenerator::base64_avx2_shuffle_addr() {
1616   __ align32();
1617   StubCodeMark mark(this, "StubRoutines", "avx2_shuffle_base64");
1618   address start = __ pc();
1619 
1620   __ emit_data64(0x0809070805060405, relocInfo::none);
1621   __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1622   __ emit_data64(0x0405030401020001, relocInfo::none);
1623   __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1624 
1625   return start;
1626 }
1627 
1628 address StubGenerator::base64_avx2_input_mask_addr() {
1629   __ align32();
1630   StubCodeMark mark(this, "StubRoutines", "avx2_input_mask_base64");
1631   address start = __ pc();
1632 
1633   __ emit_data64(0x8000000000000000, relocInfo::none);
1634   __ emit_data64(0x8000000080000000, relocInfo::none);
1635   __ emit_data64(0x8000000080000000, relocInfo::none);
1636   __ emit_data64(0x8000000080000000, relocInfo::none);
1637 
1638   return start;
1639 }
1640 
1641 address StubGenerator::base64_avx2_lut_addr() {
1642   __ align32();
1643   StubCodeMark mark(this, "StubRoutines", "avx2_lut_base64");
1644   address start = __ pc();
1645 
1646   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1647   __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1648   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1649   __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1650 
1651   // URL LUT
1652   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1653   __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1654   __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1655   __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1656 
1657   return start;
1658 }
1659 
1660 address StubGenerator::base64_encoding_table_addr() {
1661   __ align64();
1662   StubCodeMark mark(this, "StubRoutines", "encoding_table_base64");
1663   address start = __ pc();
1664 
1665   assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1666   __ emit_data64(0x4847464544434241, relocInfo::none);
1667   __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1668   __ emit_data64(0x5857565554535251, relocInfo::none);
1669   __ emit_data64(0x6665646362615a59, relocInfo::none);
1670   __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1671   __ emit_data64(0x767574737271706f, relocInfo::none);
1672   __ emit_data64(0x333231307a797877, relocInfo::none);
1673   __ emit_data64(0x2f2b393837363534, relocInfo::none);
1674 
1675   // URL table
1676   __ emit_data64(0x4847464544434241, relocInfo::none);
1677   __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1678   __ emit_data64(0x5857565554535251, relocInfo::none);
1679   __ emit_data64(0x6665646362615a59, relocInfo::none);
1680   __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1681   __ emit_data64(0x767574737271706f, relocInfo::none);
1682   __ emit_data64(0x333231307a797877, relocInfo::none);
1683   __ emit_data64(0x5f2d393837363534, relocInfo::none);
1684 
1685   return start;
1686 }
1687 
1688 // Code for generating Base64 encoding.
1689 // Intrinsic function prototype in Base64.java:
1690 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1691 // boolean isURL) {
1692 address StubGenerator::generate_base64_encodeBlock()
1693 {
1694   __ align(CodeEntryAlignment);
1695   StubCodeMark mark(this, "StubRoutines", "implEncode");
1696   address start = __ pc();
1697 
1698   __ enter();
1699 
1700   // Save callee-saved registers before using them
1701   __ push(r12);
1702   __ push(r13);
1703   __ push(r14);
1704   __ push(r15);
1705 
1706   // arguments
1707   const Register source = c_rarg0;       // Source Array
1708   const Register start_offset = c_rarg1; // start offset
1709   const Register end_offset = c_rarg2;   // end offset
1710   const Register dest = c_rarg3;   // destination array
1711 
1712 #ifndef _WIN64
1713   const Register dp = c_rarg4;    // Position for writing to dest array
1714   const Register isURL = c_rarg5; // Base64 or URL character set
1715 #else
1716   const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1717   const Address isURL_mem(rbp, 7 * wordSize);
1718   const Register isURL = r10; // pick the volatile windows register
1719   const Register dp = r12;
1720   __ movl(dp, dp_mem);
1721   __ movl(isURL, isURL_mem);
1722 #endif
1723 
1724   const Register length = r14;
1725   const Register encode_table = r13;
1726   Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1727 
1728   // calculate length from offsets
1729   __ movl(length, end_offset);
1730   __ subl(length, start_offset);
1731   __ jcc(Assembler::lessEqual, L_exit);
1732 
1733   // Code for 512-bit VBMI encoding.  Encodes 48 input bytes into 64
1734   // output bytes. We read 64 input bytes and ignore the last 16, so be
1735   // sure not to read past the end of the input buffer.
1736   if (VM_Version::supports_avx512_vbmi()) {
1737     __ cmpl(length, 64); // Do not overrun input buffer.
1738     __ jcc(Assembler::below, L_not512);
1739 
1740     __ shll(isURL, 6); // index into decode table based on isURL
1741     __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1742     __ addptr(encode_table, isURL);
1743     __ shrl(isURL, 6); // restore isURL
1744 
1745     __ mov64(rax, 0x3036242a1016040aull); // Shifts
1746     __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1747     __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1748     __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1749 
1750     __ align32();
1751     __ BIND(L_vbmiLoop);
1752 
1753     __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1754     __ subl(length, 48);
1755 
1756     // Put the input bytes into the proper lanes for writing, then
1757     // encode them.
1758     __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1759     __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1760 
1761     // Write to destination
1762     __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1763 
1764     __ addptr(dest, 64);
1765     __ addptr(source, 48);
1766     __ cmpl(length, 64);
1767     __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1768 
1769     __ vzeroupper();
1770   }
1771 
1772   __ BIND(L_not512);
1773   if (VM_Version::supports_avx2()) {
1774     /*
1775     ** This AVX2 encoder is based off the paper at:
1776     **      https://dl.acm.org/doi/10.1145/3132709
1777     **
1778     ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1779     ** output bytes.
1780     **
1781     */
1782     // Lengths under 32 bytes are done with scalar routine
1783     __ cmpl(length, 31);
1784     __ jcc(Assembler::belowEqual, L_process3);
1785 
1786     // Set up supporting constant table data
1787     __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1788     // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1789     __ movl(rax, 0x0fc0fc00);
1790     __ movdl(xmm8, rax);
1791     __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1792     __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1793 
1794     // Multiplication constant for "shifting" right by 6 and 10
1795     // bits
1796     __ movl(rax, 0x04000040);
1797 
1798     __ subl(length, 24);
1799     __ movdl(xmm7, rax);
1800     __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1801 
1802     // For the first load, we mask off reading of the first 4
1803     // bytes into the register. This is so we can get 4 3-byte
1804     // chunks into each lane of the register, avoiding having to
1805     // handle end conditions.  We then shuffle these bytes into a
1806     // specific order so that manipulation is easier.
1807     //
1808     // The initial read loads the XMM register like this:
1809     //
1810     // Lower 128-bit lane:
1811     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1812     // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1813     // | C2 | D0 | D1 | D2 |
1814     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1815     //
1816     // Upper 128-bit lane:
1817     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1818     // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1819     // | XX | XX | XX | XX |
1820     // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1821     //
1822     // Where A0 is the first input byte, B0 is the fourth, etc.
1823     // The alphabetical significance denotes the 3 bytes to be
1824     // consumed and encoded into 4 bytes.
1825     //
1826     // We then shuffle the register so each 32-bit word contains
1827     // the sequence:
1828     //    A1 A0 A2 A1, B1, B0, B2, B1, etc.
1829     // Each of these byte sequences are then manipulated into 4
1830     // 6-bit values ready for encoding.
1831     //
1832     // If we focus on one set of 3-byte chunks, changing the
1833     // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1834     // shuffle such that each 24-bit chunk contains:
1835     //
1836     // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1837     // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1838     // Explain this step.
1839     // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1840     // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1841     //
1842     // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1843     // a5..a0) and shift them using a vector multiplication
1844     // operation (vpmulhuw) which effectively shifts c right by 6
1845     // bits and a right by 10 bits.  We similarly mask bits 10-15
1846     // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1847     // bits respectively.  This is done using vpmullw.  We end up
1848     // with 4 6-bit values, thus splitting the 3 input bytes,
1849     // ready for encoding:
1850     //    0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1851     //
1852     // For translation, we recognize that there are 5 distinct
1853     // ranges of legal Base64 characters as below:
1854     //
1855     //   +-------------+-------------+------------+
1856     //   | 6-bit value | ASCII range |   offset   |
1857     //   +-------------+-------------+------------+
1858     //   |    0..25    |    A..Z     |     65     |
1859     //   |   26..51    |    a..z     |     71     |
1860     //   |   52..61    |    0..9     |     -4     |
1861     //   |     62      |   + or -    | -19 or -17 |
1862     //   |     63      |   / or _    | -16 or 32  |
1863     //   +-------------+-------------+------------+
1864     //
1865     // We note that vpshufb does a parallel lookup in a
1866     // destination register using the lower 4 bits of bytes from a
1867     // source register.  If we use a saturated subtraction and
1868     // subtract 51 from each 6-bit value, bytes from [0,51]
1869     // saturate to 0, and [52,63] map to a range of [1,12].  We
1870     // distinguish the [0,25] and [26,51] ranges by assigning a
1871     // value of 13 for all 6-bit values less than 26.  We end up
1872     // with:
1873     //
1874     //   +-------------+-------------+------------+
1875     //   | 6-bit value |   Reduced   |   offset   |
1876     //   +-------------+-------------+------------+
1877     //   |    0..25    |     13      |     65     |
1878     //   |   26..51    |      0      |     71     |
1879     //   |   52..61    |    0..9     |     -4     |
1880     //   |     62      |     11      | -19 or -17 |
1881     //   |     63      |     12      | -16 or 32  |
1882     //   +-------------+-------------+------------+
1883     //
1884     // We then use a final vpshufb to add the appropriate offset,
1885     // translating the bytes.
1886     //
1887     // Load input bytes - only 28 bytes.  Mask the first load to
1888     // not load into the full register.
1889     __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1890 
1891     // Move 3-byte chunks of input (12 bytes) into 16 bytes,
1892     // ordering by:
1893     //   1, 0, 2, 1; 4, 3, 5, 4; etc.  This groups 6-bit chunks
1894     //   for easy masking
1895     __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1896 
1897     __ addl(start_offset, 24);
1898 
1899     // Load masking register for first and third (and multiples)
1900     // 6-bit values.
1901     __ movl(rax, 0x003f03f0);
1902     __ movdl(xmm6, rax);
1903     __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
1904     // Multiplication constant for "shifting" left by 4 and 8 bits
1905     __ movl(rax, 0x01000010);
1906     __ movdl(xmm5, rax);
1907     __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
1908 
1909     // Isolate 6-bit chunks of interest
1910     __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1911 
1912     // Load constants for encoding
1913     __ movl(rax, 0x19191919);
1914     __ movdl(xmm3, rax);
1915     __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
1916     __ movl(rax, 0x33333333);
1917     __ movdl(xmm4, rax);
1918     __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
1919 
1920     // Shift output bytes 0 and 2 into proper lanes
1921     __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
1922 
1923     // Mask and shift output bytes 1 and 3 into proper lanes and
1924     // combine
1925     __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1926     __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1927     __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
1928 
1929     // Find out which are 0..25.  This indicates which input
1930     // values fall in the range of 'A'-'Z', which require an
1931     // additional offset (see comments above)
1932     __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
1933     __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1934     __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
1935 
1936     // Load the proper lookup table
1937     __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
1938     __ movl(r15, isURL);
1939     __ shll(r15, 5);
1940     __ vmovdqu(xmm2, Address(r11, r15));
1941 
1942     // Shuffle the offsets based on the range calculation done
1943     // above. This allows us to add the correct offset to the
1944     // 6-bit value corresponding to the range documented above.
1945     __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1946     __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1947 
1948     // Store the encoded bytes
1949     __ vmovdqu(Address(dest, dp), xmm0);
1950     __ addl(dp, 32);
1951 
1952     __ cmpl(length, 31);
1953     __ jcc(Assembler::belowEqual, L_process3);
1954 
1955     __ align32();
1956     __ BIND(L_32byteLoop);
1957 
1958     // Get next 32 bytes
1959     __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
1960 
1961     __ subl(length, 24);
1962     __ addl(start_offset, 24);
1963 
1964     // This logic is identical to the above, with only constant
1965     // register loads removed.  Shuffle the input, mask off 6-bit
1966     // chunks, shift them into place, then add the offset to
1967     // encode.
1968     __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1969 
1970     __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1971     __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
1972     __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1973     __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1974     __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
1975     __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
1976     __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1977     __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
1978     __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1979     __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1980 
1981     // Store the encoded bytes
1982     __ vmovdqu(Address(dest, dp), xmm0);
1983     __ addl(dp, 32);
1984 
1985     __ cmpl(length, 31);
1986     __ jcc(Assembler::above, L_32byteLoop);
1987 
1988     __ BIND(L_process3);
1989     __ vzeroupper();
1990   } else {
1991     __ BIND(L_process3);
1992   }
1993 
1994   __ cmpl(length, 3);
1995   __ jcc(Assembler::below, L_exit);
1996 
1997   // Load the encoding table based on isURL
1998   __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1999   __ movl(r15, isURL);
2000   __ shll(r15, 6);
2001   __ addptr(r11, r15);
2002 
2003   __ BIND(L_processdata);
2004 
2005   // Load 3 bytes
2006   __ load_unsigned_byte(r15, Address(source, start_offset));
2007   __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
2008   __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
2009 
2010   // Build a 32-bit word with bytes 1, 2, 0, 1
2011   __ movl(rax, r10);
2012   __ shll(r10, 24);
2013   __ orl(rax, r10);
2014 
2015   __ subl(length, 3);
2016 
2017   __ shll(r15, 8);
2018   __ shll(r13, 16);
2019   __ orl(rax, r15);
2020 
2021   __ addl(start_offset, 3);
2022 
2023   __ orl(rax, r13);
2024   // At this point, rax contains | byte1 | byte2 | byte0 | byte1
2025   // r13 has byte2 << 16 - need low-order 6 bits to translate.
2026   // This translated byte is the fourth output byte.
2027   __ shrl(r13, 16);
2028   __ andl(r13, 0x3f);
2029 
2030   // The high-order 6 bits of r15 (byte0) is translated.
2031   // The translated byte is the first output byte.
2032   __ shrl(r15, 10);
2033 
2034   __ load_unsigned_byte(r13, Address(r11, r13));
2035   __ load_unsigned_byte(r15, Address(r11, r15));
2036 
2037   __ movb(Address(dest, dp, Address::times_1, 3), r13);
2038 
2039   // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
2040   // This translated byte is the second output byte.
2041   __ shrl(rax, 4);
2042   __ movl(r10, rax);
2043   __ andl(rax, 0x3f);
2044 
2045   __ movb(Address(dest, dp, Address::times_1, 0), r15);
2046 
2047   __ load_unsigned_byte(rax, Address(r11, rax));
2048 
2049   // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
2050   // This translated byte is the third output byte.
2051   __ shrl(r10, 18);
2052   __ andl(r10, 0x3f);
2053 
2054   __ load_unsigned_byte(r10, Address(r11, r10));
2055 
2056   __ movb(Address(dest, dp, Address::times_1, 1), rax);
2057   __ movb(Address(dest, dp, Address::times_1, 2), r10);
2058 
2059   __ addl(dp, 4);
2060   __ cmpl(length, 3);
2061   __ jcc(Assembler::aboveEqual, L_processdata);
2062 
2063   __ BIND(L_exit);
2064   __ pop(r15);
2065   __ pop(r14);
2066   __ pop(r13);
2067   __ pop(r12);
2068   __ leave();
2069   __ ret(0);
2070 
2071   return start;
2072 }
2073 
2074 // base64 AVX512vbmi tables
2075 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2076   __ align64();
2077   StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64");
2078   address start = __ pc();
2079 
2080   assert(((unsigned long long)start & 0x3f) == 0,
2081          "Alignment problem (0x%08llx)", (unsigned long long)start);
2082   __ emit_data64(0x8080808080808080, relocInfo::none);
2083   __ emit_data64(0x8080808080808080, relocInfo::none);
2084   __ emit_data64(0x8080808080808080, relocInfo::none);
2085   __ emit_data64(0x8080808080808080, relocInfo::none);
2086   __ emit_data64(0x8080808080808080, relocInfo::none);
2087   __ emit_data64(0x3f8080803e808080, relocInfo::none);
2088   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2089   __ emit_data64(0x8080808080803d3c, relocInfo::none);
2090 
2091   return start;
2092 }
2093 
2094 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2095   __ align64();
2096   StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64");
2097   address start = __ pc();
2098 
2099   assert(((unsigned long long)start & 0x3f) == 0,
2100          "Alignment problem (0x%08llx)", (unsigned long long)start);
2101   __ emit_data64(0x0605040302010080, relocInfo::none);
2102   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2103   __ emit_data64(0x161514131211100f, relocInfo::none);
2104   __ emit_data64(0x8080808080191817, relocInfo::none);
2105   __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2106   __ emit_data64(0x2827262524232221, relocInfo::none);
2107   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2108   __ emit_data64(0x8080808080333231, relocInfo::none);
2109 
2110   return start;
2111 }
2112 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2113   __ align64();
2114   StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64url");
2115   address start = __ pc();
2116 
2117   assert(((unsigned long long)start & 0x3f) == 0,
2118          "Alignment problem (0x%08llx)", (unsigned long long)start);
2119   __ emit_data64(0x8080808080808080, relocInfo::none);
2120   __ emit_data64(0x8080808080808080, relocInfo::none);
2121   __ emit_data64(0x8080808080808080, relocInfo::none);
2122   __ emit_data64(0x8080808080808080, relocInfo::none);
2123   __ emit_data64(0x8080808080808080, relocInfo::none);
2124   __ emit_data64(0x80803e8080808080, relocInfo::none);
2125   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2126   __ emit_data64(0x8080808080803d3c, relocInfo::none);
2127 
2128   return start;
2129 }
2130 
2131 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2132   __ align64();
2133   StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64url");
2134   address start = __ pc();
2135 
2136   assert(((unsigned long long)start & 0x3f) == 0,
2137          "Alignment problem (0x%08llx)", (unsigned long long)start);
2138   __ emit_data64(0x0605040302010080, relocInfo::none);
2139   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2140   __ emit_data64(0x161514131211100f, relocInfo::none);
2141   __ emit_data64(0x3f80808080191817, relocInfo::none);
2142   __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2143   __ emit_data64(0x2827262524232221, relocInfo::none);
2144   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2145   __ emit_data64(0x8080808080333231, relocInfo::none);
2146 
2147   return start;
2148 }
2149 
2150 address StubGenerator::base64_vbmi_pack_vec_addr() {
2151   __ align64();
2152   StubCodeMark mark(this, "StubRoutines", "pack_vec_base64");
2153   address start = __ pc();
2154 
2155   assert(((unsigned long long)start & 0x3f) == 0,
2156          "Alignment problem (0x%08llx)", (unsigned long long)start);
2157   __ emit_data64(0x090a040506000102, relocInfo::none);
2158   __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2159   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2160   __ emit_data64(0x292a242526202122, relocInfo::none);
2161   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2162   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2163   __ emit_data64(0x0000000000000000, relocInfo::none);
2164   __ emit_data64(0x0000000000000000, relocInfo::none);
2165 
2166   return start;
2167 }
2168 
2169 address StubGenerator::base64_vbmi_join_0_1_addr() {
2170   __ align64();
2171   StubCodeMark mark(this, "StubRoutines", "join_0_1_base64");
2172   address start = __ pc();
2173 
2174   assert(((unsigned long long)start & 0x3f) == 0,
2175          "Alignment problem (0x%08llx)", (unsigned long long)start);
2176   __ emit_data64(0x090a040506000102, relocInfo::none);
2177   __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2178   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2179   __ emit_data64(0x292a242526202122, relocInfo::none);
2180   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2181   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2182   __ emit_data64(0x494a444546404142, relocInfo::none);
2183   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2184 
2185   return start;
2186 }
2187 
2188 address StubGenerator::base64_vbmi_join_1_2_addr() {
2189   __ align64();
2190   StubCodeMark mark(this, "StubRoutines", "join_1_2_base64");
2191   address start = __ pc();
2192 
2193   assert(((unsigned long long)start & 0x3f) == 0,
2194          "Alignment problem (0x%08llx)", (unsigned long long)start);
2195   __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2196   __ emit_data64(0x292a242526202122, relocInfo::none);
2197   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2198   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2199   __ emit_data64(0x494a444546404142, relocInfo::none);
2200   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2201   __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2202   __ emit_data64(0x696a646566606162, relocInfo::none);
2203 
2204   return start;
2205 }
2206 
2207 address StubGenerator::base64_vbmi_join_2_3_addr() {
2208   __ align64();
2209   StubCodeMark mark(this, "StubRoutines", "join_2_3_base64");
2210   address start = __ pc();
2211 
2212   assert(((unsigned long long)start & 0x3f) == 0,
2213          "Alignment problem (0x%08llx)", (unsigned long long)start);
2214   __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2215   __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2216   __ emit_data64(0x494a444546404142, relocInfo::none);
2217   __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2218   __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2219   __ emit_data64(0x696a646566606162, relocInfo::none);
2220   __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2221   __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2222 
2223   return start;
2224 }
2225 
2226 address StubGenerator::base64_AVX2_decode_tables_addr() {
2227   __ align64();
2228   StubCodeMark mark(this, "StubRoutines", "AVX2_tables_base64");
2229   address start = __ pc();
2230 
2231   assert(((unsigned long long)start & 0x3f) == 0,
2232          "Alignment problem (0x%08llx)", (unsigned long long)start);
2233   __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2234   __ emit_data(0x5f5f5f5f, relocInfo::none, 0);  // for URL
2235 
2236   __ emit_data(0xffffffff, relocInfo::none, 0);
2237   __ emit_data(0xfcfcfcfc, relocInfo::none, 0);  // for URL
2238 
2239   // Permute table
2240   __ emit_data64(0x0000000100000000, relocInfo::none);
2241   __ emit_data64(0x0000000400000002, relocInfo::none);
2242   __ emit_data64(0x0000000600000005, relocInfo::none);
2243   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2244 
2245   // Shuffle table
2246   __ emit_data64(0x090a040506000102, relocInfo::none);
2247   __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2248   __ emit_data64(0x090a040506000102, relocInfo::none);
2249   __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2250 
2251   // merge table
2252   __ emit_data(0x01400140, relocInfo::none, 0);
2253 
2254   // merge multiplier
2255   __ emit_data(0x00011000, relocInfo::none, 0);
2256 
2257   return start;
2258 }
2259 
2260 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2261   __ align64();
2262   StubCodeMark mark(this, "StubRoutines", "AVX2_tables_URL_base64");
2263   address start = __ pc();
2264 
2265   assert(((unsigned long long)start & 0x3f) == 0,
2266          "Alignment problem (0x%08llx)", (unsigned long long)start);
2267   // lut_lo
2268   __ emit_data64(0x1111111111111115, relocInfo::none);
2269   __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2270   __ emit_data64(0x1111111111111115, relocInfo::none);
2271   __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2272 
2273   // lut_roll
2274   __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2275   __ emit_data64(0x0000000000000000, relocInfo::none);
2276   __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2277   __ emit_data64(0x0000000000000000, relocInfo::none);
2278 
2279   // lut_lo URL
2280   __ emit_data64(0x1111111111111115, relocInfo::none);
2281   __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2282   __ emit_data64(0x1111111111111115, relocInfo::none);
2283   __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2284 
2285   // lut_roll URL
2286   __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2287   __ emit_data64(0x0000000000000000, relocInfo::none);
2288   __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2289   __ emit_data64(0x0000000000000000, relocInfo::none);
2290 
2291   // lut_hi
2292   __ emit_data64(0x0804080402011010, relocInfo::none);
2293   __ emit_data64(0x1010101010101010, relocInfo::none);
2294   __ emit_data64(0x0804080402011010, relocInfo::none);
2295   __ emit_data64(0x1010101010101010, relocInfo::none);
2296 
2297   return start;
2298 }
2299 
2300 address StubGenerator::base64_decoding_table_addr() {
2301   StubCodeMark mark(this, "StubRoutines", "decoding_table_base64");
2302   address start = __ pc();
2303 
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   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2309   __ emit_data64(0x3fffffff3effffff, relocInfo::none);
2310   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2311   __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2312   __ emit_data64(0x06050403020100ff, relocInfo::none);
2313   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2314   __ emit_data64(0x161514131211100f, relocInfo::none);
2315   __ emit_data64(0xffffffffff191817, relocInfo::none);
2316   __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2317   __ emit_data64(0x2827262524232221, relocInfo::none);
2318   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2319   __ emit_data64(0xffffffffff333231, relocInfo::none);
2320   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2321   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2322   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2323   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2324   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2325   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2326   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2327   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2328   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2329   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2330   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2331   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2332   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2333   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2334   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2335   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2336 
2337   // URL table
2338   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2339   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2340   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2341   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2342   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2343   __ emit_data64(0xffff3effffffffff, relocInfo::none);
2344   __ emit_data64(0x3b3a393837363534, relocInfo::none);
2345   __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2346   __ emit_data64(0x06050403020100ff, relocInfo::none);
2347   __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2348   __ emit_data64(0x161514131211100f, relocInfo::none);
2349   __ emit_data64(0x3fffffffff191817, relocInfo::none);
2350   __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2351   __ emit_data64(0x2827262524232221, relocInfo::none);
2352   __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2353   __ emit_data64(0xffffffffff333231, relocInfo::none);
2354   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2355   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2356   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2357   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2358   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2359   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2360   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2361   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2362   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2363   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2364   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2365   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2366   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2367   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2368   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2369   __ emit_data64(0xffffffffffffffff, relocInfo::none);
2370 
2371   return start;
2372 }
2373 
2374 
2375 // Code for generating Base64 decoding.
2376 //
2377 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2378 //
2379 // Intrinsic function prototype in Base64.java:
2380 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2381 address StubGenerator::generate_base64_decodeBlock() {
2382   __ align(CodeEntryAlignment);
2383   StubCodeMark mark(this, "StubRoutines", "implDecode");
2384   address start = __ pc();
2385 
2386   __ enter();
2387 
2388   // Save callee-saved registers before using them
2389   __ push(r12);
2390   __ push(r13);
2391   __ push(r14);
2392   __ push(r15);
2393   __ push(rbx);
2394 
2395   // arguments
2396   const Register source = c_rarg0; // Source Array
2397   const Register start_offset = c_rarg1; // start offset
2398   const Register end_offset = c_rarg2; // end offset
2399   const Register dest = c_rarg3; // destination array
2400   const Register isMIME = rbx;
2401 
2402 #ifndef _WIN64
2403   const Register dp = c_rarg4;  // Position for writing to dest array
2404   const Register isURL = c_rarg5;// Base64 or URL character set
2405   __ movl(isMIME, Address(rbp, 2 * wordSize));
2406 #else
2407   const Address dp_mem(rbp, 6 * wordSize);  // length is on stack on Win64
2408   const Address isURL_mem(rbp, 7 * wordSize);
2409   const Register isURL = r10;      // pick the volatile windows register
2410   const Register dp = r12;
2411   __ movl(dp, dp_mem);
2412   __ movl(isURL, isURL_mem);
2413   __ movl(isMIME, Address(rbp, 8 * wordSize));
2414 #endif
2415 
2416   const XMMRegister lookup_lo = xmm5;
2417   const XMMRegister lookup_hi = xmm6;
2418   const XMMRegister errorvec = xmm7;
2419   const XMMRegister pack16_op = xmm9;
2420   const XMMRegister pack32_op = xmm8;
2421   const XMMRegister input0 = xmm3;
2422   const XMMRegister input1 = xmm20;
2423   const XMMRegister input2 = xmm21;
2424   const XMMRegister input3 = xmm19;
2425   const XMMRegister join01 = xmm12;
2426   const XMMRegister join12 = xmm11;
2427   const XMMRegister join23 = xmm10;
2428   const XMMRegister translated0 = xmm2;
2429   const XMMRegister translated1 = xmm1;
2430   const XMMRegister translated2 = xmm0;
2431   const XMMRegister translated3 = xmm4;
2432 
2433   const XMMRegister merged0 = xmm2;
2434   const XMMRegister merged1 = xmm1;
2435   const XMMRegister merged2 = xmm0;
2436   const XMMRegister merged3 = xmm4;
2437   const XMMRegister merge_ab_bc0 = xmm2;
2438   const XMMRegister merge_ab_bc1 = xmm1;
2439   const XMMRegister merge_ab_bc2 = xmm0;
2440   const XMMRegister merge_ab_bc3 = xmm4;
2441 
2442   const XMMRegister pack24bits = xmm4;
2443 
2444   const Register length = r14;
2445   const Register output_size = r13;
2446   const Register output_mask = r15;
2447   const KRegister input_mask = k1;
2448 
2449   const XMMRegister input_initial_valid_b64 = xmm0;
2450   const XMMRegister tmp = xmm10;
2451   const XMMRegister mask = xmm0;
2452   const XMMRegister invalid_b64 = xmm1;
2453 
2454   Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2455   Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2456   Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2457 
2458   // calculate length from offsets
2459   __ movl(length, end_offset);
2460   __ subl(length, start_offset);
2461   __ push(dest);          // Save for return value calc
2462 
2463   // If AVX512 VBMI not supported, just compile non-AVX code
2464   if(VM_Version::supports_avx512_vbmi() &&
2465      VM_Version::supports_avx512bw()) {
2466     __ cmpl(length, 31);     // 32-bytes is break-even for AVX-512
2467     __ jcc(Assembler::lessEqual, L_lastChunk);
2468 
2469     __ cmpl(isMIME, 0);
2470     __ jcc(Assembler::notEqual, L_lastChunk);
2471 
2472     // Load lookup tables based on isURL
2473     __ cmpl(isURL, 0);
2474     __ jcc(Assembler::notZero, L_loadURL);
2475 
2476     __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2477     __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2478 
2479     __ BIND(L_continue);
2480 
2481     __ movl(r15, 0x01400140);
2482     __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2483 
2484     __ movl(r15, 0x00011000);
2485     __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2486 
2487     __ cmpl(length, 0xff);
2488     __ jcc(Assembler::lessEqual, L_process64);
2489 
2490     // load masks required for decoding data
2491     __ BIND(L_processdata);
2492     __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2493     __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2494     __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2495 
2496     __ align32();
2497     __ BIND(L_process256);
2498     // Grab input data
2499     __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2500     __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2501     __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2502     __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2503 
2504     // Copy the low part of the lookup table into the destination of the permutation
2505     __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2506     __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2507     __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2508     __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2509 
2510     // Translate the base64 input into "decoded" bytes
2511     __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2512     __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2513     __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2514     __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2515 
2516     // OR all of the translations together to check for errors (high-order bit of byte set)
2517     __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2518 
2519     __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2520     __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2521     __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2522 
2523     // Check if there was an error - if so, try 64-byte chunks
2524     __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2525     __ kortestql(k3, k3);
2526     __ jcc(Assembler::notZero, L_process64);
2527 
2528     // The merging and shuffling happens here
2529     // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2530     // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2531     // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2532     __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2533     __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2534     __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2535     __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2536 
2537     // Now do the same with packed 16-bit values.
2538     // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2539     // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2540     // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2541     __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2542     __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2543     __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2544     __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2545 
2546     // The join vectors specify which byte from which vector goes into the outputs
2547     // One of every 4 bytes in the extended vector is zero, so we pack them into their
2548     // final positions in the register for storing (256 bytes in, 192 bytes out)
2549     __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2550     __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2551     __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2552 
2553     // Store result
2554     __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2555     __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2556     __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2557 
2558     __ addptr(source, 0x100);
2559     __ addptr(dest, 0xc0);
2560     __ subl(length, 0x100);
2561     __ cmpl(length, 64 * 4);
2562     __ jcc(Assembler::greaterEqual, L_process256);
2563 
2564     // At this point, we've decoded 64 * 4 * n bytes.
2565     // The remaining length will be <= 64 * 4 - 1.
2566     // UNLESS there was an error decoding the first 256-byte chunk.  In this
2567     // case, the length will be arbitrarily long.
2568     //
2569     // Note that this will be the path for MIME-encoded strings.
2570 
2571     __ BIND(L_process64);
2572 
2573     __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2574 
2575     __ cmpl(length, 63);
2576     __ jcc(Assembler::lessEqual, L_finalBit);
2577 
2578     __ mov64(rax, 0x0000ffffffffffff);
2579     __ kmovql(k2, rax);
2580 
2581     __ align32();
2582     __ BIND(L_process64Loop);
2583 
2584     // Handle first 64-byte block
2585 
2586     __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2587     __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2588     __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2589 
2590     __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2591 
2592     // Check for error and bomb out before updating dest
2593     __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2594     __ kortestql(k3, k3);
2595     __ jcc(Assembler::notZero, L_exit);
2596 
2597     // Pack output register, selecting correct byte ordering
2598     __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2599     __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2600     __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2601 
2602     __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2603 
2604     __ subl(length, 64);
2605     __ addptr(source, 64);
2606     __ addptr(dest, 48);
2607 
2608     __ cmpl(length, 64);
2609     __ jcc(Assembler::greaterEqual, L_process64Loop);
2610 
2611     __ cmpl(length, 0);
2612     __ jcc(Assembler::lessEqual, L_exit);
2613 
2614     __ BIND(L_finalBit);
2615     // Now have 1 to 63 bytes left to decode
2616 
2617     // I was going to let Java take care of the final fragment
2618     // however it will repeatedly call this routine for every 4 bytes
2619     // of input data, so handle the rest here.
2620     __ movq(rax, -1);
2621     __ bzhiq(rax, rax, length);    // Input mask in rax
2622 
2623     __ movl(output_size, length);
2624     __ shrl(output_size, 2);   // Find (len / 4) * 3 (output length)
2625     __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2626     // output_size in r13
2627 
2628     // Strip pad characters, if any, and adjust length and mask
2629     __ addq(length, start_offset);
2630     __ cmpb(Address(source, length, Address::times_1, -1), '=');
2631     __ jcc(Assembler::equal, L_padding);
2632 
2633     __ BIND(L_donePadding);
2634     __ subq(length, start_offset);
2635 
2636     // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2637     __ kmovql(input_mask, rax);
2638     __ movq(output_mask, -1);
2639     __ bzhiq(output_mask, output_mask, output_size);
2640 
2641     // Load initial input with all valid base64 characters.  Will be used
2642     // in merging source bytes to avoid masking when determining if an error occurred.
2643     __ movl(rax, 0x61616161);
2644     __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2645 
2646     // A register containing all invalid base64 decoded values
2647     __ movl(rax, 0x80808080);
2648     __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2649 
2650     // input_mask is in k1
2651     // output_size is in r13
2652     // output_mask is in r15
2653     // zmm0 - free
2654     // zmm1 - 0x00011000
2655     // zmm2 - 0x01400140
2656     // zmm3 - errorvec
2657     // zmm4 - pack vector
2658     // zmm5 - lookup_lo
2659     // zmm6 - lookup_hi
2660     // zmm7 - errorvec
2661     // zmm8 - 0x61616161
2662     // zmm9 - 0x80808080
2663 
2664     // Load only the bytes from source, merging into our "fully-valid" register
2665     __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2666 
2667     // Decode all bytes within our merged input
2668     __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2669     __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2670     __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2671 
2672     // Check for error.  Compare (decoded | initial) to all invalid.
2673     // If any bytes have their high-order bit set, then we have an error.
2674     __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2675     __ kortestql(k2, k2);
2676 
2677     // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2678     __ jcc(Assembler::notZero, L_bruteForce);
2679 
2680     // Shuffle output bytes
2681     __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2682     __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2683 
2684     __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2685     __ kmovql(k1, output_mask);
2686     __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2687 
2688     __ addptr(dest, output_size);
2689 
2690     __ BIND(L_exit);
2691     __ vzeroupper();
2692     __ pop(rax);             // Get original dest value
2693     __ subptr(dest, rax);      // Number of bytes converted
2694     __ movptr(rax, dest);
2695     __ pop(rbx);
2696     __ pop(r15);
2697     __ pop(r14);
2698     __ pop(r13);
2699     __ pop(r12);
2700     __ leave();
2701     __ ret(0);
2702 
2703     __ BIND(L_loadURL);
2704     __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2705     __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2706     __ jmp(L_continue);
2707 
2708     __ BIND(L_padding);
2709     __ decrementq(output_size, 1);
2710     __ shrq(rax, 1);
2711 
2712     __ cmpb(Address(source, length, Address::times_1, -2), '=');
2713     __ jcc(Assembler::notEqual, L_donePadding);
2714 
2715     __ decrementq(output_size, 1);
2716     __ shrq(rax, 1);
2717     __ jmp(L_donePadding);
2718 
2719     __ align32();
2720     __ BIND(L_bruteForce);
2721   }   // End of if(avx512_vbmi)
2722 
2723   if (VM_Version::supports_avx2()) {
2724     Label L_tailProc, L_topLoop, L_enterLoop;
2725 
2726     __ cmpl(isMIME, 0);
2727     __ jcc(Assembler::notEqual, L_lastChunk);
2728 
2729     // Check for buffer too small (for algorithm)
2730     __ subl(length, 0x2c);
2731     __ jcc(Assembler::less, L_tailProc);
2732 
2733     __ shll(isURL, 2);
2734 
2735     // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2736     // Encoding and Decoding using AVX2 Instructions".  URL modifications added.
2737 
2738     // Set up constants
2739     __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2740     __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit);  // 2F or 5F
2741     __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit);  // -1 or -4
2742     __ vmovdqu(xmm12, Address(r13, 0x10));  // permute
2743     __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2744     __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit);  // merge
2745     __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit);  // merge mult
2746 
2747     __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2748     __ shll(isURL, 4);
2749     __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00));  // lut_lo
2750     __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2751     __ shrl(isURL, 6);  // restore isURL
2752     __ vmovdqu(xmm9, Address(r13, 0x80));  // lut_hi
2753     __ jmp(L_enterLoop);
2754 
2755     __ align32();
2756     __ bind(L_topLoop);
2757     // Add in the offset value (roll) to get 6-bit out values
2758     __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2759     // Merge and permute the output bits into appropriate output byte lanes
2760     __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2761     __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2762     __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2763     __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2764     // Store the output bytes
2765     __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2766     __ addptr(source, 0x20);
2767     __ addptr(dest, 0x18);
2768     __ subl(length, 0x20);
2769     __ jcc(Assembler::less, L_tailProc);
2770 
2771     __ bind(L_enterLoop);
2772 
2773     // Load in encoded string (32 bytes)
2774     __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2775     // Extract the high nibble for indexing into the lut tables.  High 4 bits are don't care.
2776     __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2777     __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2778     // Extract the low nibble. 5F/2F will isolate the low-order 4 bits.  High 4 bits are don't care.
2779     __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2780     // Check for special-case (0x2F or 0x5F (URL))
2781     __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2782     // Get the bitset based on the low nibble.  vpshufb uses low-order 4 bits only.
2783     __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2784     // Get the bit value of the high nibble
2785     __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2786     // Make sure 2F / 5F shows as valid
2787     __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2788     // Make adjustment for roll index.  For non-URL, this is a no-op,
2789     // for URL, this adjusts by -4.  This is to properly index the
2790     // roll value for 2F / 5F.
2791     __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2792     // If the and of the two is non-zero, we have an invalid input character
2793     __ vptest(xmm3, xmm5);
2794     // Extract the "roll" value - value to add to the input to get 6-bit out value
2795     __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2796     __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2797     __ jcc(Assembler::equal, L_topLoop);  // Fall through on error
2798 
2799     __ bind(L_tailProc);
2800 
2801     __ addl(length, 0x2c);
2802 
2803     __ vzeroupper();
2804   }
2805 
2806   // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2807 
2808   // Register state (Linux):
2809   // r12-15 - saved on stack
2810   // rdi - src
2811   // rsi - sp
2812   // rdx - sl
2813   // rcx - dst
2814   // r8 - dp
2815   // r9 - isURL
2816 
2817   // Register state (Windows):
2818   // r12-15 - saved on stack
2819   // rcx - src
2820   // rdx - sp
2821   // r8 - sl
2822   // r9 - dst
2823   // r12 - dp
2824   // r10 - isURL
2825 
2826   // Registers (common):
2827   // length (r14) - bytes in src
2828 
2829   const Register decode_table = r11;
2830   const Register out_byte_count = rbx;
2831   const Register byte1 = r13;
2832   const Register byte2 = r15;
2833   const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2834   const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2835 
2836   __ bind(L_lastChunk);
2837 
2838   __ shrl(length, 2);    // Multiple of 4 bytes only - length is # 4-byte chunks
2839   __ cmpl(length, 0);
2840   __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2841 
2842   __ shll(isURL, 8);    // index into decode table based on isURL
2843   __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2844   __ addptr(decode_table, isURL);
2845 
2846   __ jmp(L_bottomLoop);
2847 
2848   __ align32();
2849   __ BIND(L_forceLoop);
2850   __ shll(byte1, 18);
2851   __ shll(byte2, 12);
2852   __ shll(byte3, 6);
2853   __ orl(byte1, byte2);
2854   __ orl(byte1, byte3);
2855   __ orl(byte1, byte4);
2856 
2857   __ addptr(source, 4);
2858 
2859   __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2860   __ shrl(byte1, 8);
2861   __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2862   __ shrl(byte1, 8);
2863   __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2864 
2865   __ addptr(dest, 3);
2866   __ decrementl(length, 1);
2867   __ jcc(Assembler::zero, L_exit_no_vzero);
2868 
2869   __ BIND(L_bottomLoop);
2870   __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2871   __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2872   __ load_signed_byte(byte1, Address(decode_table, byte1));
2873   __ load_signed_byte(byte2, Address(decode_table, byte2));
2874   __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2875   __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2876   __ load_signed_byte(byte3, Address(decode_table, byte3));
2877   __ load_signed_byte(byte4, Address(decode_table, byte4));
2878 
2879   __ mov(rax, byte1);
2880   __ orl(rax, byte2);
2881   __ orl(rax, byte3);
2882   __ orl(rax, byte4);
2883   __ jcc(Assembler::positive, L_forceLoop);
2884 
2885   __ BIND(L_exit_no_vzero);
2886   __ pop(rax);             // Get original dest value
2887   __ subptr(dest, rax);      // Number of bytes converted
2888   __ movptr(rax, dest);
2889   __ pop(rbx);
2890   __ pop(r15);
2891   __ pop(r14);
2892   __ pop(r13);
2893   __ pop(r12);
2894   __ leave();
2895   __ ret(0);
2896 
2897   return start;
2898 }
2899 
2900 
2901 /**
2902  *  Arguments:
2903  *
2904  * Inputs:
2905  *   c_rarg0   - int crc
2906  *   c_rarg1   - byte* buf
2907  *   c_rarg2   - int length
2908  *
2909  * Output:
2910  *       rax   - int crc result
2911  */
2912 address StubGenerator::generate_updateBytesCRC32() {
2913   assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2914 
2915   __ align(CodeEntryAlignment);
2916   StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2917 
2918   address start = __ pc();
2919 
2920   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
2921   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
2922   // rscratch1: r10
2923   const Register crc   = c_rarg0;  // crc
2924   const Register buf   = c_rarg1;  // source java byte array address
2925   const Register len   = c_rarg2;  // length
2926   const Register table = c_rarg3;  // crc_table address (reuse register)
2927   const Register tmp1   = r11;
2928   const Register tmp2   = r10;
2929   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
2930 
2931   BLOCK_COMMENT("Entry:");
2932   __ enter(); // required for proper stackwalking of RuntimeStub frame
2933 
2934   if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2935       VM_Version::supports_avx512bw() &&
2936       VM_Version::supports_avx512vl()) {
2937       // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
2938       // However, the constant table for CRC32-C assumes the original crc value.  Account for this
2939       // difference before calling and after returning.
2940     __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
2941     __ notl(crc);
2942     __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
2943     __ notl(crc);
2944   } else {
2945     __ kernel_crc32(crc, buf, len, table, tmp1);
2946   }
2947 
2948   __ movl(rax, crc);
2949   __ vzeroupper();
2950   __ leave(); // required for proper stackwalking of RuntimeStub frame
2951   __ ret(0);
2952 
2953   return start;
2954 }
2955 
2956 /**
2957 *  Arguments:
2958 *
2959 * Inputs:
2960 *   c_rarg0   - int crc
2961 *   c_rarg1   - byte* buf
2962 *   c_rarg2   - long length
2963 *   c_rarg3   - table_start - optional (present only when doing a library_call,
2964 *              not used by x86 algorithm)
2965 *
2966 * Output:
2967 *       rax   - int crc result
2968 */
2969 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
2970   assert(UseCRC32CIntrinsics, "need SSE4_2");
2971   __ align(CodeEntryAlignment);
2972   StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
2973   address start = __ pc();
2974 
2975   //reg.arg        int#0        int#1        int#2        int#3        int#4        int#5        float regs
2976   //Windows        RCX          RDX          R8           R9           none         none         XMM0..XMM3
2977   //Lin / Sol      RDI          RSI          RDX          RCX          R8           R9           XMM0..XMM7
2978   const Register crc = c_rarg0;  // crc
2979   const Register buf = c_rarg1;  // source java byte array address
2980   const Register len = c_rarg2;  // length
2981   const Register a = rax;
2982   const Register j = r9;
2983   const Register k = r10;
2984   const Register l = r11;
2985 #ifdef _WIN64
2986   const Register y = rdi;
2987   const Register z = rsi;
2988 #else
2989   const Register y = rcx;
2990   const Register z = r8;
2991 #endif
2992   assert_different_registers(crc, buf, len, a, j, k, l, y, z);
2993 
2994   BLOCK_COMMENT("Entry:");
2995   __ enter(); // required for proper stackwalking of RuntimeStub frame
2996   Label L_continue;
2997 
2998   if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2999       VM_Version::supports_avx512bw() &&
3000       VM_Version::supports_avx512vl()) {
3001     Label L_doSmall;
3002 
3003     __ cmpl(len, 384);
3004     __ jcc(Assembler::lessEqual, L_doSmall);
3005 
3006     __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
3007     __ kernel_crc32_avx512(crc, buf, len, j, l, k);
3008 
3009     __ jmp(L_continue);
3010 
3011     __ bind(L_doSmall);
3012   }
3013 #ifdef _WIN64
3014   __ push(y);
3015   __ push(z);
3016 #endif
3017   __ crc32c_ipl_alg2_alt2(crc, buf, len,
3018                           a, j, k,
3019                           l, y, z,
3020                           c_farg0, c_farg1, c_farg2,
3021                           is_pclmulqdq_supported);
3022 #ifdef _WIN64
3023   __ pop(z);
3024   __ pop(y);
3025 #endif
3026 
3027   __ bind(L_continue);
3028   __ movl(rax, crc);
3029   __ vzeroupper();
3030   __ leave(); // required for proper stackwalking of RuntimeStub frame
3031   __ ret(0);
3032 
3033   return start;
3034 }
3035 
3036 
3037 /**
3038  *  Arguments:
3039  *
3040  *  Input:
3041  *    c_rarg0   - x address
3042  *    c_rarg1   - x length
3043  *    c_rarg2   - y address
3044  *    c_rarg3   - y length
3045  * not Win64
3046  *    c_rarg4   - z address
3047  * Win64
3048  *    rsp+40    - z address
3049  */
3050 address StubGenerator::generate_multiplyToLen() {
3051   __ align(CodeEntryAlignment);
3052   StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3053   address start = __ pc();
3054 
3055   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3056   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3057   const Register x     = rdi;
3058   const Register xlen  = rax;
3059   const Register y     = rsi;
3060   const Register ylen  = rcx;
3061   const Register z     = r8;
3062 
3063   // Next registers will be saved on stack in multiply_to_len().
3064   const Register tmp0  = r11;
3065   const Register tmp1  = r12;
3066   const Register tmp2  = r13;
3067   const Register tmp3  = r14;
3068   const Register tmp4  = r15;
3069   const Register tmp5  = rbx;
3070 
3071   BLOCK_COMMENT("Entry:");
3072   __ enter(); // required for proper stackwalking of RuntimeStub frame
3073 
3074   setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3075                      // ylen => rcx, z => r8
3076                      // r9 and r10 may be used to save non-volatile registers
3077 #ifdef _WIN64
3078   // last argument (#4) is on stack on Win64
3079   __ movptr(z, Address(rsp, 6 * wordSize));
3080 #endif
3081 
3082   __ movptr(xlen, rsi);
3083   __ movptr(y,    rdx);
3084   __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
3085 
3086   restore_arg_regs();
3087 
3088   __ leave(); // required for proper stackwalking of RuntimeStub frame
3089   __ ret(0);
3090 
3091   return start;
3092 }
3093 
3094 /**
3095 *  Arguments:
3096 *
3097 *  Input:
3098 *    c_rarg0   - obja     address
3099 *    c_rarg1   - objb     address
3100 *    c_rarg3   - length   length
3101 *    c_rarg4   - scale    log2_array_indxscale
3102 *
3103 *  Output:
3104 *        rax   - int >= mismatched index, < 0 bitwise complement of tail
3105 */
3106 address StubGenerator::generate_vectorizedMismatch() {
3107   __ align(CodeEntryAlignment);
3108   StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
3109   address start = __ pc();
3110 
3111   BLOCK_COMMENT("Entry:");
3112   __ enter();
3113 
3114 #ifdef _WIN64  // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3115   const Register scale = c_rarg0;  //rcx, will exchange with r9
3116   const Register objb = c_rarg1;   //rdx
3117   const Register length = c_rarg2; //r8
3118   const Register obja = c_rarg3;   //r9
3119   __ xchgq(obja, scale);  //now obja and scale contains the correct contents
3120 
3121   const Register tmp1 = r10;
3122   const Register tmp2 = r11;
3123 #endif
3124 #ifndef _WIN64 // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3125   const Register obja = c_rarg0;   //U:rdi
3126   const Register objb = c_rarg1;   //U:rsi
3127   const Register length = c_rarg2; //U:rdx
3128   const Register scale = c_rarg3;  //U:rcx
3129   const Register tmp1 = r8;
3130   const Register tmp2 = r9;
3131 #endif
3132   const Register result = rax; //return value
3133   const XMMRegister vec0 = xmm0;
3134   const XMMRegister vec1 = xmm1;
3135   const XMMRegister vec2 = xmm2;
3136 
3137   __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3138 
3139   __ vzeroupper();
3140   __ leave();
3141   __ ret(0);
3142 
3143   return start;
3144 }
3145 
3146 /**
3147  *  Arguments:
3148  *
3149 //  Input:
3150 //    c_rarg0   - x address
3151 //    c_rarg1   - x length
3152 //    c_rarg2   - z address
3153 //    c_rarg3   - z length
3154  *
3155  */
3156 address StubGenerator::generate_squareToLen() {
3157 
3158   __ align(CodeEntryAlignment);
3159   StubCodeMark mark(this, "StubRoutines", "squareToLen");
3160   address start = __ pc();
3161 
3162   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3163   // Unix:  rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3164   const Register x      = rdi;
3165   const Register len    = rsi;
3166   const Register z      = r8;
3167   const Register zlen   = rcx;
3168 
3169  const Register tmp1      = r12;
3170  const Register tmp2      = r13;
3171  const Register tmp3      = r14;
3172  const Register tmp4      = r15;
3173  const Register tmp5      = rbx;
3174 
3175   BLOCK_COMMENT("Entry:");
3176   __ enter(); // required for proper stackwalking of RuntimeStub frame
3177 
3178   setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3179                      // zlen => rcx
3180                      // r9 and r10 may be used to save non-volatile registers
3181   __ movptr(r8, rdx);
3182   __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3183 
3184   restore_arg_regs();
3185 
3186   __ leave(); // required for proper stackwalking of RuntimeStub frame
3187   __ ret(0);
3188 
3189   return start;
3190 }
3191 
3192 address StubGenerator::generate_method_entry_barrier() {
3193   __ align(CodeEntryAlignment);
3194   StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3195   address start = __ pc();
3196 
3197   Label deoptimize_label;
3198 
3199   __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3200 
3201   BLOCK_COMMENT("Entry:");
3202   __ enter(); // save rbp
3203 
3204   // save c_rarg0, because we want to use that value.
3205   // We could do without it but then we depend on the number of slots used by pusha
3206   __ push(c_rarg0);
3207 
3208   __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3209 
3210   __ pusha();
3211 
3212   // The method may have floats as arguments, and we must spill them before calling
3213   // the VM runtime.
3214   assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3215   const int xmm_size = wordSize * 2;
3216   const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3217   __ subptr(rsp, xmm_spill_size);
3218   __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3219   __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3220   __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3221   __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3222   __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3223   __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3224   __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3225   __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3226 
3227   __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3228 
3229   __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3230   __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3231   __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3232   __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3233   __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3234   __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3235   __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3236   __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3237   __ addptr(rsp, xmm_spill_size);
3238 
3239   __ cmpl(rax, 1); // 1 means deoptimize
3240   __ jcc(Assembler::equal, deoptimize_label);
3241 
3242   __ popa();
3243   __ pop(c_rarg0);
3244 
3245   __ leave();
3246 
3247   __ addptr(rsp, 1 * wordSize); // cookie
3248   __ ret(0);
3249 
3250 
3251   __ BIND(deoptimize_label);
3252 
3253   __ popa();
3254   __ pop(c_rarg0);
3255 
3256   __ leave();
3257 
3258   // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3259   // here while still having a correct stack is valuable
3260   __ testptr(rsp, Address(rsp, 0));
3261 
3262   __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3263   __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3264 
3265   return start;
3266 }
3267 
3268  /**
3269  *  Arguments:
3270  *
3271  *  Input:
3272  *    c_rarg0   - out address
3273  *    c_rarg1   - in address
3274  *    c_rarg2   - offset
3275  *    c_rarg3   - len
3276  * not Win64
3277  *    c_rarg4   - k
3278  * Win64
3279  *    rsp+40    - k
3280  */
3281 address StubGenerator::generate_mulAdd() {
3282   __ align(CodeEntryAlignment);
3283   StubCodeMark mark(this, "StubRoutines", "mulAdd");
3284   address start = __ pc();
3285 
3286   // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3287   // Unix:  rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3288   const Register out     = rdi;
3289   const Register in      = rsi;
3290   const Register offset  = r11;
3291   const Register len     = rcx;
3292   const Register k       = r8;
3293 
3294   // Next registers will be saved on stack in mul_add().
3295   const Register tmp1  = r12;
3296   const Register tmp2  = r13;
3297   const Register tmp3  = r14;
3298   const Register tmp4  = r15;
3299   const Register tmp5  = rbx;
3300 
3301   BLOCK_COMMENT("Entry:");
3302   __ enter(); // required for proper stackwalking of RuntimeStub frame
3303 
3304   setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3305                      // len => rcx, k => r8
3306                      // r9 and r10 may be used to save non-volatile registers
3307 #ifdef _WIN64
3308   // last argument is on stack on Win64
3309   __ movl(k, Address(rsp, 6 * wordSize));
3310 #endif
3311   __ movptr(r11, rdx);  // move offset in rdx to offset(r11)
3312   __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3313 
3314   restore_arg_regs();
3315 
3316   __ leave(); // required for proper stackwalking of RuntimeStub frame
3317   __ ret(0);
3318 
3319   return start;
3320 }
3321 
3322 address StubGenerator::generate_bigIntegerRightShift() {
3323   __ align(CodeEntryAlignment);
3324   StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
3325   address start = __ pc();
3326 
3327   Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3328   // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3329   const Register newArr = rdi;
3330   const Register oldArr = rsi;
3331   const Register newIdx = rdx;
3332   const Register shiftCount = rcx;  // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3333   const Register totalNumIter = r8;
3334 
3335   // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3336   // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3337   const Register tmp1 = r11;                    // Caller save.
3338   const Register tmp2 = rax;                    // Caller save.
3339   const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9);   // Windows: Callee save. Linux: Caller save.
3340   const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10);  // Windows: Callee save. Linux: Caller save.
3341   const Register tmp5 = r14;                    // Callee save.
3342   const Register tmp6 = r15;
3343 
3344   const XMMRegister x0 = xmm0;
3345   const XMMRegister x1 = xmm1;
3346   const XMMRegister x2 = xmm2;
3347 
3348   BLOCK_COMMENT("Entry:");
3349   __ enter(); // required for proper stackwalking of RuntimeStub frame
3350 
3351 #ifdef _WIN64
3352   setup_arg_regs(4);
3353   // For windows, since last argument is on stack, we need to move it to the appropriate register.
3354   __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3355   // Save callee save registers.
3356   __ push(tmp3);
3357   __ push(tmp4);
3358 #endif
3359   __ push(tmp5);
3360 
3361   // Rename temps used throughout the code.
3362   const Register idx = tmp1;
3363   const Register nIdx = tmp2;
3364 
3365   __ xorl(idx, idx);
3366 
3367   // Start right shift from end of the array.
3368   // For example, if #iteration = 4 and newIdx = 1
3369   // then dest[4] = src[4] >> shiftCount  | src[3] <<< (shiftCount - 32)
3370   // if #iteration = 4 and newIdx = 0
3371   // then dest[3] = src[4] >> shiftCount  | src[3] <<< (shiftCount - 32)
3372   __ movl(idx, totalNumIter);
3373   __ movl(nIdx, idx);
3374   __ addl(nIdx, newIdx);
3375 
3376   // If vectorization is enabled, check if the number of iterations is at least 64
3377   // If not, then go to ShifTwo processing 2 iterations
3378   if (VM_Version::supports_avx512_vbmi2()) {
3379     __ cmpptr(totalNumIter, (AVX3Threshold/64));
3380     __ jcc(Assembler::less, ShiftTwo);
3381 
3382     if (AVX3Threshold < 16 * 64) {
3383       __ cmpl(totalNumIter, 16);
3384       __ jcc(Assembler::less, ShiftTwo);
3385     }
3386     __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3387     __ subl(idx, 16);
3388     __ subl(nIdx, 16);
3389     __ BIND(Shift512Loop);
3390     __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
3391     __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3392     __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
3393     __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
3394     __ subl(nIdx, 16);
3395     __ subl(idx, 16);
3396     __ jcc(Assembler::greaterEqual, Shift512Loop);
3397     __ addl(idx, 16);
3398     __ addl(nIdx, 16);
3399   }
3400   __ BIND(ShiftTwo);
3401   __ cmpl(idx, 2);
3402   __ jcc(Assembler::less, ShiftOne);
3403   __ subl(idx, 2);
3404   __ subl(nIdx, 2);
3405   __ BIND(ShiftTwoLoop);
3406   __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
3407   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3408   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3409   __ shrdl(tmp5, tmp4);
3410   __ shrdl(tmp4, tmp3);
3411   __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
3412   __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3413   __ subl(nIdx, 2);
3414   __ subl(idx, 2);
3415   __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3416   __ addl(idx, 2);
3417   __ addl(nIdx, 2);
3418 
3419   // Do the last iteration
3420   __ BIND(ShiftOne);
3421   __ cmpl(idx, 1);
3422   __ jcc(Assembler::less, Exit);
3423   __ subl(idx, 1);
3424   __ subl(nIdx, 1);
3425   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3426   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3427   __ shrdl(tmp4, tmp3);
3428   __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3429   __ BIND(Exit);
3430   __ vzeroupper();
3431   // Restore callee save registers.
3432   __ pop(tmp5);
3433 #ifdef _WIN64
3434   __ pop(tmp4);
3435   __ pop(tmp3);
3436   restore_arg_regs();
3437 #endif
3438   __ leave(); // required for proper stackwalking of RuntimeStub frame
3439   __ ret(0);
3440 
3441   return start;
3442 }
3443 
3444  /**
3445  *  Arguments:
3446  *
3447  *  Input:
3448  *    c_rarg0   - newArr address
3449  *    c_rarg1   - oldArr address
3450  *    c_rarg2   - newIdx
3451  *    c_rarg3   - shiftCount
3452  * not Win64
3453  *    c_rarg4   - numIter
3454  * Win64
3455  *    rsp40    - numIter
3456  */
3457 address StubGenerator::generate_bigIntegerLeftShift() {
3458   __ align(CodeEntryAlignment);
3459   StubCodeMark mark(this,  "StubRoutines", "bigIntegerLeftShiftWorker");
3460   address start = __ pc();
3461 
3462   Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3463   // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3464   const Register newArr = rdi;
3465   const Register oldArr = rsi;
3466   const Register newIdx = rdx;
3467   const Register shiftCount = rcx;  // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3468   const Register totalNumIter = r8;
3469   // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3470   // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3471   const Register tmp1 = r11;                    // Caller save.
3472   const Register tmp2 = rax;                    // Caller save.
3473   const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9);   // Windows: Callee save. Linux: Caller save.
3474   const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10);  // Windows: Callee save. Linux: Caller save.
3475   const Register tmp5 = r14;                    // Callee save.
3476 
3477   const XMMRegister x0 = xmm0;
3478   const XMMRegister x1 = xmm1;
3479   const XMMRegister x2 = xmm2;
3480   BLOCK_COMMENT("Entry:");
3481   __ enter(); // required for proper stackwalking of RuntimeStub frame
3482 
3483 #ifdef _WIN64
3484   setup_arg_regs(4);
3485   // For windows, since last argument is on stack, we need to move it to the appropriate register.
3486   __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3487   // Save callee save registers.
3488   __ push(tmp3);
3489   __ push(tmp4);
3490 #endif
3491   __ push(tmp5);
3492 
3493   // Rename temps used throughout the code
3494   const Register idx = tmp1;
3495   const Register numIterTmp = tmp2;
3496 
3497   // Start idx from zero.
3498   __ xorl(idx, idx);
3499   // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
3500   __ lea(newArr, Address(newArr, newIdx, Address::times_4));
3501   __ movl(numIterTmp, totalNumIter);
3502 
3503   // If vectorization is enabled, check if the number of iterations is at least 64
3504   // If not, then go to ShiftTwo shifting two numbers at a time
3505   if (VM_Version::supports_avx512_vbmi2()) {
3506     __ cmpl(totalNumIter, (AVX3Threshold/64));
3507     __ jcc(Assembler::less, ShiftTwo);
3508 
3509     if (AVX3Threshold < 16 * 64) {
3510       __ cmpl(totalNumIter, 16);
3511       __ jcc(Assembler::less, ShiftTwo);
3512     }
3513     __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3514     __ subl(numIterTmp, 16);
3515     __ BIND(Shift512Loop);
3516     __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3517     __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
3518     __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
3519     __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
3520     __ addl(idx, 16);
3521     __ subl(numIterTmp, 16);
3522     __ jcc(Assembler::greaterEqual, Shift512Loop);
3523     __ addl(numIterTmp, 16);
3524   }
3525   __ BIND(ShiftTwo);
3526   __ cmpl(totalNumIter, 1);
3527   __ jcc(Assembler::less, Exit);
3528   __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3529   __ subl(numIterTmp, 2);
3530   __ jcc(Assembler::less, ShiftOne);
3531 
3532   __ BIND(ShiftTwoLoop);
3533   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3534   __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
3535   __ shldl(tmp3, tmp4);
3536   __ shldl(tmp4, tmp5);
3537   __ movl(Address(newArr, idx, Address::times_4), tmp3);
3538   __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
3539   __ movl(tmp3, tmp5);
3540   __ addl(idx, 2);
3541   __ subl(numIterTmp, 2);
3542   __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3543 
3544   // Do the last iteration
3545   __ BIND(ShiftOne);
3546   __ addl(numIterTmp, 2);
3547   __ cmpl(numIterTmp, 1);
3548   __ jcc(Assembler::less, Exit);
3549   __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3550   __ shldl(tmp3, tmp4);
3551   __ movl(Address(newArr, idx, Address::times_4), tmp3);
3552 
3553   __ BIND(Exit);
3554   __ vzeroupper();
3555   // Restore callee save registers.
3556   __ pop(tmp5);
3557 #ifdef _WIN64
3558   __ pop(tmp4);
3559   __ pop(tmp3);
3560   restore_arg_regs();
3561 #endif
3562   __ leave(); // required for proper stackwalking of RuntimeStub frame
3563   __ ret(0);
3564 
3565   return start;
3566 }
3567 
3568 void StubGenerator::generate_libm_stubs() {
3569   if (UseLibmIntrinsic && InlineIntrinsics) {
3570     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3571       StubRoutines::_dsin = generate_libmSin(); // from stubGenerator_x86_64_sin.cpp
3572     }
3573     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3574       StubRoutines::_dcos = generate_libmCos(); // from stubGenerator_x86_64_cos.cpp
3575     }
3576     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3577       StubRoutines::_dtan = generate_libmTan(); // from stubGenerator_x86_64_tan.cpp
3578     }
3579     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3580       StubRoutines::_dexp = generate_libmExp(); // from stubGenerator_x86_64_exp.cpp
3581     }
3582     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3583       StubRoutines::_dpow = generate_libmPow(); // from stubGenerator_x86_64_pow.cpp
3584     }
3585     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3586       StubRoutines::_dlog = generate_libmLog(); // from stubGenerator_x86_64_log.cpp
3587     }
3588     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3589       StubRoutines::_dlog10 = generate_libmLog10(); // from stubGenerator_x86_64_log.cpp
3590     }
3591   }
3592 }
3593 
3594 /**
3595 *  Arguments:
3596 *
3597 *  Input:
3598 *    c_rarg0   - float16  jshort
3599 *
3600 *  Output:
3601 *       xmm0   - float
3602 */
3603 address StubGenerator::generate_float16ToFloat() {
3604   StubCodeMark mark(this, "StubRoutines", "float16ToFloat");
3605 
3606   address start = __ pc();
3607 
3608   BLOCK_COMMENT("Entry:");
3609   // No need for RuntimeStub frame since it is called only during JIT compilation
3610 
3611   // Load value into xmm0 and convert
3612   __ flt16_to_flt(xmm0, c_rarg0);
3613 
3614   __ ret(0);
3615 
3616   return start;
3617 }
3618 
3619 /**
3620 *  Arguments:
3621 *
3622 *  Input:
3623 *       xmm0   - float
3624 *
3625 *  Output:
3626 *        rax   - float16  jshort
3627 */
3628 address StubGenerator::generate_floatToFloat16() {
3629   StubCodeMark mark(this, "StubRoutines", "floatToFloat16");
3630 
3631   address start = __ pc();
3632 
3633   BLOCK_COMMENT("Entry:");
3634   // No need for RuntimeStub frame since it is called only during JIT compilation
3635 
3636   // Convert and put result into rax
3637   __ flt_to_flt16(rax, xmm0, xmm1);
3638 
3639   __ ret(0);
3640 
3641   return start;
3642 }
3643 
3644 address StubGenerator::generate_cont_thaw(const char* label, Continuation::thaw_kind kind) {
3645   if (!Continuations::enabled()) return nullptr;
3646 
3647   bool return_barrier = Continuation::is_thaw_return_barrier(kind);
3648   bool return_barrier_exception = Continuation::is_thaw_return_barrier_exception(kind);
3649 
3650   StubCodeMark mark(this, "StubRoutines", label);
3651   address start = __ pc();
3652 
3653   // TODO: Handle Valhalla return types. May require generating different return barriers.
3654 
3655   if (!return_barrier) {
3656     // Pop return address. If we don't do this, we get a drift,
3657     // where the bottom-most frozen frame continuously grows.
3658     __ pop(c_rarg3);
3659   } else {
3660     __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3661   }
3662 
3663 #ifdef ASSERT
3664   {
3665     Label L_good_sp;
3666     __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3667     __ jcc(Assembler::equal, L_good_sp);
3668     __ stop("Incorrect rsp at thaw entry");
3669     __ BIND(L_good_sp);
3670   }
3671 #endif // ASSERT
3672 
3673   if (return_barrier) {
3674     // Preserve possible return value from a method returning to the return barrier.
3675     __ push(rax);
3676     __ push_d(xmm0);
3677   }
3678 
3679   __ movptr(c_rarg0, r15_thread);
3680   __ movptr(c_rarg1, (return_barrier ? 1 : 0));
3681   __ call_VM_leaf(CAST_FROM_FN_PTR(address, Continuation::prepare_thaw), 2);
3682   __ movptr(rbx, rax);
3683 
3684   if (return_barrier) {
3685     // Restore return value from a method returning to the return barrier.
3686     // No safepoint in the call to thaw, so even an oop return value should be OK.
3687     __ pop_d(xmm0);
3688     __ pop(rax);
3689   }
3690 
3691 #ifdef ASSERT
3692   {
3693     Label L_good_sp;
3694     __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3695     __ jcc(Assembler::equal, L_good_sp);
3696     __ stop("Incorrect rsp after prepare thaw");
3697     __ BIND(L_good_sp);
3698   }
3699 #endif // ASSERT
3700 
3701   // rbx contains the size of the frames to thaw, 0 if overflow or no more frames
3702   Label L_thaw_success;
3703   __ testptr(rbx, rbx);
3704   __ jccb(Assembler::notZero, L_thaw_success);
3705   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
3706   __ bind(L_thaw_success);
3707 
3708   // Make room for the thawed frames and align the stack.
3709   __ subptr(rsp, rbx);
3710   __ andptr(rsp, -StackAlignmentInBytes);
3711 
3712   if (return_barrier) {
3713     // Preserve possible return value from a method returning to the return barrier. (Again.)
3714     __ push(rax);
3715     __ push_d(xmm0);
3716   }
3717 
3718   // If we want, we can templatize thaw by kind, and have three different entries.
3719   __ movptr(c_rarg0, r15_thread);
3720   __ movptr(c_rarg1, kind);
3721   __ call_VM_leaf(Continuation::thaw_entry(), 2);
3722   __ movptr(rbx, rax);
3723 
3724   if (return_barrier) {
3725     // Restore return value from a method returning to the return barrier. (Again.)
3726     // No safepoint in the call to thaw, so even an oop return value should be OK.
3727     __ pop_d(xmm0);
3728     __ pop(rax);
3729   } else {
3730     // Return 0 (success) from doYield.
3731     __ xorptr(rax, rax);
3732   }
3733 
3734   // After thawing, rbx is the SP of the yielding frame.
3735   // Move there, and then to saved RBP slot.
3736   __ movptr(rsp, rbx);
3737   __ subptr(rsp, 2*wordSize);
3738 
3739   if (return_barrier_exception) {
3740     __ movptr(c_rarg0, r15_thread);
3741     __ movptr(c_rarg1, Address(rsp, wordSize)); // return address
3742 
3743     // rax still holds the original exception oop, save it before the call
3744     __ push(rax);
3745 
3746     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 2);
3747     __ movptr(rbx, rax);
3748 
3749     // Continue at exception handler:
3750     //   rax: exception oop
3751     //   rbx: exception handler
3752     //   rdx: exception pc
3753     __ pop(rax);
3754     __ verify_oop(rax);
3755     __ pop(rbp); // pop out RBP here too
3756     __ pop(rdx);
3757     __ jmp(rbx);
3758   } else {
3759     // We are "returning" into the topmost thawed frame; see Thaw::push_return_frame
3760     __ pop(rbp);
3761     __ ret(0);
3762   }
3763 
3764   return start;
3765 }
3766 
3767 address StubGenerator::generate_cont_thaw() {
3768   return generate_cont_thaw("Cont thaw", Continuation::thaw_top);
3769 }
3770 
3771 // TODO: will probably need multiple return barriers depending on return type
3772 
3773 address StubGenerator::generate_cont_returnBarrier() {
3774   return generate_cont_thaw("Cont thaw return barrier", Continuation::thaw_return_barrier);
3775 }
3776 
3777 address StubGenerator::generate_cont_returnBarrier_exception() {
3778   return generate_cont_thaw("Cont thaw return barrier exception", Continuation::thaw_return_barrier_exception);
3779 }
3780 
3781 #if INCLUDE_JFR
3782 
3783 // For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
3784 // It returns a jobject handle to the event writer.
3785 // The handle is dereferenced and the return value is the event writer oop.
3786 RuntimeStub* StubGenerator::generate_jfr_write_checkpoint() {
3787   enum layout {
3788     rbp_off,
3789     rbpH_off,
3790     return_off,
3791     return_off2,
3792     framesize // inclusive of return address
3793   };
3794 
3795   CodeBuffer code("jfr_write_checkpoint", 1024, 64);
3796   MacroAssembler* _masm = new MacroAssembler(&code);
3797   address start = __ pc();
3798 
3799   __ enter();
3800   address the_pc = __ pc();
3801 
3802   int frame_complete = the_pc - start;
3803 
3804   __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3805   __ movptr(c_rarg0, r15_thread);
3806   __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
3807   __ reset_last_Java_frame(true);
3808 
3809   // rax is jobject handle result, unpack and process it through a barrier.
3810   __ resolve_global_jobject(rax, r15_thread, c_rarg0);
3811 
3812   __ leave();
3813   __ ret(0);
3814 
3815   OopMapSet* oop_maps = new OopMapSet();
3816   OopMap* map = new OopMap(framesize, 1);
3817   oop_maps->add_gc_map(frame_complete, map);
3818 
3819   RuntimeStub* stub =
3820     RuntimeStub::new_runtime_stub(code.name(),
3821                                   &code,
3822                                   frame_complete,
3823                                   (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3824                                   oop_maps,
3825                                   false);
3826   return stub;
3827 }
3828 
3829 // For c2: call to return a leased buffer.
3830 RuntimeStub* StubGenerator::generate_jfr_return_lease() {
3831   enum layout {
3832     rbp_off,
3833     rbpH_off,
3834     return_off,
3835     return_off2,
3836     framesize // inclusive of return address
3837   };
3838 
3839   CodeBuffer code("jfr_return_lease", 1024, 64);
3840   MacroAssembler* _masm = new MacroAssembler(&code);
3841   address start = __ pc();
3842 
3843   __ enter();
3844   address the_pc = __ pc();
3845 
3846   int frame_complete = the_pc - start;
3847 
3848   __ set_last_Java_frame(rsp, rbp, the_pc, rscratch2);
3849   __ movptr(c_rarg0, r15_thread);
3850   __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::return_lease), 1);
3851   __ reset_last_Java_frame(true);
3852 
3853   __ leave();
3854   __ ret(0);
3855 
3856   OopMapSet* oop_maps = new OopMapSet();
3857   OopMap* map = new OopMap(framesize, 1);
3858   oop_maps->add_gc_map(frame_complete, map);
3859 
3860   RuntimeStub* stub =
3861     RuntimeStub::new_runtime_stub(code.name(),
3862                                   &code,
3863                                   frame_complete,
3864                                   (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3865                                   oop_maps,
3866                                   false);
3867   return stub;
3868 }
3869 
3870 #endif // INCLUDE_JFR
3871 
3872 // Continuation point for throwing of implicit exceptions that are
3873 // not handled in the current activation. Fabricates an exception
3874 // oop and initiates normal exception dispatching in this
3875 // frame. Since we need to preserve callee-saved values (currently
3876 // only for C2, but done for C1 as well) we need a callee-saved oop
3877 // map and therefore have to make these stubs into RuntimeStubs
3878 // rather than BufferBlobs.  If the compiler needs all registers to
3879 // be preserved between the fault point and the exception handler
3880 // then it must assume responsibility for that in
3881 // AbstractCompiler::continuation_for_implicit_null_exception or
3882 // continuation_for_implicit_division_by_zero_exception. All other
3883 // implicit exceptions (e.g., NullPointerException or
3884 // AbstractMethodError on entry) are either at call sites or
3885 // otherwise assume that stack unwinding will be initiated, so
3886 // caller saved registers were assumed volatile in the compiler.
3887 address StubGenerator::generate_throw_exception(const char* name,
3888                                                 address runtime_entry,
3889                                                 Register arg1,
3890                                                 Register arg2) {
3891   // Information about frame layout at time of blocking runtime call.
3892   // Note that we only have to preserve callee-saved registers since
3893   // the compilers are responsible for supplying a continuation point
3894   // if they expect all registers to be preserved.
3895   enum layout {
3896     rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3897     rbp_off2,
3898     return_off,
3899     return_off2,
3900     framesize // inclusive of return address
3901   };
3902 
3903   int insts_size = 512;
3904   int locs_size  = 64;
3905 
3906   CodeBuffer code(name, insts_size, locs_size);
3907   OopMapSet* oop_maps  = new OopMapSet();
3908   MacroAssembler* _masm = new MacroAssembler(&code);
3909 
3910   address start = __ pc();
3911 
3912   // This is an inlined and slightly modified version of call_VM
3913   // which has the ability to fetch the return PC out of
3914   // thread-local storage and also sets up last_Java_sp slightly
3915   // differently than the real call_VM
3916 
3917   __ enter(); // required for proper stackwalking of RuntimeStub frame
3918 
3919   assert(is_even(framesize/2), "sp not 16-byte aligned");
3920 
3921   // return address and rbp are already in place
3922   __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3923 
3924   int frame_complete = __ pc() - start;
3925 
3926   // Set up last_Java_sp and last_Java_fp
3927   address the_pc = __ pc();
3928   __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3929   __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
3930 
3931   // Call runtime
3932   if (arg1 != noreg) {
3933     assert(arg2 != c_rarg1, "clobbered");
3934     __ movptr(c_rarg1, arg1);
3935   }
3936   if (arg2 != noreg) {
3937     __ movptr(c_rarg2, arg2);
3938   }
3939   __ movptr(c_rarg0, r15_thread);
3940   BLOCK_COMMENT("call runtime_entry");
3941   __ call(RuntimeAddress(runtime_entry));
3942 
3943   // Generate oop map
3944   OopMap* map = new OopMap(framesize, 0);
3945 
3946   oop_maps->add_gc_map(the_pc - start, map);
3947 
3948   __ reset_last_Java_frame(true);
3949 
3950   __ leave(); // required for proper stackwalking of RuntimeStub frame
3951 
3952   // check for pending exceptions
3953 #ifdef ASSERT
3954   Label L;
3955   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
3956   __ jcc(Assembler::notEqual, L);
3957   __ should_not_reach_here();
3958   __ bind(L);
3959 #endif // ASSERT
3960   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3961 
3962 
3963   // codeBlob framesize is in words (not VMRegImpl::slot_size)
3964   RuntimeStub* stub =
3965     RuntimeStub::new_runtime_stub(name,
3966                                   &code,
3967                                   frame_complete,
3968                                   (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3969                                   oop_maps, false);
3970   return stub->entry_point();
3971 }
3972 
3973 // exception handler for upcall stubs
3974 address StubGenerator::generate_upcall_stub_exception_handler() {
3975   StubCodeMark mark(this, "StubRoutines", "upcall stub exception handler");
3976   address start = __ pc();
3977 
3978   // native caller has no idea how to handle exceptions
3979   // we just crash here. Up to callee to catch exceptions.
3980   __ verify_oop(rax);
3981   __ vzeroupper();
3982   __ mov(c_rarg0, rax);
3983   __ andptr(rsp, -StackAlignmentInBytes); // align stack as required by ABI
3984   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
3985   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, UpcallLinker::handle_uncaught_exception)));
3986   __ should_not_reach_here();
3987 
3988   return start;
3989 }
3990 
3991 address StubGenerator::generate_lookup_secondary_supers_table_stub(u1 super_klass_index) {
3992   StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table");
3993 
3994   address start = __ pc();
3995 
3996   const Register
3997       r_super_klass = rax,
3998       r_sub_klass   = rsi,
3999       result        = rdi;
4000 
4001   __ lookup_secondary_supers_table(r_sub_klass, r_super_klass,
4002                                    rdx, rcx, rbx, r11, // temps
4003                                    result,
4004                                    super_klass_index);
4005   __ ret(0);
4006 
4007   return start;
4008 }
4009 
4010 // Slow path implementation for UseSecondarySupersTable.
4011 address StubGenerator::generate_lookup_secondary_supers_table_slow_path_stub() {
4012   StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table");
4013 
4014   address start = __ pc();
4015 
4016   const Register
4017       r_super_klass  = rax,
4018       r_array_base   = rbx,
4019       r_array_index  = rdx,
4020       r_sub_klass    = rsi,
4021       r_bitmap       = r11,
4022       result         = rdi;
4023 
4024   Label L_success;
4025   __ lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index, r_bitmap,
4026                                              rcx, rdi, // temps
4027                                              &L_success);
4028   // bind(L_failure);
4029   __ movl(result, 1);
4030   __ ret(0);
4031 
4032   __ bind(L_success);
4033   __ movl(result, 0);
4034   __ ret(0);
4035 
4036   return start;
4037 }
4038 
4039 void StubGenerator::create_control_words() {
4040   // Round to nearest, 64-bit mode, exceptions masked, flags specialized
4041   StubRoutines::x86::_mxcsr_std = EnableX86ECoreOpts ? 0x1FBF : 0x1F80;
4042   // Round to zero, 64-bit mode, exceptions masked, flags specialized
4043   StubRoutines::x86::_mxcsr_rz = EnableX86ECoreOpts ? 0x7FBF : 0x7F80;
4044 }
4045 
4046 // Initialization
4047 void StubGenerator::generate_initial_stubs() {
4048   // Generates all stubs and initializes the entry points
4049 
4050   // This platform-specific settings are needed by generate_call_stub()
4051   create_control_words();
4052 
4053   // Initialize table for unsafe copy memeory check.
4054   if (UnsafeMemoryAccess::_table == nullptr) {
4055     UnsafeMemoryAccess::create_table(16 + 4); // 16 for copyMemory; 4 for setMemory
4056   }
4057 
4058   // entry points that exist in all platforms Note: This is code
4059   // that could be shared among different platforms - however the
4060   // benefit seems to be smaller than the disadvantage of having a
4061   // much more complicated generator structure. See also comment in
4062   // stubRoutines.hpp.
4063 
4064   StubRoutines::_forward_exception_entry = generate_forward_exception();
4065 
4066   StubRoutines::_call_stub_entry =
4067     generate_call_stub(StubRoutines::_call_stub_return_address);
4068 
4069   // is referenced by megamorphic call
4070   StubRoutines::_catch_exception_entry = generate_catch_exception();
4071 
4072   // atomic calls
4073   StubRoutines::_fence_entry                = generate_orderaccess_fence();
4074 
4075   // platform dependent
4076   StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4077 
4078   StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
4079 
4080   StubRoutines::x86::_f2i_fixup             = generate_f2i_fixup();
4081   StubRoutines::x86::_f2l_fixup             = generate_f2l_fixup();
4082   StubRoutines::x86::_d2i_fixup             = generate_d2i_fixup();
4083   StubRoutines::x86::_d2l_fixup             = generate_d2l_fixup();
4084 
4085   StubRoutines::x86::_float_sign_mask       = generate_fp_mask("float_sign_mask",  0x7FFFFFFF7FFFFFFF);
4086   StubRoutines::x86::_float_sign_flip       = generate_fp_mask("float_sign_flip",  0x8000000080000000);
4087   StubRoutines::x86::_double_sign_mask      = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4088   StubRoutines::x86::_double_sign_flip      = generate_fp_mask("double_sign_flip", 0x8000000000000000);
4089 
4090   // Build this early so it's available for the interpreter.
4091   StubRoutines::_throw_StackOverflowError_entry =
4092     generate_throw_exception("StackOverflowError throw_exception",
4093                              CAST_FROM_FN_PTR(address,
4094                                               SharedRuntime::
4095                                               throw_StackOverflowError));
4096   StubRoutines::_throw_delayed_StackOverflowError_entry =
4097     generate_throw_exception("delayed StackOverflowError throw_exception",
4098                              CAST_FROM_FN_PTR(address,
4099                                               SharedRuntime::
4100                                               throw_delayed_StackOverflowError));
4101   if (UseCRC32Intrinsics) {
4102     // set table address before stub generation which use it
4103     StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4104     StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4105   }
4106 
4107   if (UseCRC32CIntrinsics) {
4108     bool supports_clmul = VM_Version::supports_clmul();
4109     StubRoutines::x86::generate_CRC32C_table(supports_clmul);
4110     StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
4111     StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
4112   }
4113 
4114   if (VM_Version::supports_float16()) {
4115     // For results consistency both intrinsics should be enabled.
4116     // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
4117     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
4118         vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
4119       StubRoutines::_hf2f = generate_float16ToFloat();
4120       StubRoutines::_f2hf = generate_floatToFloat16();
4121     }
4122   }
4123 
4124   generate_libm_stubs();
4125 
4126   StubRoutines::_fmod = generate_libmFmod(); // from stubGenerator_x86_64_fmod.cpp
4127 }
4128 
4129 void StubGenerator::generate_continuation_stubs() {
4130   // Continuation stubs:
4131   StubRoutines::_cont_thaw          = generate_cont_thaw();
4132   StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
4133   StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
4134 
4135   JFR_ONLY(generate_jfr_stubs();)
4136 }
4137 
4138 #if INCLUDE_JFR
4139 void StubGenerator::generate_jfr_stubs() {
4140   StubRoutines::_jfr_write_checkpoint_stub = generate_jfr_write_checkpoint();
4141   StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();
4142   StubRoutines::_jfr_return_lease_stub = generate_jfr_return_lease();
4143   StubRoutines::_jfr_return_lease = StubRoutines::_jfr_return_lease_stub->entry_point();
4144 }
4145 #endif
4146 
4147 void StubGenerator::generate_final_stubs() {
4148   // Generates the rest of stubs and initializes the entry points
4149 
4150   // These entry points require SharedInfo::stack0 to be set up in
4151   // non-core builds and need to be relocatable, so they each
4152   // fabricate a RuntimeStub internally.
4153   StubRoutines::_throw_AbstractMethodError_entry =
4154     generate_throw_exception("AbstractMethodError throw_exception",
4155                              CAST_FROM_FN_PTR(address,
4156                                               SharedRuntime::
4157                                               throw_AbstractMethodError));
4158 
4159   StubRoutines::_throw_IncompatibleClassChangeError_entry =
4160     generate_throw_exception("IncompatibleClassChangeError throw_exception",
4161                              CAST_FROM_FN_PTR(address,
4162                                               SharedRuntime::
4163                                               throw_IncompatibleClassChangeError));
4164 
4165   StubRoutines::_throw_NullPointerException_at_call_entry =
4166     generate_throw_exception("NullPointerException at call throw_exception",
4167                              CAST_FROM_FN_PTR(address,
4168                                               SharedRuntime::
4169                                               throw_NullPointerException_at_call));
4170 
4171   // support for verify_oop (must happen after universe_init)
4172   if (VerifyOops) {
4173     StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4174   }
4175 
4176   // data cache line writeback
4177   StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4178   StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4179 
4180   // arraycopy stubs used by compilers
4181   generate_arraycopy_stubs();
4182 
4183   BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4184   if (bs_nm != nullptr) {
4185     StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
4186   }
4187 
4188   if (UseVectorizedMismatchIntrinsic) {
4189     StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4190   }
4191 
4192   StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
4193 }
4194 
4195 void StubGenerator::generate_compiler_stubs() {
4196 #if COMPILER2_OR_JVMCI
4197 
4198   // Entry points that are C2 compiler specific.
4199 
4200   StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4201   StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
4202   StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4203   StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
4204   StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFFFFFFFFFF);
4205   StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask("vector_int_mask_cmp_bits", 0x0000000100000001);
4206   StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
4207   StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
4208   StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff000000ff);
4209   StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff0000ffff);
4210   StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
4211                                                                       0xFFFFFFFF, 0, 0, 0);
4212   StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
4213                                                                       0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4214   StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x0302010003020100);
4215   StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask("vector_byte_shuffle_mask");
4216   StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x0100010001000100);
4217   StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask("vector_long_shuffle_mask", 0x0000000100000000);
4218   StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
4219   StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
4220   StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut("count_leading_zeros_lut");
4221   StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut("reverse_bit_lut");
4222   StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long("perm_mask_long");
4223   StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int("perm_mask_int");
4224   StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short("perm_mask_short");
4225 
4226   if (VM_Version::supports_avx2() && !VM_Version::supports_avx512vl()) {
4227     StubRoutines::x86::_compress_perm_table32 = generate_compress_perm_table("compress_perm_table32", 32);
4228     StubRoutines::x86::_compress_perm_table64 = generate_compress_perm_table("compress_perm_table64", 64);
4229     StubRoutines::x86::_expand_perm_table32 = generate_expand_perm_table("expand_perm_table32", 32);
4230     StubRoutines::x86::_expand_perm_table64 = generate_expand_perm_table("expand_perm_table64", 64);
4231   }
4232 
4233   if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4234     // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4235     StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut("popcount_lut");
4236   }
4237 
4238   generate_aes_stubs();
4239 
4240   generate_ghash_stubs();
4241 
4242   generate_chacha_stubs();
4243 
4244 #ifdef COMPILER2
4245   if ((UseAVX == 2) && EnableX86ECoreOpts && !UseCompactObjectHeaders) {
4246     generate_string_indexof(StubRoutines::_string_indexof_array);
4247   }
4248 #endif
4249 
4250   if (UseAdler32Intrinsics) {
4251      StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4252   }
4253 
4254   if (UsePoly1305Intrinsics) {
4255     StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4256   }
4257 
4258   if (UseIntPolyIntrinsics) {
4259     StubRoutines::_intpoly_montgomeryMult_P256 = generate_intpoly_montgomeryMult_P256();
4260     StubRoutines::_intpoly_assign = generate_intpoly_assign();
4261   }
4262 
4263   if (UseMD5Intrinsics) {
4264     StubRoutines::_md5_implCompress = generate_md5_implCompress(false, "md5_implCompress");
4265     StubRoutines::_md5_implCompressMB = generate_md5_implCompress(true, "md5_implCompressMB");
4266   }
4267 
4268   if (UseSHA1Intrinsics) {
4269     StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4270     StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4271     StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
4272     StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
4273   }
4274 
4275   if (UseSHA256Intrinsics) {
4276     StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4277     char* dst = (char*)StubRoutines::x86::_k256_W;
4278     char* src = (char*)StubRoutines::x86::_k256;
4279     for (int ii = 0; ii < 16; ++ii) {
4280       memcpy(dst + 32 * ii,      src + 16 * ii, 16);
4281       memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4282     }
4283     StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4284     StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4285     StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4286     StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4287   }
4288 
4289   if (UseSHA512Intrinsics) {
4290     StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4291     StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4292     StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
4293     StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
4294   }
4295 
4296   if (UseBASE64Intrinsics) {
4297     if(VM_Version::supports_avx2()) {
4298       StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4299       StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4300       StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4301       StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4302       StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4303     }
4304     StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4305     if (VM_Version::supports_avx512_vbmi()) {
4306       StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4307       StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4308       StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4309       StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4310       StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4311       StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4312       StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4313       StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4314       StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4315     }
4316     StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4317     StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4318     StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4319   }
4320 
4321 #ifdef COMPILER2
4322   if (UseMultiplyToLenIntrinsic) {
4323     StubRoutines::_multiplyToLen = generate_multiplyToLen();
4324   }
4325   if (UseSquareToLenIntrinsic) {
4326     StubRoutines::_squareToLen = generate_squareToLen();
4327   }
4328   if (UseMulAddIntrinsic) {
4329     StubRoutines::_mulAdd = generate_mulAdd();
4330   }
4331   if (VM_Version::supports_avx512_vbmi2()) {
4332     StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4333     StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4334   }
4335   if (UseSecondarySupersTable) {
4336     StubRoutines::_lookup_secondary_supers_table_slow_path_stub = generate_lookup_secondary_supers_table_slow_path_stub();
4337     if (! InlineSecondarySupersTest) {
4338       for (int slot = 0; slot < Klass::SECONDARY_SUPERS_TABLE_SIZE; slot++) {
4339         StubRoutines::_lookup_secondary_supers_table_stubs[slot] = generate_lookup_secondary_supers_table_stub(slot);
4340       }
4341     }
4342   }
4343   if (UseMontgomeryMultiplyIntrinsic) {
4344     StubRoutines::_montgomeryMultiply
4345       = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4346   }
4347   if (UseMontgomerySquareIntrinsic) {
4348     StubRoutines::_montgomerySquare
4349       = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4350   }
4351 
4352   // Load x86_64_sort library on supported hardware to enable SIMD sort and partition intrinsics
4353 
4354   if (VM_Version::is_intel() && (VM_Version::supports_avx512dq() || VM_Version::supports_avx2())) {
4355     void *libsimdsort = nullptr;
4356     char ebuf_[1024];
4357     char dll_name_simd_sort[JVM_MAXPATHLEN];
4358     if (os::dll_locate_lib(dll_name_simd_sort, sizeof(dll_name_simd_sort), Arguments::get_dll_dir(), "simdsort")) {
4359       libsimdsort = os::dll_load(dll_name_simd_sort, ebuf_, sizeof ebuf_);
4360     }
4361     // Get addresses for SIMD sort and partition routines
4362     if (libsimdsort != nullptr) {
4363       log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "simdsort" JNI_LIB_SUFFIX, p2i(libsimdsort));
4364 
4365       snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_sort" : "avx2_sort");
4366       StubRoutines::_array_sort = (address)os::dll_lookup(libsimdsort, ebuf_);
4367 
4368       snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_partition" : "avx2_partition");
4369       StubRoutines::_array_partition = (address)os::dll_lookup(libsimdsort, ebuf_);
4370     }
4371   }
4372 
4373   // Get svml stub routine addresses
4374   void *libjsvml = nullptr;
4375   char ebuf[1024];
4376   char dll_name[JVM_MAXPATHLEN];
4377   if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4378     libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4379   }
4380   if (libjsvml != nullptr) {
4381     // SVML method naming convention
4382     //   All the methods are named as __jsvml_op<T><N>_ha_<VV>
4383     //   Where:
4384     //      ha stands for high accuracy
4385     //      <T> is optional to indicate float/double
4386     //              Set to f for vector float operation
4387     //              Omitted for vector double operation
4388     //      <N> is the number of elements in the vector
4389     //              1, 2, 4, 8, 16
4390     //              e.g. 128 bit float vector has 4 float elements
4391     //      <VV> indicates the avx/sse level:
4392     //              z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4393     //      e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4394     //           __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4395 
4396     log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4397     if (UseAVX > 2) {
4398       for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4399         int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4400         if ((!VM_Version::supports_avx512dq()) &&
4401             (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4402           continue;
4403         }
4404         snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::svmlname[op]);
4405         StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4406 
4407         snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::svmlname[op]);
4408         StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4409       }
4410     }
4411     const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4412     for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4413       int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4414       if (vop == VectorSupport::VECTOR_OP_POW) {
4415         continue;
4416       }
4417       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4418       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4419 
4420       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4421       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4422 
4423       snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4424       StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4425 
4426       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4427       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4428 
4429       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4430       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4431 
4432       snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4433       StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4434     }
4435   }
4436 
4437 #endif // COMPILER2
4438 #endif // COMPILER2_OR_JVMCI
4439 }
4440 
4441 StubGenerator::StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
4442     DEBUG_ONLY( _regs_in_thread = false; )
4443     switch(kind) {
4444     case Initial_stubs:
4445       generate_initial_stubs();
4446       break;
4447      case Continuation_stubs:
4448       generate_continuation_stubs();
4449       break;
4450     case Compiler_stubs:
4451       generate_compiler_stubs();
4452       break;
4453     case Final_stubs:
4454       generate_final_stubs();
4455       break;
4456     default:
4457       fatal("unexpected stubs kind: %d", kind);
4458       break;
4459     };
4460 }
4461 
4462 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
4463   StubGenerator g(code, kind);
4464 }
4465 
4466 #undef __