1 /*
   2  * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "compiler/disassembler.hpp"
  29 #include "gc/shared/cardTableModRefBS.hpp"
  30 #include "gc/shared/collectedHeap.inline.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "memory/resourceArea.hpp"
  33 #include "memory/universe.hpp"
  34 #include "oops/klass.inline.hpp"
  35 #include "prims/methodHandles.hpp"
  36 #include "runtime/biasedLocking.hpp"
  37 #include "runtime/interfaceSupport.hpp"
  38 #include "runtime/objectMonitor.hpp"
  39 #include "runtime/os.hpp"
  40 #include "runtime/sharedRuntime.hpp"
  41 #include "runtime/stubRoutines.hpp"
  42 #include "runtime/thread.hpp"
  43 #include "utilities/macros.hpp"
  44 #if INCLUDE_ALL_GCS
  45 #include "gc/g1/g1CollectedHeap.inline.hpp"
  46 #include "gc/g1/g1SATBCardTableModRefBS.hpp"
  47 #include "gc/g1/heapRegion.hpp"
  48 #endif // INCLUDE_ALL_GCS
  49 #include "crc32c.h"
  50 #ifdef COMPILER2
  51 #include "opto/intrinsicnode.hpp"
  52 #endif
  53 
  54 #ifdef PRODUCT
  55 #define BLOCK_COMMENT(str) /* nothing */
  56 #define STOP(error) stop(error)
  57 #else
  58 #define BLOCK_COMMENT(str) block_comment(str)
  59 #define STOP(error) block_comment(error); stop(error)
  60 #endif
  61 
  62 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  63 
  64 #ifdef ASSERT
  65 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
  66 #endif
  67 
  68 static Assembler::Condition reverse[] = {
  69     Assembler::noOverflow     /* overflow      = 0x0 */ ,
  70     Assembler::overflow       /* noOverflow    = 0x1 */ ,
  71     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
  72     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
  73     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
  74     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
  75     Assembler::above          /* belowEqual    = 0x6 */ ,
  76     Assembler::belowEqual     /* above         = 0x7 */ ,
  77     Assembler::positive       /* negative      = 0x8 */ ,
  78     Assembler::negative       /* positive      = 0x9 */ ,
  79     Assembler::noParity       /* parity        = 0xa */ ,
  80     Assembler::parity         /* noParity      = 0xb */ ,
  81     Assembler::greaterEqual   /* less          = 0xc */ ,
  82     Assembler::less           /* greaterEqual  = 0xd */ ,
  83     Assembler::greater        /* lessEqual     = 0xe */ ,
  84     Assembler::lessEqual      /* greater       = 0xf, */
  85 
  86 };
  87 
  88 
  89 // Implementation of MacroAssembler
  90 
  91 // First all the versions that have distinct versions depending on 32/64 bit
  92 // Unless the difference is trivial (1 line or so).
  93 
  94 #ifndef _LP64
  95 
  96 // 32bit versions
  97 
  98 Address MacroAssembler::as_Address(AddressLiteral adr) {
  99   return Address(adr.target(), adr.rspec());
 100 }
 101 
 102 Address MacroAssembler::as_Address(ArrayAddress adr) {
 103   return Address::make_array(adr);
 104 }
 105 
 106 void MacroAssembler::call_VM_leaf_base(address entry_point,
 107                                        int number_of_arguments) {
 108   call(RuntimeAddress(entry_point));
 109   increment(rsp, number_of_arguments * wordSize);
 110 }
 111 
 112 void MacroAssembler::cmpklass(Address src1, Metadata* obj) {
 113   cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
 114 }
 115 
 116 void MacroAssembler::cmpklass(Register src1, Metadata* obj) {
 117   cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
 118 }
 119 
 120 void MacroAssembler::cmpoop(Address src1, jobject obj) {
 121   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
 122 }
 123 
 124 void MacroAssembler::cmpoop(Register src1, jobject obj) {
 125   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
 126 }
 127 
 128 void MacroAssembler::extend_sign(Register hi, Register lo) {
 129   // According to Intel Doc. AP-526, "Integer Divide", p.18.
 130   if (VM_Version::is_P6() && hi == rdx && lo == rax) {
 131     cdql();
 132   } else {
 133     movl(hi, lo);
 134     sarl(hi, 31);
 135   }
 136 }
 137 
 138 void MacroAssembler::jC2(Register tmp, Label& L) {
 139   // set parity bit if FPU flag C2 is set (via rax)
 140   save_rax(tmp);
 141   fwait(); fnstsw_ax();
 142   sahf();
 143   restore_rax(tmp);
 144   // branch
 145   jcc(Assembler::parity, L);
 146 }
 147 
 148 void MacroAssembler::jnC2(Register tmp, Label& L) {
 149   // set parity bit if FPU flag C2 is set (via rax)
 150   save_rax(tmp);
 151   fwait(); fnstsw_ax();
 152   sahf();
 153   restore_rax(tmp);
 154   // branch
 155   jcc(Assembler::noParity, L);
 156 }
 157 
 158 // 32bit can do a case table jump in one instruction but we no longer allow the base
 159 // to be installed in the Address class
 160 void MacroAssembler::jump(ArrayAddress entry) {
 161   jmp(as_Address(entry));
 162 }
 163 
 164 // Note: y_lo will be destroyed
 165 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
 166   // Long compare for Java (semantics as described in JVM spec.)
 167   Label high, low, done;
 168 
 169   cmpl(x_hi, y_hi);
 170   jcc(Assembler::less, low);
 171   jcc(Assembler::greater, high);
 172   // x_hi is the return register
 173   xorl(x_hi, x_hi);
 174   cmpl(x_lo, y_lo);
 175   jcc(Assembler::below, low);
 176   jcc(Assembler::equal, done);
 177 
 178   bind(high);
 179   xorl(x_hi, x_hi);
 180   increment(x_hi);
 181   jmp(done);
 182 
 183   bind(low);
 184   xorl(x_hi, x_hi);
 185   decrementl(x_hi);
 186 
 187   bind(done);
 188 }
 189 
 190 void MacroAssembler::lea(Register dst, AddressLiteral src) {
 191     mov_literal32(dst, (int32_t)src.target(), src.rspec());
 192 }
 193 
 194 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
 195   // leal(dst, as_Address(adr));
 196   // see note in movl as to why we must use a move
 197   mov_literal32(dst, (int32_t) adr.target(), adr.rspec());
 198 }
 199 
 200 void MacroAssembler::leave() {
 201   mov(rsp, rbp);
 202   pop(rbp);
 203 }
 204 
 205 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) {
 206   // Multiplication of two Java long values stored on the stack
 207   // as illustrated below. Result is in rdx:rax.
 208   //
 209   // rsp ---> [  ??  ] \               \
 210   //            ....    | y_rsp_offset  |
 211   //          [ y_lo ] /  (in bytes)    | x_rsp_offset
 212   //          [ y_hi ]                  | (in bytes)
 213   //            ....                    |
 214   //          [ x_lo ]                 /
 215   //          [ x_hi ]
 216   //            ....
 217   //
 218   // Basic idea: lo(result) = lo(x_lo * y_lo)
 219   //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
 220   Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset);
 221   Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset);
 222   Label quick;
 223   // load x_hi, y_hi and check if quick
 224   // multiplication is possible
 225   movl(rbx, x_hi);
 226   movl(rcx, y_hi);
 227   movl(rax, rbx);
 228   orl(rbx, rcx);                                 // rbx, = 0 <=> x_hi = 0 and y_hi = 0
 229   jcc(Assembler::zero, quick);                   // if rbx, = 0 do quick multiply
 230   // do full multiplication
 231   // 1st step
 232   mull(y_lo);                                    // x_hi * y_lo
 233   movl(rbx, rax);                                // save lo(x_hi * y_lo) in rbx,
 234   // 2nd step
 235   movl(rax, x_lo);
 236   mull(rcx);                                     // x_lo * y_hi
 237   addl(rbx, rax);                                // add lo(x_lo * y_hi) to rbx,
 238   // 3rd step
 239   bind(quick);                                   // note: rbx, = 0 if quick multiply!
 240   movl(rax, x_lo);
 241   mull(y_lo);                                    // x_lo * y_lo
 242   addl(rdx, rbx);                                // correct hi(x_lo * y_lo)
 243 }
 244 
 245 void MacroAssembler::lneg(Register hi, Register lo) {
 246   negl(lo);
 247   adcl(hi, 0);
 248   negl(hi);
 249 }
 250 
 251 void MacroAssembler::lshl(Register hi, Register lo) {
 252   // Java shift left long support (semantics as described in JVM spec., p.305)
 253   // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n))
 254   // shift value is in rcx !
 255   assert(hi != rcx, "must not use rcx");
 256   assert(lo != rcx, "must not use rcx");
 257   const Register s = rcx;                        // shift count
 258   const int      n = BitsPerWord;
 259   Label L;
 260   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
 261   cmpl(s, n);                                    // if (s < n)
 262   jcc(Assembler::less, L);                       // else (s >= n)
 263   movl(hi, lo);                                  // x := x << n
 264   xorl(lo, lo);
 265   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
 266   bind(L);                                       // s (mod n) < n
 267   shldl(hi, lo);                                 // x := x << s
 268   shll(lo);
 269 }
 270 
 271 
 272 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) {
 273   // Java shift right long support (semantics as described in JVM spec., p.306 & p.310)
 274   // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n))
 275   assert(hi != rcx, "must not use rcx");
 276   assert(lo != rcx, "must not use rcx");
 277   const Register s = rcx;                        // shift count
 278   const int      n = BitsPerWord;
 279   Label L;
 280   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
 281   cmpl(s, n);                                    // if (s < n)
 282   jcc(Assembler::less, L);                       // else (s >= n)
 283   movl(lo, hi);                                  // x := x >> n
 284   if (sign_extension) sarl(hi, 31);
 285   else                xorl(hi, hi);
 286   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
 287   bind(L);                                       // s (mod n) < n
 288   shrdl(lo, hi);                                 // x := x >> s
 289   if (sign_extension) sarl(hi);
 290   else                shrl(hi);
 291 }
 292 
 293 void MacroAssembler::movoop(Register dst, jobject obj) {
 294   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
 295 }
 296 
 297 void MacroAssembler::movoop(Address dst, jobject obj) {
 298   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
 299 }
 300 
 301 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
 302   mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
 303 }
 304 
 305 void MacroAssembler::mov_metadata(Address dst, Metadata* obj) {
 306   mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
 307 }
 308 
 309 void MacroAssembler::movptr(Register dst, AddressLiteral src, Register scratch) {
 310   // scratch register is not used,
 311   // it is defined to match parameters of 64-bit version of this method.
 312   if (src.is_lval()) {
 313     mov_literal32(dst, (intptr_t)src.target(), src.rspec());
 314   } else {
 315     movl(dst, as_Address(src));
 316   }
 317 }
 318 
 319 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
 320   movl(as_Address(dst), src);
 321 }
 322 
 323 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
 324   movl(dst, as_Address(src));
 325 }
 326 
 327 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 328 void MacroAssembler::movptr(Address dst, intptr_t src) {
 329   movl(dst, src);
 330 }
 331 
 332 
 333 void MacroAssembler::pop_callee_saved_registers() {
 334   pop(rcx);
 335   pop(rdx);
 336   pop(rdi);
 337   pop(rsi);
 338 }
 339 
 340 void MacroAssembler::pop_fTOS() {
 341   fld_d(Address(rsp, 0));
 342   addl(rsp, 2 * wordSize);
 343 }
 344 
 345 void MacroAssembler::push_callee_saved_registers() {
 346   push(rsi);
 347   push(rdi);
 348   push(rdx);
 349   push(rcx);
 350 }
 351 
 352 void MacroAssembler::push_fTOS() {
 353   subl(rsp, 2 * wordSize);
 354   fstp_d(Address(rsp, 0));
 355 }
 356 
 357 
 358 void MacroAssembler::pushoop(jobject obj) {
 359   push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
 360 }
 361 
 362 void MacroAssembler::pushklass(Metadata* obj) {
 363   push_literal32((int32_t)obj, metadata_Relocation::spec_for_immediate());
 364 }
 365 
 366 void MacroAssembler::pushptr(AddressLiteral src) {
 367   if (src.is_lval()) {
 368     push_literal32((int32_t)src.target(), src.rspec());
 369   } else {
 370     pushl(as_Address(src));
 371   }
 372 }
 373 
 374 void MacroAssembler::set_word_if_not_zero(Register dst) {
 375   xorl(dst, dst);
 376   set_byte_if_not_zero(dst);
 377 }
 378 
 379 static void pass_arg0(MacroAssembler* masm, Register arg) {
 380   masm->push(arg);
 381 }
 382 
 383 static void pass_arg1(MacroAssembler* masm, Register arg) {
 384   masm->push(arg);
 385 }
 386 
 387 static void pass_arg2(MacroAssembler* masm, Register arg) {
 388   masm->push(arg);
 389 }
 390 
 391 static void pass_arg3(MacroAssembler* masm, Register arg) {
 392   masm->push(arg);
 393 }
 394 
 395 #ifndef PRODUCT
 396 extern "C" void findpc(intptr_t x);
 397 #endif
 398 
 399 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) {
 400   // In order to get locks to work, we need to fake a in_VM state
 401   JavaThread* thread = JavaThread::current();
 402   JavaThreadState saved_state = thread->thread_state();
 403   thread->set_thread_state(_thread_in_vm);
 404   if (ShowMessageBoxOnError) {
 405     JavaThread* thread = JavaThread::current();
 406     JavaThreadState saved_state = thread->thread_state();
 407     thread->set_thread_state(_thread_in_vm);
 408     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
 409       ttyLocker ttyl;
 410       BytecodeCounter::print();
 411     }
 412     // To see where a verify_oop failed, get $ebx+40/X for this frame.
 413     // This is the value of eip which points to where verify_oop will return.
 414     if (os::message_box(msg, "Execution stopped, print registers?")) {
 415       print_state32(rdi, rsi, rbp, rsp, rbx, rdx, rcx, rax, eip);
 416       BREAKPOINT;
 417     }
 418   } else {
 419     ttyLocker ttyl;
 420     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
 421   }
 422   // Don't assert holding the ttyLock
 423     assert(false, "DEBUG MESSAGE: %s", msg);
 424   ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
 425 }
 426 
 427 void MacroAssembler::print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip) {
 428   ttyLocker ttyl;
 429   FlagSetting fs(Debugging, true);
 430   tty->print_cr("eip = 0x%08x", eip);
 431 #ifndef PRODUCT
 432   if ((WizardMode || Verbose) && PrintMiscellaneous) {
 433     tty->cr();
 434     findpc(eip);
 435     tty->cr();
 436   }
 437 #endif
 438 #define PRINT_REG(rax) \
 439   { tty->print("%s = ", #rax); os::print_location(tty, rax); }
 440   PRINT_REG(rax);
 441   PRINT_REG(rbx);
 442   PRINT_REG(rcx);
 443   PRINT_REG(rdx);
 444   PRINT_REG(rdi);
 445   PRINT_REG(rsi);
 446   PRINT_REG(rbp);
 447   PRINT_REG(rsp);
 448 #undef PRINT_REG
 449   // Print some words near top of staack.
 450   int* dump_sp = (int*) rsp;
 451   for (int col1 = 0; col1 < 8; col1++) {
 452     tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
 453     os::print_location(tty, *dump_sp++);
 454   }
 455   for (int row = 0; row < 16; row++) {
 456     tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
 457     for (int col = 0; col < 8; col++) {
 458       tty->print(" 0x%08x", *dump_sp++);
 459     }
 460     tty->cr();
 461   }
 462   // Print some instructions around pc:
 463   Disassembler::decode((address)eip-64, (address)eip);
 464   tty->print_cr("--------");
 465   Disassembler::decode((address)eip, (address)eip+32);
 466 }
 467 
 468 void MacroAssembler::stop(const char* msg) {
 469   ExternalAddress message((address)msg);
 470   // push address of message
 471   pushptr(message.addr());
 472   { Label L; call(L, relocInfo::none); bind(L); }     // push eip
 473   pusha();                                            // push registers
 474   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
 475   hlt();
 476 }
 477 
 478 void MacroAssembler::warn(const char* msg) {
 479   push_CPU_state();
 480 
 481   ExternalAddress message((address) msg);
 482   // push address of message
 483   pushptr(message.addr());
 484 
 485   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
 486   addl(rsp, wordSize);       // discard argument
 487   pop_CPU_state();
 488 }
 489 
 490 void MacroAssembler::print_state() {
 491   { Label L; call(L, relocInfo::none); bind(L); }     // push eip
 492   pusha();                                            // push registers
 493 
 494   push_CPU_state();
 495   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::print_state32)));
 496   pop_CPU_state();
 497 
 498   popa();
 499   addl(rsp, wordSize);
 500 }
 501 
 502 #else // _LP64
 503 
 504 // 64 bit versions
 505 
 506 Address MacroAssembler::as_Address(AddressLiteral adr) {
 507   // amd64 always does this as a pc-rel
 508   // we can be absolute or disp based on the instruction type
 509   // jmp/call are displacements others are absolute
 510   assert(!adr.is_lval(), "must be rval");
 511   assert(reachable(adr), "must be");
 512   return Address((int32_t)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc());
 513 
 514 }
 515 
 516 Address MacroAssembler::as_Address(ArrayAddress adr) {
 517   AddressLiteral base = adr.base();
 518   lea(rscratch1, base);
 519   Address index = adr.index();
 520   assert(index._disp == 0, "must not have disp"); // maybe it can?
 521   Address array(rscratch1, index._index, index._scale, index._disp);
 522   return array;
 523 }
 524 
 525 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
 526   Label L, E;
 527 
 528 #ifdef _WIN64
 529   // Windows always allocates space for it's register args
 530   assert(num_args <= 4, "only register arguments supported");
 531   subq(rsp,  frame::arg_reg_save_area_bytes);
 532 #endif
 533 
 534   // Align stack if necessary
 535   testl(rsp, 15);
 536   jcc(Assembler::zero, L);
 537 
 538   subq(rsp, 8);
 539   {
 540     call(RuntimeAddress(entry_point));
 541   }
 542   addq(rsp, 8);
 543   jmp(E);
 544 
 545   bind(L);
 546   {
 547     call(RuntimeAddress(entry_point));
 548   }
 549 
 550   bind(E);
 551 
 552 #ifdef _WIN64
 553   // restore stack pointer
 554   addq(rsp, frame::arg_reg_save_area_bytes);
 555 #endif
 556 
 557 }
 558 
 559 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) {
 560   assert(!src2.is_lval(), "should use cmpptr");
 561 
 562   if (reachable(src2)) {
 563     cmpq(src1, as_Address(src2));
 564   } else {
 565     lea(rscratch1, src2);
 566     Assembler::cmpq(src1, Address(rscratch1, 0));
 567   }
 568 }
 569 
 570 int MacroAssembler::corrected_idivq(Register reg) {
 571   // Full implementation of Java ldiv and lrem; checks for special
 572   // case as described in JVM spec., p.243 & p.271.  The function
 573   // returns the (pc) offset of the idivl instruction - may be needed
 574   // for implicit exceptions.
 575   //
 576   //         normal case                           special case
 577   //
 578   // input : rax: dividend                         min_long
 579   //         reg: divisor   (may not be eax/edx)   -1
 580   //
 581   // output: rax: quotient  (= rax idiv reg)       min_long
 582   //         rdx: remainder (= rax irem reg)       0
 583   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
 584   static const int64_t min_long = 0x8000000000000000;
 585   Label normal_case, special_case;
 586 
 587   // check for special case
 588   cmp64(rax, ExternalAddress((address) &min_long));
 589   jcc(Assembler::notEqual, normal_case);
 590   xorl(rdx, rdx); // prepare rdx for possible special case (where
 591                   // remainder = 0)
 592   cmpq(reg, -1);
 593   jcc(Assembler::equal, special_case);
 594 
 595   // handle normal case
 596   bind(normal_case);
 597   cdqq();
 598   int idivq_offset = offset();
 599   idivq(reg);
 600 
 601   // normal and special case exit
 602   bind(special_case);
 603 
 604   return idivq_offset;
 605 }
 606 
 607 void MacroAssembler::decrementq(Register reg, int value) {
 608   if (value == min_jint) { subq(reg, value); return; }
 609   if (value <  0) { incrementq(reg, -value); return; }
 610   if (value == 0) {                        ; return; }
 611   if (value == 1 && UseIncDec) { decq(reg) ; return; }
 612   /* else */      { subq(reg, value)       ; return; }
 613 }
 614 
 615 void MacroAssembler::decrementq(Address dst, int value) {
 616   if (value == min_jint) { subq(dst, value); return; }
 617   if (value <  0) { incrementq(dst, -value); return; }
 618   if (value == 0) {                        ; return; }
 619   if (value == 1 && UseIncDec) { decq(dst) ; return; }
 620   /* else */      { subq(dst, value)       ; return; }
 621 }
 622 
 623 void MacroAssembler::incrementq(AddressLiteral dst) {
 624   if (reachable(dst)) {
 625     incrementq(as_Address(dst));
 626   } else {
 627     lea(rscratch1, dst);
 628     incrementq(Address(rscratch1, 0));
 629   }
 630 }
 631 
 632 void MacroAssembler::incrementq(Register reg, int value) {
 633   if (value == min_jint) { addq(reg, value); return; }
 634   if (value <  0) { decrementq(reg, -value); return; }
 635   if (value == 0) {                        ; return; }
 636   if (value == 1 && UseIncDec) { incq(reg) ; return; }
 637   /* else */      { addq(reg, value)       ; return; }
 638 }
 639 
 640 void MacroAssembler::incrementq(Address dst, int value) {
 641   if (value == min_jint) { addq(dst, value); return; }
 642   if (value <  0) { decrementq(dst, -value); return; }
 643   if (value == 0) {                        ; return; }
 644   if (value == 1 && UseIncDec) { incq(dst) ; return; }
 645   /* else */      { addq(dst, value)       ; return; }
 646 }
 647 
 648 // 32bit can do a case table jump in one instruction but we no longer allow the base
 649 // to be installed in the Address class
 650 void MacroAssembler::jump(ArrayAddress entry) {
 651   lea(rscratch1, entry.base());
 652   Address dispatch = entry.index();
 653   assert(dispatch._base == noreg, "must be");
 654   dispatch._base = rscratch1;
 655   jmp(dispatch);
 656 }
 657 
 658 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
 659   ShouldNotReachHere(); // 64bit doesn't use two regs
 660   cmpq(x_lo, y_lo);
 661 }
 662 
 663 void MacroAssembler::lea(Register dst, AddressLiteral src) {
 664     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
 665 }
 666 
 667 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
 668   mov_literal64(rscratch1, (intptr_t)adr.target(), adr.rspec());
 669   movptr(dst, rscratch1);
 670 }
 671 
 672 void MacroAssembler::leave() {
 673   // %%% is this really better? Why not on 32bit too?
 674   emit_int8((unsigned char)0xC9); // LEAVE
 675 }
 676 
 677 void MacroAssembler::lneg(Register hi, Register lo) {
 678   ShouldNotReachHere(); // 64bit doesn't use two regs
 679   negq(lo);
 680 }
 681 
 682 void MacroAssembler::movoop(Register dst, jobject obj) {
 683   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
 684 }
 685 
 686 void MacroAssembler::movoop(Address dst, jobject obj) {
 687   mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate());
 688   movq(dst, rscratch1);
 689 }
 690 
 691 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
 692   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
 693 }
 694 
 695 void MacroAssembler::mov_metadata(Address dst, Metadata* obj) {
 696   mov_literal64(rscratch1, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
 697   movq(dst, rscratch1);
 698 }
 699 
 700 void MacroAssembler::movptr(Register dst, AddressLiteral src, Register scratch) {
 701   if (src.is_lval()) {
 702     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
 703   } else {
 704     if (reachable(src)) {
 705       movq(dst, as_Address(src));
 706     } else {
 707       lea(scratch, src);
 708       movq(dst, Address(scratch, 0));
 709     }
 710   }
 711 }
 712 
 713 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
 714   movq(as_Address(dst), src);
 715 }
 716 
 717 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
 718   movq(dst, as_Address(src));
 719 }
 720 
 721 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 722 void MacroAssembler::movptr(Address dst, intptr_t src) {
 723   mov64(rscratch1, src);
 724   movq(dst, rscratch1);
 725 }
 726 
 727 // These are mostly for initializing NULL
 728 void MacroAssembler::movptr(Address dst, int32_t src) {
 729   movslq(dst, src);
 730 }
 731 
 732 void MacroAssembler::movptr(Register dst, int32_t src) {
 733   mov64(dst, (intptr_t)src);
 734 }
 735 
 736 void MacroAssembler::pushoop(jobject obj) {
 737   movoop(rscratch1, obj);
 738   push(rscratch1);
 739 }
 740 
 741 void MacroAssembler::pushklass(Metadata* obj) {
 742   mov_metadata(rscratch1, obj);
 743   push(rscratch1);
 744 }
 745 
 746 void MacroAssembler::pushptr(AddressLiteral src) {
 747   lea(rscratch1, src);
 748   if (src.is_lval()) {
 749     push(rscratch1);
 750   } else {
 751     pushq(Address(rscratch1, 0));
 752   }
 753 }
 754 
 755 void MacroAssembler::reset_last_Java_frame(bool clear_fp,
 756                                            bool clear_pc) {
 757   // we must set sp to zero to clear frame
 758   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 759   // must clear fp, so that compiled frames are not confused; it is
 760   // possible that we need it only for debugging
 761   if (clear_fp) {
 762     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 763   }
 764 
 765   if (clear_pc) {
 766     movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 767   }
 768 }
 769 
 770 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 771                                          Register last_java_fp,
 772                                          address  last_java_pc) {
 773   // determine last_java_sp register
 774   if (!last_java_sp->is_valid()) {
 775     last_java_sp = rsp;
 776   }
 777 
 778   // last_java_fp is optional
 779   if (last_java_fp->is_valid()) {
 780     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()),
 781            last_java_fp);
 782   }
 783 
 784   // last_java_pc is optional
 785   if (last_java_pc != NULL) {
 786     Address java_pc(r15_thread,
 787                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 788     lea(rscratch1, InternalAddress(last_java_pc));
 789     movptr(java_pc, rscratch1);
 790   }
 791 
 792   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 793 }
 794 
 795 static void pass_arg0(MacroAssembler* masm, Register arg) {
 796   if (c_rarg0 != arg ) {
 797     masm->mov(c_rarg0, arg);
 798   }
 799 }
 800 
 801 static void pass_arg1(MacroAssembler* masm, Register arg) {
 802   if (c_rarg1 != arg ) {
 803     masm->mov(c_rarg1, arg);
 804   }
 805 }
 806 
 807 static void pass_arg2(MacroAssembler* masm, Register arg) {
 808   if (c_rarg2 != arg ) {
 809     masm->mov(c_rarg2, arg);
 810   }
 811 }
 812 
 813 static void pass_arg3(MacroAssembler* masm, Register arg) {
 814   if (c_rarg3 != arg ) {
 815     masm->mov(c_rarg3, arg);
 816   }
 817 }
 818 
 819 void MacroAssembler::stop(const char* msg) {
 820   address rip = pc();
 821   pusha(); // get regs on stack
 822   lea(c_rarg0, ExternalAddress((address) msg));
 823   lea(c_rarg1, InternalAddress(rip));
 824   movq(c_rarg2, rsp); // pass pointer to regs array
 825   andq(rsp, -16); // align stack as required by ABI
 826   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
 827   hlt();
 828 }
 829 
 830 void MacroAssembler::warn(const char* msg) {
 831   push(rbp);
 832   movq(rbp, rsp);
 833   andq(rsp, -16);     // align stack as required by push_CPU_state and call
 834   push_CPU_state();   // keeps alignment at 16 bytes
 835   lea(c_rarg0, ExternalAddress((address) msg));
 836   call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0);
 837   pop_CPU_state();
 838   mov(rsp, rbp);
 839   pop(rbp);
 840 }
 841 
 842 void MacroAssembler::print_state() {
 843   address rip = pc();
 844   pusha();            // get regs on stack
 845   push(rbp);
 846   movq(rbp, rsp);
 847   andq(rsp, -16);     // align stack as required by push_CPU_state and call
 848   push_CPU_state();   // keeps alignment at 16 bytes
 849 
 850   lea(c_rarg0, InternalAddress(rip));
 851   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
 852   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
 853 
 854   pop_CPU_state();
 855   mov(rsp, rbp);
 856   pop(rbp);
 857   popa();
 858 }
 859 
 860 #ifndef PRODUCT
 861 extern "C" void findpc(intptr_t x);
 862 #endif
 863 
 864 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
 865   // In order to get locks to work, we need to fake a in_VM state
 866   if (ShowMessageBoxOnError) {
 867     JavaThread* thread = JavaThread::current();
 868     JavaThreadState saved_state = thread->thread_state();
 869     thread->set_thread_state(_thread_in_vm);
 870 #ifndef PRODUCT
 871     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
 872       ttyLocker ttyl;
 873       BytecodeCounter::print();
 874     }
 875 #endif
 876     // To see where a verify_oop failed, get $ebx+40/X for this frame.
 877     // XXX correct this offset for amd64
 878     // This is the value of eip which points to where verify_oop will return.
 879     if (os::message_box(msg, "Execution stopped, print registers?")) {
 880       print_state64(pc, regs);
 881       BREAKPOINT;
 882       assert(false, "start up GDB");
 883     }
 884     ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
 885   } else {
 886     ttyLocker ttyl;
 887     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n",
 888                     msg);
 889     assert(false, "DEBUG MESSAGE: %s", msg);
 890   }
 891 }
 892 
 893 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
 894   ttyLocker ttyl;
 895   FlagSetting fs(Debugging, true);
 896   tty->print_cr("rip = 0x%016lx", pc);
 897 #ifndef PRODUCT
 898   tty->cr();
 899   findpc(pc);
 900   tty->cr();
 901 #endif
 902 #define PRINT_REG(rax, value) \
 903   { tty->print("%s = ", #rax); os::print_location(tty, value); }
 904   PRINT_REG(rax, regs[15]);
 905   PRINT_REG(rbx, regs[12]);
 906   PRINT_REG(rcx, regs[14]);
 907   PRINT_REG(rdx, regs[13]);
 908   PRINT_REG(rdi, regs[8]);
 909   PRINT_REG(rsi, regs[9]);
 910   PRINT_REG(rbp, regs[10]);
 911   PRINT_REG(rsp, regs[11]);
 912   PRINT_REG(r8 , regs[7]);
 913   PRINT_REG(r9 , regs[6]);
 914   PRINT_REG(r10, regs[5]);
 915   PRINT_REG(r11, regs[4]);
 916   PRINT_REG(r12, regs[3]);
 917   PRINT_REG(r13, regs[2]);
 918   PRINT_REG(r14, regs[1]);
 919   PRINT_REG(r15, regs[0]);
 920 #undef PRINT_REG
 921   // Print some words near top of staack.
 922   int64_t* rsp = (int64_t*) regs[11];
 923   int64_t* dump_sp = rsp;
 924   for (int col1 = 0; col1 < 8; col1++) {
 925     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (int64_t)dump_sp);
 926     os::print_location(tty, *dump_sp++);
 927   }
 928   for (int row = 0; row < 25; row++) {
 929     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (int64_t)dump_sp);
 930     for (int col = 0; col < 4; col++) {
 931       tty->print(" 0x%016lx", *dump_sp++);
 932     }
 933     tty->cr();
 934   }
 935   // Print some instructions around pc:
 936   Disassembler::decode((address)pc-64, (address)pc);
 937   tty->print_cr("--------");
 938   Disassembler::decode((address)pc, (address)pc+32);
 939 }
 940 
 941 #endif // _LP64
 942 
 943 // Now versions that are common to 32/64 bit
 944 
 945 void MacroAssembler::addptr(Register dst, int32_t imm32) {
 946   LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32));
 947 }
 948 
 949 void MacroAssembler::addptr(Register dst, Register src) {
 950   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
 951 }
 952 
 953 void MacroAssembler::addptr(Address dst, Register src) {
 954   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
 955 }
 956 
 957 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src) {
 958   if (reachable(src)) {
 959     Assembler::addsd(dst, as_Address(src));
 960   } else {
 961     lea(rscratch1, src);
 962     Assembler::addsd(dst, Address(rscratch1, 0));
 963   }
 964 }
 965 
 966 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src) {
 967   if (reachable(src)) {
 968     addss(dst, as_Address(src));
 969   } else {
 970     lea(rscratch1, src);
 971     addss(dst, Address(rscratch1, 0));
 972   }
 973 }
 974 
 975 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src) {
 976   if (reachable(src)) {
 977     Assembler::addpd(dst, as_Address(src));
 978   } else {
 979     lea(rscratch1, src);
 980     Assembler::addpd(dst, Address(rscratch1, 0));
 981   }
 982 }
 983 
 984 void MacroAssembler::align(int modulus) {
 985   align(modulus, offset());
 986 }
 987 
 988 void MacroAssembler::align(int modulus, int target) {
 989   if (target % modulus != 0) {
 990     nop(modulus - (target % modulus));
 991   }
 992 }
 993 
 994 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) {
 995   // Used in sign-masking with aligned address.
 996   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 997   if (reachable(src)) {
 998     Assembler::andpd(dst, as_Address(src));
 999   } else {
1000     lea(rscratch1, src);
1001     Assembler::andpd(dst, Address(rscratch1, 0));
1002   }
1003 }
1004 
1005 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src) {
1006   // Used in sign-masking with aligned address.
1007   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
1008   if (reachable(src)) {
1009     Assembler::andps(dst, as_Address(src));
1010   } else {
1011     lea(rscratch1, src);
1012     Assembler::andps(dst, Address(rscratch1, 0));
1013   }
1014 }
1015 
1016 void MacroAssembler::andptr(Register dst, int32_t imm32) {
1017   LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32));
1018 }
1019 
1020 void MacroAssembler::atomic_incl(Address counter_addr) {
1021   if (os::is_MP())
1022     lock();
1023   incrementl(counter_addr);
1024 }
1025 
1026 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register scr) {
1027   if (reachable(counter_addr)) {
1028     atomic_incl(as_Address(counter_addr));
1029   } else {
1030     lea(scr, counter_addr);
1031     atomic_incl(Address(scr, 0));
1032   }
1033 }
1034 
1035 #ifdef _LP64
1036 void MacroAssembler::atomic_incq(Address counter_addr) {
1037   if (os::is_MP())
1038     lock();
1039   incrementq(counter_addr);
1040 }
1041 
1042 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register scr) {
1043   if (reachable(counter_addr)) {
1044     atomic_incq(as_Address(counter_addr));
1045   } else {
1046     lea(scr, counter_addr);
1047     atomic_incq(Address(scr, 0));
1048   }
1049 }
1050 #endif
1051 
1052 // Writes to stack successive pages until offset reached to check for
1053 // stack overflow + shadow pages.  This clobbers tmp.
1054 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
1055   movptr(tmp, rsp);
1056   // Bang stack for total size given plus shadow page size.
1057   // Bang one page at a time because large size can bang beyond yellow and
1058   // red zones.
1059   Label loop;
1060   bind(loop);
1061   movl(Address(tmp, (-os::vm_page_size())), size );
1062   subptr(tmp, os::vm_page_size());
1063   subl(size, os::vm_page_size());
1064   jcc(Assembler::greater, loop);
1065 
1066   // Bang down shadow pages too.
1067   // At this point, (tmp-0) is the last address touched, so don't
1068   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
1069   // was post-decremented.)  Skip this address by starting at i=1, and
1070   // touch a few more pages below.  N.B.  It is important to touch all
1071   // the way down including all pages in the shadow zone.
1072   for (int i = 1; i < ((int)JavaThread::stack_shadow_zone_size() / os::vm_page_size()); i++) {
1073     // this could be any sized move but this is can be a debugging crumb
1074     // so the bigger the better.
1075     movptr(Address(tmp, (-i*os::vm_page_size())), size );
1076   }
1077 }
1078 
1079 void MacroAssembler::reserved_stack_check() {
1080     // testing if reserved zone needs to be enabled
1081     Label no_reserved_zone_enabling;
1082     Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
1083     NOT_LP64(get_thread(rsi);)
1084 
1085     cmpptr(rsp, Address(thread, JavaThread::reserved_stack_activation_offset()));
1086     jcc(Assembler::below, no_reserved_zone_enabling);
1087 
1088     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), thread);
1089     jump(RuntimeAddress(StubRoutines::throw_delayed_StackOverflowError_entry()));
1090     should_not_reach_here();
1091 
1092     bind(no_reserved_zone_enabling);
1093 }
1094 
1095 int MacroAssembler::biased_locking_enter(Register lock_reg,
1096                                          Register obj_reg,
1097                                          Register swap_reg,
1098                                          Register tmp_reg,
1099                                          bool swap_reg_contains_mark,
1100                                          Label& done,
1101                                          Label* slow_case,
1102                                          BiasedLockingCounters* counters) {
1103   assert(UseBiasedLocking, "why call this otherwise?");
1104   assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq");
1105   assert(tmp_reg != noreg, "tmp_reg must be supplied");
1106   assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
1107   assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
1108   Address mark_addr      (obj_reg, oopDesc::mark_offset_in_bytes());
1109   Address saved_mark_addr(lock_reg, 0);
1110 
1111   if (PrintBiasedLockingStatistics && counters == NULL) {
1112     counters = BiasedLocking::counters();
1113   }
1114   // Biased locking
1115   // See whether the lock is currently biased toward our thread and
1116   // whether the epoch is still valid
1117   // Note that the runtime guarantees sufficient alignment of JavaThread
1118   // pointers to allow age to be placed into low bits
1119   // First check to see whether biasing is even enabled for this object
1120   Label cas_label;
1121   int null_check_offset = -1;
1122   if (!swap_reg_contains_mark) {
1123     null_check_offset = offset();
1124     movptr(swap_reg, mark_addr);
1125   }
1126   movptr(tmp_reg, swap_reg);
1127   andptr(tmp_reg, markOopDesc::biased_lock_mask_in_place);
1128   cmpptr(tmp_reg, markOopDesc::biased_lock_pattern);
1129   jcc(Assembler::notEqual, cas_label);
1130   // The bias pattern is present in the object's header. Need to check
1131   // whether the bias owner and the epoch are both still current.
1132 #ifndef _LP64
1133   // Note that because there is no current thread register on x86_32 we
1134   // need to store off the mark word we read out of the object to
1135   // avoid reloading it and needing to recheck invariants below. This
1136   // store is unfortunate but it makes the overall code shorter and
1137   // simpler.
1138   movptr(saved_mark_addr, swap_reg);
1139 #endif
1140   if (swap_reg_contains_mark) {
1141     null_check_offset = offset();
1142   }
1143   load_prototype_header(tmp_reg, obj_reg);
1144 #ifdef _LP64
1145   orptr(tmp_reg, r15_thread);
1146   xorptr(tmp_reg, swap_reg);
1147   Register header_reg = tmp_reg;
1148 #else
1149   xorptr(tmp_reg, swap_reg);
1150   get_thread(swap_reg);
1151   xorptr(swap_reg, tmp_reg);
1152   Register header_reg = swap_reg;
1153 #endif
1154   andptr(header_reg, ~((int) markOopDesc::age_mask_in_place));
1155   if (counters != NULL) {
1156     cond_inc32(Assembler::zero,
1157                ExternalAddress((address) counters->biased_lock_entry_count_addr()));
1158   }
1159   jcc(Assembler::equal, done);
1160 
1161   Label try_revoke_bias;
1162   Label try_rebias;
1163 
1164   // At this point we know that the header has the bias pattern and
1165   // that we are not the bias owner in the current epoch. We need to
1166   // figure out more details about the state of the header in order to
1167   // know what operations can be legally performed on the object's
1168   // header.
1169 
1170   // If the low three bits in the xor result aren't clear, that means
1171   // the prototype header is no longer biased and we have to revoke
1172   // the bias on this object.
1173   testptr(header_reg, markOopDesc::biased_lock_mask_in_place);
1174   jccb(Assembler::notZero, try_revoke_bias);
1175 
1176   // Biasing is still enabled for this data type. See whether the
1177   // epoch of the current bias is still valid, meaning that the epoch
1178   // bits of the mark word are equal to the epoch bits of the
1179   // prototype header. (Note that the prototype header's epoch bits
1180   // only change at a safepoint.) If not, attempt to rebias the object
1181   // toward the current thread. Note that we must be absolutely sure
1182   // that the current epoch is invalid in order to do this because
1183   // otherwise the manipulations it performs on the mark word are
1184   // illegal.
1185   testptr(header_reg, markOopDesc::epoch_mask_in_place);
1186   jccb(Assembler::notZero, try_rebias);
1187 
1188   // The epoch of the current bias is still valid but we know nothing
1189   // about the owner; it might be set or it might be clear. Try to
1190   // acquire the bias of the object using an atomic operation. If this
1191   // fails we will go in to the runtime to revoke the object's bias.
1192   // Note that we first construct the presumed unbiased header so we
1193   // don't accidentally blow away another thread's valid bias.
1194   NOT_LP64( movptr(swap_reg, saved_mark_addr); )
1195   andptr(swap_reg,
1196          markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
1197 #ifdef _LP64
1198   movptr(tmp_reg, swap_reg);
1199   orptr(tmp_reg, r15_thread);
1200 #else
1201   get_thread(tmp_reg);
1202   orptr(tmp_reg, swap_reg);
1203 #endif
1204   if (os::is_MP()) {
1205     lock();
1206   }
1207   cmpxchgptr(tmp_reg, mark_addr); // compare tmp_reg and swap_reg
1208   // If the biasing toward our thread failed, this means that
1209   // another thread succeeded in biasing it toward itself and we
1210   // need to revoke that bias. The revocation will occur in the
1211   // interpreter runtime in the slow case.
1212   if (counters != NULL) {
1213     cond_inc32(Assembler::zero,
1214                ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
1215   }
1216   if (slow_case != NULL) {
1217     jcc(Assembler::notZero, *slow_case);
1218   }
1219   jmp(done);
1220 
1221   bind(try_rebias);
1222   // At this point we know the epoch has expired, meaning that the
1223   // current "bias owner", if any, is actually invalid. Under these
1224   // circumstances _only_, we are allowed to use the current header's
1225   // value as the comparison value when doing the cas to acquire the
1226   // bias in the current epoch. In other words, we allow transfer of
1227   // the bias from one thread to another directly in this situation.
1228   //
1229   // FIXME: due to a lack of registers we currently blow away the age
1230   // bits in this situation. Should attempt to preserve them.
1231   load_prototype_header(tmp_reg, obj_reg);
1232 #ifdef _LP64
1233   orptr(tmp_reg, r15_thread);
1234 #else
1235   get_thread(swap_reg);
1236   orptr(tmp_reg, swap_reg);
1237   movptr(swap_reg, saved_mark_addr);
1238 #endif
1239   if (os::is_MP()) {
1240     lock();
1241   }
1242   cmpxchgptr(tmp_reg, mark_addr); // compare tmp_reg and swap_reg
1243   // If the biasing toward our thread failed, then another thread
1244   // succeeded in biasing it toward itself and we need to revoke that
1245   // bias. The revocation will occur in the runtime in the slow case.
1246   if (counters != NULL) {
1247     cond_inc32(Assembler::zero,
1248                ExternalAddress((address) counters->rebiased_lock_entry_count_addr()));
1249   }
1250   if (slow_case != NULL) {
1251     jcc(Assembler::notZero, *slow_case);
1252   }
1253   jmp(done);
1254 
1255   bind(try_revoke_bias);
1256   // The prototype mark in the klass doesn't have the bias bit set any
1257   // more, indicating that objects of this data type are not supposed
1258   // to be biased any more. We are going to try to reset the mark of
1259   // this object to the prototype value and fall through to the
1260   // CAS-based locking scheme. Note that if our CAS fails, it means
1261   // that another thread raced us for the privilege of revoking the
1262   // bias of this particular object, so it's okay to continue in the
1263   // normal locking code.
1264   //
1265   // FIXME: due to a lack of registers we currently blow away the age
1266   // bits in this situation. Should attempt to preserve them.
1267   NOT_LP64( movptr(swap_reg, saved_mark_addr); )
1268   load_prototype_header(tmp_reg, obj_reg);
1269   if (os::is_MP()) {
1270     lock();
1271   }
1272   cmpxchgptr(tmp_reg, mark_addr); // compare tmp_reg and swap_reg
1273   // Fall through to the normal CAS-based lock, because no matter what
1274   // the result of the above CAS, some thread must have succeeded in
1275   // removing the bias bit from the object's header.
1276   if (counters != NULL) {
1277     cond_inc32(Assembler::zero,
1278                ExternalAddress((address) counters->revoked_lock_entry_count_addr()));
1279   }
1280 
1281   bind(cas_label);
1282 
1283   return null_check_offset;
1284 }
1285 
1286 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) {
1287   assert(UseBiasedLocking, "why call this otherwise?");
1288 
1289   // Check for biased locking unlock case, which is a no-op
1290   // Note: we do not have to check the thread ID for two reasons.
1291   // First, the interpreter checks for IllegalMonitorStateException at
1292   // a higher level. Second, if the bias was revoked while we held the
1293   // lock, the object could not be rebiased toward another thread, so
1294   // the bias bit would be clear.
1295   movptr(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1296   andptr(temp_reg, markOopDesc::biased_lock_mask_in_place);
1297   cmpptr(temp_reg, markOopDesc::biased_lock_pattern);
1298   jcc(Assembler::equal, done);
1299 }
1300 
1301 #ifdef COMPILER2
1302 
1303 #if INCLUDE_RTM_OPT
1304 
1305 // Update rtm_counters based on abort status
1306 // input: abort_status
1307 //        rtm_counters (RTMLockingCounters*)
1308 // flags are killed
1309 void MacroAssembler::rtm_counters_update(Register abort_status, Register rtm_counters) {
1310 
1311   atomic_incptr(Address(rtm_counters, RTMLockingCounters::abort_count_offset()));
1312   if (PrintPreciseRTMLockingStatistics) {
1313     for (int i = 0; i < RTMLockingCounters::ABORT_STATUS_LIMIT; i++) {
1314       Label check_abort;
1315       testl(abort_status, (1<<i));
1316       jccb(Assembler::equal, check_abort);
1317       atomic_incptr(Address(rtm_counters, RTMLockingCounters::abortX_count_offset() + (i * sizeof(uintx))));
1318       bind(check_abort);
1319     }
1320   }
1321 }
1322 
1323 // Branch if (random & (count-1) != 0), count is 2^n
1324 // tmp, scr and flags are killed
1325 void MacroAssembler::branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel) {
1326   assert(tmp == rax, "");
1327   assert(scr == rdx, "");
1328   rdtsc(); // modifies EDX:EAX
1329   andptr(tmp, count-1);
1330   jccb(Assembler::notZero, brLabel);
1331 }
1332 
1333 // Perform abort ratio calculation, set no_rtm bit if high ratio
1334 // input:  rtm_counters_Reg (RTMLockingCounters* address)
1335 // tmpReg, rtm_counters_Reg and flags are killed
1336 void MacroAssembler::rtm_abort_ratio_calculation(Register tmpReg,
1337                                                  Register rtm_counters_Reg,
1338                                                  RTMLockingCounters* rtm_counters,
1339                                                  Metadata* method_data) {
1340   Label L_done, L_check_always_rtm1, L_check_always_rtm2;
1341 
1342   if (RTMLockingCalculationDelay > 0) {
1343     // Delay calculation
1344     movptr(tmpReg, ExternalAddress((address) RTMLockingCounters::rtm_calculation_flag_addr()), tmpReg);
1345     testptr(tmpReg, tmpReg);
1346     jccb(Assembler::equal, L_done);
1347   }
1348   // Abort ratio calculation only if abort_count > RTMAbortThreshold
1349   //   Aborted transactions = abort_count * 100
1350   //   All transactions = total_count *  RTMTotalCountIncrRate
1351   //   Set no_rtm bit if (Aborted transactions >= All transactions * RTMAbortRatio)
1352 
1353   movptr(tmpReg, Address(rtm_counters_Reg, RTMLockingCounters::abort_count_offset()));
1354   cmpptr(tmpReg, RTMAbortThreshold);
1355   jccb(Assembler::below, L_check_always_rtm2);
1356   imulptr(tmpReg, tmpReg, 100);
1357 
1358   Register scrReg = rtm_counters_Reg;
1359   movptr(scrReg, Address(rtm_counters_Reg, RTMLockingCounters::total_count_offset()));
1360   imulptr(scrReg, scrReg, RTMTotalCountIncrRate);
1361   imulptr(scrReg, scrReg, RTMAbortRatio);
1362   cmpptr(tmpReg, scrReg);
1363   jccb(Assembler::below, L_check_always_rtm1);
1364   if (method_data != NULL) {
1365     // set rtm_state to "no rtm" in MDO
1366     mov_metadata(tmpReg, method_data);
1367     if (os::is_MP()) {
1368       lock();
1369     }
1370     orl(Address(tmpReg, MethodData::rtm_state_offset_in_bytes()), NoRTM);
1371   }
1372   jmpb(L_done);
1373   bind(L_check_always_rtm1);
1374   // Reload RTMLockingCounters* address
1375   lea(rtm_counters_Reg, ExternalAddress((address)rtm_counters));
1376   bind(L_check_always_rtm2);
1377   movptr(tmpReg, Address(rtm_counters_Reg, RTMLockingCounters::total_count_offset()));
1378   cmpptr(tmpReg, RTMLockingThreshold / RTMTotalCountIncrRate);
1379   jccb(Assembler::below, L_done);
1380   if (method_data != NULL) {
1381     // set rtm_state to "always rtm" in MDO
1382     mov_metadata(tmpReg, method_data);
1383     if (os::is_MP()) {
1384       lock();
1385     }
1386     orl(Address(tmpReg, MethodData::rtm_state_offset_in_bytes()), UseRTM);
1387   }
1388   bind(L_done);
1389 }
1390 
1391 // Update counters and perform abort ratio calculation
1392 // input:  abort_status_Reg
1393 // rtm_counters_Reg, flags are killed
1394 void MacroAssembler::rtm_profiling(Register abort_status_Reg,
1395                                    Register rtm_counters_Reg,
1396                                    RTMLockingCounters* rtm_counters,
1397                                    Metadata* method_data,
1398                                    bool profile_rtm) {
1399 
1400   assert(rtm_counters != NULL, "should not be NULL when profiling RTM");
1401   // update rtm counters based on rax value at abort
1402   // reads abort_status_Reg, updates flags
1403   lea(rtm_counters_Reg, ExternalAddress((address)rtm_counters));
1404   rtm_counters_update(abort_status_Reg, rtm_counters_Reg);
1405   if (profile_rtm) {
1406     // Save abort status because abort_status_Reg is used by following code.
1407     if (RTMRetryCount > 0) {
1408       push(abort_status_Reg);
1409     }
1410     assert(rtm_counters != NULL, "should not be NULL when profiling RTM");
1411     rtm_abort_ratio_calculation(abort_status_Reg, rtm_counters_Reg, rtm_counters, method_data);
1412     // restore abort status
1413     if (RTMRetryCount > 0) {
1414       pop(abort_status_Reg);
1415     }
1416   }
1417 }
1418 
1419 // Retry on abort if abort's status is 0x6: can retry (0x2) | memory conflict (0x4)
1420 // inputs: retry_count_Reg
1421 //       : abort_status_Reg
1422 // output: retry_count_Reg decremented by 1
1423 // flags are killed
1424 void MacroAssembler::rtm_retry_lock_on_abort(Register retry_count_Reg, Register abort_status_Reg, Label& retryLabel) {
1425   Label doneRetry;
1426   assert(abort_status_Reg == rax, "");
1427   // The abort reason bits are in eax (see all states in rtmLocking.hpp)
1428   // 0x6 = conflict on which we can retry (0x2) | memory conflict (0x4)
1429   // if reason is in 0x6 and retry count != 0 then retry
1430   andptr(abort_status_Reg, 0x6);
1431   jccb(Assembler::zero, doneRetry);
1432   testl(retry_count_Reg, retry_count_Reg);
1433   jccb(Assembler::zero, doneRetry);
1434   pause();
1435   decrementl(retry_count_Reg);
1436   jmp(retryLabel);
1437   bind(doneRetry);
1438 }
1439 
1440 // Spin and retry if lock is busy,
1441 // inputs: box_Reg (monitor address)
1442 //       : retry_count_Reg
1443 // output: retry_count_Reg decremented by 1
1444 //       : clear z flag if retry count exceeded
1445 // tmp_Reg, scr_Reg, flags are killed
1446 void MacroAssembler::rtm_retry_lock_on_busy(Register retry_count_Reg, Register box_Reg,
1447                                             Register tmp_Reg, Register scr_Reg, Label& retryLabel) {
1448   Label SpinLoop, SpinExit, doneRetry;
1449   int owner_offset = OM_OFFSET_NO_MONITOR_VALUE_TAG(owner);
1450 
1451   testl(retry_count_Reg, retry_count_Reg);
1452   jccb(Assembler::zero, doneRetry);
1453   decrementl(retry_count_Reg);
1454   movptr(scr_Reg, RTMSpinLoopCount);
1455 
1456   bind(SpinLoop);
1457   pause();
1458   decrementl(scr_Reg);
1459   jccb(Assembler::lessEqual, SpinExit);
1460   movptr(tmp_Reg, Address(box_Reg, owner_offset));
1461   testptr(tmp_Reg, tmp_Reg);
1462   jccb(Assembler::notZero, SpinLoop);
1463 
1464   bind(SpinExit);
1465   jmp(retryLabel);
1466   bind(doneRetry);
1467   incrementl(retry_count_Reg); // clear z flag
1468 }
1469 
1470 // Use RTM for normal stack locks
1471 // Input: objReg (object to lock)
1472 void MacroAssembler::rtm_stack_locking(Register objReg, Register tmpReg, Register scrReg,
1473                                        Register retry_on_abort_count_Reg,
1474                                        RTMLockingCounters* stack_rtm_counters,
1475                                        Metadata* method_data, bool profile_rtm,
1476                                        Label& DONE_LABEL, Label& IsInflated) {
1477   assert(UseRTMForStackLocks, "why call this otherwise?");
1478   assert(!UseBiasedLocking, "Biased locking is not supported with RTM locking");
1479   assert(tmpReg == rax, "");
1480   assert(scrReg == rdx, "");
1481   Label L_rtm_retry, L_decrement_retry, L_on_abort;
1482 
1483   if (RTMRetryCount > 0) {
1484     movl(retry_on_abort_count_Reg, RTMRetryCount); // Retry on abort
1485     bind(L_rtm_retry);
1486   }
1487   movptr(tmpReg, Address(objReg, 0));
1488   testptr(tmpReg, markOopDesc::monitor_value);  // inflated vs stack-locked|neutral|biased
1489   jcc(Assembler::notZero, IsInflated);
1490 
1491   if (PrintPreciseRTMLockingStatistics || profile_rtm) {
1492     Label L_noincrement;
1493     if (RTMTotalCountIncrRate > 1) {
1494       // tmpReg, scrReg and flags are killed
1495       branch_on_random_using_rdtsc(tmpReg, scrReg, (int)RTMTotalCountIncrRate, L_noincrement);
1496     }
1497     assert(stack_rtm_counters != NULL, "should not be NULL when profiling RTM");
1498     atomic_incptr(ExternalAddress((address)stack_rtm_counters->total_count_addr()), scrReg);
1499     bind(L_noincrement);
1500   }
1501   xbegin(L_on_abort);
1502   movptr(tmpReg, Address(objReg, 0));       // fetch markword
1503   andptr(tmpReg, markOopDesc::biased_lock_mask_in_place); // look at 3 lock bits
1504   cmpptr(tmpReg, markOopDesc::unlocked_value);            // bits = 001 unlocked
1505   jcc(Assembler::equal, DONE_LABEL);        // all done if unlocked
1506 
1507   Register abort_status_Reg = tmpReg; // status of abort is stored in RAX
1508   if (UseRTMXendForLockBusy) {
1509     xend();
1510     movptr(abort_status_Reg, 0x2);   // Set the abort status to 2 (so we can retry)
1511     jmp(L_decrement_retry);
1512   }
1513   else {
1514     xabort(0);
1515   }
1516   bind(L_on_abort);
1517   if (PrintPreciseRTMLockingStatistics || profile_rtm) {
1518     rtm_profiling(abort_status_Reg, scrReg, stack_rtm_counters, method_data, profile_rtm);
1519   }
1520   bind(L_decrement_retry);
1521   if (RTMRetryCount > 0) {
1522     // retry on lock abort if abort status is 'can retry' (0x2) or 'memory conflict' (0x4)
1523     rtm_retry_lock_on_abort(retry_on_abort_count_Reg, abort_status_Reg, L_rtm_retry);
1524   }
1525 }
1526 
1527 // Use RTM for inflating locks
1528 // inputs: objReg (object to lock)
1529 //         boxReg (on-stack box address (displaced header location) - KILLED)
1530 //         tmpReg (ObjectMonitor address + markOopDesc::monitor_value)
1531 void MacroAssembler::rtm_inflated_locking(Register objReg, Register boxReg, Register tmpReg,
1532                                           Register scrReg, Register retry_on_busy_count_Reg,
1533                                           Register retry_on_abort_count_Reg,
1534                                           RTMLockingCounters* rtm_counters,
1535                                           Metadata* method_data, bool profile_rtm,
1536                                           Label& DONE_LABEL) {
1537   assert(UseRTMLocking, "why call this otherwise?");
1538   assert(tmpReg == rax, "");
1539   assert(scrReg == rdx, "");
1540   Label L_rtm_retry, L_decrement_retry, L_on_abort;
1541   int owner_offset = OM_OFFSET_NO_MONITOR_VALUE_TAG(owner);
1542 
1543   // Without cast to int32_t a movptr will destroy r10 which is typically obj
1544   movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark()));
1545   movptr(boxReg, tmpReg); // Save ObjectMonitor address
1546 
1547   if (RTMRetryCount > 0) {
1548     movl(retry_on_busy_count_Reg, RTMRetryCount);  // Retry on lock busy
1549     movl(retry_on_abort_count_Reg, RTMRetryCount); // Retry on abort
1550     bind(L_rtm_retry);
1551   }
1552   if (PrintPreciseRTMLockingStatistics || profile_rtm) {
1553     Label L_noincrement;
1554     if (RTMTotalCountIncrRate > 1) {
1555       // tmpReg, scrReg and flags are killed
1556       branch_on_random_using_rdtsc(tmpReg, scrReg, (int)RTMTotalCountIncrRate, L_noincrement);
1557     }
1558     assert(rtm_counters != NULL, "should not be NULL when profiling RTM");
1559     atomic_incptr(ExternalAddress((address)rtm_counters->total_count_addr()), scrReg);
1560     bind(L_noincrement);
1561   }
1562   xbegin(L_on_abort);
1563   movptr(tmpReg, Address(objReg, 0));
1564   movptr(tmpReg, Address(tmpReg, owner_offset));
1565   testptr(tmpReg, tmpReg);
1566   jcc(Assembler::zero, DONE_LABEL);
1567   if (UseRTMXendForLockBusy) {
1568     xend();
1569     jmp(L_decrement_retry);
1570   }
1571   else {
1572     xabort(0);
1573   }
1574   bind(L_on_abort);
1575   Register abort_status_Reg = tmpReg; // status of abort is stored in RAX
1576   if (PrintPreciseRTMLockingStatistics || profile_rtm) {
1577     rtm_profiling(abort_status_Reg, scrReg, rtm_counters, method_data, profile_rtm);
1578   }
1579   if (RTMRetryCount > 0) {
1580     // retry on lock abort if abort status is 'can retry' (0x2) or 'memory conflict' (0x4)
1581     rtm_retry_lock_on_abort(retry_on_abort_count_Reg, abort_status_Reg, L_rtm_retry);
1582   }
1583 
1584   movptr(tmpReg, Address(boxReg, owner_offset)) ;
1585   testptr(tmpReg, tmpReg) ;
1586   jccb(Assembler::notZero, L_decrement_retry) ;
1587 
1588   // Appears unlocked - try to swing _owner from null to non-null.
1589   // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
1590 #ifdef _LP64
1591   Register threadReg = r15_thread;
1592 #else
1593   get_thread(scrReg);
1594   Register threadReg = scrReg;
1595 #endif
1596   if (os::is_MP()) {
1597     lock();
1598   }
1599   cmpxchgptr(threadReg, Address(boxReg, owner_offset)); // Updates tmpReg
1600 
1601   if (RTMRetryCount > 0) {
1602     // success done else retry
1603     jccb(Assembler::equal, DONE_LABEL) ;
1604     bind(L_decrement_retry);
1605     // Spin and retry if lock is busy.
1606     rtm_retry_lock_on_busy(retry_on_busy_count_Reg, boxReg, tmpReg, scrReg, L_rtm_retry);
1607   }
1608   else {
1609     bind(L_decrement_retry);
1610   }
1611 }
1612 
1613 #endif //  INCLUDE_RTM_OPT
1614 
1615 // Fast_Lock and Fast_Unlock used by C2
1616 
1617 // Because the transitions from emitted code to the runtime
1618 // monitorenter/exit helper stubs are so slow it's critical that
1619 // we inline both the stack-locking fast-path and the inflated fast path.
1620 //
1621 // See also: cmpFastLock and cmpFastUnlock.
1622 //
1623 // What follows is a specialized inline transliteration of the code
1624 // in slow_enter() and slow_exit().  If we're concerned about I$ bloat
1625 // another option would be to emit TrySlowEnter and TrySlowExit methods
1626 // at startup-time.  These methods would accept arguments as
1627 // (rax,=Obj, rbx=Self, rcx=box, rdx=Scratch) and return success-failure
1628 // indications in the icc.ZFlag.  Fast_Lock and Fast_Unlock would simply
1629 // marshal the arguments and emit calls to TrySlowEnter and TrySlowExit.
1630 // In practice, however, the # of lock sites is bounded and is usually small.
1631 // Besides the call overhead, TrySlowEnter and TrySlowExit might suffer
1632 // if the processor uses simple bimodal branch predictors keyed by EIP
1633 // Since the helper routines would be called from multiple synchronization
1634 // sites.
1635 //
1636 // An even better approach would be write "MonitorEnter()" and "MonitorExit()"
1637 // in java - using j.u.c and unsafe - and just bind the lock and unlock sites
1638 // to those specialized methods.  That'd give us a mostly platform-independent
1639 // implementation that the JITs could optimize and inline at their pleasure.
1640 // Done correctly, the only time we'd need to cross to native could would be
1641 // to park() or unpark() threads.  We'd also need a few more unsafe operators
1642 // to (a) prevent compiler-JIT reordering of non-volatile accesses, and
1643 // (b) explicit barriers or fence operations.
1644 //
1645 // TODO:
1646 //
1647 // *  Arrange for C2 to pass "Self" into Fast_Lock and Fast_Unlock in one of the registers (scr).
1648 //    This avoids manifesting the Self pointer in the Fast_Lock and Fast_Unlock terminals.
1649 //    Given TLAB allocation, Self is usually manifested in a register, so passing it into
1650 //    the lock operators would typically be faster than reifying Self.
1651 //
1652 // *  Ideally I'd define the primitives as:
1653 //       fast_lock   (nax Obj, nax box, EAX tmp, nax scr) where box, tmp and scr are KILLED.
1654 //       fast_unlock (nax Obj, EAX box, nax tmp) where box and tmp are KILLED
1655 //    Unfortunately ADLC bugs prevent us from expressing the ideal form.
1656 //    Instead, we're stuck with a rather awkward and brittle register assignments below.
1657 //    Furthermore the register assignments are overconstrained, possibly resulting in
1658 //    sub-optimal code near the synchronization site.
1659 //
1660 // *  Eliminate the sp-proximity tests and just use "== Self" tests instead.
1661 //    Alternately, use a better sp-proximity test.
1662 //
1663 // *  Currently ObjectMonitor._Owner can hold either an sp value or a (THREAD *) value.
1664 //    Either one is sufficient to uniquely identify a thread.
1665 //    TODO: eliminate use of sp in _owner and use get_thread(tr) instead.
1666 //
1667 // *  Intrinsify notify() and notifyAll() for the common cases where the
1668 //    object is locked by the calling thread but the waitlist is empty.
1669 //    avoid the expensive JNI call to JVM_Notify() and JVM_NotifyAll().
1670 //
1671 // *  use jccb and jmpb instead of jcc and jmp to improve code density.
1672 //    But beware of excessive branch density on AMD Opterons.
1673 //
1674 // *  Both Fast_Lock and Fast_Unlock set the ICC.ZF to indicate success
1675 //    or failure of the fast-path.  If the fast-path fails then we pass
1676 //    control to the slow-path, typically in C.  In Fast_Lock and
1677 //    Fast_Unlock we often branch to DONE_LABEL, just to find that C2
1678 //    will emit a conditional branch immediately after the node.
1679 //    So we have branches to branches and lots of ICC.ZF games.
1680 //    Instead, it might be better to have C2 pass a "FailureLabel"
1681 //    into Fast_Lock and Fast_Unlock.  In the case of success, control
1682 //    will drop through the node.  ICC.ZF is undefined at exit.
1683 //    In the case of failure, the node will branch directly to the
1684 //    FailureLabel
1685 
1686 
1687 // obj: object to lock
1688 // box: on-stack box address (displaced header location) - KILLED
1689 // rax,: tmp -- KILLED
1690 // scr: tmp -- KILLED
1691 void MacroAssembler::fast_lock(Register objReg, Register boxReg, Register tmpReg,
1692                                Register scrReg, Register cx1Reg, Register cx2Reg,
1693                                BiasedLockingCounters* counters,
1694                                RTMLockingCounters* rtm_counters,
1695                                RTMLockingCounters* stack_rtm_counters,
1696                                Metadata* method_data,
1697                                bool use_rtm, bool profile_rtm) {
1698   // Ensure the register assignents are disjoint
1699   assert(tmpReg == rax, "");
1700 
1701   if (use_rtm) {
1702     assert_different_registers(objReg, boxReg, tmpReg, scrReg, cx1Reg, cx2Reg);
1703   } else {
1704     assert(cx1Reg == noreg, "");
1705     assert(cx2Reg == noreg, "");
1706     assert_different_registers(objReg, boxReg, tmpReg, scrReg);
1707   }
1708 
1709   if (counters != NULL) {
1710     atomic_incl(ExternalAddress((address)counters->total_entry_count_addr()), scrReg);
1711   }
1712   if (EmitSync & 1) {
1713       // set box->dhw = markOopDesc::unused_mark()
1714       // Force all sync thru slow-path: slow_enter() and slow_exit()
1715       movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark()));
1716       cmpptr (rsp, (int32_t)NULL_WORD);
1717   } else {
1718     // Possible cases that we'll encounter in fast_lock
1719     // ------------------------------------------------
1720     // * Inflated
1721     //    -- unlocked
1722     //    -- Locked
1723     //       = by self
1724     //       = by other
1725     // * biased
1726     //    -- by Self
1727     //    -- by other
1728     // * neutral
1729     // * stack-locked
1730     //    -- by self
1731     //       = sp-proximity test hits
1732     //       = sp-proximity test generates false-negative
1733     //    -- by other
1734     //
1735 
1736     Label IsInflated, DONE_LABEL;
1737 
1738     // it's stack-locked, biased or neutral
1739     // TODO: optimize away redundant LDs of obj->mark and improve the markword triage
1740     // order to reduce the number of conditional branches in the most common cases.
1741     // Beware -- there's a subtle invariant that fetch of the markword
1742     // at [FETCH], below, will never observe a biased encoding (*101b).
1743     // If this invariant is not held we risk exclusion (safety) failure.
1744     if (UseBiasedLocking && !UseOptoBiasInlining) {
1745       biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, counters);
1746     }
1747 
1748 #if INCLUDE_RTM_OPT
1749     if (UseRTMForStackLocks && use_rtm) {
1750       rtm_stack_locking(objReg, tmpReg, scrReg, cx2Reg,
1751                         stack_rtm_counters, method_data, profile_rtm,
1752                         DONE_LABEL, IsInflated);
1753     }
1754 #endif // INCLUDE_RTM_OPT
1755 
1756     movptr(tmpReg, Address(objReg, 0));          // [FETCH]
1757     testptr(tmpReg, markOopDesc::monitor_value); // inflated vs stack-locked|neutral|biased
1758     jccb(Assembler::notZero, IsInflated);
1759 
1760     // Attempt stack-locking ...
1761     orptr (tmpReg, markOopDesc::unlocked_value);
1762     movptr(Address(boxReg, 0), tmpReg);          // Anticipate successful CAS
1763     if (os::is_MP()) {
1764       lock();
1765     }
1766     cmpxchgptr(boxReg, Address(objReg, 0));      // Updates tmpReg
1767     if (counters != NULL) {
1768       cond_inc32(Assembler::equal,
1769                  ExternalAddress((address)counters->fast_path_entry_count_addr()));
1770     }
1771     jcc(Assembler::equal, DONE_LABEL);           // Success
1772 
1773     // Recursive locking.
1774     // The object is stack-locked: markword contains stack pointer to BasicLock.
1775     // Locked by current thread if difference with current SP is less than one page.
1776     subptr(tmpReg, rsp);
1777     // Next instruction set ZFlag == 1 (Success) if difference is less then one page.
1778     andptr(tmpReg, (int32_t) (NOT_LP64(0xFFFFF003) LP64_ONLY(7 - os::vm_page_size())) );
1779     movptr(Address(boxReg, 0), tmpReg);
1780     if (counters != NULL) {
1781       cond_inc32(Assembler::equal,
1782                  ExternalAddress((address)counters->fast_path_entry_count_addr()));
1783     }
1784     jmp(DONE_LABEL);
1785 
1786     bind(IsInflated);
1787     // The object is inflated. tmpReg contains pointer to ObjectMonitor* + markOopDesc::monitor_value
1788 
1789 #if INCLUDE_RTM_OPT
1790     // Use the same RTM locking code in 32- and 64-bit VM.
1791     if (use_rtm) {
1792       rtm_inflated_locking(objReg, boxReg, tmpReg, scrReg, cx1Reg, cx2Reg,
1793                            rtm_counters, method_data, profile_rtm, DONE_LABEL);
1794     } else {
1795 #endif // INCLUDE_RTM_OPT
1796 
1797 #ifndef _LP64
1798     // The object is inflated.
1799 
1800     // boxReg refers to the on-stack BasicLock in the current frame.
1801     // We'd like to write:
1802     //   set box->_displaced_header = markOopDesc::unused_mark().  Any non-0 value suffices.
1803     // This is convenient but results a ST-before-CAS penalty.  The following CAS suffers
1804     // additional latency as we have another ST in the store buffer that must drain.
1805 
1806     if (EmitSync & 8192) {
1807        movptr(Address(boxReg, 0), 3);            // results in ST-before-CAS penalty
1808        get_thread (scrReg);
1809        movptr(boxReg, tmpReg);                    // consider: LEA box, [tmp-2]
1810        movptr(tmpReg, NULL_WORD);                 // consider: xor vs mov
1811        if (os::is_MP()) {
1812          lock();
1813        }
1814        cmpxchgptr(scrReg, Address(boxReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1815     } else
1816     if ((EmitSync & 128) == 0) {                      // avoid ST-before-CAS
1817        // register juggle because we need tmpReg for cmpxchgptr below
1818        movptr(scrReg, boxReg);
1819        movptr(boxReg, tmpReg);                   // consider: LEA box, [tmp-2]
1820 
1821        // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
1822        if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
1823           // prefetchw [eax + Offset(_owner)-2]
1824           prefetchw(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1825        }
1826 
1827        if ((EmitSync & 64) == 0) {
1828          // Optimistic form: consider XORL tmpReg,tmpReg
1829          movptr(tmpReg, NULL_WORD);
1830        } else {
1831          // Can suffer RTS->RTO upgrades on shared or cold $ lines
1832          // Test-And-CAS instead of CAS
1833          movptr(tmpReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));   // rax, = m->_owner
1834          testptr(tmpReg, tmpReg);                   // Locked ?
1835          jccb  (Assembler::notZero, DONE_LABEL);
1836        }
1837 
1838        // Appears unlocked - try to swing _owner from null to non-null.
1839        // Ideally, I'd manifest "Self" with get_thread and then attempt
1840        // to CAS the register containing Self into m->Owner.
1841        // But we don't have enough registers, so instead we can either try to CAS
1842        // rsp or the address of the box (in scr) into &m->owner.  If the CAS succeeds
1843        // we later store "Self" into m->Owner.  Transiently storing a stack address
1844        // (rsp or the address of the box) into  m->owner is harmless.
1845        // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
1846        if (os::is_MP()) {
1847          lock();
1848        }
1849        cmpxchgptr(scrReg, Address(boxReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1850        movptr(Address(scrReg, 0), 3);          // box->_displaced_header = 3
1851        // If we weren't able to swing _owner from NULL to the BasicLock
1852        // then take the slow path.
1853        jccb  (Assembler::notZero, DONE_LABEL);
1854        // update _owner from BasicLock to thread
1855        get_thread (scrReg);                    // beware: clobbers ICCs
1856        movptr(Address(boxReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), scrReg);
1857        xorptr(boxReg, boxReg);                 // set icc.ZFlag = 1 to indicate success
1858 
1859        // If the CAS fails we can either retry or pass control to the slow-path.
1860        // We use the latter tactic.
1861        // Pass the CAS result in the icc.ZFlag into DONE_LABEL
1862        // If the CAS was successful ...
1863        //   Self has acquired the lock
1864        //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
1865        // Intentional fall-through into DONE_LABEL ...
1866     } else {
1867        movptr(Address(boxReg, 0), intptr_t(markOopDesc::unused_mark()));  // results in ST-before-CAS penalty
1868        movptr(boxReg, tmpReg);
1869 
1870        // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
1871        if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
1872           // prefetchw [eax + Offset(_owner)-2]
1873           prefetchw(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1874        }
1875 
1876        if ((EmitSync & 64) == 0) {
1877          // Optimistic form
1878          xorptr  (tmpReg, tmpReg);
1879        } else {
1880          // Can suffer RTS->RTO upgrades on shared or cold $ lines
1881          movptr(tmpReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));   // rax, = m->_owner
1882          testptr(tmpReg, tmpReg);                   // Locked ?
1883          jccb  (Assembler::notZero, DONE_LABEL);
1884        }
1885 
1886        // Appears unlocked - try to swing _owner from null to non-null.
1887        // Use either "Self" (in scr) or rsp as thread identity in _owner.
1888        // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
1889        get_thread (scrReg);
1890        if (os::is_MP()) {
1891          lock();
1892        }
1893        cmpxchgptr(scrReg, Address(boxReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1894 
1895        // If the CAS fails we can either retry or pass control to the slow-path.
1896        // We use the latter tactic.
1897        // Pass the CAS result in the icc.ZFlag into DONE_LABEL
1898        // If the CAS was successful ...
1899        //   Self has acquired the lock
1900        //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
1901        // Intentional fall-through into DONE_LABEL ...
1902     }
1903 #else // _LP64
1904     // It's inflated
1905     movq(scrReg, tmpReg);
1906     xorq(tmpReg, tmpReg);
1907 
1908     if (os::is_MP()) {
1909       lock();
1910     }
1911     cmpxchgptr(r15_thread, Address(scrReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
1912     // Unconditionally set box->_displaced_header = markOopDesc::unused_mark().
1913     // Without cast to int32_t movptr will destroy r10 which is typically obj.
1914     movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark()));
1915     // Intentional fall-through into DONE_LABEL ...
1916     // Propagate ICC.ZF from CAS above into DONE_LABEL.
1917 #endif // _LP64
1918 #if INCLUDE_RTM_OPT
1919     } // use_rtm()
1920 #endif
1921     // DONE_LABEL is a hot target - we'd really like to place it at the
1922     // start of cache line by padding with NOPs.
1923     // See the AMD and Intel software optimization manuals for the
1924     // most efficient "long" NOP encodings.
1925     // Unfortunately none of our alignment mechanisms suffice.
1926     bind(DONE_LABEL);
1927 
1928     // At DONE_LABEL the icc ZFlag is set as follows ...
1929     // Fast_Unlock uses the same protocol.
1930     // ZFlag == 1 -> Success
1931     // ZFlag == 0 -> Failure - force control through the slow-path
1932   }
1933 }
1934 
1935 // obj: object to unlock
1936 // box: box address (displaced header location), killed.  Must be EAX.
1937 // tmp: killed, cannot be obj nor box.
1938 //
1939 // Some commentary on balanced locking:
1940 //
1941 // Fast_Lock and Fast_Unlock are emitted only for provably balanced lock sites.
1942 // Methods that don't have provably balanced locking are forced to run in the
1943 // interpreter - such methods won't be compiled to use fast_lock and fast_unlock.
1944 // The interpreter provides two properties:
1945 // I1:  At return-time the interpreter automatically and quietly unlocks any
1946 //      objects acquired the current activation (frame).  Recall that the
1947 //      interpreter maintains an on-stack list of locks currently held by
1948 //      a frame.
1949 // I2:  If a method attempts to unlock an object that is not held by the
1950 //      the frame the interpreter throws IMSX.
1951 //
1952 // Lets say A(), which has provably balanced locking, acquires O and then calls B().
1953 // B() doesn't have provably balanced locking so it runs in the interpreter.
1954 // Control returns to A() and A() unlocks O.  By I1 and I2, above, we know that O
1955 // is still locked by A().
1956 //
1957 // The only other source of unbalanced locking would be JNI.  The "Java Native Interface:
1958 // Programmer's Guide and Specification" claims that an object locked by jni_monitorenter
1959 // should not be unlocked by "normal" java-level locking and vice-versa.  The specification
1960 // doesn't specify what will occur if a program engages in such mixed-mode locking, however.
1961 // Arguably given that the spec legislates the JNI case as undefined our implementation
1962 // could reasonably *avoid* checking owner in Fast_Unlock().
1963 // In the interest of performance we elide m->Owner==Self check in unlock.
1964 // A perfectly viable alternative is to elide the owner check except when
1965 // Xcheck:jni is enabled.
1966 
1967 void MacroAssembler::fast_unlock(Register objReg, Register boxReg, Register tmpReg, bool use_rtm) {
1968   assert(boxReg == rax, "");
1969   assert_different_registers(objReg, boxReg, tmpReg);
1970 
1971   if (EmitSync & 4) {
1972     // Disable - inhibit all inlining.  Force control through the slow-path
1973     cmpptr (rsp, 0);
1974   } else {
1975     Label DONE_LABEL, Stacked, CheckSucc;
1976 
1977     // Critically, the biased locking test must have precedence over
1978     // and appear before the (box->dhw == 0) recursive stack-lock test.
1979     if (UseBiasedLocking && !UseOptoBiasInlining) {
1980        biased_locking_exit(objReg, tmpReg, DONE_LABEL);
1981     }
1982 
1983 #if INCLUDE_RTM_OPT
1984     if (UseRTMForStackLocks && use_rtm) {
1985       assert(!UseBiasedLocking, "Biased locking is not supported with RTM locking");
1986       Label L_regular_unlock;
1987       movptr(tmpReg, Address(objReg, 0));           // fetch markword
1988       andptr(tmpReg, markOopDesc::biased_lock_mask_in_place); // look at 3 lock bits
1989       cmpptr(tmpReg, markOopDesc::unlocked_value);            // bits = 001 unlocked
1990       jccb(Assembler::notEqual, L_regular_unlock);  // if !HLE RegularLock
1991       xend();                                       // otherwise end...
1992       jmp(DONE_LABEL);                              // ... and we're done
1993       bind(L_regular_unlock);
1994     }
1995 #endif
1996 
1997     cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD); // Examine the displaced header
1998     jcc   (Assembler::zero, DONE_LABEL);            // 0 indicates recursive stack-lock
1999     movptr(tmpReg, Address(objReg, 0));             // Examine the object's markword
2000     testptr(tmpReg, markOopDesc::monitor_value);    // Inflated?
2001     jccb  (Assembler::zero, Stacked);
2002 
2003     // It's inflated.
2004 #if INCLUDE_RTM_OPT
2005     if (use_rtm) {
2006       Label L_regular_inflated_unlock;
2007       int owner_offset = OM_OFFSET_NO_MONITOR_VALUE_TAG(owner);
2008       movptr(boxReg, Address(tmpReg, owner_offset));
2009       testptr(boxReg, boxReg);
2010       jccb(Assembler::notZero, L_regular_inflated_unlock);
2011       xend();
2012       jmpb(DONE_LABEL);
2013       bind(L_regular_inflated_unlock);
2014     }
2015 #endif
2016 
2017     // Despite our balanced locking property we still check that m->_owner == Self
2018     // as java routines or native JNI code called by this thread might
2019     // have released the lock.
2020     // Refer to the comments in synchronizer.cpp for how we might encode extra
2021     // state in _succ so we can avoid fetching EntryList|cxq.
2022     //
2023     // I'd like to add more cases in fast_lock() and fast_unlock() --
2024     // such as recursive enter and exit -- but we have to be wary of
2025     // I$ bloat, T$ effects and BP$ effects.
2026     //
2027     // If there's no contention try a 1-0 exit.  That is, exit without
2028     // a costly MEMBAR or CAS.  See synchronizer.cpp for details on how
2029     // we detect and recover from the race that the 1-0 exit admits.
2030     //
2031     // Conceptually Fast_Unlock() must execute a STST|LDST "release" barrier
2032     // before it STs null into _owner, releasing the lock.  Updates
2033     // to data protected by the critical section must be visible before
2034     // we drop the lock (and thus before any other thread could acquire
2035     // the lock and observe the fields protected by the lock).
2036     // IA32's memory-model is SPO, so STs are ordered with respect to
2037     // each other and there's no need for an explicit barrier (fence).
2038     // See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
2039 #ifndef _LP64
2040     get_thread (boxReg);
2041     if ((EmitSync & 4096) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
2042       // prefetchw [ebx + Offset(_owner)-2]
2043       prefetchw(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
2044     }
2045 
2046     // Note that we could employ various encoding schemes to reduce
2047     // the number of loads below (currently 4) to just 2 or 3.
2048     // Refer to the comments in synchronizer.cpp.
2049     // In practice the chain of fetches doesn't seem to impact performance, however.
2050     xorptr(boxReg, boxReg);
2051     if ((EmitSync & 65536) == 0 && (EmitSync & 256)) {
2052        // Attempt to reduce branch density - AMD's branch predictor.
2053        orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(recursions)));
2054        orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(EntryList)));
2055        orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(cxq)));
2056        jccb  (Assembler::notZero, DONE_LABEL);
2057        movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), NULL_WORD);
2058        jmpb  (DONE_LABEL);
2059     } else {
2060        orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(recursions)));
2061        jccb  (Assembler::notZero, DONE_LABEL);
2062        movptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(EntryList)));
2063        orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(cxq)));
2064        jccb  (Assembler::notZero, CheckSucc);
2065        movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), NULL_WORD);
2066        jmpb  (DONE_LABEL);
2067     }
2068 
2069     // The Following code fragment (EmitSync & 65536) improves the performance of
2070     // contended applications and contended synchronization microbenchmarks.
2071     // Unfortunately the emission of the code - even though not executed - causes regressions
2072     // in scimark and jetstream, evidently because of $ effects.  Replacing the code
2073     // with an equal number of never-executed NOPs results in the same regression.
2074     // We leave it off by default.
2075 
2076     if ((EmitSync & 65536) != 0) {
2077        Label LSuccess, LGoSlowPath ;
2078 
2079        bind  (CheckSucc);
2080 
2081        // Optional pre-test ... it's safe to elide this
2082        cmpptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(succ)), (int32_t)NULL_WORD);
2083        jccb(Assembler::zero, LGoSlowPath);
2084 
2085        // We have a classic Dekker-style idiom:
2086        //    ST m->_owner = 0 ; MEMBAR; LD m->_succ
2087        // There are a number of ways to implement the barrier:
2088        // (1) lock:andl &m->_owner, 0
2089        //     is fast, but mask doesn't currently support the "ANDL M,IMM32" form.
2090        //     LOCK: ANDL [ebx+Offset(_Owner)-2], 0
2091        //     Encodes as 81 31 OFF32 IMM32 or 83 63 OFF8 IMM8
2092        // (2) If supported, an explicit MFENCE is appealing.
2093        //     In older IA32 processors MFENCE is slower than lock:add or xchg
2094        //     particularly if the write-buffer is full as might be the case if
2095        //     if stores closely precede the fence or fence-equivalent instruction.
2096        //     See https://blogs.oracle.com/dave/entry/instruction_selection_for_volatile_fences
2097        //     as the situation has changed with Nehalem and Shanghai.
2098        // (3) In lieu of an explicit fence, use lock:addl to the top-of-stack
2099        //     The $lines underlying the top-of-stack should be in M-state.
2100        //     The locked add instruction is serializing, of course.
2101        // (4) Use xchg, which is serializing
2102        //     mov boxReg, 0; xchgl boxReg, [tmpReg + Offset(_owner)-2] also works
2103        // (5) ST m->_owner = 0 and then execute lock:orl &m->_succ, 0.
2104        //     The integer condition codes will tell us if succ was 0.
2105        //     Since _succ and _owner should reside in the same $line and
2106        //     we just stored into _owner, it's likely that the $line
2107        //     remains in M-state for the lock:orl.
2108        //
2109        // We currently use (3), although it's likely that switching to (2)
2110        // is correct for the future.
2111 
2112        movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), NULL_WORD);
2113        if (os::is_MP()) {
2114          lock(); addptr(Address(rsp, 0), 0);
2115        }
2116        // Ratify _succ remains non-null
2117        cmpptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(succ)), 0);
2118        jccb  (Assembler::notZero, LSuccess);
2119 
2120        xorptr(boxReg, boxReg);                  // box is really EAX
2121        if (os::is_MP()) { lock(); }
2122        cmpxchgptr(rsp, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
2123        // There's no successor so we tried to regrab the lock with the
2124        // placeholder value. If that didn't work, then another thread
2125        // grabbed the lock so we're done (and exit was a success).
2126        jccb  (Assembler::notEqual, LSuccess);
2127        // Since we're low on registers we installed rsp as a placeholding in _owner.
2128        // Now install Self over rsp.  This is safe as we're transitioning from
2129        // non-null to non=null
2130        get_thread (boxReg);
2131        movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), boxReg);
2132        // Intentional fall-through into LGoSlowPath ...
2133 
2134        bind  (LGoSlowPath);
2135        orptr(boxReg, 1);                      // set ICC.ZF=0 to indicate failure
2136        jmpb  (DONE_LABEL);
2137 
2138        bind  (LSuccess);
2139        xorptr(boxReg, boxReg);                 // set ICC.ZF=1 to indicate success
2140        jmpb  (DONE_LABEL);
2141     }
2142 
2143     bind (Stacked);
2144     // It's not inflated and it's not recursively stack-locked and it's not biased.
2145     // It must be stack-locked.
2146     // Try to reset the header to displaced header.
2147     // The "box" value on the stack is stable, so we can reload
2148     // and be assured we observe the same value as above.
2149     movptr(tmpReg, Address(boxReg, 0));
2150     if (os::is_MP()) {
2151       lock();
2152     }
2153     cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
2154     // Intention fall-thru into DONE_LABEL
2155 
2156     // DONE_LABEL is a hot target - we'd really like to place it at the
2157     // start of cache line by padding with NOPs.
2158     // See the AMD and Intel software optimization manuals for the
2159     // most efficient "long" NOP encodings.
2160     // Unfortunately none of our alignment mechanisms suffice.
2161     if ((EmitSync & 65536) == 0) {
2162        bind (CheckSucc);
2163     }
2164 #else // _LP64
2165     // It's inflated
2166     if (EmitSync & 1024) {
2167       // Emit code to check that _owner == Self
2168       // We could fold the _owner test into subsequent code more efficiently
2169       // than using a stand-alone check, but since _owner checking is off by
2170       // default we don't bother. We also might consider predicating the
2171       // _owner==Self check on Xcheck:jni or running on a debug build.
2172       movptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
2173       xorptr(boxReg, r15_thread);
2174     } else {
2175       xorptr(boxReg, boxReg);
2176     }
2177     orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(recursions)));
2178     jccb  (Assembler::notZero, DONE_LABEL);
2179     movptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(cxq)));
2180     orptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(EntryList)));
2181     jccb  (Assembler::notZero, CheckSucc);
2182     movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), (int32_t)NULL_WORD);
2183     jmpb  (DONE_LABEL);
2184 
2185     if ((EmitSync & 65536) == 0) {
2186       // Try to avoid passing control into the slow_path ...
2187       Label LSuccess, LGoSlowPath ;
2188       bind  (CheckSucc);
2189 
2190       // The following optional optimization can be elided if necessary
2191       // Effectively: if (succ == null) goto SlowPath
2192       // The code reduces the window for a race, however,
2193       // and thus benefits performance.
2194       cmpptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(succ)), (int32_t)NULL_WORD);
2195       jccb  (Assembler::zero, LGoSlowPath);
2196 
2197       if ((EmitSync & 16) && os::is_MP()) {
2198         orptr(boxReg, boxReg);
2199         xchgptr(boxReg, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
2200       } else {
2201         movptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)), (int32_t)NULL_WORD);
2202         if (os::is_MP()) {
2203           // Memory barrier/fence
2204           // Dekker pivot point -- fulcrum : ST Owner; MEMBAR; LD Succ
2205           // Instead of MFENCE we use a dummy locked add of 0 to the top-of-stack.
2206           // This is faster on Nehalem and AMD Shanghai/Barcelona.
2207           // See https://blogs.oracle.com/dave/entry/instruction_selection_for_volatile_fences
2208           // We might also restructure (ST Owner=0;barrier;LD _Succ) to
2209           // (mov box,0; xchgq box, &m->Owner; LD _succ) .
2210           lock(); addl(Address(rsp, 0), 0);
2211         }
2212       }
2213       cmpptr(Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(succ)), (int32_t)NULL_WORD);
2214       jccb  (Assembler::notZero, LSuccess);
2215 
2216       // Rare inopportune interleaving - race.
2217       // The successor vanished in the small window above.
2218       // The lock is contended -- (cxq|EntryList) != null -- and there's no apparent successor.
2219       // We need to ensure progress and succession.
2220       // Try to reacquire the lock.
2221       // If that fails then the new owner is responsible for succession and this
2222       // thread needs to take no further action and can exit via the fast path (success).
2223       // If the re-acquire succeeds then pass control into the slow path.
2224       // As implemented, this latter mode is horrible because we generated more
2225       // coherence traffic on the lock *and* artifically extended the critical section
2226       // length while by virtue of passing control into the slow path.
2227 
2228       // box is really RAX -- the following CMPXCHG depends on that binding
2229       // cmpxchg R,[M] is equivalent to rax = CAS(M,rax,R)
2230       movptr(boxReg, (int32_t)NULL_WORD);
2231       if (os::is_MP()) { lock(); }
2232       cmpxchgptr(r15_thread, Address(tmpReg, OM_OFFSET_NO_MONITOR_VALUE_TAG(owner)));
2233       // There's no successor so we tried to regrab the lock.
2234       // If that didn't work, then another thread grabbed the
2235       // lock so we're done (and exit was a success).
2236       jccb  (Assembler::notEqual, LSuccess);
2237       // Intentional fall-through into slow-path
2238 
2239       bind  (LGoSlowPath);
2240       orl   (boxReg, 1);                      // set ICC.ZF=0 to indicate failure
2241       jmpb  (DONE_LABEL);
2242 
2243       bind  (LSuccess);
2244       testl (boxReg, 0);                      // set ICC.ZF=1 to indicate success
2245       jmpb  (DONE_LABEL);
2246     }
2247 
2248     bind  (Stacked);
2249     movptr(tmpReg, Address (boxReg, 0));      // re-fetch
2250     if (os::is_MP()) { lock(); }
2251     cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
2252 
2253     if (EmitSync & 65536) {
2254        bind (CheckSucc);
2255     }
2256 #endif
2257     bind(DONE_LABEL);
2258   }
2259 }
2260 #endif // COMPILER2
2261 
2262 void MacroAssembler::c2bool(Register x) {
2263   // implements x == 0 ? 0 : 1
2264   // note: must only look at least-significant byte of x
2265   //       since C-style booleans are stored in one byte
2266   //       only! (was bug)
2267   andl(x, 0xFF);
2268   setb(Assembler::notZero, x);
2269 }
2270 
2271 // Wouldn't need if AddressLiteral version had new name
2272 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
2273   Assembler::call(L, rtype);
2274 }
2275 
2276 void MacroAssembler::call(Register entry) {
2277   Assembler::call(entry);
2278 }
2279 
2280 void MacroAssembler::call(AddressLiteral entry) {
2281   if (reachable(entry)) {
2282     Assembler::call_literal(entry.target(), entry.rspec());
2283   } else {
2284     lea(rscratch1, entry);
2285     Assembler::call(rscratch1);
2286   }
2287 }
2288 
2289 void MacroAssembler::ic_call(address entry, jint method_index) {
2290   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
2291   movptr(rax, (intptr_t)Universe::non_oop_word());
2292   call(AddressLiteral(entry, rh));
2293 }
2294 
2295 // Implementation of call_VM versions
2296 
2297 void MacroAssembler::call_VM(Register oop_result,
2298                              address entry_point,
2299                              bool check_exceptions) {
2300   Label C, E;
2301   call(C, relocInfo::none);
2302   jmp(E);
2303 
2304   bind(C);
2305   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
2306   ret(0);
2307 
2308   bind(E);
2309 }
2310 
2311 void MacroAssembler::call_VM(Register oop_result,
2312                              address entry_point,
2313                              Register arg_1,
2314                              bool check_exceptions) {
2315   Label C, E;
2316   call(C, relocInfo::none);
2317   jmp(E);
2318 
2319   bind(C);
2320   pass_arg1(this, arg_1);
2321   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
2322   ret(0);
2323 
2324   bind(E);
2325 }
2326 
2327 void MacroAssembler::call_VM(Register oop_result,
2328                              address entry_point,
2329                              Register arg_1,
2330                              Register arg_2,
2331                              bool check_exceptions) {
2332   Label C, E;
2333   call(C, relocInfo::none);
2334   jmp(E);
2335 
2336   bind(C);
2337 
2338   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2339 
2340   pass_arg2(this, arg_2);
2341   pass_arg1(this, arg_1);
2342   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
2343   ret(0);
2344 
2345   bind(E);
2346 }
2347 
2348 void MacroAssembler::call_VM(Register oop_result,
2349                              address entry_point,
2350                              Register arg_1,
2351                              Register arg_2,
2352                              Register arg_3,
2353                              bool check_exceptions) {
2354   Label C, E;
2355   call(C, relocInfo::none);
2356   jmp(E);
2357 
2358   bind(C);
2359 
2360   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
2361   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
2362   pass_arg3(this, arg_3);
2363 
2364   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2365   pass_arg2(this, arg_2);
2366 
2367   pass_arg1(this, arg_1);
2368   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
2369   ret(0);
2370 
2371   bind(E);
2372 }
2373 
2374 void MacroAssembler::call_VM(Register oop_result,
2375                              Register last_java_sp,
2376                              address entry_point,
2377                              int number_of_arguments,
2378                              bool check_exceptions) {
2379   Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2380   call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
2381 }
2382 
2383 void MacroAssembler::call_VM(Register oop_result,
2384                              Register last_java_sp,
2385                              address entry_point,
2386                              Register arg_1,
2387                              bool check_exceptions) {
2388   pass_arg1(this, arg_1);
2389   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
2390 }
2391 
2392 void MacroAssembler::call_VM(Register oop_result,
2393                              Register last_java_sp,
2394                              address entry_point,
2395                              Register arg_1,
2396                              Register arg_2,
2397                              bool check_exceptions) {
2398 
2399   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2400   pass_arg2(this, arg_2);
2401   pass_arg1(this, arg_1);
2402   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
2403 }
2404 
2405 void MacroAssembler::call_VM(Register oop_result,
2406                              Register last_java_sp,
2407                              address entry_point,
2408                              Register arg_1,
2409                              Register arg_2,
2410                              Register arg_3,
2411                              bool check_exceptions) {
2412   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
2413   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
2414   pass_arg3(this, arg_3);
2415   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2416   pass_arg2(this, arg_2);
2417   pass_arg1(this, arg_1);
2418   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
2419 }
2420 
2421 void MacroAssembler::super_call_VM(Register oop_result,
2422                                    Register last_java_sp,
2423                                    address entry_point,
2424                                    int number_of_arguments,
2425                                    bool check_exceptions) {
2426   Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
2427   MacroAssembler::call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
2428 }
2429 
2430 void MacroAssembler::super_call_VM(Register oop_result,
2431                                    Register last_java_sp,
2432                                    address entry_point,
2433                                    Register arg_1,
2434                                    bool check_exceptions) {
2435   pass_arg1(this, arg_1);
2436   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
2437 }
2438 
2439 void MacroAssembler::super_call_VM(Register oop_result,
2440                                    Register last_java_sp,
2441                                    address entry_point,
2442                                    Register arg_1,
2443                                    Register arg_2,
2444                                    bool check_exceptions) {
2445 
2446   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2447   pass_arg2(this, arg_2);
2448   pass_arg1(this, arg_1);
2449   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
2450 }
2451 
2452 void MacroAssembler::super_call_VM(Register oop_result,
2453                                    Register last_java_sp,
2454                                    address entry_point,
2455                                    Register arg_1,
2456                                    Register arg_2,
2457                                    Register arg_3,
2458                                    bool check_exceptions) {
2459   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
2460   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
2461   pass_arg3(this, arg_3);
2462   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2463   pass_arg2(this, arg_2);
2464   pass_arg1(this, arg_1);
2465   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
2466 }
2467 
2468 void MacroAssembler::call_VM_base(Register oop_result,
2469                                   Register java_thread,
2470                                   Register last_java_sp,
2471                                   address  entry_point,
2472                                   int      number_of_arguments,
2473                                   bool     check_exceptions) {
2474   // determine java_thread register
2475   if (!java_thread->is_valid()) {
2476 #ifdef _LP64
2477     java_thread = r15_thread;
2478 #else
2479     java_thread = rdi;
2480     get_thread(java_thread);
2481 #endif // LP64
2482   }
2483   // determine last_java_sp register
2484   if (!last_java_sp->is_valid()) {
2485     last_java_sp = rsp;
2486   }
2487   // debugging support
2488   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
2489   LP64_ONLY(assert(java_thread == r15_thread, "unexpected register"));
2490 #ifdef ASSERT
2491   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
2492   // r12 is the heapbase.
2493   LP64_ONLY(if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");)
2494 #endif // ASSERT
2495 
2496   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
2497   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
2498 
2499   // push java thread (becomes first argument of C function)
2500 
2501   NOT_LP64(push(java_thread); number_of_arguments++);
2502   LP64_ONLY(mov(c_rarg0, r15_thread));
2503 
2504   // set last Java frame before call
2505   assert(last_java_sp != rbp, "can't use ebp/rbp");
2506 
2507   // Only interpreter should have to set fp
2508   set_last_Java_frame(java_thread, last_java_sp, rbp, NULL);
2509 
2510   // do the call, remove parameters
2511   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
2512 
2513   // restore the thread (cannot use the pushed argument since arguments
2514   // may be overwritten by C code generated by an optimizing compiler);
2515   // however can use the register value directly if it is callee saved.
2516   if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) {
2517     // rdi & rsi (also r15) are callee saved -> nothing to do
2518 #ifdef ASSERT
2519     guarantee(java_thread != rax, "change this code");
2520     push(rax);
2521     { Label L;
2522       get_thread(rax);
2523       cmpptr(java_thread, rax);
2524       jcc(Assembler::equal, L);
2525       STOP("MacroAssembler::call_VM_base: rdi not callee saved?");
2526       bind(L);
2527     }
2528     pop(rax);
2529 #endif
2530   } else {
2531     get_thread(java_thread);
2532   }
2533   // reset last Java frame
2534   // Only interpreter should have to clear fp
2535   reset_last_Java_frame(java_thread, true, false);
2536 
2537    // C++ interp handles this in the interpreter
2538   check_and_handle_popframe(java_thread);
2539   check_and_handle_earlyret(java_thread);
2540 
2541   if (check_exceptions) {
2542     // check for pending exceptions (java_thread is set upon return)
2543     cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
2544 #ifndef _LP64
2545     jump_cc(Assembler::notEqual,
2546             RuntimeAddress(StubRoutines::forward_exception_entry()));
2547 #else
2548     // This used to conditionally jump to forward_exception however it is
2549     // possible if we relocate that the branch will not reach. So we must jump
2550     // around so we can always reach
2551 
2552     Label ok;
2553     jcc(Assembler::equal, ok);
2554     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2555     bind(ok);
2556 #endif // LP64
2557   }
2558 
2559   // get oop result if there is one and reset the value in the thread
2560   if (oop_result->is_valid()) {
2561     get_vm_result(oop_result, java_thread);
2562   }
2563 }
2564 
2565 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
2566 
2567   // Calculate the value for last_Java_sp
2568   // somewhat subtle. call_VM does an intermediate call
2569   // which places a return address on the stack just under the
2570   // stack pointer as the user finsihed with it. This allows
2571   // use to retrieve last_Java_pc from last_Java_sp[-1].
2572   // On 32bit we then have to push additional args on the stack to accomplish
2573   // the actual requested call. On 64bit call_VM only can use register args
2574   // so the only extra space is the return address that call_VM created.
2575   // This hopefully explains the calculations here.
2576 
2577 #ifdef _LP64
2578   // We've pushed one address, correct last_Java_sp
2579   lea(rax, Address(rsp, wordSize));
2580 #else
2581   lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize));
2582 #endif // LP64
2583 
2584   call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions);
2585 
2586 }
2587 
2588 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2589   call_VM_leaf_base(entry_point, number_of_arguments);
2590 }
2591 
2592 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2593   pass_arg0(this, arg_0);
2594   call_VM_leaf(entry_point, 1);
2595 }
2596 
2597 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2598 
2599   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
2600   pass_arg1(this, arg_1);
2601   pass_arg0(this, arg_0);
2602   call_VM_leaf(entry_point, 2);
2603 }
2604 
2605 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2606   LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
2607   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2608   pass_arg2(this, arg_2);
2609   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
2610   pass_arg1(this, arg_1);
2611   pass_arg0(this, arg_0);
2612   call_VM_leaf(entry_point, 3);
2613 }
2614 
2615 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2616   pass_arg0(this, arg_0);
2617   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2618 }
2619 
2620 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2621 
2622   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
2623   pass_arg1(this, arg_1);
2624   pass_arg0(this, arg_0);
2625   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2626 }
2627 
2628 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2629   LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
2630   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2631   pass_arg2(this, arg_2);
2632   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
2633   pass_arg1(this, arg_1);
2634   pass_arg0(this, arg_0);
2635   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2636 }
2637 
2638 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2639   LP64_ONLY(assert(arg_0 != c_rarg3, "smashed arg"));
2640   LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
2641   LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
2642   pass_arg3(this, arg_3);
2643   LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
2644   LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
2645   pass_arg2(this, arg_2);
2646   LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
2647   pass_arg1(this, arg_1);
2648   pass_arg0(this, arg_0);
2649   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2650 }
2651 
2652 void MacroAssembler::get_vm_result(Register oop_result, Register java_thread) {
2653   movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
2654   movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD);
2655   verify_oop(oop_result, "broken oop in call_VM_base");
2656 }
2657 
2658 void MacroAssembler::get_vm_result_2(Register metadata_result, Register java_thread) {
2659   movptr(metadata_result, Address(java_thread, JavaThread::vm_result_2_offset()));
2660   movptr(Address(java_thread, JavaThread::vm_result_2_offset()), NULL_WORD);
2661 }
2662 
2663 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
2664 }
2665 
2666 void MacroAssembler::check_and_handle_popframe(Register java_thread) {
2667 }
2668 
2669 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) {
2670   if (reachable(src1)) {
2671     cmpl(as_Address(src1), imm);
2672   } else {
2673     lea(rscratch1, src1);
2674     cmpl(Address(rscratch1, 0), imm);
2675   }
2676 }
2677 
2678 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) {
2679   assert(!src2.is_lval(), "use cmpptr");
2680   if (reachable(src2)) {
2681     cmpl(src1, as_Address(src2));
2682   } else {
2683     lea(rscratch1, src2);
2684     cmpl(src1, Address(rscratch1, 0));
2685   }
2686 }
2687 
2688 void MacroAssembler::cmp32(Register src1, int32_t imm) {
2689   Assembler::cmpl(src1, imm);
2690 }
2691 
2692 void MacroAssembler::cmp32(Register src1, Address src2) {
2693   Assembler::cmpl(src1, src2);
2694 }
2695 
2696 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
2697   ucomisd(opr1, opr2);
2698 
2699   Label L;
2700   if (unordered_is_less) {
2701     movl(dst, -1);
2702     jcc(Assembler::parity, L);
2703     jcc(Assembler::below , L);
2704     movl(dst, 0);
2705     jcc(Assembler::equal , L);
2706     increment(dst);
2707   } else { // unordered is greater
2708     movl(dst, 1);
2709     jcc(Assembler::parity, L);
2710     jcc(Assembler::above , L);
2711     movl(dst, 0);
2712     jcc(Assembler::equal , L);
2713     decrementl(dst);
2714   }
2715   bind(L);
2716 }
2717 
2718 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
2719   ucomiss(opr1, opr2);
2720 
2721   Label L;
2722   if (unordered_is_less) {
2723     movl(dst, -1);
2724     jcc(Assembler::parity, L);
2725     jcc(Assembler::below , L);
2726     movl(dst, 0);
2727     jcc(Assembler::equal , L);
2728     increment(dst);
2729   } else { // unordered is greater
2730     movl(dst, 1);
2731     jcc(Assembler::parity, L);
2732     jcc(Assembler::above , L);
2733     movl(dst, 0);
2734     jcc(Assembler::equal , L);
2735     decrementl(dst);
2736   }
2737   bind(L);
2738 }
2739 
2740 
2741 void MacroAssembler::cmp8(AddressLiteral src1, int imm) {
2742   if (reachable(src1)) {
2743     cmpb(as_Address(src1), imm);
2744   } else {
2745     lea(rscratch1, src1);
2746     cmpb(Address(rscratch1, 0), imm);
2747   }
2748 }
2749 
2750 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) {
2751 #ifdef _LP64
2752   if (src2.is_lval()) {
2753     movptr(rscratch1, src2);
2754     Assembler::cmpq(src1, rscratch1);
2755   } else if (reachable(src2)) {
2756     cmpq(src1, as_Address(src2));
2757   } else {
2758     lea(rscratch1, src2);
2759     Assembler::cmpq(src1, Address(rscratch1, 0));
2760   }
2761 #else
2762   if (src2.is_lval()) {
2763     cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
2764   } else {
2765     cmpl(src1, as_Address(src2));
2766   }
2767 #endif // _LP64
2768 }
2769 
2770 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) {
2771   assert(src2.is_lval(), "not a mem-mem compare");
2772 #ifdef _LP64
2773   // moves src2's literal address
2774   movptr(rscratch1, src2);
2775   Assembler::cmpq(src1, rscratch1);
2776 #else
2777   cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
2778 #endif // _LP64
2779 }
2780 
2781 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr) {
2782   if (reachable(adr)) {
2783     if (os::is_MP())
2784       lock();
2785     cmpxchgptr(reg, as_Address(adr));
2786   } else {
2787     lea(rscratch1, adr);
2788     if (os::is_MP())
2789       lock();
2790     cmpxchgptr(reg, Address(rscratch1, 0));
2791   }
2792 }
2793 
2794 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
2795   LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr));
2796 }
2797 
2798 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) {
2799   if (reachable(src)) {
2800     Assembler::comisd(dst, as_Address(src));
2801   } else {
2802     lea(rscratch1, src);
2803     Assembler::comisd(dst, Address(rscratch1, 0));
2804   }
2805 }
2806 
2807 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) {
2808   if (reachable(src)) {
2809     Assembler::comiss(dst, as_Address(src));
2810   } else {
2811     lea(rscratch1, src);
2812     Assembler::comiss(dst, Address(rscratch1, 0));
2813   }
2814 }
2815 
2816 
2817 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) {
2818   Condition negated_cond = negate_condition(cond);
2819   Label L;
2820   jcc(negated_cond, L);
2821   pushf(); // Preserve flags
2822   atomic_incl(counter_addr);
2823   popf();
2824   bind(L);
2825 }
2826 
2827 int MacroAssembler::corrected_idivl(Register reg) {
2828   // Full implementation of Java idiv and irem; checks for
2829   // special case as described in JVM spec., p.243 & p.271.
2830   // The function returns the (pc) offset of the idivl
2831   // instruction - may be needed for implicit exceptions.
2832   //
2833   //         normal case                           special case
2834   //
2835   // input : rax,: dividend                         min_int
2836   //         reg: divisor   (may not be rax,/rdx)   -1
2837   //
2838   // output: rax,: quotient  (= rax, idiv reg)       min_int
2839   //         rdx: remainder (= rax, irem reg)       0
2840   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
2841   const int min_int = 0x80000000;
2842   Label normal_case, special_case;
2843 
2844   // check for special case
2845   cmpl(rax, min_int);
2846   jcc(Assembler::notEqual, normal_case);
2847   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
2848   cmpl(reg, -1);
2849   jcc(Assembler::equal, special_case);
2850 
2851   // handle normal case
2852   bind(normal_case);
2853   cdql();
2854   int idivl_offset = offset();
2855   idivl(reg);
2856 
2857   // normal and special case exit
2858   bind(special_case);
2859 
2860   return idivl_offset;
2861 }
2862 
2863 
2864 
2865 void MacroAssembler::decrementl(Register reg, int value) {
2866   if (value == min_jint) {subl(reg, value) ; return; }
2867   if (value <  0) { incrementl(reg, -value); return; }
2868   if (value == 0) {                        ; return; }
2869   if (value == 1 && UseIncDec) { decl(reg) ; return; }
2870   /* else */      { subl(reg, value)       ; return; }
2871 }
2872 
2873 void MacroAssembler::decrementl(Address dst, int value) {
2874   if (value == min_jint) {subl(dst, value) ; return; }
2875   if (value <  0) { incrementl(dst, -value); return; }
2876   if (value == 0) {                        ; return; }
2877   if (value == 1 && UseIncDec) { decl(dst) ; return; }
2878   /* else */      { subl(dst, value)       ; return; }
2879 }
2880 
2881 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
2882   assert (shift_value > 0, "illegal shift value");
2883   Label _is_positive;
2884   testl (reg, reg);
2885   jcc (Assembler::positive, _is_positive);
2886   int offset = (1 << shift_value) - 1 ;
2887 
2888   if (offset == 1) {
2889     incrementl(reg);
2890   } else {
2891     addl(reg, offset);
2892   }
2893 
2894   bind (_is_positive);
2895   sarl(reg, shift_value);
2896 }
2897 
2898 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src) {
2899   if (reachable(src)) {
2900     Assembler::divsd(dst, as_Address(src));
2901   } else {
2902     lea(rscratch1, src);
2903     Assembler::divsd(dst, Address(rscratch1, 0));
2904   }
2905 }
2906 
2907 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src) {
2908   if (reachable(src)) {
2909     Assembler::divss(dst, as_Address(src));
2910   } else {
2911     lea(rscratch1, src);
2912     Assembler::divss(dst, Address(rscratch1, 0));
2913   }
2914 }
2915 
2916 // !defined(COMPILER2) is because of stupid core builds
2917 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) || INCLUDE_JVMCI
2918 void MacroAssembler::empty_FPU_stack() {
2919   if (VM_Version::supports_mmx()) {
2920     emms();
2921   } else {
2922     for (int i = 8; i-- > 0; ) ffree(i);
2923   }
2924 }
2925 #endif // !LP64 || C1 || !C2 || INCLUDE_JVMCI
2926 
2927 
2928 // Defines obj, preserves var_size_in_bytes
2929 void MacroAssembler::eden_allocate(Register obj,
2930                                    Register var_size_in_bytes,
2931                                    int con_size_in_bytes,
2932                                    Register t1,
2933                                    Label& slow_case) {
2934   assert(obj == rax, "obj must be in rax, for cmpxchg");
2935   assert_different_registers(obj, var_size_in_bytes, t1);
2936   if (!Universe::heap()->supports_inline_contig_alloc()) {
2937     jmp(slow_case);
2938   } else {
2939     Register end = t1;
2940     Label retry;
2941     bind(retry);
2942     ExternalAddress heap_top((address) Universe::heap()->top_addr());
2943     movptr(obj, heap_top);
2944     if (var_size_in_bytes == noreg) {
2945       lea(end, Address(obj, con_size_in_bytes));
2946     } else {
2947       lea(end, Address(obj, var_size_in_bytes, Address::times_1));
2948     }
2949     // if end < obj then we wrapped around => object too long => slow case
2950     cmpptr(end, obj);
2951     jcc(Assembler::below, slow_case);
2952     cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr()));
2953     jcc(Assembler::above, slow_case);
2954     // Compare obj with the top addr, and if still equal, store the new top addr in
2955     // end at the address of the top addr pointer. Sets ZF if was equal, and clears
2956     // it otherwise. Use lock prefix for atomicity on MPs.
2957     locked_cmpxchgptr(end, heap_top);
2958     jcc(Assembler::notEqual, retry);
2959   }
2960 }
2961 
2962 void MacroAssembler::enter() {
2963   push(rbp);
2964   mov(rbp, rsp);
2965 }
2966 
2967 // A 5 byte nop that is safe for patching (see patch_verified_entry)
2968 void MacroAssembler::fat_nop() {
2969   if (UseAddressNop) {
2970     addr_nop_5();
2971   } else {
2972     emit_int8(0x26); // es:
2973     emit_int8(0x2e); // cs:
2974     emit_int8(0x64); // fs:
2975     emit_int8(0x65); // gs:
2976     emit_int8((unsigned char)0x90);
2977   }
2978 }
2979 
2980 void MacroAssembler::fcmp(Register tmp) {
2981   fcmp(tmp, 1, true, true);
2982 }
2983 
2984 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) {
2985   assert(!pop_right || pop_left, "usage error");
2986   if (VM_Version::supports_cmov()) {
2987     assert(tmp == noreg, "unneeded temp");
2988     if (pop_left) {
2989       fucomip(index);
2990     } else {
2991       fucomi(index);
2992     }
2993     if (pop_right) {
2994       fpop();
2995     }
2996   } else {
2997     assert(tmp != noreg, "need temp");
2998     if (pop_left) {
2999       if (pop_right) {
3000         fcompp();
3001       } else {
3002         fcomp(index);
3003       }
3004     } else {
3005       fcom(index);
3006     }
3007     // convert FPU condition into eflags condition via rax,
3008     save_rax(tmp);
3009     fwait(); fnstsw_ax();
3010     sahf();
3011     restore_rax(tmp);
3012   }
3013   // condition codes set as follows:
3014   //
3015   // CF (corresponds to C0) if x < y
3016   // PF (corresponds to C2) if unordered
3017   // ZF (corresponds to C3) if x = y
3018 }
3019 
3020 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) {
3021   fcmp2int(dst, unordered_is_less, 1, true, true);
3022 }
3023 
3024 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) {
3025   fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right);
3026   Label L;
3027   if (unordered_is_less) {
3028     movl(dst, -1);
3029     jcc(Assembler::parity, L);
3030     jcc(Assembler::below , L);
3031     movl(dst, 0);
3032     jcc(Assembler::equal , L);
3033     increment(dst);
3034   } else { // unordered is greater
3035     movl(dst, 1);
3036     jcc(Assembler::parity, L);
3037     jcc(Assembler::above , L);
3038     movl(dst, 0);
3039     jcc(Assembler::equal , L);
3040     decrementl(dst);
3041   }
3042   bind(L);
3043 }
3044 
3045 void MacroAssembler::fld_d(AddressLiteral src) {
3046   fld_d(as_Address(src));
3047 }
3048 
3049 void MacroAssembler::fld_s(AddressLiteral src) {
3050   fld_s(as_Address(src));
3051 }
3052 
3053 void MacroAssembler::fld_x(AddressLiteral src) {
3054   Assembler::fld_x(as_Address(src));
3055 }
3056 
3057 void MacroAssembler::fldcw(AddressLiteral src) {
3058   Assembler::fldcw(as_Address(src));
3059 }
3060 
3061 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src) {
3062   if (reachable(src)) {
3063     Assembler::mulpd(dst, as_Address(src));
3064   } else {
3065     lea(rscratch1, src);
3066     Assembler::mulpd(dst, Address(rscratch1, 0));
3067   }
3068 }
3069 
3070 void MacroAssembler::increase_precision() {
3071   subptr(rsp, BytesPerWord);
3072   fnstcw(Address(rsp, 0));
3073   movl(rax, Address(rsp, 0));
3074   orl(rax, 0x300);
3075   push(rax);
3076   fldcw(Address(rsp, 0));
3077   pop(rax);
3078 }
3079 
3080 void MacroAssembler::restore_precision() {
3081   fldcw(Address(rsp, 0));
3082   addptr(rsp, BytesPerWord);
3083 }
3084 
3085 void MacroAssembler::fpop() {
3086   ffree();
3087   fincstp();
3088 }
3089 
3090 void MacroAssembler::load_float(Address src) {
3091   if (UseSSE >= 1) {
3092     movflt(xmm0, src);
3093   } else {
3094     LP64_ONLY(ShouldNotReachHere());
3095     NOT_LP64(fld_s(src));
3096   }
3097 }
3098 
3099 void MacroAssembler::store_float(Address dst) {
3100   if (UseSSE >= 1) {
3101     movflt(dst, xmm0);
3102   } else {
3103     LP64_ONLY(ShouldNotReachHere());
3104     NOT_LP64(fstp_s(dst));
3105   }
3106 }
3107 
3108 void MacroAssembler::load_double(Address src) {
3109   if (UseSSE >= 2) {
3110     movdbl(xmm0, src);
3111   } else {
3112     LP64_ONLY(ShouldNotReachHere());
3113     NOT_LP64(fld_d(src));
3114   }
3115 }
3116 
3117 void MacroAssembler::store_double(Address dst) {
3118   if (UseSSE >= 2) {
3119     movdbl(dst, xmm0);
3120   } else {
3121     LP64_ONLY(ShouldNotReachHere());
3122     NOT_LP64(fstp_d(dst));
3123   }
3124 }
3125 
3126 void MacroAssembler::fremr(Register tmp) {
3127   save_rax(tmp);
3128   { Label L;
3129     bind(L);
3130     fprem();
3131     fwait(); fnstsw_ax();
3132 #ifdef _LP64
3133     testl(rax, 0x400);
3134     jcc(Assembler::notEqual, L);
3135 #else
3136     sahf();
3137     jcc(Assembler::parity, L);
3138 #endif // _LP64
3139   }
3140   restore_rax(tmp);
3141   // Result is in ST0.
3142   // Note: fxch & fpop to get rid of ST1
3143   // (otherwise FPU stack could overflow eventually)
3144   fxch(1);
3145   fpop();
3146 }
3147 
3148 
3149 void MacroAssembler::incrementl(AddressLiteral dst) {
3150   if (reachable(dst)) {
3151     incrementl(as_Address(dst));
3152   } else {
3153     lea(rscratch1, dst);
3154     incrementl(Address(rscratch1, 0));
3155   }
3156 }
3157 
3158 void MacroAssembler::incrementl(ArrayAddress dst) {
3159   incrementl(as_Address(dst));
3160 }
3161 
3162 void MacroAssembler::incrementl(Register reg, int value) {
3163   if (value == min_jint) {addl(reg, value) ; return; }
3164   if (value <  0) { decrementl(reg, -value); return; }
3165   if (value == 0) {                        ; return; }
3166   if (value == 1 && UseIncDec) { incl(reg) ; return; }
3167   /* else */      { addl(reg, value)       ; return; }
3168 }
3169 
3170 void MacroAssembler::incrementl(Address dst, int value) {
3171   if (value == min_jint) {addl(dst, value) ; return; }
3172   if (value <  0) { decrementl(dst, -value); return; }
3173   if (value == 0) {                        ; return; }
3174   if (value == 1 && UseIncDec) { incl(dst) ; return; }
3175   /* else */      { addl(dst, value)       ; return; }
3176 }
3177 
3178 void MacroAssembler::jump(AddressLiteral dst) {
3179   if (reachable(dst)) {
3180     jmp_literal(dst.target(), dst.rspec());
3181   } else {
3182     lea(rscratch1, dst);
3183     jmp(rscratch1);
3184   }
3185 }
3186 
3187 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) {
3188   if (reachable(dst)) {
3189     InstructionMark im(this);
3190     relocate(dst.reloc());
3191     const int short_size = 2;
3192     const int long_size = 6;
3193     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
3194     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
3195       // 0111 tttn #8-bit disp
3196       emit_int8(0x70 | cc);
3197       emit_int8((offs - short_size) & 0xFF);
3198     } else {
3199       // 0000 1111 1000 tttn #32-bit disp
3200       emit_int8(0x0F);
3201       emit_int8((unsigned char)(0x80 | cc));
3202       emit_int32(offs - long_size);
3203     }
3204   } else {
3205 #ifdef ASSERT
3206     warning("reversing conditional branch");
3207 #endif /* ASSERT */
3208     Label skip;
3209     jccb(reverse[cc], skip);
3210     lea(rscratch1, dst);
3211     Assembler::jmp(rscratch1);
3212     bind(skip);
3213   }
3214 }
3215 
3216 void MacroAssembler::ldmxcsr(AddressLiteral src) {
3217   if (reachable(src)) {
3218     Assembler::ldmxcsr(as_Address(src));
3219   } else {
3220     lea(rscratch1, src);
3221     Assembler::ldmxcsr(Address(rscratch1, 0));
3222   }
3223 }
3224 
3225 int MacroAssembler::load_signed_byte(Register dst, Address src) {
3226   int off;
3227   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
3228     off = offset();
3229     movsbl(dst, src); // movsxb
3230   } else {
3231     off = load_unsigned_byte(dst, src);
3232     shll(dst, 24);
3233     sarl(dst, 24);
3234   }
3235   return off;
3236 }
3237 
3238 // Note: load_signed_short used to be called load_signed_word.
3239 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
3240 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
3241 // The term "word" in HotSpot means a 32- or 64-bit machine word.
3242 int MacroAssembler::load_signed_short(Register dst, Address src) {
3243   int off;
3244   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
3245     // This is dubious to me since it seems safe to do a signed 16 => 64 bit
3246     // version but this is what 64bit has always done. This seems to imply
3247     // that users are only using 32bits worth.
3248     off = offset();
3249     movswl(dst, src); // movsxw
3250   } else {
3251     off = load_unsigned_short(dst, src);
3252     shll(dst, 16);
3253     sarl(dst, 16);
3254   }
3255   return off;
3256 }
3257 
3258 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
3259   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
3260   // and "3.9 Partial Register Penalties", p. 22).
3261   int off;
3262   if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) {
3263     off = offset();
3264     movzbl(dst, src); // movzxb
3265   } else {
3266     xorl(dst, dst);
3267     off = offset();
3268     movb(dst, src);
3269   }
3270   return off;
3271 }
3272 
3273 // Note: load_unsigned_short used to be called load_unsigned_word.
3274 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
3275   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
3276   // and "3.9 Partial Register Penalties", p. 22).
3277   int off;
3278   if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) {
3279     off = offset();
3280     movzwl(dst, src); // movzxw
3281   } else {
3282     xorl(dst, dst);
3283     off = offset();
3284     movw(dst, src);
3285   }
3286   return off;
3287 }
3288 
3289 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
3290   switch (size_in_bytes) {
3291 #ifndef _LP64
3292   case  8:
3293     assert(dst2 != noreg, "second dest register required");
3294     movl(dst,  src);
3295     movl(dst2, src.plus_disp(BytesPerInt));
3296     break;
3297 #else
3298   case  8:  movq(dst, src); break;
3299 #endif
3300   case  4:  movl(dst, src); break;
3301   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
3302   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
3303   default:  ShouldNotReachHere();
3304   }
3305 }
3306 
3307 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
3308   switch (size_in_bytes) {
3309 #ifndef _LP64
3310   case  8:
3311     assert(src2 != noreg, "second source register required");
3312     movl(dst,                        src);
3313     movl(dst.plus_disp(BytesPerInt), src2);
3314     break;
3315 #else
3316   case  8:  movq(dst, src); break;
3317 #endif
3318   case  4:  movl(dst, src); break;
3319   case  2:  movw(dst, src); break;
3320   case  1:  movb(dst, src); break;
3321   default:  ShouldNotReachHere();
3322   }
3323 }
3324 
3325 void MacroAssembler::mov32(AddressLiteral dst, Register src) {
3326   if (reachable(dst)) {
3327     movl(as_Address(dst), src);
3328   } else {
3329     lea(rscratch1, dst);
3330     movl(Address(rscratch1, 0), src);
3331   }
3332 }
3333 
3334 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
3335   if (reachable(src)) {
3336     movl(dst, as_Address(src));
3337   } else {
3338     lea(rscratch1, src);
3339     movl(dst, Address(rscratch1, 0));
3340   }
3341 }
3342 
3343 // C++ bool manipulation
3344 
3345 void MacroAssembler::movbool(Register dst, Address src) {
3346   if(sizeof(bool) == 1)
3347     movb(dst, src);
3348   else if(sizeof(bool) == 2)
3349     movw(dst, src);
3350   else if(sizeof(bool) == 4)
3351     movl(dst, src);
3352   else
3353     // unsupported
3354     ShouldNotReachHere();
3355 }
3356 
3357 void MacroAssembler::movbool(Address dst, bool boolconst) {
3358   if(sizeof(bool) == 1)
3359     movb(dst, (int) boolconst);
3360   else if(sizeof(bool) == 2)
3361     movw(dst, (int) boolconst);
3362   else if(sizeof(bool) == 4)
3363     movl(dst, (int) boolconst);
3364   else
3365     // unsupported
3366     ShouldNotReachHere();
3367 }
3368 
3369 void MacroAssembler::movbool(Address dst, Register src) {
3370   if(sizeof(bool) == 1)
3371     movb(dst, src);
3372   else if(sizeof(bool) == 2)
3373     movw(dst, src);
3374   else if(sizeof(bool) == 4)
3375     movl(dst, src);
3376   else
3377     // unsupported
3378     ShouldNotReachHere();
3379 }
3380 
3381 void MacroAssembler::movbyte(ArrayAddress dst, int src) {
3382   movb(as_Address(dst), src);
3383 }
3384 
3385 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src) {
3386   if (reachable(src)) {
3387     movdl(dst, as_Address(src));
3388   } else {
3389     lea(rscratch1, src);
3390     movdl(dst, Address(rscratch1, 0));
3391   }
3392 }
3393 
3394 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src) {
3395   if (reachable(src)) {
3396     movq(dst, as_Address(src));
3397   } else {
3398     lea(rscratch1, src);
3399     movq(dst, Address(rscratch1, 0));
3400   }
3401 }
3402 
3403 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) {
3404   if (reachable(src)) {
3405     if (UseXmmLoadAndClearUpper) {
3406       movsd (dst, as_Address(src));
3407     } else {
3408       movlpd(dst, as_Address(src));
3409     }
3410   } else {
3411     lea(rscratch1, src);
3412     if (UseXmmLoadAndClearUpper) {
3413       movsd (dst, Address(rscratch1, 0));
3414     } else {
3415       movlpd(dst, Address(rscratch1, 0));
3416     }
3417   }
3418 }
3419 
3420 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) {
3421   if (reachable(src)) {
3422     movss(dst, as_Address(src));
3423   } else {
3424     lea(rscratch1, src);
3425     movss(dst, Address(rscratch1, 0));
3426   }
3427 }
3428 
3429 void MacroAssembler::movptr(Register dst, Register src) {
3430   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
3431 }
3432 
3433 void MacroAssembler::movptr(Register dst, Address src) {
3434   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
3435 }
3436 
3437 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
3438 void MacroAssembler::movptr(Register dst, intptr_t src) {
3439   LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src));
3440 }
3441 
3442 void MacroAssembler::movptr(Address dst, Register src) {
3443   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
3444 }
3445 
3446 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
3447   if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (src->encoding() > 15)) {
3448     Assembler::vextractf32x4(dst, src, 0);
3449   } else {
3450     Assembler::movdqu(dst, src);
3451   }
3452 }
3453 
3454 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
3455   if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (dst->encoding() > 15)) {
3456     Assembler::vinsertf32x4(dst, dst, src, 0);
3457   } else {
3458     Assembler::movdqu(dst, src);
3459   }
3460 }
3461 
3462 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
3463   if (UseAVX > 2 && !VM_Version::supports_avx512vl()) {
3464     Assembler::evmovdqul(dst, src, Assembler::AVX_512bit);
3465   } else {
3466     Assembler::movdqu(dst, src);
3467   }
3468 }
3469 
3470 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src) {
3471   if (reachable(src)) {
3472     movdqu(dst, as_Address(src));
3473   } else {
3474     lea(rscratch1, src);
3475     movdqu(dst, Address(rscratch1, 0));
3476   }
3477 }
3478 
3479 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
3480   if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (src->encoding() > 15)) {
3481     vextractf64x4_low(dst, src);
3482   } else {
3483     Assembler::vmovdqu(dst, src);
3484   }
3485 }
3486 
3487 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
3488   if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (dst->encoding() > 15)) {
3489     vinsertf64x4_low(dst, src);
3490   } else {
3491     Assembler::vmovdqu(dst, src);
3492   }
3493 }
3494 
3495 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
3496   if (UseAVX > 2 && !VM_Version::supports_avx512vl()) {
3497     Assembler::evmovdqul(dst, src, Assembler::AVX_512bit);
3498   }
3499   else {
3500     Assembler::vmovdqu(dst, src);
3501   }
3502 }
3503 
3504 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src) {
3505   if (reachable(src)) {
3506     vmovdqu(dst, as_Address(src));
3507   }
3508   else {
3509     lea(rscratch1, src);
3510     vmovdqu(dst, Address(rscratch1, 0));
3511   }
3512 }
3513 
3514 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src) {
3515   if (reachable(src)) {
3516     Assembler::movdqa(dst, as_Address(src));
3517   } else {
3518     lea(rscratch1, src);
3519     Assembler::movdqa(dst, Address(rscratch1, 0));
3520   }
3521 }
3522 
3523 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src) {
3524   if (reachable(src)) {
3525     Assembler::movsd(dst, as_Address(src));
3526   } else {
3527     lea(rscratch1, src);
3528     Assembler::movsd(dst, Address(rscratch1, 0));
3529   }
3530 }
3531 
3532 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) {
3533   if (reachable(src)) {
3534     Assembler::movss(dst, as_Address(src));
3535   } else {
3536     lea(rscratch1, src);
3537     Assembler::movss(dst, Address(rscratch1, 0));
3538   }
3539 }
3540 
3541 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src) {
3542   if (reachable(src)) {
3543     Assembler::mulsd(dst, as_Address(src));
3544   } else {
3545     lea(rscratch1, src);
3546     Assembler::mulsd(dst, Address(rscratch1, 0));
3547   }
3548 }
3549 
3550 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src) {
3551   if (reachable(src)) {
3552     Assembler::mulss(dst, as_Address(src));
3553   } else {
3554     lea(rscratch1, src);
3555     Assembler::mulss(dst, Address(rscratch1, 0));
3556   }
3557 }
3558 
3559 void MacroAssembler::null_check(Register reg, int offset) {
3560   if (needs_explicit_null_check(offset)) {
3561     // provoke OS NULL exception if reg = NULL by
3562     // accessing M[reg] w/o changing any (non-CC) registers
3563     // NOTE: cmpl is plenty here to provoke a segv
3564     cmpptr(rax, Address(reg, 0));
3565     // Note: should probably use testl(rax, Address(reg, 0));
3566     //       may be shorter code (however, this version of
3567     //       testl needs to be implemented first)
3568   } else {
3569     // nothing to do, (later) access of M[reg + offset]
3570     // will provoke OS NULL exception if reg = NULL
3571   }
3572 }
3573 
3574 void MacroAssembler::os_breakpoint() {
3575   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
3576   // (e.g., MSVC can't call ps() otherwise)
3577   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
3578 }
3579 
3580 #ifdef _LP64
3581 #define XSTATE_BV 0x200
3582 #endif
3583 
3584 void MacroAssembler::pop_CPU_state() {
3585   pop_FPU_state();
3586   pop_IU_state();
3587 }
3588 
3589 void MacroAssembler::pop_FPU_state() {
3590 #ifndef _LP64
3591   frstor(Address(rsp, 0));
3592 #else
3593   fxrstor(Address(rsp, 0));
3594 #endif
3595   addptr(rsp, FPUStateSizeInWords * wordSize);
3596 }
3597 
3598 void MacroAssembler::pop_IU_state() {
3599   popa();
3600   LP64_ONLY(addq(rsp, 8));
3601   popf();
3602 }
3603 
3604 // Save Integer and Float state
3605 // Warning: Stack must be 16 byte aligned (64bit)
3606 void MacroAssembler::push_CPU_state() {
3607   push_IU_state();
3608   push_FPU_state();
3609 }
3610 
3611 void MacroAssembler::push_FPU_state() {
3612   subptr(rsp, FPUStateSizeInWords * wordSize);
3613 #ifndef _LP64
3614   fnsave(Address(rsp, 0));
3615   fwait();
3616 #else
3617   fxsave(Address(rsp, 0));
3618 #endif // LP64
3619 }
3620 
3621 void MacroAssembler::push_IU_state() {
3622   // Push flags first because pusha kills them
3623   pushf();
3624   // Make sure rsp stays 16-byte aligned
3625   LP64_ONLY(subq(rsp, 8));
3626   pusha();
3627 }
3628 
3629 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) {
3630   // determine java_thread register
3631   if (!java_thread->is_valid()) {
3632     java_thread = rdi;
3633     get_thread(java_thread);
3634   }
3635   // we must set sp to zero to clear frame
3636   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
3637   if (clear_fp) {
3638     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
3639   }
3640 
3641   if (clear_pc)
3642     movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
3643 
3644 }
3645 
3646 void MacroAssembler::restore_rax(Register tmp) {
3647   if (tmp == noreg) pop(rax);
3648   else if (tmp != rax) mov(rax, tmp);
3649 }
3650 
3651 void MacroAssembler::round_to(Register reg, int modulus) {
3652   addptr(reg, modulus - 1);
3653   andptr(reg, -modulus);
3654 }
3655 
3656 void MacroAssembler::save_rax(Register tmp) {
3657   if (tmp == noreg) push(rax);
3658   else if (tmp != rax) mov(tmp, rax);
3659 }
3660 
3661 // Write serialization page so VM thread can do a pseudo remote membar.
3662 // We use the current thread pointer to calculate a thread specific
3663 // offset to write to within the page. This minimizes bus traffic
3664 // due to cache line collision.
3665 void MacroAssembler::serialize_memory(Register thread, Register tmp) {
3666   movl(tmp, thread);
3667   shrl(tmp, os::get_serialize_page_shift_count());
3668   andl(tmp, (os::vm_page_size() - sizeof(int)));
3669 
3670   Address index(noreg, tmp, Address::times_1);
3671   ExternalAddress page(os::get_memory_serialize_page());
3672 
3673   // Size of store must match masking code above
3674   movl(as_Address(ArrayAddress(page, index)), tmp);
3675 }
3676 
3677 // Calls to C land
3678 //
3679 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
3680 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
3681 // has to be reset to 0. This is required to allow proper stack traversal.
3682 void MacroAssembler::set_last_Java_frame(Register java_thread,
3683                                          Register last_java_sp,
3684                                          Register last_java_fp,
3685                                          address  last_java_pc) {
3686   // determine java_thread register
3687   if (!java_thread->is_valid()) {
3688     java_thread = rdi;
3689     get_thread(java_thread);
3690   }
3691   // determine last_java_sp register
3692   if (!last_java_sp->is_valid()) {
3693     last_java_sp = rsp;
3694   }
3695 
3696   // last_java_fp is optional
3697 
3698   if (last_java_fp->is_valid()) {
3699     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
3700   }
3701 
3702   // last_java_pc is optional
3703 
3704   if (last_java_pc != NULL) {
3705     lea(Address(java_thread,
3706                  JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()),
3707         InternalAddress(last_java_pc));
3708 
3709   }
3710   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
3711 }
3712 
3713 void MacroAssembler::shlptr(Register dst, int imm8) {
3714   LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8));
3715 }
3716 
3717 void MacroAssembler::shrptr(Register dst, int imm8) {
3718   LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8));
3719 }
3720 
3721 void MacroAssembler::sign_extend_byte(Register reg) {
3722   if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) {
3723     movsbl(reg, reg); // movsxb
3724   } else {
3725     shll(reg, 24);
3726     sarl(reg, 24);
3727   }
3728 }
3729 
3730 void MacroAssembler::sign_extend_short(Register reg) {
3731   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
3732     movswl(reg, reg); // movsxw
3733   } else {
3734     shll(reg, 16);
3735     sarl(reg, 16);
3736   }
3737 }
3738 
3739 void MacroAssembler::testl(Register dst, AddressLiteral src) {
3740   assert(reachable(src), "Address should be reachable");
3741   testl(dst, as_Address(src));
3742 }
3743 
3744 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
3745   int dst_enc = dst->encoding();
3746   int src_enc = src->encoding();
3747   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
3748     Assembler::pcmpeqb(dst, src);
3749   } else if ((dst_enc < 16) && (src_enc < 16)) {
3750     Assembler::pcmpeqb(dst, src);
3751   } else if (src_enc < 16) {
3752     subptr(rsp, 64);
3753     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3754     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3755     Assembler::pcmpeqb(xmm0, src);
3756     movdqu(dst, xmm0);
3757     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3758     addptr(rsp, 64);
3759   } else if (dst_enc < 16) {
3760     subptr(rsp, 64);
3761     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3762     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3763     Assembler::pcmpeqb(dst, xmm0);
3764     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3765     addptr(rsp, 64);
3766   } else {
3767     subptr(rsp, 64);
3768     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3769     subptr(rsp, 64);
3770     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
3771     movdqu(xmm0, src);
3772     movdqu(xmm1, dst);
3773     Assembler::pcmpeqb(xmm1, xmm0);
3774     movdqu(dst, xmm1);
3775     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
3776     addptr(rsp, 64);
3777     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3778     addptr(rsp, 64);
3779   }
3780 }
3781 
3782 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
3783   int dst_enc = dst->encoding();
3784   int src_enc = src->encoding();
3785   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
3786     Assembler::pcmpeqw(dst, src);
3787   } else if ((dst_enc < 16) && (src_enc < 16)) {
3788     Assembler::pcmpeqw(dst, src);
3789   } else if (src_enc < 16) {
3790     subptr(rsp, 64);
3791     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3792     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3793     Assembler::pcmpeqw(xmm0, src);
3794     movdqu(dst, xmm0);
3795     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3796     addptr(rsp, 64);
3797   } else if (dst_enc < 16) {
3798     subptr(rsp, 64);
3799     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3800     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3801     Assembler::pcmpeqw(dst, xmm0);
3802     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3803     addptr(rsp, 64);
3804   } else {
3805     subptr(rsp, 64);
3806     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3807     subptr(rsp, 64);
3808     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
3809     movdqu(xmm0, src);
3810     movdqu(xmm1, dst);
3811     Assembler::pcmpeqw(xmm1, xmm0);
3812     movdqu(dst, xmm1);
3813     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
3814     addptr(rsp, 64);
3815     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3816     addptr(rsp, 64);
3817   }
3818 }
3819 
3820 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
3821   int dst_enc = dst->encoding();
3822   if (dst_enc < 16) {
3823     Assembler::pcmpestri(dst, src, imm8);
3824   } else {
3825     subptr(rsp, 64);
3826     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3827     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3828     Assembler::pcmpestri(xmm0, src, imm8);
3829     movdqu(dst, xmm0);
3830     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3831     addptr(rsp, 64);
3832   }
3833 }
3834 
3835 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
3836   int dst_enc = dst->encoding();
3837   int src_enc = src->encoding();
3838   if ((dst_enc < 16) && (src_enc < 16)) {
3839     Assembler::pcmpestri(dst, src, imm8);
3840   } else if (src_enc < 16) {
3841     subptr(rsp, 64);
3842     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3843     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3844     Assembler::pcmpestri(xmm0, src, imm8);
3845     movdqu(dst, xmm0);
3846     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3847     addptr(rsp, 64);
3848   } else if (dst_enc < 16) {
3849     subptr(rsp, 64);
3850     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3851     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3852     Assembler::pcmpestri(dst, xmm0, imm8);
3853     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3854     addptr(rsp, 64);
3855   } else {
3856     subptr(rsp, 64);
3857     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3858     subptr(rsp, 64);
3859     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
3860     movdqu(xmm0, src);
3861     movdqu(xmm1, dst);
3862     Assembler::pcmpestri(xmm1, xmm0, imm8);
3863     movdqu(dst, xmm1);
3864     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
3865     addptr(rsp, 64);
3866     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3867     addptr(rsp, 64);
3868   }
3869 }
3870 
3871 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
3872   int dst_enc = dst->encoding();
3873   int src_enc = src->encoding();
3874   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
3875     Assembler::pmovzxbw(dst, src);
3876   } else if ((dst_enc < 16) && (src_enc < 16)) {
3877     Assembler::pmovzxbw(dst, src);
3878   } else if (src_enc < 16) {
3879     subptr(rsp, 64);
3880     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3881     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3882     Assembler::pmovzxbw(xmm0, src);
3883     movdqu(dst, xmm0);
3884     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3885     addptr(rsp, 64);
3886   } else if (dst_enc < 16) {
3887     subptr(rsp, 64);
3888     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3889     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3890     Assembler::pmovzxbw(dst, xmm0);
3891     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3892     addptr(rsp, 64);
3893   } else {
3894     subptr(rsp, 64);
3895     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3896     subptr(rsp, 64);
3897     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
3898     movdqu(xmm0, src);
3899     movdqu(xmm1, dst);
3900     Assembler::pmovzxbw(xmm1, xmm0);
3901     movdqu(dst, xmm1);
3902     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
3903     addptr(rsp, 64);
3904     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3905     addptr(rsp, 64);
3906   }
3907 }
3908 
3909 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
3910   int dst_enc = dst->encoding();
3911   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
3912     Assembler::pmovzxbw(dst, src);
3913   } else if (dst_enc < 16) {
3914     Assembler::pmovzxbw(dst, src);
3915   } else {
3916     subptr(rsp, 64);
3917     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3918     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3919     Assembler::pmovzxbw(xmm0, src);
3920     movdqu(dst, xmm0);
3921     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3922     addptr(rsp, 64);
3923   }
3924 }
3925 
3926 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
3927   int src_enc = src->encoding();
3928   if (src_enc < 16) {
3929     Assembler::pmovmskb(dst, src);
3930   } else {
3931     subptr(rsp, 64);
3932     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3933     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3934     Assembler::pmovmskb(dst, xmm0);
3935     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3936     addptr(rsp, 64);
3937   }
3938 }
3939 
3940 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
3941   int dst_enc = dst->encoding();
3942   int src_enc = src->encoding();
3943   if ((dst_enc < 16) && (src_enc < 16)) {
3944     Assembler::ptest(dst, src);
3945   } else if (src_enc < 16) {
3946     subptr(rsp, 64);
3947     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3948     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
3949     Assembler::ptest(xmm0, src);
3950     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3951     addptr(rsp, 64);
3952   } else if (dst_enc < 16) {
3953     subptr(rsp, 64);
3954     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3955     evmovdqul(xmm0, src, Assembler::AVX_512bit);
3956     Assembler::ptest(dst, xmm0);
3957     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3958     addptr(rsp, 64);
3959   } else {
3960     subptr(rsp, 64);
3961     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
3962     subptr(rsp, 64);
3963     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
3964     movdqu(xmm0, src);
3965     movdqu(xmm1, dst);
3966     Assembler::ptest(xmm1, xmm0);
3967     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
3968     addptr(rsp, 64);
3969     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
3970     addptr(rsp, 64);
3971   }
3972 }
3973 
3974 void MacroAssembler::sqrtsd(XMMRegister dst, AddressLiteral src) {
3975   if (reachable(src)) {
3976     Assembler::sqrtsd(dst, as_Address(src));
3977   } else {
3978     lea(rscratch1, src);
3979     Assembler::sqrtsd(dst, Address(rscratch1, 0));
3980   }
3981 }
3982 
3983 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src) {
3984   if (reachable(src)) {
3985     Assembler::sqrtss(dst, as_Address(src));
3986   } else {
3987     lea(rscratch1, src);
3988     Assembler::sqrtss(dst, Address(rscratch1, 0));
3989   }
3990 }
3991 
3992 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src) {
3993   if (reachable(src)) {
3994     Assembler::subsd(dst, as_Address(src));
3995   } else {
3996     lea(rscratch1, src);
3997     Assembler::subsd(dst, Address(rscratch1, 0));
3998   }
3999 }
4000 
4001 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src) {
4002   if (reachable(src)) {
4003     Assembler::subss(dst, as_Address(src));
4004   } else {
4005     lea(rscratch1, src);
4006     Assembler::subss(dst, Address(rscratch1, 0));
4007   }
4008 }
4009 
4010 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) {
4011   if (reachable(src)) {
4012     Assembler::ucomisd(dst, as_Address(src));
4013   } else {
4014     lea(rscratch1, src);
4015     Assembler::ucomisd(dst, Address(rscratch1, 0));
4016   }
4017 }
4018 
4019 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) {
4020   if (reachable(src)) {
4021     Assembler::ucomiss(dst, as_Address(src));
4022   } else {
4023     lea(rscratch1, src);
4024     Assembler::ucomiss(dst, Address(rscratch1, 0));
4025   }
4026 }
4027 
4028 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) {
4029   // Used in sign-bit flipping with aligned address.
4030   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
4031   if (reachable(src)) {
4032     Assembler::xorpd(dst, as_Address(src));
4033   } else {
4034     lea(rscratch1, src);
4035     Assembler::xorpd(dst, Address(rscratch1, 0));
4036   }
4037 }
4038 
4039 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
4040   if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
4041     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
4042   }
4043   else {
4044     Assembler::xorpd(dst, src);
4045   }
4046 }
4047 
4048 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
4049   if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
4050     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
4051   } else {
4052     Assembler::xorps(dst, src);
4053   }
4054 }
4055 
4056 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) {
4057   // Used in sign-bit flipping with aligned address.
4058   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
4059   if (reachable(src)) {
4060     Assembler::xorps(dst, as_Address(src));
4061   } else {
4062     lea(rscratch1, src);
4063     Assembler::xorps(dst, Address(rscratch1, 0));
4064   }
4065 }
4066 
4067 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src) {
4068   // Used in sign-bit flipping with aligned address.
4069   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
4070   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
4071   if (reachable(src)) {
4072     Assembler::pshufb(dst, as_Address(src));
4073   } else {
4074     lea(rscratch1, src);
4075     Assembler::pshufb(dst, Address(rscratch1, 0));
4076   }
4077 }
4078 
4079 // AVX 3-operands instructions
4080 
4081 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
4082   if (reachable(src)) {
4083     vaddsd(dst, nds, as_Address(src));
4084   } else {
4085     lea(rscratch1, src);
4086     vaddsd(dst, nds, Address(rscratch1, 0));
4087   }
4088 }
4089 
4090 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
4091   if (reachable(src)) {
4092     vaddss(dst, nds, as_Address(src));
4093   } else {
4094     lea(rscratch1, src);
4095     vaddss(dst, nds, Address(rscratch1, 0));
4096   }
4097 }
4098 
4099 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len) {
4100   int dst_enc = dst->encoding();
4101   int nds_enc = nds->encoding();
4102   int src_enc = src->encoding();
4103   if ((dst_enc < 16) && (nds_enc < 16)) {
4104     vandps(dst, nds, negate_field, vector_len);
4105   } else if ((src_enc < 16) && (dst_enc < 16)) {
4106     movss(src, nds);
4107     vandps(dst, src, negate_field, vector_len);
4108   } else if (src_enc < 16) {
4109     movss(src, nds);
4110     vandps(src, src, negate_field, vector_len);
4111     movss(dst, src);
4112   } else if (dst_enc < 16) {
4113     movdqu(src, xmm0);
4114     movss(xmm0, nds);
4115     vandps(dst, xmm0, negate_field, vector_len);
4116     movdqu(xmm0, src);
4117   } else if (nds_enc < 16) {
4118     movdqu(src, xmm0);
4119     vandps(xmm0, nds, negate_field, vector_len);
4120     movss(dst, xmm0);
4121     movdqu(xmm0, src);
4122   } else {
4123     movdqu(src, xmm0);
4124     movss(xmm0, nds);
4125     vandps(xmm0, xmm0, negate_field, vector_len);
4126     movss(dst, xmm0);
4127     movdqu(xmm0, src);
4128   }
4129 }
4130 
4131 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len) {
4132   int dst_enc = dst->encoding();
4133   int nds_enc = nds->encoding();
4134   int src_enc = src->encoding();
4135   if ((dst_enc < 16) && (nds_enc < 16)) {
4136     vandpd(dst, nds, negate_field, vector_len);
4137   } else if ((src_enc < 16) && (dst_enc < 16)) {
4138     movsd(src, nds);
4139     vandpd(dst, src, negate_field, vector_len);
4140   } else if (src_enc < 16) {
4141     movsd(src, nds);
4142     vandpd(src, src, negate_field, vector_len);
4143     movsd(dst, src);
4144   } else if (dst_enc < 16) {
4145     movdqu(src, xmm0);
4146     movsd(xmm0, nds);
4147     vandpd(dst, xmm0, negate_field, vector_len);
4148     movdqu(xmm0, src);
4149   } else if (nds_enc < 16) {
4150     movdqu(src, xmm0);
4151     vandpd(xmm0, nds, negate_field, vector_len);
4152     movsd(dst, xmm0);
4153     movdqu(xmm0, src);
4154   } else {
4155     movdqu(src, xmm0);
4156     movsd(xmm0, nds);
4157     vandpd(xmm0, xmm0, negate_field, vector_len);
4158     movsd(dst, xmm0);
4159     movdqu(xmm0, src);
4160   }
4161 }
4162 
4163 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4164   int dst_enc = dst->encoding();
4165   int nds_enc = nds->encoding();
4166   int src_enc = src->encoding();
4167   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4168     Assembler::vpaddb(dst, nds, src, vector_len);
4169   } else if ((dst_enc < 16) && (src_enc < 16)) {
4170     Assembler::vpaddb(dst, dst, src, vector_len);
4171   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4172     // use nds as scratch for src
4173     evmovdqul(nds, src, Assembler::AVX_512bit);
4174     Assembler::vpaddb(dst, dst, nds, vector_len);
4175   } else if ((src_enc < 16) && (nds_enc < 16)) {
4176     // use nds as scratch for dst
4177     evmovdqul(nds, dst, Assembler::AVX_512bit);
4178     Assembler::vpaddb(nds, nds, src, vector_len);
4179     evmovdqul(dst, nds, Assembler::AVX_512bit);
4180   } else if (dst_enc < 16) {
4181     // use nds as scatch for xmm0 to hold src
4182     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4183     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4184     Assembler::vpaddb(dst, dst, xmm0, vector_len);
4185     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4186   } else {
4187     // worse case scenario, all regs are in the upper bank
4188     subptr(rsp, 64);
4189     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4190     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4191     evmovdqul(xmm1, src, Assembler::AVX_512bit);
4192     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4193     Assembler::vpaddb(xmm0, xmm0, xmm1, vector_len);
4194     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4195     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4196     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4197     addptr(rsp, 64);
4198   }
4199 }
4200 
4201 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
4202   int dst_enc = dst->encoding();
4203   int nds_enc = nds->encoding();
4204   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4205     Assembler::vpaddb(dst, nds, src, vector_len);
4206   } else if (dst_enc < 16) {
4207     Assembler::vpaddb(dst, dst, src, vector_len);
4208   } else if (nds_enc < 16) {
4209     // implies dst_enc in upper bank with src as scratch
4210     evmovdqul(nds, dst, Assembler::AVX_512bit);
4211     Assembler::vpaddb(nds, nds, src, vector_len);
4212     evmovdqul(dst, nds, Assembler::AVX_512bit);
4213   } else {
4214     // worse case scenario, all regs in upper bank
4215     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4216     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4217     Assembler::vpaddb(xmm0, xmm0, src, vector_len);
4218     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4219   }
4220 }
4221 
4222 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4223   int dst_enc = dst->encoding();
4224   int nds_enc = nds->encoding();
4225   int src_enc = src->encoding();
4226   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4227     Assembler::vpaddw(dst, nds, src, vector_len);
4228   } else if ((dst_enc < 16) && (src_enc < 16)) {
4229     Assembler::vpaddw(dst, dst, src, vector_len);
4230   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4231     // use nds as scratch for src
4232     evmovdqul(nds, src, Assembler::AVX_512bit);
4233     Assembler::vpaddw(dst, dst, nds, vector_len);
4234   } else if ((src_enc < 16) && (nds_enc < 16)) {
4235     // use nds as scratch for dst
4236     evmovdqul(nds, dst, Assembler::AVX_512bit);
4237     Assembler::vpaddw(nds, nds, src, vector_len);
4238     evmovdqul(dst, nds, Assembler::AVX_512bit);
4239   } else if (dst_enc < 16) {
4240     // use nds as scatch for xmm0 to hold src
4241     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4242     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4243     Assembler::vpaddw(dst, dst, xmm0, vector_len);
4244     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4245   } else {
4246     // worse case scenario, all regs are in the upper bank
4247     subptr(rsp, 64);
4248     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4249     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4250     evmovdqul(xmm1, src, Assembler::AVX_512bit);
4251     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4252     Assembler::vpaddw(xmm0, xmm0, xmm1, vector_len);
4253     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4254     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4255     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4256     addptr(rsp, 64);
4257   }
4258 }
4259 
4260 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
4261   int dst_enc = dst->encoding();
4262   int nds_enc = nds->encoding();
4263   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4264     Assembler::vpaddw(dst, nds, src, vector_len);
4265   } else if (dst_enc < 16) {
4266     Assembler::vpaddw(dst, dst, src, vector_len);
4267   } else if (nds_enc < 16) {
4268     // implies dst_enc in upper bank with src as scratch
4269     evmovdqul(nds, dst, Assembler::AVX_512bit);
4270     Assembler::vpaddw(nds, nds, src, vector_len);
4271     evmovdqul(dst, nds, Assembler::AVX_512bit);
4272   } else {
4273     // worse case scenario, all regs in upper bank
4274     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4275     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4276     Assembler::vpaddw(xmm0, xmm0, src, vector_len);
4277     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4278   }
4279 }
4280 
4281 void MacroAssembler::vpbroadcastw(XMMRegister dst, XMMRegister src) {
4282   int dst_enc = dst->encoding();
4283   int src_enc = src->encoding();
4284   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4285     Assembler::vpbroadcastw(dst, src);
4286   } else if ((dst_enc < 16) && (src_enc < 16)) {
4287     Assembler::vpbroadcastw(dst, src);
4288   } else if (src_enc < 16) {
4289     subptr(rsp, 64);
4290     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4291     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4292     Assembler::vpbroadcastw(xmm0, src);
4293     movdqu(dst, xmm0);
4294     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4295     addptr(rsp, 64);
4296   } else if (dst_enc < 16) {
4297     subptr(rsp, 64);
4298     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4299     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4300     Assembler::vpbroadcastw(dst, xmm0);
4301     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4302     addptr(rsp, 64);
4303   } else {
4304     subptr(rsp, 64);
4305     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4306     subptr(rsp, 64);
4307     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4308     movdqu(xmm0, src);
4309     movdqu(xmm1, dst);
4310     Assembler::vpbroadcastw(xmm1, xmm0);
4311     movdqu(dst, xmm1);
4312     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4313     addptr(rsp, 64);
4314     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4315     addptr(rsp, 64);
4316   }
4317 }
4318 
4319 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4320   int dst_enc = dst->encoding();
4321   int nds_enc = nds->encoding();
4322   int src_enc = src->encoding();
4323   assert(dst_enc == nds_enc, "");
4324   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4325     Assembler::vpcmpeqb(dst, nds, src, vector_len);
4326   } else if ((dst_enc < 16) && (src_enc < 16)) {
4327     Assembler::vpcmpeqb(dst, nds, src, vector_len);
4328   } else if (src_enc < 16) {
4329     subptr(rsp, 64);
4330     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4331     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4332     Assembler::vpcmpeqb(xmm0, xmm0, src, vector_len);
4333     movdqu(dst, xmm0);
4334     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4335     addptr(rsp, 64);
4336   } else if (dst_enc < 16) {
4337     subptr(rsp, 64);
4338     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4339     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4340     Assembler::vpcmpeqb(dst, dst, xmm0, vector_len);
4341     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4342     addptr(rsp, 64);
4343   } else {
4344     subptr(rsp, 64);
4345     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4346     subptr(rsp, 64);
4347     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4348     movdqu(xmm0, src);
4349     movdqu(xmm1, dst);
4350     Assembler::vpcmpeqb(xmm1, xmm1, xmm0, vector_len);
4351     movdqu(dst, xmm1);
4352     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4353     addptr(rsp, 64);
4354     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4355     addptr(rsp, 64);
4356   }
4357 }
4358 
4359 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4360   int dst_enc = dst->encoding();
4361   int nds_enc = nds->encoding();
4362   int src_enc = src->encoding();
4363   assert(dst_enc == nds_enc, "");
4364   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4365     Assembler::vpcmpeqw(dst, nds, src, vector_len);
4366   } else if ((dst_enc < 16) && (src_enc < 16)) {
4367     Assembler::vpcmpeqw(dst, nds, src, vector_len);
4368   } else if (src_enc < 16) {
4369     subptr(rsp, 64);
4370     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4371     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4372     Assembler::vpcmpeqw(xmm0, xmm0, src, vector_len);
4373     movdqu(dst, xmm0);
4374     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4375     addptr(rsp, 64);
4376   } else if (dst_enc < 16) {
4377     subptr(rsp, 64);
4378     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4379     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4380     Assembler::vpcmpeqw(dst, dst, xmm0, vector_len);
4381     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4382     addptr(rsp, 64);
4383   } else {
4384     subptr(rsp, 64);
4385     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4386     subptr(rsp, 64);
4387     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4388     movdqu(xmm0, src);
4389     movdqu(xmm1, dst);
4390     Assembler::vpcmpeqw(xmm1, xmm1, xmm0, vector_len);
4391     movdqu(dst, xmm1);
4392     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4393     addptr(rsp, 64);
4394     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4395     addptr(rsp, 64);
4396   }
4397 }
4398 
4399 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
4400   int dst_enc = dst->encoding();
4401   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4402     Assembler::vpmovzxbw(dst, src, vector_len);
4403   } else if (dst_enc < 16) {
4404     Assembler::vpmovzxbw(dst, src, vector_len);
4405   } else {
4406     subptr(rsp, 64);
4407     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4408     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4409     Assembler::vpmovzxbw(xmm0, src, vector_len);
4410     movdqu(dst, xmm0);
4411     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4412     addptr(rsp, 64);
4413   }
4414 }
4415 
4416 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src) {
4417   int src_enc = src->encoding();
4418   if (src_enc < 16) {
4419     Assembler::vpmovmskb(dst, src);
4420   } else {
4421     subptr(rsp, 64);
4422     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4423     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4424     Assembler::vpmovmskb(dst, xmm0);
4425     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4426     addptr(rsp, 64);
4427   }
4428 }
4429 
4430 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4431   int dst_enc = dst->encoding();
4432   int nds_enc = nds->encoding();
4433   int src_enc = src->encoding();
4434   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4435     Assembler::vpmullw(dst, nds, src, vector_len);
4436   } else if ((dst_enc < 16) && (src_enc < 16)) {
4437     Assembler::vpmullw(dst, dst, src, vector_len);
4438   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4439     // use nds as scratch for src
4440     evmovdqul(nds, src, Assembler::AVX_512bit);
4441     Assembler::vpmullw(dst, dst, nds, vector_len);
4442   } else if ((src_enc < 16) && (nds_enc < 16)) {
4443     // use nds as scratch for dst
4444     evmovdqul(nds, dst, Assembler::AVX_512bit);
4445     Assembler::vpmullw(nds, nds, src, vector_len);
4446     evmovdqul(dst, nds, Assembler::AVX_512bit);
4447   } else if (dst_enc < 16) {
4448     // use nds as scatch for xmm0 to hold src
4449     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4450     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4451     Assembler::vpmullw(dst, dst, xmm0, vector_len);
4452     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4453   } else {
4454     // worse case scenario, all regs are in the upper bank
4455     subptr(rsp, 64);
4456     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4457     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4458     evmovdqul(xmm1, src, Assembler::AVX_512bit);
4459     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4460     Assembler::vpmullw(xmm0, xmm0, xmm1, vector_len);
4461     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4462     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4463     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4464     addptr(rsp, 64);
4465   }
4466 }
4467 
4468 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
4469   int dst_enc = dst->encoding();
4470   int nds_enc = nds->encoding();
4471   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4472     Assembler::vpmullw(dst, nds, src, vector_len);
4473   } else if (dst_enc < 16) {
4474     Assembler::vpmullw(dst, dst, src, vector_len);
4475   } else if (nds_enc < 16) {
4476     // implies dst_enc in upper bank with src as scratch
4477     evmovdqul(nds, dst, Assembler::AVX_512bit);
4478     Assembler::vpmullw(nds, nds, src, vector_len);
4479     evmovdqul(dst, nds, Assembler::AVX_512bit);
4480   } else {
4481     // worse case scenario, all regs in upper bank
4482     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4483     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4484     Assembler::vpmullw(xmm0, xmm0, src, vector_len);
4485     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4486   }
4487 }
4488 
4489 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4490   int dst_enc = dst->encoding();
4491   int nds_enc = nds->encoding();
4492   int src_enc = src->encoding();
4493   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4494     Assembler::vpsubb(dst, nds, src, vector_len);
4495   } else if ((dst_enc < 16) && (src_enc < 16)) {
4496     Assembler::vpsubb(dst, dst, src, vector_len);
4497   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4498     // use nds as scratch for src
4499     evmovdqul(nds, src, Assembler::AVX_512bit);
4500     Assembler::vpsubb(dst, dst, nds, vector_len);
4501   } else if ((src_enc < 16) && (nds_enc < 16)) {
4502     // use nds as scratch for dst
4503     evmovdqul(nds, dst, Assembler::AVX_512bit);
4504     Assembler::vpsubb(nds, nds, src, vector_len);
4505     evmovdqul(dst, nds, Assembler::AVX_512bit);
4506   } else if (dst_enc < 16) {
4507     // use nds as scatch for xmm0 to hold src
4508     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4509     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4510     Assembler::vpsubb(dst, dst, xmm0, vector_len);
4511     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4512   } else {
4513     // worse case scenario, all regs are in the upper bank
4514     subptr(rsp, 64);
4515     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4516     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4517     evmovdqul(xmm1, src, Assembler::AVX_512bit);
4518     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4519     Assembler::vpsubb(xmm0, xmm0, xmm1, vector_len);
4520     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4521     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4522     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4523     addptr(rsp, 64);
4524   }
4525 }
4526 
4527 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
4528   int dst_enc = dst->encoding();
4529   int nds_enc = nds->encoding();
4530   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4531     Assembler::vpsubb(dst, nds, src, vector_len);
4532   } else if (dst_enc < 16) {
4533     Assembler::vpsubb(dst, dst, src, vector_len);
4534   } else if (nds_enc < 16) {
4535     // implies dst_enc in upper bank with src as scratch
4536     evmovdqul(nds, dst, Assembler::AVX_512bit);
4537     Assembler::vpsubb(nds, nds, src, vector_len);
4538     evmovdqul(dst, nds, Assembler::AVX_512bit);
4539   } else {
4540     // worse case scenario, all regs in upper bank
4541     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4542     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4543     Assembler::vpsubw(xmm0, xmm0, src, vector_len);
4544     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4545   }
4546 }
4547 
4548 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
4549   int dst_enc = dst->encoding();
4550   int nds_enc = nds->encoding();
4551   int src_enc = src->encoding();
4552   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4553     Assembler::vpsubw(dst, nds, src, vector_len);
4554   } else if ((dst_enc < 16) && (src_enc < 16)) {
4555     Assembler::vpsubw(dst, dst, src, vector_len);
4556   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4557     // use nds as scratch for src
4558     evmovdqul(nds, src, Assembler::AVX_512bit);
4559     Assembler::vpsubw(dst, dst, nds, vector_len);
4560   } else if ((src_enc < 16) && (nds_enc < 16)) {
4561     // use nds as scratch for dst
4562     evmovdqul(nds, dst, Assembler::AVX_512bit);
4563     Assembler::vpsubw(nds, nds, src, vector_len);
4564     evmovdqul(dst, nds, Assembler::AVX_512bit);
4565   } else if (dst_enc < 16) {
4566     // use nds as scatch for xmm0 to hold src
4567     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4568     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4569     Assembler::vpsubw(dst, dst, xmm0, vector_len);
4570     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4571   } else {
4572     // worse case scenario, all regs are in the upper bank
4573     subptr(rsp, 64);
4574     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4575     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4576     evmovdqul(xmm1, src, Assembler::AVX_512bit);
4577     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4578     Assembler::vpsubw(xmm0, xmm0, xmm1, vector_len);
4579     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4580     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4581     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4582     addptr(rsp, 64);
4583   }
4584 }
4585 
4586 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
4587   int dst_enc = dst->encoding();
4588   int nds_enc = nds->encoding();
4589   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4590     Assembler::vpsubw(dst, nds, src, vector_len);
4591   } else if (dst_enc < 16) {
4592     Assembler::vpsubw(dst, dst, src, vector_len);
4593   } else if (nds_enc < 16) {
4594     // implies dst_enc in upper bank with src as scratch
4595     evmovdqul(nds, dst, Assembler::AVX_512bit);
4596     Assembler::vpsubw(nds, nds, src, vector_len);
4597     evmovdqul(dst, nds, Assembler::AVX_512bit);
4598   } else {
4599     // worse case scenario, all regs in upper bank
4600     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4601     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4602     Assembler::vpsubw(xmm0, xmm0, src, vector_len);
4603     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4604   }
4605 }
4606 
4607 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
4608   int dst_enc = dst->encoding();
4609   int nds_enc = nds->encoding();
4610   int shift_enc = shift->encoding();
4611   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4612     Assembler::vpsraw(dst, nds, shift, vector_len);
4613   } else if ((dst_enc < 16) && (shift_enc < 16)) {
4614     Assembler::vpsraw(dst, dst, shift, vector_len);
4615   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4616     // use nds_enc as scratch with shift
4617     evmovdqul(nds, shift, Assembler::AVX_512bit);
4618     Assembler::vpsraw(dst, dst, nds, vector_len);
4619   } else if ((shift_enc < 16) && (nds_enc < 16)) {
4620     // use nds as scratch with dst
4621     evmovdqul(nds, dst, Assembler::AVX_512bit);
4622     Assembler::vpsraw(nds, nds, shift, vector_len);
4623     evmovdqul(dst, nds, Assembler::AVX_512bit);
4624   } else if (dst_enc < 16) {
4625     // use nds to save a copy of xmm0 and hold shift
4626     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4627     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4628     Assembler::vpsraw(dst, dst, xmm0, vector_len);
4629     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4630   } else if (nds_enc < 16) {
4631     // use nds as dest as temps
4632     evmovdqul(nds, dst, Assembler::AVX_512bit);
4633     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4634     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4635     Assembler::vpsraw(nds, nds, xmm0, vector_len);
4636     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4637     evmovdqul(dst, nds, Assembler::AVX_512bit);
4638   } else {
4639     // worse case scenario, all regs are in the upper bank
4640     subptr(rsp, 64);
4641     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4642     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4643     evmovdqul(xmm1, shift, Assembler::AVX_512bit);
4644     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4645     Assembler::vpsllw(xmm0, xmm0, xmm1, vector_len);
4646     evmovdqul(xmm1, dst, Assembler::AVX_512bit);
4647     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4648     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4649     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4650     addptr(rsp, 64);
4651   }
4652 }
4653 
4654 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
4655   int dst_enc = dst->encoding();
4656   int nds_enc = nds->encoding();
4657   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4658     Assembler::vpsraw(dst, nds, shift, vector_len);
4659   } else if (dst_enc < 16) {
4660     Assembler::vpsraw(dst, dst, shift, vector_len);
4661   } else if (nds_enc < 16) {
4662     // use nds as scratch
4663     evmovdqul(nds, dst, Assembler::AVX_512bit);
4664     Assembler::vpsraw(nds, nds, shift, vector_len);
4665     evmovdqul(dst, nds, Assembler::AVX_512bit);
4666   } else {
4667     // use nds as scratch for xmm0
4668     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4669     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4670     Assembler::vpsraw(xmm0, xmm0, shift, vector_len);
4671     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4672   }
4673 }
4674 
4675 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
4676   int dst_enc = dst->encoding();
4677   int nds_enc = nds->encoding();
4678   int shift_enc = shift->encoding();
4679   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4680     Assembler::vpsrlw(dst, nds, shift, vector_len);
4681   } else if ((dst_enc < 16) && (shift_enc < 16)) {
4682     Assembler::vpsrlw(dst, dst, shift, vector_len);
4683   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4684     // use nds_enc as scratch with shift
4685     evmovdqul(nds, shift, Assembler::AVX_512bit);
4686     Assembler::vpsrlw(dst, dst, nds, vector_len);
4687   } else if ((shift_enc < 16) && (nds_enc < 16)) {
4688     // use nds as scratch with dst
4689     evmovdqul(nds, dst, Assembler::AVX_512bit);
4690     Assembler::vpsrlw(nds, nds, shift, vector_len);
4691     evmovdqul(dst, nds, Assembler::AVX_512bit);
4692   } else if (dst_enc < 16) {
4693     // use nds to save a copy of xmm0 and hold shift
4694     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4695     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4696     Assembler::vpsrlw(dst, dst, xmm0, vector_len);
4697     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4698   } else if (nds_enc < 16) {
4699     // use nds as dest as temps
4700     evmovdqul(nds, dst, Assembler::AVX_512bit);
4701     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4702     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4703     Assembler::vpsrlw(nds, nds, xmm0, vector_len);
4704     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4705     evmovdqul(dst, nds, Assembler::AVX_512bit);
4706   } else {
4707     // worse case scenario, all regs are in the upper bank
4708     subptr(rsp, 64);
4709     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4710     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4711     evmovdqul(xmm1, shift, Assembler::AVX_512bit);
4712     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4713     Assembler::vpsllw(xmm0, xmm0, xmm1, vector_len);
4714     evmovdqul(xmm1, dst, Assembler::AVX_512bit);
4715     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4716     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4717     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4718     addptr(rsp, 64);
4719   }
4720 }
4721 
4722 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
4723   int dst_enc = dst->encoding();
4724   int nds_enc = nds->encoding();
4725   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4726     Assembler::vpsrlw(dst, nds, shift, vector_len);
4727   } else if (dst_enc < 16) {
4728     Assembler::vpsrlw(dst, dst, shift, vector_len);
4729   } else if (nds_enc < 16) {
4730     // use nds as scratch
4731     evmovdqul(nds, dst, Assembler::AVX_512bit);
4732     Assembler::vpsrlw(nds, nds, shift, vector_len);
4733     evmovdqul(dst, nds, Assembler::AVX_512bit);
4734   } else {
4735     // use nds as scratch for xmm0
4736     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4737     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4738     Assembler::vpsrlw(xmm0, xmm0, shift, vector_len);
4739     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4740   }
4741 }
4742 
4743 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
4744   int dst_enc = dst->encoding();
4745   int nds_enc = nds->encoding();
4746   int shift_enc = shift->encoding();
4747   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4748     Assembler::vpsllw(dst, nds, shift, vector_len);
4749   } else if ((dst_enc < 16) && (shift_enc < 16)) {
4750     Assembler::vpsllw(dst, dst, shift, vector_len);
4751   } else if ((dst_enc < 16) && (nds_enc < 16)) {
4752     // use nds_enc as scratch with shift
4753     evmovdqul(nds, shift, Assembler::AVX_512bit);
4754     Assembler::vpsllw(dst, dst, nds, vector_len);
4755   } else if ((shift_enc < 16) && (nds_enc < 16)) {
4756     // use nds as scratch with dst
4757     evmovdqul(nds, dst, Assembler::AVX_512bit);
4758     Assembler::vpsllw(nds, nds, shift, vector_len);
4759     evmovdqul(dst, nds, Assembler::AVX_512bit);
4760   } else if (dst_enc < 16) {
4761     // use nds to save a copy of xmm0 and hold shift
4762     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4763     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4764     Assembler::vpsllw(dst, dst, xmm0, vector_len);
4765     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4766   } else if (nds_enc < 16) {
4767     // use nds as dest as temps
4768     evmovdqul(nds, dst, Assembler::AVX_512bit);
4769     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4770     evmovdqul(xmm0, shift, Assembler::AVX_512bit);
4771     Assembler::vpsllw(nds, nds, xmm0, vector_len);
4772     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4773     evmovdqul(dst, nds, Assembler::AVX_512bit);
4774   } else {
4775     // worse case scenario, all regs are in the upper bank
4776     subptr(rsp, 64);
4777     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4778     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4779     evmovdqul(xmm1, shift, Assembler::AVX_512bit);
4780     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4781     Assembler::vpsllw(xmm0, xmm0, xmm1, vector_len);
4782     evmovdqul(xmm1, dst, Assembler::AVX_512bit);
4783     evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4784     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4785     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4786     addptr(rsp, 64);
4787   }
4788 }
4789 
4790 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
4791   int dst_enc = dst->encoding();
4792   int nds_enc = nds->encoding();
4793   if (VM_Version::supports_avxonly() || VM_Version::supports_avx512bw()) {
4794     Assembler::vpsllw(dst, nds, shift, vector_len);
4795   } else if (dst_enc < 16) {
4796     Assembler::vpsllw(dst, dst, shift, vector_len);
4797   } else if (nds_enc < 16) {
4798     // use nds as scratch
4799     evmovdqul(nds, dst, Assembler::AVX_512bit);
4800     Assembler::vpsllw(nds, nds, shift, vector_len);
4801     evmovdqul(dst, nds, Assembler::AVX_512bit);
4802   } else {
4803     // use nds as scratch for xmm0
4804     evmovdqul(nds, xmm0, Assembler::AVX_512bit);
4805     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4806     Assembler::vpsllw(xmm0, xmm0, shift, vector_len);
4807     evmovdqul(xmm0, nds, Assembler::AVX_512bit);
4808   }
4809 }
4810 
4811 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
4812   int dst_enc = dst->encoding();
4813   int src_enc = src->encoding();
4814   if ((dst_enc < 16) && (src_enc < 16)) {
4815     Assembler::vptest(dst, src);
4816   } else if (src_enc < 16) {
4817     subptr(rsp, 64);
4818     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4819     evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4820     Assembler::vptest(xmm0, src);
4821     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4822     addptr(rsp, 64);
4823   } else if (dst_enc < 16) {
4824     subptr(rsp, 64);
4825     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4826     evmovdqul(xmm0, src, Assembler::AVX_512bit);
4827     Assembler::vptest(dst, xmm0);
4828     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4829     addptr(rsp, 64);
4830   } else {
4831     subptr(rsp, 64);
4832     evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4833     subptr(rsp, 64);
4834     evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4835     movdqu(xmm0, src);
4836     movdqu(xmm1, dst);
4837     Assembler::vptest(xmm1, xmm0);
4838     evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4839     addptr(rsp, 64);
4840     evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4841     addptr(rsp, 64);
4842   }
4843 }
4844 
4845 // This instruction exists within macros, ergo we cannot control its input
4846 // when emitted through those patterns.
4847 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
4848   if (VM_Version::supports_avx512nobw()) {
4849     int dst_enc = dst->encoding();
4850     int src_enc = src->encoding();
4851     if (dst_enc == src_enc) {
4852       if (dst_enc < 16) {
4853         Assembler::punpcklbw(dst, src);
4854       } else {
4855         subptr(rsp, 64);
4856         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4857         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4858         Assembler::punpcklbw(xmm0, xmm0);
4859         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4860         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4861         addptr(rsp, 64);
4862       }
4863     } else {
4864       if ((src_enc < 16) && (dst_enc < 16)) {
4865         Assembler::punpcklbw(dst, src);
4866       } else if (src_enc < 16) {
4867         subptr(rsp, 64);
4868         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4869         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4870         Assembler::punpcklbw(xmm0, src);
4871         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4872         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4873         addptr(rsp, 64);
4874       } else if (dst_enc < 16) {
4875         subptr(rsp, 64);
4876         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4877         evmovdqul(xmm0, src, Assembler::AVX_512bit);
4878         Assembler::punpcklbw(dst, xmm0);
4879         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4880         addptr(rsp, 64);
4881       } else {
4882         subptr(rsp, 64);
4883         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4884         subptr(rsp, 64);
4885         evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4886         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4887         evmovdqul(xmm1, src, Assembler::AVX_512bit);
4888         Assembler::punpcklbw(xmm0, xmm1);
4889         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4890         evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4891         addptr(rsp, 64);
4892         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4893         addptr(rsp, 64);
4894       }
4895     }
4896   } else {
4897     Assembler::punpcklbw(dst, src);
4898   }
4899 }
4900 
4901 // This instruction exists within macros, ergo we cannot control its input
4902 // when emitted through those patterns.
4903 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
4904   if (VM_Version::supports_avx512nobw()) {
4905     int dst_enc = dst->encoding();
4906     int src_enc = src->encoding();
4907     if (dst_enc == src_enc) {
4908       if (dst_enc < 16) {
4909         Assembler::pshuflw(dst, src, mode);
4910       } else {
4911         subptr(rsp, 64);
4912         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4913         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4914         Assembler::pshuflw(xmm0, xmm0, mode);
4915         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4916         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4917         addptr(rsp, 64);
4918       }
4919     } else {
4920       if ((src_enc < 16) && (dst_enc < 16)) {
4921         Assembler::pshuflw(dst, src, mode);
4922       } else if (src_enc < 16) {
4923         subptr(rsp, 64);
4924         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4925         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4926         Assembler::pshuflw(xmm0, src, mode);
4927         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4928         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4929         addptr(rsp, 64);
4930       } else if (dst_enc < 16) {
4931         subptr(rsp, 64);
4932         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4933         evmovdqul(xmm0, src, Assembler::AVX_512bit);
4934         Assembler::pshuflw(dst, xmm0, mode);
4935         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4936         addptr(rsp, 64);
4937       } else {
4938         subptr(rsp, 64);
4939         evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
4940         subptr(rsp, 64);
4941         evmovdqul(Address(rsp, 0), xmm1, Assembler::AVX_512bit);
4942         evmovdqul(xmm0, dst, Assembler::AVX_512bit);
4943         evmovdqul(xmm1, src, Assembler::AVX_512bit);
4944         Assembler::pshuflw(xmm0, xmm1, mode);
4945         evmovdqul(dst, xmm0, Assembler::AVX_512bit);
4946         evmovdqul(xmm1, Address(rsp, 0), Assembler::AVX_512bit);
4947         addptr(rsp, 64);
4948         evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
4949         addptr(rsp, 64);
4950       }
4951     }
4952   } else {
4953     Assembler::pshuflw(dst, src, mode);
4954   }
4955 }
4956 
4957 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len) {
4958   if (reachable(src)) {
4959     vandpd(dst, nds, as_Address(src), vector_len);
4960   } else {
4961     lea(rscratch1, src);
4962     vandpd(dst, nds, Address(rscratch1, 0), vector_len);
4963   }
4964 }
4965 
4966 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len) {
4967   if (reachable(src)) {
4968     vandps(dst, nds, as_Address(src), vector_len);
4969   } else {
4970     lea(rscratch1, src);
4971     vandps(dst, nds, Address(rscratch1, 0), vector_len);
4972   }
4973 }
4974 
4975 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
4976   if (reachable(src)) {
4977     vdivsd(dst, nds, as_Address(src));
4978   } else {
4979     lea(rscratch1, src);
4980     vdivsd(dst, nds, Address(rscratch1, 0));
4981   }
4982 }
4983 
4984 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
4985   if (reachable(src)) {
4986     vdivss(dst, nds, as_Address(src));
4987   } else {
4988     lea(rscratch1, src);
4989     vdivss(dst, nds, Address(rscratch1, 0));
4990   }
4991 }
4992 
4993 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
4994   if (reachable(src)) {
4995     vmulsd(dst, nds, as_Address(src));
4996   } else {
4997     lea(rscratch1, src);
4998     vmulsd(dst, nds, Address(rscratch1, 0));
4999   }
5000 }
5001 
5002 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
5003   if (reachable(src)) {
5004     vmulss(dst, nds, as_Address(src));
5005   } else {
5006     lea(rscratch1, src);
5007     vmulss(dst, nds, Address(rscratch1, 0));
5008   }
5009 }
5010 
5011 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
5012   if (reachable(src)) {
5013     vsubsd(dst, nds, as_Address(src));
5014   } else {
5015     lea(rscratch1, src);
5016     vsubsd(dst, nds, Address(rscratch1, 0));
5017   }
5018 }
5019 
5020 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
5021   if (reachable(src)) {
5022     vsubss(dst, nds, as_Address(src));
5023   } else {
5024     lea(rscratch1, src);
5025     vsubss(dst, nds, Address(rscratch1, 0));
5026   }
5027 }
5028 
5029 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
5030   int nds_enc = nds->encoding();
5031   int dst_enc = dst->encoding();
5032   bool dst_upper_bank = (dst_enc > 15);
5033   bool nds_upper_bank = (nds_enc > 15);
5034   if (VM_Version::supports_avx512novl() &&
5035       (nds_upper_bank || dst_upper_bank)) {
5036     if (dst_upper_bank) {
5037       subptr(rsp, 64);
5038       evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
5039       movflt(xmm0, nds);
5040       vxorps(xmm0, xmm0, src, Assembler::AVX_128bit);
5041       movflt(dst, xmm0);
5042       evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
5043       addptr(rsp, 64);
5044     } else {
5045       movflt(dst, nds);
5046       vxorps(dst, dst, src, Assembler::AVX_128bit);
5047     }
5048   } else {
5049     vxorps(dst, nds, src, Assembler::AVX_128bit);
5050   }
5051 }
5052 
5053 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src) {
5054   int nds_enc = nds->encoding();
5055   int dst_enc = dst->encoding();
5056   bool dst_upper_bank = (dst_enc > 15);
5057   bool nds_upper_bank = (nds_enc > 15);
5058   if (VM_Version::supports_avx512novl() &&
5059       (nds_upper_bank || dst_upper_bank)) {
5060     if (dst_upper_bank) {
5061       subptr(rsp, 64);
5062       evmovdqul(Address(rsp, 0), xmm0, Assembler::AVX_512bit);
5063       movdbl(xmm0, nds);
5064       vxorpd(xmm0, xmm0, src, Assembler::AVX_128bit);
5065       movdbl(dst, xmm0);
5066       evmovdqul(xmm0, Address(rsp, 0), Assembler::AVX_512bit);
5067       addptr(rsp, 64);
5068     } else {
5069       movdbl(dst, nds);
5070       vxorpd(dst, dst, src, Assembler::AVX_128bit);
5071     }
5072   } else {
5073     vxorpd(dst, nds, src, Assembler::AVX_128bit);
5074   }
5075 }
5076 
5077 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len) {
5078   if (reachable(src)) {
5079     vxorpd(dst, nds, as_Address(src), vector_len);
5080   } else {
5081     lea(rscratch1, src);
5082     vxorpd(dst, nds, Address(rscratch1, 0), vector_len);
5083   }
5084 }
5085 
5086 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len) {
5087   if (reachable(src)) {
5088     vxorps(dst, nds, as_Address(src), vector_len);
5089   } else {
5090     lea(rscratch1, src);
5091     vxorps(dst, nds, Address(rscratch1, 0), vector_len);
5092   }
5093 }
5094 
5095 
5096 //////////////////////////////////////////////////////////////////////////////////
5097 #if INCLUDE_ALL_GCS
5098 
5099 void MacroAssembler::g1_write_barrier_pre(Register obj,
5100                                           Register pre_val,
5101                                           Register thread,
5102                                           Register tmp,
5103                                           bool tosca_live,
5104                                           bool expand_call) {
5105 
5106   // If expand_call is true then we expand the call_VM_leaf macro
5107   // directly to skip generating the check by
5108   // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
5109 
5110 #ifdef _LP64
5111   assert(thread == r15_thread, "must be");
5112 #endif // _LP64
5113 
5114   Label done;
5115   Label runtime;
5116 
5117   assert(pre_val != noreg, "check this code");
5118 
5119   if (obj != noreg) {
5120     assert_different_registers(obj, pre_val, tmp);
5121     assert(pre_val != rax, "check this code");
5122   }
5123 
5124   Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
5125                                        SATBMarkQueue::byte_offset_of_active()));
5126   Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
5127                                        SATBMarkQueue::byte_offset_of_index()));
5128   Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
5129                                        SATBMarkQueue::byte_offset_of_buf()));
5130 
5131 
5132   // Is marking active?
5133   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
5134     cmpl(in_progress, 0);
5135   } else {
5136     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
5137     cmpb(in_progress, 0);
5138   }
5139   jcc(Assembler::equal, done);
5140 
5141   // Do we need to load the previous value?
5142   if (obj != noreg) {
5143     load_heap_oop(pre_val, Address(obj, 0));
5144   }
5145 
5146   // Is the previous value null?
5147   cmpptr(pre_val, (int32_t) NULL_WORD);
5148   jcc(Assembler::equal, done);
5149 
5150   // Can we store original value in the thread's buffer?
5151   // Is index == 0?
5152   // (The index field is typed as size_t.)
5153 
5154   movptr(tmp, index);                   // tmp := *index_adr
5155   cmpptr(tmp, 0);                       // tmp == 0?
5156   jcc(Assembler::equal, runtime);       // If yes, goto runtime
5157 
5158   subptr(tmp, wordSize);                // tmp := tmp - wordSize
5159   movptr(index, tmp);                   // *index_adr := tmp
5160   addptr(tmp, buffer);                  // tmp := tmp + *buffer_adr
5161 
5162   // Record the previous value
5163   movptr(Address(tmp, 0), pre_val);
5164   jmp(done);
5165 
5166   bind(runtime);
5167   // save the live input values
5168   if(tosca_live) push(rax);
5169 
5170   if (obj != noreg && obj != rax)
5171     push(obj);
5172 
5173   if (pre_val != rax)
5174     push(pre_val);
5175 
5176   // Calling the runtime using the regular call_VM_leaf mechanism generates
5177   // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
5178   // that checks that the *(ebp+frame::interpreter_frame_last_sp) == NULL.
5179   //
5180   // If we care generating the pre-barrier without a frame (e.g. in the
5181   // intrinsified Reference.get() routine) then ebp might be pointing to
5182   // the caller frame and so this check will most likely fail at runtime.
5183   //
5184   // Expanding the call directly bypasses the generation of the check.
5185   // So when we do not have have a full interpreter frame on the stack
5186   // expand_call should be passed true.
5187 
5188   NOT_LP64( push(thread); )
5189 
5190   if (expand_call) {
5191     LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); )
5192     pass_arg1(this, thread);
5193     pass_arg0(this, pre_val);
5194     MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), 2);
5195   } else {
5196     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), pre_val, thread);
5197   }
5198 
5199   NOT_LP64( pop(thread); )
5200 
5201   // save the live input values
5202   if (pre_val != rax)
5203     pop(pre_val);
5204 
5205   if (obj != noreg && obj != rax)
5206     pop(obj);
5207 
5208   if(tosca_live) pop(rax);
5209 
5210   bind(done);
5211 }
5212 
5213 void MacroAssembler::g1_write_barrier_post(Register store_addr,
5214                                            Register new_val,
5215                                            Register thread,
5216                                            Register tmp,
5217                                            Register tmp2) {
5218 #ifdef _LP64
5219   assert(thread == r15_thread, "must be");
5220 #endif // _LP64
5221 
5222   Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
5223                                        DirtyCardQueue::byte_offset_of_index()));
5224   Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
5225                                        DirtyCardQueue::byte_offset_of_buf()));
5226 
5227   CardTableModRefBS* ct =
5228     barrier_set_cast<CardTableModRefBS>(Universe::heap()->barrier_set());
5229   assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
5230 
5231   Label done;
5232   Label runtime;
5233 
5234   // Does store cross heap regions?
5235 
5236   movptr(tmp, store_addr);
5237   xorptr(tmp, new_val);
5238   shrptr(tmp, HeapRegion::LogOfHRGrainBytes);
5239   jcc(Assembler::equal, done);
5240 
5241   // crosses regions, storing NULL?
5242 
5243   cmpptr(new_val, (int32_t) NULL_WORD);
5244   jcc(Assembler::equal, done);
5245 
5246   // storing region crossing non-NULL, is card already dirty?
5247 
5248   const Register card_addr = tmp;
5249   const Register cardtable = tmp2;
5250 
5251   movptr(card_addr, store_addr);
5252   shrptr(card_addr, CardTableModRefBS::card_shift);
5253   // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
5254   // a valid address and therefore is not properly handled by the relocation code.
5255   movptr(cardtable, (intptr_t)ct->byte_map_base);
5256   addptr(card_addr, cardtable);
5257 
5258   cmpb(Address(card_addr, 0), (int)G1SATBCardTableModRefBS::g1_young_card_val());
5259   jcc(Assembler::equal, done);
5260 
5261   membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
5262   cmpb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
5263   jcc(Assembler::equal, done);
5264 
5265 
5266   // storing a region crossing, non-NULL oop, card is clean.
5267   // dirty card and log.
5268 
5269   movb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
5270 
5271   cmpl(queue_index, 0);
5272   jcc(Assembler::equal, runtime);
5273   subl(queue_index, wordSize);
5274   movptr(tmp2, buffer);
5275 #ifdef _LP64
5276   movslq(rscratch1, queue_index);
5277   addq(tmp2, rscratch1);
5278   movq(Address(tmp2, 0), card_addr);
5279 #else
5280   addl(tmp2, queue_index);
5281   movl(Address(tmp2, 0), card_addr);
5282 #endif
5283   jmp(done);
5284 
5285   bind(runtime);
5286   // save the live input values
5287   push(store_addr);
5288   push(new_val);
5289 #ifdef _LP64
5290   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, r15_thread);
5291 #else
5292   push(thread);
5293   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
5294   pop(thread);
5295 #endif
5296   pop(new_val);
5297   pop(store_addr);
5298 
5299   bind(done);
5300 }
5301 
5302 #endif // INCLUDE_ALL_GCS
5303 //////////////////////////////////////////////////////////////////////////////////
5304 
5305 
5306 void MacroAssembler::store_check(Register obj, Address dst) {
5307   store_check(obj);
5308 }
5309 
5310 void MacroAssembler::store_check(Register obj) {
5311   // Does a store check for the oop in register obj. The content of
5312   // register obj is destroyed afterwards.
5313   BarrierSet* bs = Universe::heap()->barrier_set();
5314   assert(bs->kind() == BarrierSet::CardTableForRS ||
5315          bs->kind() == BarrierSet::CardTableExtension,
5316          "Wrong barrier set kind");
5317 
5318   CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
5319   assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
5320 
5321   shrptr(obj, CardTableModRefBS::card_shift);
5322 
5323   Address card_addr;
5324 
5325   // The calculation for byte_map_base is as follows:
5326   // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift);
5327   // So this essentially converts an address to a displacement and it will
5328   // never need to be relocated. On 64bit however the value may be too
5329   // large for a 32bit displacement.
5330   intptr_t disp = (intptr_t) ct->byte_map_base;
5331   if (is_simm32(disp)) {
5332     card_addr = Address(noreg, obj, Address::times_1, disp);
5333   } else {
5334     // By doing it as an ExternalAddress 'disp' could be converted to a rip-relative
5335     // displacement and done in a single instruction given favorable mapping and a
5336     // smarter version of as_Address. However, 'ExternalAddress' generates a relocation
5337     // entry and that entry is not properly handled by the relocation code.
5338     AddressLiteral cardtable((address)ct->byte_map_base, relocInfo::none);
5339     Address index(noreg, obj, Address::times_1);
5340     card_addr = as_Address(ArrayAddress(cardtable, index));
5341   }
5342 
5343   int dirty = CardTableModRefBS::dirty_card_val();
5344   if (UseCondCardMark) {
5345     Label L_already_dirty;
5346     if (UseConcMarkSweepGC) {
5347       membar(Assembler::StoreLoad);
5348     }
5349     cmpb(card_addr, dirty);
5350     jcc(Assembler::equal, L_already_dirty);
5351     movb(card_addr, dirty);
5352     bind(L_already_dirty);
5353   } else {
5354     movb(card_addr, dirty);
5355   }
5356 }
5357 
5358 void MacroAssembler::subptr(Register dst, int32_t imm32) {
5359   LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32));
5360 }
5361 
5362 // Force generation of a 4 byte immediate value even if it fits into 8bit
5363 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
5364   LP64_ONLY(subq_imm32(dst, imm32)) NOT_LP64(subl_imm32(dst, imm32));
5365 }
5366 
5367 void MacroAssembler::subptr(Register dst, Register src) {
5368   LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src));
5369 }
5370 
5371 // C++ bool manipulation
5372 void MacroAssembler::testbool(Register dst) {
5373   if(sizeof(bool) == 1)
5374     testb(dst, 0xff);
5375   else if(sizeof(bool) == 2) {
5376     // testw implementation needed for two byte bools
5377     ShouldNotReachHere();
5378   } else if(sizeof(bool) == 4)
5379     testl(dst, dst);
5380   else
5381     // unsupported
5382     ShouldNotReachHere();
5383 }
5384 
5385 void MacroAssembler::testptr(Register dst, Register src) {
5386   LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
5387 }
5388 
5389 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5390 void MacroAssembler::tlab_allocate(Register obj,
5391                                    Register var_size_in_bytes,
5392                                    int con_size_in_bytes,
5393                                    Register t1,
5394                                    Register t2,
5395                                    Label& slow_case) {
5396   assert_different_registers(obj, t1, t2);
5397   assert_different_registers(obj, var_size_in_bytes, t1);
5398   Register end = t2;
5399   Register thread = NOT_LP64(t1) LP64_ONLY(r15_thread);
5400 
5401   verify_tlab();
5402 
5403   NOT_LP64(get_thread(thread));
5404 
5405   movptr(obj, Address(thread, JavaThread::tlab_top_offset()));
5406   if (var_size_in_bytes == noreg) {
5407     lea(end, Address(obj, con_size_in_bytes));
5408   } else {
5409     lea(end, Address(obj, var_size_in_bytes, Address::times_1));
5410   }
5411   cmpptr(end, Address(thread, JavaThread::tlab_end_offset()));
5412   jcc(Assembler::above, slow_case);
5413 
5414   // update the tlab top pointer
5415   movptr(Address(thread, JavaThread::tlab_top_offset()), end);
5416 
5417   // recover var_size_in_bytes if necessary
5418   if (var_size_in_bytes == end) {
5419     subptr(var_size_in_bytes, obj);
5420   }
5421   verify_tlab();
5422 }
5423 
5424 // Preserves rbx, and rdx.
5425 Register MacroAssembler::tlab_refill(Label& retry,
5426                                      Label& try_eden,
5427                                      Label& slow_case) {
5428   Register top = rax;
5429   Register t1  = rcx; // object size
5430   Register t2  = rsi;
5431   Register thread_reg = NOT_LP64(rdi) LP64_ONLY(r15_thread);
5432   assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx);
5433   Label do_refill, discard_tlab;
5434 
5435   if (!Universe::heap()->supports_inline_contig_alloc()) {
5436     // No allocation in the shared eden.
5437     jmp(slow_case);
5438   }
5439 
5440   NOT_LP64(get_thread(thread_reg));
5441 
5442   movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
5443   movptr(t1,  Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
5444 
5445   // calculate amount of free space
5446   subptr(t1, top);
5447   shrptr(t1, LogHeapWordSize);
5448 
5449   // Retain tlab and allocate object in shared space if
5450   // the amount free in the tlab is too large to discard.
5451   cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())));
5452   jcc(Assembler::lessEqual, discard_tlab);
5453 
5454   // Retain
5455   // %%% yuck as movptr...
5456   movptr(t2, (int32_t) ThreadLocalAllocBuffer::refill_waste_limit_increment());
5457   addptr(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2);
5458   if (TLABStats) {
5459     // increment number of slow_allocations
5460     addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1);
5461   }
5462   jmp(try_eden);
5463 
5464   bind(discard_tlab);
5465   if (TLABStats) {
5466     // increment number of refills
5467     addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1);
5468     // accumulate wastage -- t1 is amount free in tlab
5469     addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1);
5470   }
5471 
5472   // if tlab is currently allocated (top or end != null) then
5473   // fill [top, end + alignment_reserve) with array object
5474   testptr(top, top);
5475   jcc(Assembler::zero, do_refill);
5476 
5477   // set up the mark word
5478   movptr(Address(top, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::prototype()->copy_set_hash(0x2));
5479   // set the length to the remaining space
5480   subptr(t1, typeArrayOopDesc::header_size(T_INT));
5481   addptr(t1, (int32_t)ThreadLocalAllocBuffer::alignment_reserve());
5482   shlptr(t1, log2_intptr(HeapWordSize/sizeof(jint)));
5483   movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1);
5484   // set klass to intArrayKlass
5485   // dubious reloc why not an oop reloc?
5486   movptr(t1, ExternalAddress((address)Universe::intArrayKlassObj_addr()));
5487   // store klass last.  concurrent gcs assumes klass length is valid if
5488   // klass field is not null.
5489   store_klass(top, t1);
5490 
5491   movptr(t1, top);
5492   subptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
5493   incr_allocated_bytes(thread_reg, t1, 0);
5494 
5495   // refill the tlab with an eden allocation
5496   bind(do_refill);
5497   movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
5498   shlptr(t1, LogHeapWordSize);
5499   // allocate new tlab, address returned in top
5500   eden_allocate(top, t1, 0, t2, slow_case);
5501 
5502   // Check that t1 was preserved in eden_allocate.
5503 #ifdef ASSERT
5504   if (UseTLAB) {
5505     Label ok;
5506     Register tsize = rsi;
5507     assert_different_registers(tsize, thread_reg, t1);
5508     push(tsize);
5509     movptr(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
5510     shlptr(tsize, LogHeapWordSize);
5511     cmpptr(t1, tsize);
5512     jcc(Assembler::equal, ok);
5513     STOP("assert(t1 != tlab size)");
5514     should_not_reach_here();
5515 
5516     bind(ok);
5517     pop(tsize);
5518   }
5519 #endif
5520   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top);
5521   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top);
5522   addptr(top, t1);
5523   subptr(top, (int32_t)ThreadLocalAllocBuffer::alignment_reserve_in_bytes());
5524   movptr(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top);
5525 
5526   if (ZeroTLAB) {
5527     // This is a fast TLAB refill, therefore the GC is not notified of it.
5528     // So compiled code must fill the new TLAB with zeroes.
5529     movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
5530     zero_memory(top, t1, 0, t2);
5531   }
5532 
5533   verify_tlab();
5534   jmp(retry);
5535 
5536   return thread_reg; // for use by caller
5537 }
5538 
5539 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
5540 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
5541   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
5542   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
5543   Label done;
5544 
5545   testptr(length_in_bytes, length_in_bytes);
5546   jcc(Assembler::zero, done);
5547 
5548   // initialize topmost word, divide index by 2, check if odd and test if zero
5549   // note: for the remaining code to work, index must be a multiple of BytesPerWord
5550 #ifdef ASSERT
5551   {
5552     Label L;
5553     testptr(length_in_bytes, BytesPerWord - 1);
5554     jcc(Assembler::zero, L);
5555     stop("length must be a multiple of BytesPerWord");
5556     bind(L);
5557   }
5558 #endif
5559   Register index = length_in_bytes;
5560   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
5561   if (UseIncDec) {
5562     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
5563   } else {
5564     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
5565     shrptr(index, 1);
5566   }
5567 #ifndef _LP64
5568   // index could have not been a multiple of 8 (i.e., bit 2 was set)
5569   {
5570     Label even;
5571     // note: if index was a multiple of 8, then it cannot
5572     //       be 0 now otherwise it must have been 0 before
5573     //       => if it is even, we don't need to check for 0 again
5574     jcc(Assembler::carryClear, even);
5575     // clear topmost word (no jump would be needed if conditional assignment worked here)
5576     movptr(Address(address, index, Address::times_8, offset_in_bytes - 0*BytesPerWord), temp);
5577     // index could be 0 now, must check again
5578     jcc(Assembler::zero, done);
5579     bind(even);
5580   }
5581 #endif // !_LP64
5582   // initialize remaining object fields: index is a multiple of 2 now
5583   {
5584     Label loop;
5585     bind(loop);
5586     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
5587     NOT_LP64(movptr(Address(address, index, Address::times_8, offset_in_bytes - 2*BytesPerWord), temp);)
5588     decrement(index);
5589     jcc(Assembler::notZero, loop);
5590   }
5591 
5592   bind(done);
5593 }
5594 
5595 void MacroAssembler::incr_allocated_bytes(Register thread,
5596                                           Register var_size_in_bytes,
5597                                           int con_size_in_bytes,
5598                                           Register t1) {
5599   if (!thread->is_valid()) {
5600 #ifdef _LP64
5601     thread = r15_thread;
5602 #else
5603     assert(t1->is_valid(), "need temp reg");
5604     thread = t1;
5605     get_thread(thread);
5606 #endif
5607   }
5608 
5609 #ifdef _LP64
5610   if (var_size_in_bytes->is_valid()) {
5611     addq(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), var_size_in_bytes);
5612   } else {
5613     addq(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), con_size_in_bytes);
5614   }
5615 #else
5616   if (var_size_in_bytes->is_valid()) {
5617     addl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), var_size_in_bytes);
5618   } else {
5619     addl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())), con_size_in_bytes);
5620   }
5621   adcl(Address(thread, in_bytes(JavaThread::allocated_bytes_offset())+4), 0);
5622 #endif
5623 }
5624 
5625 void MacroAssembler::fp_runtime_fallback(address runtime_entry, int nb_args, int num_fpu_regs_in_use) {
5626   pusha();
5627 
5628   // if we are coming from c1, xmm registers may be live
5629   int num_xmm_regs = LP64_ONLY(16) NOT_LP64(8);
5630   if (UseAVX > 2) {
5631     num_xmm_regs = LP64_ONLY(32) NOT_LP64(8);
5632   }
5633 
5634   if (UseSSE == 1)  {
5635     subptr(rsp, sizeof(jdouble)*8);
5636     for (int n = 0; n < 8; n++) {
5637       movflt(Address(rsp, n*sizeof(jdouble)), as_XMMRegister(n));
5638     }
5639   } else if (UseSSE >= 2)  {
5640     if (UseAVX > 2) {
5641       push(rbx);
5642       movl(rbx, 0xffff);
5643       kmovwl(k1, rbx);
5644       pop(rbx);
5645     }
5646 #ifdef COMPILER2
5647     if (MaxVectorSize > 16) {
5648       if(UseAVX > 2) {
5649         // Save upper half of ZMM registers
5650         subptr(rsp, 32*num_xmm_regs);
5651         for (int n = 0; n < num_xmm_regs; n++) {
5652           vextractf64x4_high(Address(rsp, n*32), as_XMMRegister(n));
5653         }
5654       }
5655       assert(UseAVX > 0, "256 bit vectors are supported only with AVX");
5656       // Save upper half of YMM registers
5657       subptr(rsp, 16*num_xmm_regs);
5658       for (int n = 0; n < num_xmm_regs; n++) {
5659         vextractf128_high(Address(rsp, n*16), as_XMMRegister(n));
5660       }
5661     }
5662 #endif
5663     // Save whole 128bit (16 bytes) XMM registers
5664     subptr(rsp, 16*num_xmm_regs);
5665 #ifdef _LP64
5666     if (VM_Version::supports_evex()) {
5667       for (int n = 0; n < num_xmm_regs; n++) {
5668         vextractf32x4(Address(rsp, n*16), as_XMMRegister(n), 0);
5669       }
5670     } else {
5671       for (int n = 0; n < num_xmm_regs; n++) {
5672         movdqu(Address(rsp, n*16), as_XMMRegister(n));
5673       }
5674     }
5675 #else
5676     for (int n = 0; n < num_xmm_regs; n++) {
5677       movdqu(Address(rsp, n*16), as_XMMRegister(n));
5678     }
5679 #endif
5680   }
5681 
5682   // Preserve registers across runtime call
5683   int incoming_argument_and_return_value_offset = -1;
5684   if (num_fpu_regs_in_use > 1) {
5685     // Must preserve all other FPU regs (could alternatively convert
5686     // SharedRuntime::dsin, dcos etc. into assembly routines known not to trash
5687     // FPU state, but can not trust C compiler)
5688     NEEDS_CLEANUP;
5689     // NOTE that in this case we also push the incoming argument(s) to
5690     // the stack and restore it later; we also use this stack slot to
5691     // hold the return value from dsin, dcos etc.
5692     for (int i = 0; i < num_fpu_regs_in_use; i++) {
5693       subptr(rsp, sizeof(jdouble));
5694       fstp_d(Address(rsp, 0));
5695     }
5696     incoming_argument_and_return_value_offset = sizeof(jdouble)*(num_fpu_regs_in_use-1);
5697     for (int i = nb_args-1; i >= 0; i--) {
5698       fld_d(Address(rsp, incoming_argument_and_return_value_offset-i*sizeof(jdouble)));
5699     }
5700   }
5701 
5702   subptr(rsp, nb_args*sizeof(jdouble));
5703   for (int i = 0; i < nb_args; i++) {
5704     fstp_d(Address(rsp, i*sizeof(jdouble)));
5705   }
5706 
5707 #ifdef _LP64
5708   if (nb_args > 0) {
5709     movdbl(xmm0, Address(rsp, 0));
5710   }
5711   if (nb_args > 1) {
5712     movdbl(xmm1, Address(rsp, sizeof(jdouble)));
5713   }
5714   assert(nb_args <= 2, "unsupported number of args");
5715 #endif // _LP64
5716 
5717   // NOTE: we must not use call_VM_leaf here because that requires a
5718   // complete interpreter frame in debug mode -- same bug as 4387334
5719   // MacroAssembler::call_VM_leaf_base is perfectly safe and will
5720   // do proper 64bit abi
5721 
5722   NEEDS_CLEANUP;
5723   // Need to add stack banging before this runtime call if it needs to
5724   // be taken; however, there is no generic stack banging routine at
5725   // the MacroAssembler level
5726 
5727   MacroAssembler::call_VM_leaf_base(runtime_entry, 0);
5728 
5729 #ifdef _LP64
5730   movsd(Address(rsp, 0), xmm0);
5731   fld_d(Address(rsp, 0));
5732 #endif // _LP64
5733   addptr(rsp, sizeof(jdouble)*nb_args);
5734   if (num_fpu_regs_in_use > 1) {
5735     // Must save return value to stack and then restore entire FPU
5736     // stack except incoming arguments
5737     fstp_d(Address(rsp, incoming_argument_and_return_value_offset));
5738     for (int i = 0; i < num_fpu_regs_in_use - nb_args; i++) {
5739       fld_d(Address(rsp, 0));
5740       addptr(rsp, sizeof(jdouble));
5741     }
5742     fld_d(Address(rsp, (nb_args-1)*sizeof(jdouble)));
5743     addptr(rsp, sizeof(jdouble)*nb_args);
5744   }
5745 
5746   if (UseSSE == 1)  {
5747     for (int n = 0; n < 8; n++) {
5748       movflt(as_XMMRegister(n), Address(rsp, n*sizeof(jdouble)));
5749     }
5750     addptr(rsp, sizeof(jdouble)*8);
5751   } else if (UseSSE >= 2)  {
5752     // Restore whole 128bit (16 bytes) XMM registers
5753 #ifdef _LP64
5754   if (VM_Version::supports_evex()) {
5755     for (int n = 0; n < num_xmm_regs; n++) {
5756       vinsertf32x4(as_XMMRegister(n), as_XMMRegister(n), Address(rsp, n*16), 0);
5757     }
5758   } else {
5759     for (int n = 0; n < num_xmm_regs; n++) {
5760       movdqu(as_XMMRegister(n), Address(rsp, n*16));
5761     }
5762   }
5763 #else
5764   for (int n = 0; n < num_xmm_regs; n++) {
5765     movdqu(as_XMMRegister(n), Address(rsp, n*16));
5766   }
5767 #endif
5768     addptr(rsp, 16*num_xmm_regs);
5769 
5770 #ifdef COMPILER2
5771     if (MaxVectorSize > 16) {
5772       // Restore upper half of YMM registers.
5773       for (int n = 0; n < num_xmm_regs; n++) {
5774         vinsertf128_high(as_XMMRegister(n), Address(rsp, n*16));
5775       }
5776       addptr(rsp, 16*num_xmm_regs);
5777       if(UseAVX > 2) {
5778         for (int n = 0; n < num_xmm_regs; n++) {
5779           vinsertf64x4_high(as_XMMRegister(n), Address(rsp, n*32));
5780         }
5781         addptr(rsp, 32*num_xmm_regs);
5782       }
5783     }
5784 #endif
5785   }
5786   popa();
5787 }
5788 
5789 static const double     pi_4 =  0.7853981633974483;
5790 
5791 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) {
5792   // A hand-coded argument reduction for values in fabs(pi/4, pi/2)
5793   // was attempted in this code; unfortunately it appears that the
5794   // switch to 80-bit precision and back causes this to be
5795   // unprofitable compared with simply performing a runtime call if
5796   // the argument is out of the (-pi/4, pi/4) range.
5797 
5798   Register tmp = noreg;
5799   if (!VM_Version::supports_cmov()) {
5800     // fcmp needs a temporary so preserve rbx,
5801     tmp = rbx;
5802     push(tmp);
5803   }
5804 
5805   Label slow_case, done;
5806   if (trig == 't') {
5807     ExternalAddress pi4_adr = (address)&pi_4;
5808     if (reachable(pi4_adr)) {
5809       // x ?<= pi/4
5810       fld_d(pi4_adr);
5811       fld_s(1);                // Stack:  X  PI/4  X
5812       fabs();                  // Stack: |X| PI/4  X
5813       fcmp(tmp);
5814       jcc(Assembler::above, slow_case);
5815 
5816       // fastest case: -pi/4 <= x <= pi/4
5817       ftan();
5818 
5819       jmp(done);
5820     }
5821   }
5822   // slow case: runtime call
5823   bind(slow_case);
5824 
5825   switch(trig) {
5826   case 's':
5827     {
5828       fp_runtime_fallback(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), 1, num_fpu_regs_in_use);
5829     }
5830     break;
5831   case 'c':
5832     {
5833       fp_runtime_fallback(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), 1, num_fpu_regs_in_use);
5834     }
5835     break;
5836   case 't':
5837     {
5838       fp_runtime_fallback(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 1, num_fpu_regs_in_use);
5839     }
5840     break;
5841   default:
5842     assert(false, "bad intrinsic");
5843     break;
5844   }
5845 
5846   // Come here with result in F-TOS
5847   bind(done);
5848 
5849   if (tmp != noreg) {
5850     pop(tmp);
5851   }
5852 }
5853 
5854 // Look up the method for a megamorphic invokeinterface call.
5855 // The target method is determined by <intf_klass, itable_index>.
5856 // The receiver klass is in recv_klass.
5857 // On success, the result will be in method_result, and execution falls through.
5858 // On failure, execution transfers to the given label.
5859 void MacroAssembler::lookup_interface_method(Register recv_klass,
5860                                              Register intf_klass,
5861                                              RegisterOrConstant itable_index,
5862                                              Register method_result,
5863                                              Register scan_temp,
5864                                              Label& L_no_such_interface) {
5865   assert_different_registers(recv_klass, intf_klass, method_result, scan_temp);
5866   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
5867          "caller must use same register for non-constant itable index as for method");
5868 
5869   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
5870   int vtable_base = in_bytes(Klass::vtable_start_offset());
5871   int itentry_off = itableMethodEntry::method_offset_in_bytes();
5872   int scan_step   = itableOffsetEntry::size() * wordSize;
5873   int vte_size    = vtableEntry::size_in_bytes();
5874   Address::ScaleFactor times_vte_scale = Address::times_ptr;
5875   assert(vte_size == wordSize, "else adjust times_vte_scale");
5876 
5877   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
5878 
5879   // %%% Could store the aligned, prescaled offset in the klassoop.
5880   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
5881 
5882   // Adjust recv_klass by scaled itable_index, so we can free itable_index.
5883   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
5884   lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
5885 
5886   // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) {
5887   //   if (scan->interface() == intf) {
5888   //     result = (klass + scan->offset() + itable_index);
5889   //   }
5890   // }
5891   Label search, found_method;
5892 
5893   for (int peel = 1; peel >= 0; peel--) {
5894     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset_in_bytes()));
5895     cmpptr(intf_klass, method_result);
5896 
5897     if (peel) {
5898       jccb(Assembler::equal, found_method);
5899     } else {
5900       jccb(Assembler::notEqual, search);
5901       // (invert the test to fall through to found_method...)
5902     }
5903 
5904     if (!peel)  break;
5905 
5906     bind(search);
5907 
5908     // Check that the previous entry is non-null.  A null entry means that
5909     // the receiver class doesn't implement the interface, and wasn't the
5910     // same as when the caller was compiled.
5911     testptr(method_result, method_result);
5912     jcc(Assembler::zero, L_no_such_interface);
5913     addptr(scan_temp, scan_step);
5914   }
5915 
5916   bind(found_method);
5917 
5918   // Got a hit.
5919   movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset_in_bytes()));
5920   movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
5921 }
5922 
5923 
5924 // virtual method calling
5925 void MacroAssembler::lookup_virtual_method(Register recv_klass,
5926                                            RegisterOrConstant vtable_index,
5927                                            Register method_result) {
5928   const int base = in_bytes(Klass::vtable_start_offset());
5929   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
5930   Address vtable_entry_addr(recv_klass,
5931                             vtable_index, Address::times_ptr,
5932                             base + vtableEntry::method_offset_in_bytes());
5933   movptr(method_result, vtable_entry_addr);
5934 }
5935 
5936 
5937 void MacroAssembler::check_klass_subtype(Register sub_klass,
5938                            Register super_klass,
5939                            Register temp_reg,
5940                            Label& L_success) {
5941   Label L_failure;
5942   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, NULL);
5943   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL);
5944   bind(L_failure);
5945 }
5946 
5947 
5948 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
5949                                                    Register super_klass,
5950                                                    Register temp_reg,
5951                                                    Label* L_success,
5952                                                    Label* L_failure,
5953                                                    Label* L_slow_path,
5954                                         RegisterOrConstant super_check_offset) {
5955   assert_different_registers(sub_klass, super_klass, temp_reg);
5956   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
5957   if (super_check_offset.is_register()) {
5958     assert_different_registers(sub_klass, super_klass,
5959                                super_check_offset.as_register());
5960   } else if (must_load_sco) {
5961     assert(temp_reg != noreg, "supply either a temp or a register offset");
5962   }
5963 
5964   Label L_fallthrough;
5965   int label_nulls = 0;
5966   if (L_success == NULL)   { L_success   = &L_fallthrough; label_nulls++; }
5967   if (L_failure == NULL)   { L_failure   = &L_fallthrough; label_nulls++; }
5968   if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; }
5969   assert(label_nulls <= 1, "at most one NULL in the batch");
5970 
5971   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
5972   int sco_offset = in_bytes(Klass::super_check_offset_offset());
5973   Address super_check_offset_addr(super_klass, sco_offset);
5974 
5975   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
5976   // range of a jccb.  If this routine grows larger, reconsider at
5977   // least some of these.
5978 #define local_jcc(assembler_cond, label)                                \
5979   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
5980   else                             jcc( assembler_cond, label) /*omit semi*/
5981 
5982   // Hacked jmp, which may only be used just before L_fallthrough.
5983 #define final_jmp(label)                                                \
5984   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
5985   else                            jmp(label)                /*omit semi*/
5986 
5987   // If the pointers are equal, we are done (e.g., String[] elements).
5988   // This self-check enables sharing of secondary supertype arrays among
5989   // non-primary types such as array-of-interface.  Otherwise, each such
5990   // type would need its own customized SSA.
5991   // We move this check to the front of the fast path because many
5992   // type checks are in fact trivially successful in this manner,
5993   // so we get a nicely predicted branch right at the start of the check.
5994   cmpptr(sub_klass, super_klass);
5995   local_jcc(Assembler::equal, *L_success);
5996 
5997   // Check the supertype display:
5998   if (must_load_sco) {
5999     // Positive movl does right thing on LP64.
6000     movl(temp_reg, super_check_offset_addr);
6001     super_check_offset = RegisterOrConstant(temp_reg);
6002   }
6003   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
6004   cmpptr(super_klass, super_check_addr); // load displayed supertype
6005 
6006   // This check has worked decisively for primary supers.
6007   // Secondary supers are sought in the super_cache ('super_cache_addr').
6008   // (Secondary supers are interfaces and very deeply nested subtypes.)
6009   // This works in the same check above because of a tricky aliasing
6010   // between the super_cache and the primary super display elements.
6011   // (The 'super_check_addr' can address either, as the case requires.)
6012   // Note that the cache is updated below if it does not help us find
6013   // what we need immediately.
6014   // So if it was a primary super, we can just fail immediately.
6015   // Otherwise, it's the slow path for us (no success at this point).
6016 
6017   if (super_check_offset.is_register()) {
6018     local_jcc(Assembler::equal, *L_success);
6019     cmpl(super_check_offset.as_register(), sc_offset);
6020     if (L_failure == &L_fallthrough) {
6021       local_jcc(Assembler::equal, *L_slow_path);
6022     } else {
6023       local_jcc(Assembler::notEqual, *L_failure);
6024       final_jmp(*L_slow_path);
6025     }
6026   } else if (super_check_offset.as_constant() == sc_offset) {
6027     // Need a slow path; fast failure is impossible.
6028     if (L_slow_path == &L_fallthrough) {
6029       local_jcc(Assembler::equal, *L_success);
6030     } else {
6031       local_jcc(Assembler::notEqual, *L_slow_path);
6032       final_jmp(*L_success);
6033     }
6034   } else {
6035     // No slow path; it's a fast decision.
6036     if (L_failure == &L_fallthrough) {
6037       local_jcc(Assembler::equal, *L_success);
6038     } else {
6039       local_jcc(Assembler::notEqual, *L_failure);
6040       final_jmp(*L_success);
6041     }
6042   }
6043 
6044   bind(L_fallthrough);
6045 
6046 #undef local_jcc
6047 #undef final_jmp
6048 }
6049 
6050 
6051 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
6052                                                    Register super_klass,
6053                                                    Register temp_reg,
6054                                                    Register temp2_reg,
6055                                                    Label* L_success,
6056                                                    Label* L_failure,
6057                                                    bool set_cond_codes) {
6058   assert_different_registers(sub_klass, super_klass, temp_reg);
6059   if (temp2_reg != noreg)
6060     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
6061 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
6062 
6063   Label L_fallthrough;
6064   int label_nulls = 0;
6065   if (L_success == NULL)   { L_success   = &L_fallthrough; label_nulls++; }
6066   if (L_failure == NULL)   { L_failure   = &L_fallthrough; label_nulls++; }
6067   assert(label_nulls <= 1, "at most one NULL in the batch");
6068 
6069   // a couple of useful fields in sub_klass:
6070   int ss_offset = in_bytes(Klass::secondary_supers_offset());
6071   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
6072   Address secondary_supers_addr(sub_klass, ss_offset);
6073   Address super_cache_addr(     sub_klass, sc_offset);
6074 
6075   // Do a linear scan of the secondary super-klass chain.
6076   // This code is rarely used, so simplicity is a virtue here.
6077   // The repne_scan instruction uses fixed registers, which we must spill.
6078   // Don't worry too much about pre-existing connections with the input regs.
6079 
6080   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
6081   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
6082 
6083   // Get super_klass value into rax (even if it was in rdi or rcx).
6084   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
6085   if (super_klass != rax || UseCompressedOops) {
6086     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
6087     mov(rax, super_klass);
6088   }
6089   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
6090   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
6091 
6092 #ifndef PRODUCT
6093   int* pst_counter = &SharedRuntime::_partial_subtype_ctr;
6094   ExternalAddress pst_counter_addr((address) pst_counter);
6095   NOT_LP64(  incrementl(pst_counter_addr) );
6096   LP64_ONLY( lea(rcx, pst_counter_addr) );
6097   LP64_ONLY( incrementl(Address(rcx, 0)) );
6098 #endif //PRODUCT
6099 
6100   // We will consult the secondary-super array.
6101   movptr(rdi, secondary_supers_addr);
6102   // Load the array length.  (Positive movl does right thing on LP64.)
6103   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
6104   // Skip to start of data.
6105   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
6106 
6107   // Scan RCX words at [RDI] for an occurrence of RAX.
6108   // Set NZ/Z based on last compare.
6109   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
6110   // not change flags (only scas instruction which is repeated sets flags).
6111   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
6112 
6113     testptr(rax,rax); // Set Z = 0
6114     repne_scan();
6115 
6116   // Unspill the temp. registers:
6117   if (pushed_rdi)  pop(rdi);
6118   if (pushed_rcx)  pop(rcx);
6119   if (pushed_rax)  pop(rax);
6120 
6121   if (set_cond_codes) {
6122     // Special hack for the AD files:  rdi is guaranteed non-zero.
6123     assert(!pushed_rdi, "rdi must be left non-NULL");
6124     // Also, the condition codes are properly set Z/NZ on succeed/failure.
6125   }
6126 
6127   if (L_failure == &L_fallthrough)
6128         jccb(Assembler::notEqual, *L_failure);
6129   else  jcc(Assembler::notEqual, *L_failure);
6130 
6131   // Success.  Cache the super we found and proceed in triumph.
6132   movptr(super_cache_addr, super_klass);
6133 
6134   if (L_success != &L_fallthrough) {
6135     jmp(*L_success);
6136   }
6137 
6138 #undef IS_A_TEMP
6139 
6140   bind(L_fallthrough);
6141 }
6142 
6143 
6144 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
6145   if (VM_Version::supports_cmov()) {
6146     cmovl(cc, dst, src);
6147   } else {
6148     Label L;
6149     jccb(negate_condition(cc), L);
6150     movl(dst, src);
6151     bind(L);
6152   }
6153 }
6154 
6155 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
6156   if (VM_Version::supports_cmov()) {
6157     cmovl(cc, dst, src);
6158   } else {
6159     Label L;
6160     jccb(negate_condition(cc), L);
6161     movl(dst, src);
6162     bind(L);
6163   }
6164 }
6165 
6166 void MacroAssembler::verify_oop(Register reg, const char* s) {
6167   if (!VerifyOops) return;
6168 
6169   // Pass register number to verify_oop_subroutine
6170   const char* b = NULL;
6171   {
6172     ResourceMark rm;
6173     stringStream ss;
6174     ss.print("verify_oop: %s: %s", reg->name(), s);
6175     b = code_string(ss.as_string());
6176   }
6177   BLOCK_COMMENT("verify_oop {");
6178 #ifdef _LP64
6179   push(rscratch1);                    // save r10, trashed by movptr()
6180 #endif
6181   push(rax);                          // save rax,
6182   push(reg);                          // pass register argument
6183   ExternalAddress buffer((address) b);
6184   // avoid using pushptr, as it modifies scratch registers
6185   // and our contract is not to modify anything
6186   movptr(rax, buffer.addr());
6187   push(rax);
6188   // call indirectly to solve generation ordering problem
6189   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
6190   call(rax);
6191   // Caller pops the arguments (oop, message) and restores rax, r10
6192   BLOCK_COMMENT("} verify_oop");
6193 }
6194 
6195 
6196 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr,
6197                                                       Register tmp,
6198                                                       int offset) {
6199   intptr_t value = *delayed_value_addr;
6200   if (value != 0)
6201     return RegisterOrConstant(value + offset);
6202 
6203   // load indirectly to solve generation ordering problem
6204   movptr(tmp, ExternalAddress((address) delayed_value_addr));
6205 
6206 #ifdef ASSERT
6207   { Label L;
6208     testptr(tmp, tmp);
6209     if (WizardMode) {
6210       const char* buf = NULL;
6211       {
6212         ResourceMark rm;
6213         stringStream ss;
6214         ss.print("DelayedValue=" INTPTR_FORMAT, delayed_value_addr[1]);
6215         buf = code_string(ss.as_string());
6216       }
6217       jcc(Assembler::notZero, L);
6218       STOP(buf);
6219     } else {
6220       jccb(Assembler::notZero, L);
6221       hlt();
6222     }
6223     bind(L);
6224   }
6225 #endif
6226 
6227   if (offset != 0)
6228     addptr(tmp, offset);
6229 
6230   return RegisterOrConstant(tmp);
6231 }
6232 
6233 
6234 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
6235                                          int extra_slot_offset) {
6236   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
6237   int stackElementSize = Interpreter::stackElementSize;
6238   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
6239 #ifdef ASSERT
6240   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
6241   assert(offset1 - offset == stackElementSize, "correct arithmetic");
6242 #endif
6243   Register             scale_reg    = noreg;
6244   Address::ScaleFactor scale_factor = Address::no_scale;
6245   if (arg_slot.is_constant()) {
6246     offset += arg_slot.as_constant() * stackElementSize;
6247   } else {
6248     scale_reg    = arg_slot.as_register();
6249     scale_factor = Address::times(stackElementSize);
6250   }
6251   offset += wordSize;           // return PC is on stack
6252   return Address(rsp, scale_reg, scale_factor, offset);
6253 }
6254 
6255 
6256 void MacroAssembler::verify_oop_addr(Address addr, const char* s) {
6257   if (!VerifyOops) return;
6258 
6259   // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord);
6260   // Pass register number to verify_oop_subroutine
6261   const char* b = NULL;
6262   {
6263     ResourceMark rm;
6264     stringStream ss;
6265     ss.print("verify_oop_addr: %s", s);
6266     b = code_string(ss.as_string());
6267   }
6268 #ifdef _LP64
6269   push(rscratch1);                    // save r10, trashed by movptr()
6270 #endif
6271   push(rax);                          // save rax,
6272   // addr may contain rsp so we will have to adjust it based on the push
6273   // we just did (and on 64 bit we do two pushes)
6274   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
6275   // stores rax into addr which is backwards of what was intended.
6276   if (addr.uses(rsp)) {
6277     lea(rax, addr);
6278     pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
6279   } else {
6280     pushptr(addr);
6281   }
6282 
6283   ExternalAddress buffer((address) b);
6284   // pass msg argument
6285   // avoid using pushptr, as it modifies scratch registers
6286   // and our contract is not to modify anything
6287   movptr(rax, buffer.addr());
6288   push(rax);
6289 
6290   // call indirectly to solve generation ordering problem
6291   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
6292   call(rax);
6293   // Caller pops the arguments (addr, message) and restores rax, r10.
6294 }
6295 
6296 void MacroAssembler::verify_tlab() {
6297 #ifdef ASSERT
6298   if (UseTLAB && VerifyOops) {
6299     Label next, ok;
6300     Register t1 = rsi;
6301     Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread);
6302 
6303     push(t1);
6304     NOT_LP64(push(thread_reg));
6305     NOT_LP64(get_thread(thread_reg));
6306 
6307     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
6308     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
6309     jcc(Assembler::aboveEqual, next);
6310     STOP("assert(top >= start)");
6311     should_not_reach_here();
6312 
6313     bind(next);
6314     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
6315     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
6316     jcc(Assembler::aboveEqual, ok);
6317     STOP("assert(top <= end)");
6318     should_not_reach_here();
6319 
6320     bind(ok);
6321     NOT_LP64(pop(thread_reg));
6322     pop(t1);
6323   }
6324 #endif
6325 }
6326 
6327 class ControlWord {
6328  public:
6329   int32_t _value;
6330 
6331   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
6332   int  precision_control() const       { return  (_value >>  8) & 3      ; }
6333   bool precision() const               { return ((_value >>  5) & 1) != 0; }
6334   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
6335   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
6336   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
6337   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
6338   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
6339 
6340   void print() const {
6341     // rounding control
6342     const char* rc;
6343     switch (rounding_control()) {
6344       case 0: rc = "round near"; break;
6345       case 1: rc = "round down"; break;
6346       case 2: rc = "round up  "; break;
6347       case 3: rc = "chop      "; break;
6348     };
6349     // precision control
6350     const char* pc;
6351     switch (precision_control()) {
6352       case 0: pc = "24 bits "; break;
6353       case 1: pc = "reserved"; break;
6354       case 2: pc = "53 bits "; break;
6355       case 3: pc = "64 bits "; break;
6356     };
6357     // flags
6358     char f[9];
6359     f[0] = ' ';
6360     f[1] = ' ';
6361     f[2] = (precision   ()) ? 'P' : 'p';
6362     f[3] = (underflow   ()) ? 'U' : 'u';
6363     f[4] = (overflow    ()) ? 'O' : 'o';
6364     f[5] = (zero_divide ()) ? 'Z' : 'z';
6365     f[6] = (denormalized()) ? 'D' : 'd';
6366     f[7] = (invalid     ()) ? 'I' : 'i';
6367     f[8] = '\x0';
6368     // output
6369     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
6370   }
6371 
6372 };
6373 
6374 class StatusWord {
6375  public:
6376   int32_t _value;
6377 
6378   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
6379   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
6380   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
6381   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
6382   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
6383   int  top() const                     { return  (_value >> 11) & 7      ; }
6384   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
6385   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
6386   bool precision() const               { return ((_value >>  5) & 1) != 0; }
6387   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
6388   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
6389   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
6390   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
6391   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
6392 
6393   void print() const {
6394     // condition codes
6395     char c[5];
6396     c[0] = (C3()) ? '3' : '-';
6397     c[1] = (C2()) ? '2' : '-';
6398     c[2] = (C1()) ? '1' : '-';
6399     c[3] = (C0()) ? '0' : '-';
6400     c[4] = '\x0';
6401     // flags
6402     char f[9];
6403     f[0] = (error_status()) ? 'E' : '-';
6404     f[1] = (stack_fault ()) ? 'S' : '-';
6405     f[2] = (precision   ()) ? 'P' : '-';
6406     f[3] = (underflow   ()) ? 'U' : '-';
6407     f[4] = (overflow    ()) ? 'O' : '-';
6408     f[5] = (zero_divide ()) ? 'Z' : '-';
6409     f[6] = (denormalized()) ? 'D' : '-';
6410     f[7] = (invalid     ()) ? 'I' : '-';
6411     f[8] = '\x0';
6412     // output
6413     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
6414   }
6415 
6416 };
6417 
6418 class TagWord {
6419  public:
6420   int32_t _value;
6421 
6422   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
6423 
6424   void print() const {
6425     printf("%04x", _value & 0xFFFF);
6426   }
6427 
6428 };
6429 
6430 class FPU_Register {
6431  public:
6432   int32_t _m0;
6433   int32_t _m1;
6434   int16_t _ex;
6435 
6436   bool is_indefinite() const           {
6437     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
6438   }
6439 
6440   void print() const {
6441     char  sign = (_ex < 0) ? '-' : '+';
6442     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
6443     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
6444   };
6445 
6446 };
6447 
6448 class FPU_State {
6449  public:
6450   enum {
6451     register_size       = 10,
6452     number_of_registers =  8,
6453     register_mask       =  7
6454   };
6455 
6456   ControlWord  _control_word;
6457   StatusWord   _status_word;
6458   TagWord      _tag_word;
6459   int32_t      _error_offset;
6460   int32_t      _error_selector;
6461   int32_t      _data_offset;
6462   int32_t      _data_selector;
6463   int8_t       _register[register_size * number_of_registers];
6464 
6465   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
6466   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
6467 
6468   const char* tag_as_string(int tag) const {
6469     switch (tag) {
6470       case 0: return "valid";
6471       case 1: return "zero";
6472       case 2: return "special";
6473       case 3: return "empty";
6474     }
6475     ShouldNotReachHere();
6476     return NULL;
6477   }
6478 
6479   void print() const {
6480     // print computation registers
6481     { int t = _status_word.top();
6482       for (int i = 0; i < number_of_registers; i++) {
6483         int j = (i - t) & register_mask;
6484         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
6485         st(j)->print();
6486         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
6487       }
6488     }
6489     printf("\n");
6490     // print control registers
6491     printf("ctrl = "); _control_word.print(); printf("\n");
6492     printf("stat = "); _status_word .print(); printf("\n");
6493     printf("tags = "); _tag_word    .print(); printf("\n");
6494   }
6495 
6496 };
6497 
6498 class Flag_Register {
6499  public:
6500   int32_t _value;
6501 
6502   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
6503   bool direction() const               { return ((_value >> 10) & 1) != 0; }
6504   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
6505   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
6506   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
6507   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
6508   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
6509 
6510   void print() const {
6511     // flags
6512     char f[8];
6513     f[0] = (overflow       ()) ? 'O' : '-';
6514     f[1] = (direction      ()) ? 'D' : '-';
6515     f[2] = (sign           ()) ? 'S' : '-';
6516     f[3] = (zero           ()) ? 'Z' : '-';
6517     f[4] = (auxiliary_carry()) ? 'A' : '-';
6518     f[5] = (parity         ()) ? 'P' : '-';
6519     f[6] = (carry          ()) ? 'C' : '-';
6520     f[7] = '\x0';
6521     // output
6522     printf("%08x  flags = %s", _value, f);
6523   }
6524 
6525 };
6526 
6527 class IU_Register {
6528  public:
6529   int32_t _value;
6530 
6531   void print() const {
6532     printf("%08x  %11d", _value, _value);
6533   }
6534 
6535 };
6536 
6537 class IU_State {
6538  public:
6539   Flag_Register _eflags;
6540   IU_Register   _rdi;
6541   IU_Register   _rsi;
6542   IU_Register   _rbp;
6543   IU_Register   _rsp;
6544   IU_Register   _rbx;
6545   IU_Register   _rdx;
6546   IU_Register   _rcx;
6547   IU_Register   _rax;
6548 
6549   void print() const {
6550     // computation registers
6551     printf("rax,  = "); _rax.print(); printf("\n");
6552     printf("rbx,  = "); _rbx.print(); printf("\n");
6553     printf("rcx  = "); _rcx.print(); printf("\n");
6554     printf("rdx  = "); _rdx.print(); printf("\n");
6555     printf("rdi  = "); _rdi.print(); printf("\n");
6556     printf("rsi  = "); _rsi.print(); printf("\n");
6557     printf("rbp,  = "); _rbp.print(); printf("\n");
6558     printf("rsp  = "); _rsp.print(); printf("\n");
6559     printf("\n");
6560     // control registers
6561     printf("flgs = "); _eflags.print(); printf("\n");
6562   }
6563 };
6564 
6565 
6566 class CPU_State {
6567  public:
6568   FPU_State _fpu_state;
6569   IU_State  _iu_state;
6570 
6571   void print() const {
6572     printf("--------------------------------------------------\n");
6573     _iu_state .print();
6574     printf("\n");
6575     _fpu_state.print();
6576     printf("--------------------------------------------------\n");
6577   }
6578 
6579 };
6580 
6581 
6582 static void _print_CPU_state(CPU_State* state) {
6583   state->print();
6584 };
6585 
6586 
6587 void MacroAssembler::print_CPU_state() {
6588   push_CPU_state();
6589   push(rsp);                // pass CPU state
6590   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
6591   addptr(rsp, wordSize);       // discard argument
6592   pop_CPU_state();
6593 }
6594 
6595 
6596 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) {
6597   static int counter = 0;
6598   FPU_State* fs = &state->_fpu_state;
6599   counter++;
6600   // For leaf calls, only verify that the top few elements remain empty.
6601   // We only need 1 empty at the top for C2 code.
6602   if( stack_depth < 0 ) {
6603     if( fs->tag_for_st(7) != 3 ) {
6604       printf("FPR7 not empty\n");
6605       state->print();
6606       assert(false, "error");
6607       return false;
6608     }
6609     return true;                // All other stack states do not matter
6610   }
6611 
6612   assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std,
6613          "bad FPU control word");
6614 
6615   // compute stack depth
6616   int i = 0;
6617   while (i < FPU_State::number_of_registers && fs->tag_for_st(i)  < 3) i++;
6618   int d = i;
6619   while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++;
6620   // verify findings
6621   if (i != FPU_State::number_of_registers) {
6622     // stack not contiguous
6623     printf("%s: stack not contiguous at ST%d\n", s, i);
6624     state->print();
6625     assert(false, "error");
6626     return false;
6627   }
6628   // check if computed stack depth corresponds to expected stack depth
6629   if (stack_depth < 0) {
6630     // expected stack depth is -stack_depth or less
6631     if (d > -stack_depth) {
6632       // too many elements on the stack
6633       printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d);
6634       state->print();
6635       assert(false, "error");
6636       return false;
6637     }
6638   } else {
6639     // expected stack depth is stack_depth
6640     if (d != stack_depth) {
6641       // wrong stack depth
6642       printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d);
6643       state->print();
6644       assert(false, "error");
6645       return false;
6646     }
6647   }
6648   // everything is cool
6649   return true;
6650 }
6651 
6652 
6653 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
6654   if (!VerifyFPU) return;
6655   push_CPU_state();
6656   push(rsp);                // pass CPU state
6657   ExternalAddress msg((address) s);
6658   // pass message string s
6659   pushptr(msg.addr());
6660   push(stack_depth);        // pass stack depth
6661   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU)));
6662   addptr(rsp, 3 * wordSize);   // discard arguments
6663   // check for error
6664   { Label L;
6665     testl(rax, rax);
6666     jcc(Assembler::notZero, L);
6667     int3();                  // break if error condition
6668     bind(L);
6669   }
6670   pop_CPU_state();
6671 }
6672 
6673 void MacroAssembler::restore_cpu_control_state_after_jni() {
6674   // Either restore the MXCSR register after returning from the JNI Call
6675   // or verify that it wasn't changed (with -Xcheck:jni flag).
6676   if (VM_Version::supports_sse()) {
6677     if (RestoreMXCSROnJNICalls) {
6678       ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
6679     } else if (CheckJNICalls) {
6680       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
6681     }
6682   }
6683   if (VM_Version::supports_avx()) {
6684     // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
6685     vzeroupper();
6686   }
6687 
6688 #ifndef _LP64
6689   // Either restore the x87 floating pointer control word after returning
6690   // from the JNI call or verify that it wasn't changed.
6691   if (CheckJNICalls) {
6692     call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
6693   }
6694 #endif // _LP64
6695 }
6696 
6697 
6698 void MacroAssembler::load_klass(Register dst, Register src) {
6699 #ifdef _LP64
6700   if (UseCompressedClassPointers) {
6701     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
6702     decode_klass_not_null(dst);
6703   } else
6704 #endif
6705     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
6706 }
6707 
6708 void MacroAssembler::load_prototype_header(Register dst, Register src) {
6709   load_klass(dst, src);
6710   movptr(dst, Address(dst, Klass::prototype_header_offset()));
6711 }
6712 
6713 void MacroAssembler::store_klass(Register dst, Register src) {
6714 #ifdef _LP64
6715   if (UseCompressedClassPointers) {
6716     encode_klass_not_null(src);
6717     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
6718   } else
6719 #endif
6720     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
6721 }
6722 
6723 void MacroAssembler::load_heap_oop(Register dst, Address src) {
6724 #ifdef _LP64
6725   // FIXME: Must change all places where we try to load the klass.
6726   if (UseCompressedOops) {
6727     movl(dst, src);
6728     decode_heap_oop(dst);
6729   } else
6730 #endif
6731     movptr(dst, src);
6732 }
6733 
6734 // Doesn't do verfication, generates fixed size code
6735 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src) {
6736 #ifdef _LP64
6737   if (UseCompressedOops) {
6738     movl(dst, src);
6739     decode_heap_oop_not_null(dst);
6740   } else
6741 #endif
6742     movptr(dst, src);
6743 }
6744 
6745 void MacroAssembler::store_heap_oop(Address dst, Register src) {
6746 #ifdef _LP64
6747   if (UseCompressedOops) {
6748     assert(!dst.uses(src), "not enough registers");
6749     encode_heap_oop(src);
6750     movl(dst, src);
6751   } else
6752 #endif
6753     movptr(dst, src);
6754 }
6755 
6756 void MacroAssembler::cmp_heap_oop(Register src1, Address src2, Register tmp) {
6757   assert_different_registers(src1, tmp);
6758 #ifdef _LP64
6759   if (UseCompressedOops) {
6760     bool did_push = false;
6761     if (tmp == noreg) {
6762       tmp = rax;
6763       push(tmp);
6764       did_push = true;
6765       assert(!src2.uses(rsp), "can't push");
6766     }
6767     load_heap_oop(tmp, src2);
6768     cmpptr(src1, tmp);
6769     if (did_push)  pop(tmp);
6770   } else
6771 #endif
6772     cmpptr(src1, src2);
6773 }
6774 
6775 // Used for storing NULLs.
6776 void MacroAssembler::store_heap_oop_null(Address dst) {
6777 #ifdef _LP64
6778   if (UseCompressedOops) {
6779     movl(dst, (int32_t)NULL_WORD);
6780   } else {
6781     movslq(dst, (int32_t)NULL_WORD);
6782   }
6783 #else
6784   movl(dst, (int32_t)NULL_WORD);
6785 #endif
6786 }
6787 
6788 #ifdef _LP64
6789 void MacroAssembler::store_klass_gap(Register dst, Register src) {
6790   if (UseCompressedClassPointers) {
6791     // Store to klass gap in destination
6792     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
6793   }
6794 }
6795 
6796 #ifdef ASSERT
6797 void MacroAssembler::verify_heapbase(const char* msg) {
6798   assert (UseCompressedOops, "should be compressed");
6799   assert (Universe::heap() != NULL, "java heap should be initialized");
6800   if (CheckCompressedOops) {
6801     Label ok;
6802     push(rscratch1); // cmpptr trashes rscratch1
6803     cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_ptrs_base_addr()));
6804     jcc(Assembler::equal, ok);
6805     STOP(msg);
6806     bind(ok);
6807     pop(rscratch1);
6808   }
6809 }
6810 #endif
6811 
6812 // Algorithm must match oop.inline.hpp encode_heap_oop.
6813 void MacroAssembler::encode_heap_oop(Register r) {
6814 #ifdef ASSERT
6815   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
6816 #endif
6817   verify_oop(r, "broken oop in encode_heap_oop");
6818   if (Universe::narrow_oop_base() == NULL) {
6819     if (Universe::narrow_oop_shift() != 0) {
6820       assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6821       shrq(r, LogMinObjAlignmentInBytes);
6822     }
6823     return;
6824   }
6825   testq(r, r);
6826   cmovq(Assembler::equal, r, r12_heapbase);
6827   subq(r, r12_heapbase);
6828   shrq(r, LogMinObjAlignmentInBytes);
6829 }
6830 
6831 void MacroAssembler::encode_heap_oop_not_null(Register r) {
6832 #ifdef ASSERT
6833   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
6834   if (CheckCompressedOops) {
6835     Label ok;
6836     testq(r, r);
6837     jcc(Assembler::notEqual, ok);
6838     STOP("null oop passed to encode_heap_oop_not_null");
6839     bind(ok);
6840   }
6841 #endif
6842   verify_oop(r, "broken oop in encode_heap_oop_not_null");
6843   if (Universe::narrow_oop_base() != NULL) {
6844     subq(r, r12_heapbase);
6845   }
6846   if (Universe::narrow_oop_shift() != 0) {
6847     assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6848     shrq(r, LogMinObjAlignmentInBytes);
6849   }
6850 }
6851 
6852 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
6853 #ifdef ASSERT
6854   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
6855   if (CheckCompressedOops) {
6856     Label ok;
6857     testq(src, src);
6858     jcc(Assembler::notEqual, ok);
6859     STOP("null oop passed to encode_heap_oop_not_null2");
6860     bind(ok);
6861   }
6862 #endif
6863   verify_oop(src, "broken oop in encode_heap_oop_not_null2");
6864   if (dst != src) {
6865     movq(dst, src);
6866   }
6867   if (Universe::narrow_oop_base() != NULL) {
6868     subq(dst, r12_heapbase);
6869   }
6870   if (Universe::narrow_oop_shift() != 0) {
6871     assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6872     shrq(dst, LogMinObjAlignmentInBytes);
6873   }
6874 }
6875 
6876 void  MacroAssembler::decode_heap_oop(Register r) {
6877 #ifdef ASSERT
6878   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
6879 #endif
6880   if (Universe::narrow_oop_base() == NULL) {
6881     if (Universe::narrow_oop_shift() != 0) {
6882       assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6883       shlq(r, LogMinObjAlignmentInBytes);
6884     }
6885   } else {
6886     Label done;
6887     shlq(r, LogMinObjAlignmentInBytes);
6888     jccb(Assembler::equal, done);
6889     addq(r, r12_heapbase);
6890     bind(done);
6891   }
6892   verify_oop(r, "broken oop in decode_heap_oop");
6893 }
6894 
6895 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
6896   // Note: it will change flags
6897   assert (UseCompressedOops, "should only be used for compressed headers");
6898   assert (Universe::heap() != NULL, "java heap should be initialized");
6899   // Cannot assert, unverified entry point counts instructions (see .ad file)
6900   // vtableStubs also counts instructions in pd_code_size_limit.
6901   // Also do not verify_oop as this is called by verify_oop.
6902   if (Universe::narrow_oop_shift() != 0) {
6903     assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6904     shlq(r, LogMinObjAlignmentInBytes);
6905     if (Universe::narrow_oop_base() != NULL) {
6906       addq(r, r12_heapbase);
6907     }
6908   } else {
6909     assert (Universe::narrow_oop_base() == NULL, "sanity");
6910   }
6911 }
6912 
6913 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
6914   // Note: it will change flags
6915   assert (UseCompressedOops, "should only be used for compressed headers");
6916   assert (Universe::heap() != NULL, "java heap should be initialized");
6917   // Cannot assert, unverified entry point counts instructions (see .ad file)
6918   // vtableStubs also counts instructions in pd_code_size_limit.
6919   // Also do not verify_oop as this is called by verify_oop.
6920   if (Universe::narrow_oop_shift() != 0) {
6921     assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
6922     if (LogMinObjAlignmentInBytes == Address::times_8) {
6923       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
6924     } else {
6925       if (dst != src) {
6926         movq(dst, src);
6927       }
6928       shlq(dst, LogMinObjAlignmentInBytes);
6929       if (Universe::narrow_oop_base() != NULL) {
6930         addq(dst, r12_heapbase);
6931       }
6932     }
6933   } else {
6934     assert (Universe::narrow_oop_base() == NULL, "sanity");
6935     if (dst != src) {
6936       movq(dst, src);
6937     }
6938   }
6939 }
6940 
6941 void MacroAssembler::encode_klass_not_null(Register r) {
6942   if (Universe::narrow_klass_base() != NULL) {
6943     // Use r12 as a scratch register in which to temporarily load the narrow_klass_base.
6944     assert(r != r12_heapbase, "Encoding a klass in r12");
6945     mov64(r12_heapbase, (int64_t)Universe::narrow_klass_base());
6946     subq(r, r12_heapbase);
6947   }
6948   if (Universe::narrow_klass_shift() != 0) {
6949     assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
6950     shrq(r, LogKlassAlignmentInBytes);
6951   }
6952   if (Universe::narrow_klass_base() != NULL) {
6953     reinit_heapbase();
6954   }
6955 }
6956 
6957 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
6958   if (dst == src) {
6959     encode_klass_not_null(src);
6960   } else {
6961     if (Universe::narrow_klass_base() != NULL) {
6962       mov64(dst, (int64_t)Universe::narrow_klass_base());
6963       negq(dst);
6964       addq(dst, src);
6965     } else {
6966       movptr(dst, src);
6967     }
6968     if (Universe::narrow_klass_shift() != 0) {
6969       assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
6970       shrq(dst, LogKlassAlignmentInBytes);
6971     }
6972   }
6973 }
6974 
6975 // Function instr_size_for_decode_klass_not_null() counts the instructions
6976 // generated by decode_klass_not_null(register r) and reinit_heapbase(),
6977 // when (Universe::heap() != NULL).  Hence, if the instructions they
6978 // generate change, then this method needs to be updated.
6979 int MacroAssembler::instr_size_for_decode_klass_not_null() {
6980   assert (UseCompressedClassPointers, "only for compressed klass ptrs");
6981   if (Universe::narrow_klass_base() != NULL) {
6982     // mov64 + addq + shlq? + mov64  (for reinit_heapbase()).
6983     return (Universe::narrow_klass_shift() == 0 ? 20 : 24);
6984   } else {
6985     // longest load decode klass function, mov64, leaq
6986     return 16;
6987   }
6988 }
6989 
6990 // !!! If the instructions that get generated here change then function
6991 // instr_size_for_decode_klass_not_null() needs to get updated.
6992 void  MacroAssembler::decode_klass_not_null(Register r) {
6993   // Note: it will change flags
6994   assert (UseCompressedClassPointers, "should only be used for compressed headers");
6995   assert(r != r12_heapbase, "Decoding a klass in r12");
6996   // Cannot assert, unverified entry point counts instructions (see .ad file)
6997   // vtableStubs also counts instructions in pd_code_size_limit.
6998   // Also do not verify_oop as this is called by verify_oop.
6999   if (Universe::narrow_klass_shift() != 0) {
7000     assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
7001     shlq(r, LogKlassAlignmentInBytes);
7002   }
7003   // Use r12 as a scratch register in which to temporarily load the narrow_klass_base.
7004   if (Universe::narrow_klass_base() != NULL) {
7005     mov64(r12_heapbase, (int64_t)Universe::narrow_klass_base());
7006     addq(r, r12_heapbase);
7007     reinit_heapbase();
7008   }
7009 }
7010 
7011 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
7012   // Note: it will change flags
7013   assert (UseCompressedClassPointers, "should only be used for compressed headers");
7014   if (dst == src) {
7015     decode_klass_not_null(dst);
7016   } else {
7017     // Cannot assert, unverified entry point counts instructions (see .ad file)
7018     // vtableStubs also counts instructions in pd_code_size_limit.
7019     // Also do not verify_oop as this is called by verify_oop.
7020     mov64(dst, (int64_t)Universe::narrow_klass_base());
7021     if (Universe::narrow_klass_shift() != 0) {
7022       assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
7023       assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
7024       leaq(dst, Address(dst, src, Address::times_8, 0));
7025     } else {
7026       addq(dst, src);
7027     }
7028   }
7029 }
7030 
7031 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
7032   assert (UseCompressedOops, "should only be used for compressed headers");
7033   assert (Universe::heap() != NULL, "java heap should be initialized");
7034   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7035   int oop_index = oop_recorder()->find_index(obj);
7036   RelocationHolder rspec = oop_Relocation::spec(oop_index);
7037   mov_narrow_oop(dst, oop_index, rspec);
7038 }
7039 
7040 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
7041   assert (UseCompressedOops, "should only be used for compressed headers");
7042   assert (Universe::heap() != NULL, "java heap should be initialized");
7043   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7044   int oop_index = oop_recorder()->find_index(obj);
7045   RelocationHolder rspec = oop_Relocation::spec(oop_index);
7046   mov_narrow_oop(dst, oop_index, rspec);
7047 }
7048 
7049 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
7050   assert (UseCompressedClassPointers, "should only be used for compressed headers");
7051   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7052   int klass_index = oop_recorder()->find_index(k);
7053   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
7054   mov_narrow_oop(dst, Klass::encode_klass(k), rspec);
7055 }
7056 
7057 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
7058   assert (UseCompressedClassPointers, "should only be used for compressed headers");
7059   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7060   int klass_index = oop_recorder()->find_index(k);
7061   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
7062   mov_narrow_oop(dst, Klass::encode_klass(k), rspec);
7063 }
7064 
7065 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
7066   assert (UseCompressedOops, "should only be used for compressed headers");
7067   assert (Universe::heap() != NULL, "java heap should be initialized");
7068   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7069   int oop_index = oop_recorder()->find_index(obj);
7070   RelocationHolder rspec = oop_Relocation::spec(oop_index);
7071   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
7072 }
7073 
7074 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
7075   assert (UseCompressedOops, "should only be used for compressed headers");
7076   assert (Universe::heap() != NULL, "java heap should be initialized");
7077   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7078   int oop_index = oop_recorder()->find_index(obj);
7079   RelocationHolder rspec = oop_Relocation::spec(oop_index);
7080   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
7081 }
7082 
7083 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
7084   assert (UseCompressedClassPointers, "should only be used for compressed headers");
7085   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7086   int klass_index = oop_recorder()->find_index(k);
7087   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
7088   Assembler::cmp_narrow_oop(dst, Klass::encode_klass(k), rspec);
7089 }
7090 
7091 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
7092   assert (UseCompressedClassPointers, "should only be used for compressed headers");
7093   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
7094   int klass_index = oop_recorder()->find_index(k);
7095   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
7096   Assembler::cmp_narrow_oop(dst, Klass::encode_klass(k), rspec);
7097 }
7098 
7099 void MacroAssembler::reinit_heapbase() {
7100   if (UseCompressedOops || UseCompressedClassPointers) {
7101     if (Universe::heap() != NULL) {
7102       if (Universe::narrow_oop_base() == NULL) {
7103         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
7104       } else {
7105         mov64(r12_heapbase, (int64_t)Universe::narrow_ptrs_base());
7106       }
7107     } else {
7108       movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_ptrs_base_addr()));
7109     }
7110   }
7111 }
7112 
7113 #endif // _LP64
7114 
7115 
7116 // C2 compiled method's prolog code.
7117 void MacroAssembler::verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b) {
7118 
7119   // WARNING: Initial instruction MUST be 5 bytes or longer so that
7120   // NativeJump::patch_verified_entry will be able to patch out the entry
7121   // code safely. The push to verify stack depth is ok at 5 bytes,
7122   // the frame allocation can be either 3 or 6 bytes. So if we don't do
7123   // stack bang then we must use the 6 byte frame allocation even if
7124   // we have no frame. :-(
7125   assert(stack_bang_size >= framesize || stack_bang_size <= 0, "stack bang size incorrect");
7126 
7127   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
7128   // Remove word for return addr
7129   framesize -= wordSize;
7130   stack_bang_size -= wordSize;
7131 
7132   // Calls to C2R adapters often do not accept exceptional returns.
7133   // We require that their callers must bang for them.  But be careful, because
7134   // some VM calls (such as call site linkage) can use several kilobytes of
7135   // stack.  But the stack safety zone should account for that.
7136   // See bugs 4446381, 4468289, 4497237.
7137   if (stack_bang_size > 0) {
7138     generate_stack_overflow_check(stack_bang_size);
7139 
7140     // We always push rbp, so that on return to interpreter rbp, will be
7141     // restored correctly and we can correct the stack.
7142     push(rbp);
7143     // Save caller's stack pointer into RBP if the frame pointer is preserved.
7144     if (PreserveFramePointer) {
7145       mov(rbp, rsp);
7146     }
7147     // Remove word for ebp
7148     framesize -= wordSize;
7149 
7150     // Create frame
7151     if (framesize) {
7152       subptr(rsp, framesize);
7153     }
7154   } else {
7155     // Create frame (force generation of a 4 byte immediate value)
7156     subptr_imm32(rsp, framesize);
7157 
7158     // Save RBP register now.
7159     framesize -= wordSize;
7160     movptr(Address(rsp, framesize), rbp);
7161     // Save caller's stack pointer into RBP if the frame pointer is preserved.
7162     if (PreserveFramePointer) {
7163       movptr(rbp, rsp);
7164       if (framesize > 0) {
7165         addptr(rbp, framesize);
7166       }
7167     }
7168   }
7169 
7170   if (VerifyStackAtCalls) { // Majik cookie to verify stack depth
7171     framesize -= wordSize;
7172     movptr(Address(rsp, framesize), (int32_t)0xbadb100d);
7173   }
7174 
7175 #ifndef _LP64
7176   // If method sets FPU control word do it now
7177   if (fp_mode_24b) {
7178     fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
7179   }
7180   if (UseSSE >= 2 && VerifyFPU) {
7181     verify_FPU(0, "FPU stack must be clean on entry");
7182   }
7183 #endif
7184 
7185 #ifdef ASSERT
7186   if (VerifyStackAtCalls) {
7187     Label L;
7188     push(rax);
7189     mov(rax, rsp);
7190     andptr(rax, StackAlignmentInBytes-1);
7191     cmpptr(rax, StackAlignmentInBytes-wordSize);
7192     pop(rax);
7193     jcc(Assembler::equal, L);
7194     STOP("Stack is not properly aligned!");
7195     bind(L);
7196   }
7197 #endif
7198 
7199 }
7200 
7201 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, bool is_large) {
7202   // cnt - number of qwords (8-byte words).
7203   // base - start address, qword aligned.
7204   // is_large - if optimizers know cnt is larger than InitArrayShortSize
7205   assert(base==rdi, "base register must be edi for rep stos");
7206   assert(tmp==rax,   "tmp register must be eax for rep stos");
7207   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
7208   assert(InitArrayShortSize % BytesPerLong == 0,
7209     "InitArrayShortSize should be the multiple of BytesPerLong");
7210 
7211   Label DONE;
7212 
7213   xorptr(tmp, tmp);
7214 
7215   if (!is_large) {
7216     Label LOOP, LONG;
7217     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
7218     jccb(Assembler::greater, LONG);
7219 
7220     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
7221 
7222     decrement(cnt);
7223     jccb(Assembler::negative, DONE); // Zero length
7224 
7225     // Use individual pointer-sized stores for small counts:
7226     BIND(LOOP);
7227     movptr(Address(base, cnt, Address::times_ptr), tmp);
7228     decrement(cnt);
7229     jccb(Assembler::greaterEqual, LOOP);
7230     jmpb(DONE);
7231 
7232     BIND(LONG);
7233   }
7234 
7235   // Use longer rep-prefixed ops for non-small counts:
7236   if (UseFastStosb) {
7237     shlptr(cnt, 3); // convert to number of bytes
7238     rep_stosb();
7239   } else {
7240     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
7241     rep_stos();
7242   }
7243 
7244   BIND(DONE);
7245 }
7246 
7247 #ifdef COMPILER2
7248 
7249 // IndexOf for constant substrings with size >= 8 chars
7250 // which don't need to be loaded through stack.
7251 void MacroAssembler::string_indexofC8(Register str1, Register str2,
7252                                       Register cnt1, Register cnt2,
7253                                       int int_cnt2,  Register result,
7254                                       XMMRegister vec, Register tmp,
7255                                       int ae) {
7256   ShortBranchVerifier sbv(this);
7257   assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
7258   assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
7259   assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
7260 
7261   // This method uses the pcmpestri instruction with bound registers
7262   //   inputs:
7263   //     xmm - substring
7264   //     rax - substring length (elements count)
7265   //     mem - scanned string
7266   //     rdx - string length (elements count)
7267   //     0xd - mode: 1100 (substring search) + 01 (unsigned shorts)
7268   //     0xc - mode: 1100 (substring search) + 00 (unsigned bytes)
7269   //   outputs:
7270   //     rcx - matched index in string
7271   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
7272   int mode   = (ae == StrIntrinsicNode::LL) ? 0x0c : 0x0d; // bytes or shorts
7273   int stride = (ae == StrIntrinsicNode::LL) ? 16 : 8; //UU, UL -> 8
7274   Address::ScaleFactor scale1 = (ae == StrIntrinsicNode::LL) ? Address::times_1 : Address::times_2;
7275   Address::ScaleFactor scale2 = (ae == StrIntrinsicNode::UL) ? Address::times_1 : scale1;
7276 
7277   Label RELOAD_SUBSTR, SCAN_TO_SUBSTR, SCAN_SUBSTR,
7278         RET_FOUND, RET_NOT_FOUND, EXIT, FOUND_SUBSTR,
7279         MATCH_SUBSTR_HEAD, RELOAD_STR, FOUND_CANDIDATE;
7280 
7281   // Note, inline_string_indexOf() generates checks:
7282   // if (substr.count > string.count) return -1;
7283   // if (substr.count == 0) return 0;
7284   assert(int_cnt2 >= stride, "this code is used only for cnt2 >= 8 chars");
7285 
7286   // Load substring.
7287   if (ae == StrIntrinsicNode::UL) {
7288     pmovzxbw(vec, Address(str2, 0));
7289   } else {
7290     movdqu(vec, Address(str2, 0));
7291   }
7292   movl(cnt2, int_cnt2);
7293   movptr(result, str1); // string addr
7294 
7295   if (int_cnt2 > stride) {
7296     jmpb(SCAN_TO_SUBSTR);
7297 
7298     // Reload substr for rescan, this code
7299     // is executed only for large substrings (> 8 chars)
7300     bind(RELOAD_SUBSTR);
7301     if (ae == StrIntrinsicNode::UL) {
7302       pmovzxbw(vec, Address(str2, 0));
7303     } else {
7304       movdqu(vec, Address(str2, 0));
7305     }
7306     negptr(cnt2); // Jumped here with negative cnt2, convert to positive
7307 
7308     bind(RELOAD_STR);
7309     // We came here after the beginning of the substring was
7310     // matched but the rest of it was not so we need to search
7311     // again. Start from the next element after the previous match.
7312 
7313     // cnt2 is number of substring reminding elements and
7314     // cnt1 is number of string reminding elements when cmp failed.
7315     // Restored cnt1 = cnt1 - cnt2 + int_cnt2
7316     subl(cnt1, cnt2);
7317     addl(cnt1, int_cnt2);
7318     movl(cnt2, int_cnt2); // Now restore cnt2
7319 
7320     decrementl(cnt1);     // Shift to next element
7321     cmpl(cnt1, cnt2);
7322     jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
7323 
7324     addptr(result, (1<<scale1));
7325 
7326   } // (int_cnt2 > 8)
7327 
7328   // Scan string for start of substr in 16-byte vectors
7329   bind(SCAN_TO_SUBSTR);
7330   pcmpestri(vec, Address(result, 0), mode);
7331   jccb(Assembler::below, FOUND_CANDIDATE);   // CF == 1
7332   subl(cnt1, stride);
7333   jccb(Assembler::lessEqual, RET_NOT_FOUND); // Scanned full string
7334   cmpl(cnt1, cnt2);
7335   jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
7336   addptr(result, 16);
7337   jmpb(SCAN_TO_SUBSTR);
7338 
7339   // Found a potential substr
7340   bind(FOUND_CANDIDATE);
7341   // Matched whole vector if first element matched (tmp(rcx) == 0).
7342   if (int_cnt2 == stride) {
7343     jccb(Assembler::overflow, RET_FOUND);    // OF == 1
7344   } else { // int_cnt2 > 8
7345     jccb(Assembler::overflow, FOUND_SUBSTR);
7346   }
7347   // After pcmpestri tmp(rcx) contains matched element index
7348   // Compute start addr of substr
7349   lea(result, Address(result, tmp, scale1));
7350 
7351   // Make sure string is still long enough
7352   subl(cnt1, tmp);
7353   cmpl(cnt1, cnt2);
7354   if (int_cnt2 == stride) {
7355     jccb(Assembler::greaterEqual, SCAN_TO_SUBSTR);
7356   } else { // int_cnt2 > 8
7357     jccb(Assembler::greaterEqual, MATCH_SUBSTR_HEAD);
7358   }
7359   // Left less then substring.
7360 
7361   bind(RET_NOT_FOUND);
7362   movl(result, -1);
7363   jmpb(EXIT);
7364 
7365   if (int_cnt2 > stride) {
7366     // This code is optimized for the case when whole substring
7367     // is matched if its head is matched.
7368     bind(MATCH_SUBSTR_HEAD);
7369     pcmpestri(vec, Address(result, 0), mode);
7370     // Reload only string if does not match
7371     jccb(Assembler::noOverflow, RELOAD_STR); // OF == 0
7372 
7373     Label CONT_SCAN_SUBSTR;
7374     // Compare the rest of substring (> 8 chars).
7375     bind(FOUND_SUBSTR);
7376     // First 8 chars are already matched.
7377     negptr(cnt2);
7378     addptr(cnt2, stride);
7379 
7380     bind(SCAN_SUBSTR);
7381     subl(cnt1, stride);
7382     cmpl(cnt2, -stride); // Do not read beyond substring
7383     jccb(Assembler::lessEqual, CONT_SCAN_SUBSTR);
7384     // Back-up strings to avoid reading beyond substring:
7385     // cnt1 = cnt1 - cnt2 + 8
7386     addl(cnt1, cnt2); // cnt2 is negative
7387     addl(cnt1, stride);
7388     movl(cnt2, stride); negptr(cnt2);
7389     bind(CONT_SCAN_SUBSTR);
7390     if (int_cnt2 < (int)G) {
7391       int tail_off1 = int_cnt2<<scale1;
7392       int tail_off2 = int_cnt2<<scale2;
7393       if (ae == StrIntrinsicNode::UL) {
7394         pmovzxbw(vec, Address(str2, cnt2, scale2, tail_off2));
7395       } else {
7396         movdqu(vec, Address(str2, cnt2, scale2, tail_off2));
7397       }
7398       pcmpestri(vec, Address(result, cnt2, scale1, tail_off1), mode);
7399     } else {
7400       // calculate index in register to avoid integer overflow (int_cnt2*2)
7401       movl(tmp, int_cnt2);
7402       addptr(tmp, cnt2);
7403       if (ae == StrIntrinsicNode::UL) {
7404         pmovzxbw(vec, Address(str2, tmp, scale2, 0));
7405       } else {
7406         movdqu(vec, Address(str2, tmp, scale2, 0));
7407       }
7408       pcmpestri(vec, Address(result, tmp, scale1, 0), mode);
7409     }
7410     // Need to reload strings pointers if not matched whole vector
7411     jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0
7412     addptr(cnt2, stride);
7413     jcc(Assembler::negative, SCAN_SUBSTR);
7414     // Fall through if found full substring
7415 
7416   } // (int_cnt2 > 8)
7417 
7418   bind(RET_FOUND);
7419   // Found result if we matched full small substring.
7420   // Compute substr offset
7421   subptr(result, str1);
7422   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
7423     shrl(result, 1); // index
7424   }
7425   bind(EXIT);
7426 
7427 } // string_indexofC8
7428 
7429 // Small strings are loaded through stack if they cross page boundary.
7430 void MacroAssembler::string_indexof(Register str1, Register str2,
7431                                     Register cnt1, Register cnt2,
7432                                     int int_cnt2,  Register result,
7433                                     XMMRegister vec, Register tmp,
7434                                     int ae) {
7435   ShortBranchVerifier sbv(this);
7436   assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
7437   assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
7438   assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
7439 
7440   //
7441   // int_cnt2 is length of small (< 8 chars) constant substring
7442   // or (-1) for non constant substring in which case its length
7443   // is in cnt2 register.
7444   //
7445   // Note, inline_string_indexOf() generates checks:
7446   // if (substr.count > string.count) return -1;
7447   // if (substr.count == 0) return 0;
7448   //
7449   int stride = (ae == StrIntrinsicNode::LL) ? 16 : 8; //UU, UL -> 8
7450   assert(int_cnt2 == -1 || (0 < int_cnt2 && int_cnt2 < stride), "should be != 0");
7451   // This method uses the pcmpestri instruction with bound registers
7452   //   inputs:
7453   //     xmm - substring
7454   //     rax - substring length (elements count)
7455   //     mem - scanned string
7456   //     rdx - string length (elements count)
7457   //     0xd - mode: 1100 (substring search) + 01 (unsigned shorts)
7458   //     0xc - mode: 1100 (substring search) + 00 (unsigned bytes)
7459   //   outputs:
7460   //     rcx - matched index in string
7461   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
7462   int mode = (ae == StrIntrinsicNode::LL) ? 0x0c : 0x0d; // bytes or shorts
7463   Address::ScaleFactor scale1 = (ae == StrIntrinsicNode::LL) ? Address::times_1 : Address::times_2;
7464   Address::ScaleFactor scale2 = (ae == StrIntrinsicNode::UL) ? Address::times_1 : scale1;
7465 
7466   Label RELOAD_SUBSTR, SCAN_TO_SUBSTR, SCAN_SUBSTR, ADJUST_STR,
7467         RET_FOUND, RET_NOT_FOUND, CLEANUP, FOUND_SUBSTR,
7468         FOUND_CANDIDATE;
7469 
7470   { //========================================================
7471     // We don't know where these strings are located
7472     // and we can't read beyond them. Load them through stack.
7473     Label BIG_STRINGS, CHECK_STR, COPY_SUBSTR, COPY_STR;
7474 
7475     movptr(tmp, rsp); // save old SP
7476 
7477     if (int_cnt2 > 0) {     // small (< 8 chars) constant substring
7478       if (int_cnt2 == (1>>scale2)) { // One byte
7479         assert((ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UL), "Only possible for latin1 encoding");
7480         load_unsigned_byte(result, Address(str2, 0));
7481         movdl(vec, result); // move 32 bits
7482       } else if (ae == StrIntrinsicNode::LL && int_cnt2 == 3) {  // Three bytes
7483         // Not enough header space in 32-bit VM: 12+3 = 15.
7484         movl(result, Address(str2, -1));
7485         shrl(result, 8);
7486         movdl(vec, result); // move 32 bits
7487       } else if (ae != StrIntrinsicNode::UL && int_cnt2 == (2>>scale2)) {  // One char
7488         load_unsigned_short(result, Address(str2, 0));
7489         movdl(vec, result); // move 32 bits
7490       } else if (ae != StrIntrinsicNode::UL && int_cnt2 == (4>>scale2)) { // Two chars
7491         movdl(vec, Address(str2, 0)); // move 32 bits
7492       } else if (ae != StrIntrinsicNode::UL && int_cnt2 == (8>>scale2)) { // Four chars
7493         movq(vec, Address(str2, 0));  // move 64 bits
7494       } else { // cnt2 = { 3, 5, 6, 7 } || (ae == StrIntrinsicNode::UL && cnt2 ={2, ..., 7})
7495         // Array header size is 12 bytes in 32-bit VM
7496         // + 6 bytes for 3 chars == 18 bytes,
7497         // enough space to load vec and shift.
7498         assert(HeapWordSize*TypeArrayKlass::header_size() >= 12,"sanity");
7499         if (ae == StrIntrinsicNode::UL) {
7500           int tail_off = int_cnt2-8;
7501           pmovzxbw(vec, Address(str2, tail_off));
7502           psrldq(vec, -2*tail_off);
7503         }
7504         else {
7505           int tail_off = int_cnt2*(1<<scale2);
7506           movdqu(vec, Address(str2, tail_off-16));
7507           psrldq(vec, 16-tail_off);
7508         }
7509       }
7510     } else { // not constant substring
7511       cmpl(cnt2, stride);
7512       jccb(Assembler::aboveEqual, BIG_STRINGS); // Both strings are big enough
7513 
7514       // We can read beyond string if srt+16 does not cross page boundary
7515       // since heaps are aligned and mapped by pages.
7516       assert(os::vm_page_size() < (int)G, "default page should be small");
7517       movl(result, str2); // We need only low 32 bits
7518       andl(result, (os::vm_page_size()-1));
7519       cmpl(result, (os::vm_page_size()-16));
7520       jccb(Assembler::belowEqual, CHECK_STR);
7521 
7522       // Move small strings to stack to allow load 16 bytes into vec.
7523       subptr(rsp, 16);
7524       int stk_offset = wordSize-(1<<scale2);
7525       push(cnt2);
7526 
7527       bind(COPY_SUBSTR);
7528       if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UL) {
7529         load_unsigned_byte(result, Address(str2, cnt2, scale2, -1));
7530         movb(Address(rsp, cnt2, scale2, stk_offset), result);
7531       } else if (ae == StrIntrinsicNode::UU) {
7532         load_unsigned_short(result, Address(str2, cnt2, scale2, -2));
7533         movw(Address(rsp, cnt2, scale2, stk_offset), result);
7534       }
7535       decrement(cnt2);
7536       jccb(Assembler::notZero, COPY_SUBSTR);
7537 
7538       pop(cnt2);
7539       movptr(str2, rsp);  // New substring address
7540     } // non constant
7541 
7542     bind(CHECK_STR);
7543     cmpl(cnt1, stride);
7544     jccb(Assembler::aboveEqual, BIG_STRINGS);
7545 
7546     // Check cross page boundary.
7547     movl(result, str1); // We need only low 32 bits
7548     andl(result, (os::vm_page_size()-1));
7549     cmpl(result, (os::vm_page_size()-16));
7550     jccb(Assembler::belowEqual, BIG_STRINGS);
7551 
7552     subptr(rsp, 16);
7553     int stk_offset = -(1<<scale1);
7554     if (int_cnt2 < 0) { // not constant
7555       push(cnt2);
7556       stk_offset += wordSize;
7557     }
7558     movl(cnt2, cnt1);
7559 
7560     bind(COPY_STR);
7561     if (ae == StrIntrinsicNode::LL) {
7562       load_unsigned_byte(result, Address(str1, cnt2, scale1, -1));
7563       movb(Address(rsp, cnt2, scale1, stk_offset), result);
7564     } else {
7565       load_unsigned_short(result, Address(str1, cnt2, scale1, -2));
7566       movw(Address(rsp, cnt2, scale1, stk_offset), result);
7567     }
7568     decrement(cnt2);
7569     jccb(Assembler::notZero, COPY_STR);
7570 
7571     if (int_cnt2 < 0) { // not constant
7572       pop(cnt2);
7573     }
7574     movptr(str1, rsp);  // New string address
7575 
7576     bind(BIG_STRINGS);
7577     // Load substring.
7578     if (int_cnt2 < 0) { // -1
7579       if (ae == StrIntrinsicNode::UL) {
7580         pmovzxbw(vec, Address(str2, 0));
7581       } else {
7582         movdqu(vec, Address(str2, 0));
7583       }
7584       push(cnt2);       // substr count
7585       push(str2);       // substr addr
7586       push(str1);       // string addr
7587     } else {
7588       // Small (< 8 chars) constant substrings are loaded already.
7589       movl(cnt2, int_cnt2);
7590     }
7591     push(tmp);  // original SP
7592 
7593   } // Finished loading
7594 
7595   //========================================================
7596   // Start search
7597   //
7598 
7599   movptr(result, str1); // string addr
7600 
7601   if (int_cnt2  < 0) {  // Only for non constant substring
7602     jmpb(SCAN_TO_SUBSTR);
7603 
7604     // SP saved at sp+0
7605     // String saved at sp+1*wordSize
7606     // Substr saved at sp+2*wordSize
7607     // Substr count saved at sp+3*wordSize
7608 
7609     // Reload substr for rescan, this code
7610     // is executed only for large substrings (> 8 chars)
7611     bind(RELOAD_SUBSTR);
7612     movptr(str2, Address(rsp, 2*wordSize));
7613     movl(cnt2, Address(rsp, 3*wordSize));
7614     if (ae == StrIntrinsicNode::UL) {
7615       pmovzxbw(vec, Address(str2, 0));
7616     } else {
7617       movdqu(vec, Address(str2, 0));
7618     }
7619     // We came here after the beginning of the substring was
7620     // matched but the rest of it was not so we need to search
7621     // again. Start from the next element after the previous match.
7622     subptr(str1, result); // Restore counter
7623     if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
7624       shrl(str1, 1);
7625     }
7626     addl(cnt1, str1);
7627     decrementl(cnt1);   // Shift to next element
7628     cmpl(cnt1, cnt2);
7629     jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
7630 
7631     addptr(result, (1<<scale1));
7632   } // non constant
7633 
7634   // Scan string for start of substr in 16-byte vectors
7635   bind(SCAN_TO_SUBSTR);
7636   assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
7637   pcmpestri(vec, Address(result, 0), mode);
7638   jccb(Assembler::below, FOUND_CANDIDATE);   // CF == 1
7639   subl(cnt1, stride);
7640   jccb(Assembler::lessEqual, RET_NOT_FOUND); // Scanned full string
7641   cmpl(cnt1, cnt2);
7642   jccb(Assembler::negative, RET_NOT_FOUND);  // Left less then substring
7643   addptr(result, 16);
7644 
7645   bind(ADJUST_STR);
7646   cmpl(cnt1, stride); // Do not read beyond string
7647   jccb(Assembler::greaterEqual, SCAN_TO_SUBSTR);
7648   // Back-up string to avoid reading beyond string.
7649   lea(result, Address(result, cnt1, scale1, -16));
7650   movl(cnt1, stride);
7651   jmpb(SCAN_TO_SUBSTR);
7652 
7653   // Found a potential substr
7654   bind(FOUND_CANDIDATE);
7655   // After pcmpestri tmp(rcx) contains matched element index
7656 
7657   // Make sure string is still long enough
7658   subl(cnt1, tmp);
7659   cmpl(cnt1, cnt2);
7660   jccb(Assembler::greaterEqual, FOUND_SUBSTR);
7661   // Left less then substring.
7662 
7663   bind(RET_NOT_FOUND);
7664   movl(result, -1);
7665   jmpb(CLEANUP);
7666 
7667   bind(FOUND_SUBSTR);
7668   // Compute start addr of substr
7669   lea(result, Address(result, tmp, scale1));
7670   if (int_cnt2 > 0) { // Constant substring
7671     // Repeat search for small substring (< 8 chars)
7672     // from new point without reloading substring.
7673     // Have to check that we don't read beyond string.
7674     cmpl(tmp, stride-int_cnt2);
7675     jccb(Assembler::greater, ADJUST_STR);
7676     // Fall through if matched whole substring.
7677   } else { // non constant
7678     assert(int_cnt2 == -1, "should be != 0");
7679 
7680     addl(tmp, cnt2);
7681     // Found result if we matched whole substring.
7682     cmpl(tmp, stride);
7683     jccb(Assembler::lessEqual, RET_FOUND);
7684 
7685     // Repeat search for small substring (<= 8 chars)
7686     // from new point 'str1' without reloading substring.
7687     cmpl(cnt2, stride);
7688     // Have to check that we don't read beyond string.
7689     jccb(Assembler::lessEqual, ADJUST_STR);
7690 
7691     Label CHECK_NEXT, CONT_SCAN_SUBSTR, RET_FOUND_LONG;
7692     // Compare the rest of substring (> 8 chars).
7693     movptr(str1, result);
7694 
7695     cmpl(tmp, cnt2);
7696     // First 8 chars are already matched.
7697     jccb(Assembler::equal, CHECK_NEXT);
7698 
7699     bind(SCAN_SUBSTR);
7700     pcmpestri(vec, Address(str1, 0), mode);
7701     // Need to reload strings pointers if not matched whole vector
7702     jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0
7703 
7704     bind(CHECK_NEXT);
7705     subl(cnt2, stride);
7706     jccb(Assembler::lessEqual, RET_FOUND_LONG); // Found full substring
7707     addptr(str1, 16);
7708     if (ae == StrIntrinsicNode::UL) {
7709       addptr(str2, 8);
7710     } else {
7711       addptr(str2, 16);
7712     }
7713     subl(cnt1, stride);
7714     cmpl(cnt2, stride); // Do not read beyond substring
7715     jccb(Assembler::greaterEqual, CONT_SCAN_SUBSTR);
7716     // Back-up strings to avoid reading beyond substring.
7717 
7718     if (ae == StrIntrinsicNode::UL) {
7719       lea(str2, Address(str2, cnt2, scale2, -8));
7720       lea(str1, Address(str1, cnt2, scale1, -16));
7721     } else {
7722       lea(str2, Address(str2, cnt2, scale2, -16));
7723       lea(str1, Address(str1, cnt2, scale1, -16));
7724     }
7725     subl(cnt1, cnt2);
7726     movl(cnt2, stride);
7727     addl(cnt1, stride);
7728     bind(CONT_SCAN_SUBSTR);
7729     if (ae == StrIntrinsicNode::UL) {
7730       pmovzxbw(vec, Address(str2, 0));
7731     } else {
7732       movdqu(vec, Address(str2, 0));
7733     }
7734     jmpb(SCAN_SUBSTR);
7735 
7736     bind(RET_FOUND_LONG);
7737     movptr(str1, Address(rsp, wordSize));
7738   } // non constant
7739 
7740   bind(RET_FOUND);
7741   // Compute substr offset
7742   subptr(result, str1);
7743   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
7744     shrl(result, 1); // index
7745   }
7746   bind(CLEANUP);
7747   pop(rsp); // restore SP
7748 
7749 } // string_indexof
7750 
7751 void MacroAssembler::string_indexof_char(Register str1, Register cnt1, Register ch, Register result,
7752                                          XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp) {
7753   ShortBranchVerifier sbv(this);
7754   assert(UseSSE42Intrinsics, "SSE4.2 intrinsics are required");
7755   assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
7756 
7757   int stride = 8;
7758 
7759   Label FOUND_CHAR, SCAN_TO_CHAR, SCAN_TO_CHAR_LOOP,
7760         SCAN_TO_8_CHAR, SCAN_TO_8_CHAR_LOOP, SCAN_TO_16_CHAR_LOOP,
7761         RET_NOT_FOUND, SCAN_TO_8_CHAR_INIT,
7762         FOUND_SEQ_CHAR, DONE_LABEL;
7763 
7764   movptr(result, str1);
7765   if (UseAVX >= 2) {
7766     cmpl(cnt1, stride);
7767     jccb(Assembler::less, SCAN_TO_CHAR_LOOP);
7768     cmpl(cnt1, 2*stride);
7769     jccb(Assembler::less, SCAN_TO_8_CHAR_INIT);
7770     movdl(vec1, ch);
7771     vpbroadcastw(vec1, vec1);
7772     vpxor(vec2, vec2);
7773     movl(tmp, cnt1);
7774     andl(tmp, 0xFFFFFFF0);  //vector count (in chars)
7775     andl(cnt1,0x0000000F);  //tail count (in chars)
7776 
7777     bind(SCAN_TO_16_CHAR_LOOP);
7778     vmovdqu(vec3, Address(result, 0));
7779     vpcmpeqw(vec3, vec3, vec1, 1);
7780     vptest(vec2, vec3);
7781     jcc(Assembler::carryClear, FOUND_CHAR);
7782     addptr(result, 32);
7783     subl(tmp, 2*stride);
7784     jccb(Assembler::notZero, SCAN_TO_16_CHAR_LOOP);
7785     jmp(SCAN_TO_8_CHAR);
7786     bind(SCAN_TO_8_CHAR_INIT);
7787     movdl(vec1, ch);
7788     pshuflw(vec1, vec1, 0x00);
7789     pshufd(vec1, vec1, 0);
7790     pxor(vec2, vec2);
7791   }
7792   bind(SCAN_TO_8_CHAR);
7793   cmpl(cnt1, stride);
7794   if (UseAVX >= 2) {
7795     jccb(Assembler::less, SCAN_TO_CHAR);
7796   } else {
7797     jccb(Assembler::less, SCAN_TO_CHAR_LOOP);
7798     movdl(vec1, ch);
7799     pshuflw(vec1, vec1, 0x00);
7800     pshufd(vec1, vec1, 0);
7801     pxor(vec2, vec2);
7802   }
7803   movl(tmp, cnt1);
7804   andl(tmp, 0xFFFFFFF8);  //vector count (in chars)
7805   andl(cnt1,0x00000007);  //tail count (in chars)
7806 
7807   bind(SCAN_TO_8_CHAR_LOOP);
7808   movdqu(vec3, Address(result, 0));
7809   pcmpeqw(vec3, vec1);
7810   ptest(vec2, vec3);
7811   jcc(Assembler::carryClear, FOUND_CHAR);
7812   addptr(result, 16);
7813   subl(tmp, stride);
7814   jccb(Assembler::notZero, SCAN_TO_8_CHAR_LOOP);
7815   bind(SCAN_TO_CHAR);
7816   testl(cnt1, cnt1);
7817   jcc(Assembler::zero, RET_NOT_FOUND);
7818   bind(SCAN_TO_CHAR_LOOP);
7819   load_unsigned_short(tmp, Address(result, 0));
7820   cmpl(ch, tmp);
7821   jccb(Assembler::equal, FOUND_SEQ_CHAR);
7822   addptr(result, 2);
7823   subl(cnt1, 1);
7824   jccb(Assembler::zero, RET_NOT_FOUND);
7825   jmp(SCAN_TO_CHAR_LOOP);
7826 
7827   bind(RET_NOT_FOUND);
7828   movl(result, -1);
7829   jmpb(DONE_LABEL);
7830 
7831   bind(FOUND_CHAR);
7832   if (UseAVX >= 2) {
7833     vpmovmskb(tmp, vec3);
7834   } else {
7835     pmovmskb(tmp, vec3);
7836   }
7837   bsfl(ch, tmp);
7838   addl(result, ch);
7839 
7840   bind(FOUND_SEQ_CHAR);
7841   subptr(result, str1);
7842   shrl(result, 1);
7843 
7844   bind(DONE_LABEL);
7845 } // string_indexof_char
7846 
7847 // helper function for string_compare
7848 void MacroAssembler::load_next_elements(Register elem1, Register elem2, Register str1, Register str2,
7849                                         Address::ScaleFactor scale, Address::ScaleFactor scale1,
7850                                         Address::ScaleFactor scale2, Register index, int ae) {
7851   if (ae == StrIntrinsicNode::LL) {
7852     load_unsigned_byte(elem1, Address(str1, index, scale, 0));
7853     load_unsigned_byte(elem2, Address(str2, index, scale, 0));
7854   } else if (ae == StrIntrinsicNode::UU) {
7855     load_unsigned_short(elem1, Address(str1, index, scale, 0));
7856     load_unsigned_short(elem2, Address(str2, index, scale, 0));
7857   } else {
7858     load_unsigned_byte(elem1, Address(str1, index, scale1, 0));
7859     load_unsigned_short(elem2, Address(str2, index, scale2, 0));
7860   }
7861 }
7862 
7863 // Compare strings, used for char[] and byte[].
7864 void MacroAssembler::string_compare(Register str1, Register str2,
7865                                     Register cnt1, Register cnt2, Register result,
7866                                     XMMRegister vec1, int ae) {
7867   ShortBranchVerifier sbv(this);
7868   Label LENGTH_DIFF_LABEL, POP_LABEL, DONE_LABEL, WHILE_HEAD_LABEL;
7869   Label COMPARE_WIDE_VECTORS_LOOP_FAILED;  // used only _LP64 && AVX3
7870   int stride, stride2, adr_stride, adr_stride1, adr_stride2;
7871   int stride2x2 = 0x40;
7872   Address::ScaleFactor scale = Address::no_scale;
7873   Address::ScaleFactor scale1 = Address::no_scale;
7874   Address::ScaleFactor scale2 = Address::no_scale;
7875 
7876   if (ae != StrIntrinsicNode::LL) {
7877     stride2x2 = 0x20;
7878   }
7879 
7880   if (ae == StrIntrinsicNode::LU || ae == StrIntrinsicNode::UL) {
7881     shrl(cnt2, 1);
7882   }
7883   // Compute the minimum of the string lengths and the
7884   // difference of the string lengths (stack).
7885   // Do the conditional move stuff
7886   movl(result, cnt1);
7887   subl(cnt1, cnt2);
7888   push(cnt1);
7889   cmov32(Assembler::lessEqual, cnt2, result);    // cnt2 = min(cnt1, cnt2)
7890 
7891   // Is the minimum length zero?
7892   testl(cnt2, cnt2);
7893   jcc(Assembler::zero, LENGTH_DIFF_LABEL);
7894   if (ae == StrIntrinsicNode::LL) {
7895     // Load first bytes
7896     load_unsigned_byte(result, Address(str1, 0));  // result = str1[0]
7897     load_unsigned_byte(cnt1, Address(str2, 0));    // cnt1   = str2[0]
7898   } else if (ae == StrIntrinsicNode::UU) {
7899     // Load first characters
7900     load_unsigned_short(result, Address(str1, 0));
7901     load_unsigned_short(cnt1, Address(str2, 0));
7902   } else {
7903     load_unsigned_byte(result, Address(str1, 0));
7904     load_unsigned_short(cnt1, Address(str2, 0));
7905   }
7906   subl(result, cnt1);
7907   jcc(Assembler::notZero,  POP_LABEL);
7908 
7909   if (ae == StrIntrinsicNode::UU) {
7910     // Divide length by 2 to get number of chars
7911     shrl(cnt2, 1);
7912   }
7913   cmpl(cnt2, 1);
7914   jcc(Assembler::equal, LENGTH_DIFF_LABEL);
7915 
7916   // Check if the strings start at the same location and setup scale and stride
7917   if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
7918     cmpptr(str1, str2);
7919     jcc(Assembler::equal, LENGTH_DIFF_LABEL);
7920     if (ae == StrIntrinsicNode::LL) {
7921       scale = Address::times_1;
7922       stride = 16;
7923     } else {
7924       scale = Address::times_2;
7925       stride = 8;
7926     }
7927   } else {
7928     scale1 = Address::times_1;
7929     scale2 = Address::times_2;
7930     // scale not used
7931     stride = 8;
7932   }
7933 
7934   if (UseAVX >= 2 && UseSSE42Intrinsics) {
7935     assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
7936     Label COMPARE_WIDE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_WIDE_TAIL, COMPARE_SMALL_STR;
7937     Label COMPARE_WIDE_VECTORS_LOOP, COMPARE_16_CHARS, COMPARE_INDEX_CHAR;
7938     Label COMPARE_WIDE_VECTORS_LOOP_AVX2;
7939     Label COMPARE_TAIL_LONG;
7940     Label COMPARE_WIDE_VECTORS_LOOP_AVX3;  // used only _LP64 && AVX3
7941 
7942     int pcmpmask = 0x19;
7943     if (ae == StrIntrinsicNode::LL) {
7944       pcmpmask &= ~0x01;
7945     }
7946 
7947     // Setup to compare 16-chars (32-bytes) vectors,
7948     // start from first character again because it has aligned address.
7949     if (ae == StrIntrinsicNode::LL) {
7950       stride2 = 32;
7951     } else {
7952       stride2 = 16;
7953     }
7954     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
7955       adr_stride = stride << scale;
7956     } else {
7957       adr_stride1 = 8;  //stride << scale1;
7958       adr_stride2 = 16; //stride << scale2;
7959     }
7960 
7961     assert(result == rax && cnt2 == rdx && cnt1 == rcx, "pcmpestri");
7962     // rax and rdx are used by pcmpestri as elements counters
7963     movl(result, cnt2);
7964     andl(cnt2, ~(stride2-1));   // cnt2 holds the vector count
7965     jcc(Assembler::zero, COMPARE_TAIL_LONG);
7966 
7967     // fast path : compare first 2 8-char vectors.
7968     bind(COMPARE_16_CHARS);
7969     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
7970       movdqu(vec1, Address(str1, 0));
7971     } else {
7972       pmovzxbw(vec1, Address(str1, 0));
7973     }
7974     pcmpestri(vec1, Address(str2, 0), pcmpmask);
7975     jccb(Assembler::below, COMPARE_INDEX_CHAR);
7976 
7977     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
7978       movdqu(vec1, Address(str1, adr_stride));
7979       pcmpestri(vec1, Address(str2, adr_stride), pcmpmask);
7980     } else {
7981       pmovzxbw(vec1, Address(str1, adr_stride1));
7982       pcmpestri(vec1, Address(str2, adr_stride2), pcmpmask);
7983     }
7984     jccb(Assembler::aboveEqual, COMPARE_WIDE_VECTORS);
7985     addl(cnt1, stride);
7986 
7987     // Compare the characters at index in cnt1
7988     bind(COMPARE_INDEX_CHAR); // cnt1 has the offset of the mismatching character
7989     load_next_elements(result, cnt2, str1, str2, scale, scale1, scale2, cnt1, ae);
7990     subl(result, cnt2);
7991     jmp(POP_LABEL);
7992 
7993     // Setup the registers to start vector comparison loop
7994     bind(COMPARE_WIDE_VECTORS);
7995     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
7996       lea(str1, Address(str1, result, scale));
7997       lea(str2, Address(str2, result, scale));
7998     } else {
7999       lea(str1, Address(str1, result, scale1));
8000       lea(str2, Address(str2, result, scale2));
8001     }
8002     subl(result, stride2);
8003     subl(cnt2, stride2);
8004     jcc(Assembler::zero, COMPARE_WIDE_TAIL);
8005     negptr(result);
8006 
8007     //  In a loop, compare 16-chars (32-bytes) at once using (vpxor+vptest)
8008     bind(COMPARE_WIDE_VECTORS_LOOP);
8009 
8010 #ifdef _LP64
8011     if (VM_Version::supports_avx512vlbw()) { // trying 64 bytes fast loop
8012       cmpl(cnt2, stride2x2);
8013       jccb(Assembler::below, COMPARE_WIDE_VECTORS_LOOP_AVX2);
8014       testl(cnt2, stride2x2-1);   // cnt2 holds the vector count
8015       jccb(Assembler::notZero, COMPARE_WIDE_VECTORS_LOOP_AVX2);   // means we cannot subtract by 0x40
8016 
8017       bind(COMPARE_WIDE_VECTORS_LOOP_AVX3); // the hottest loop
8018       if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8019         evmovdquq(vec1, Address(str1, result, scale), Assembler::AVX_512bit);
8020         evpcmpeqb(k7, vec1, Address(str2, result, scale), Assembler::AVX_512bit); // k7 == 11..11, if operands equal, otherwise k7 has some 0
8021       } else {
8022         vpmovzxbw(vec1, Address(str1, result, scale1), Assembler::AVX_512bit);
8023         evpcmpeqb(k7, vec1, Address(str2, result, scale2), Assembler::AVX_512bit); // k7 == 11..11, if operands equal, otherwise k7 has some 0
8024       }
8025       kortestql(k7, k7);
8026       jcc(Assembler::aboveEqual, COMPARE_WIDE_VECTORS_LOOP_FAILED);     // miscompare
8027       addptr(result, stride2x2);  // update since we already compared at this addr
8028       subl(cnt2, stride2x2);      // and sub the size too
8029       jccb(Assembler::notZero, COMPARE_WIDE_VECTORS_LOOP_AVX3);
8030 
8031       vpxor(vec1, vec1);
8032       jmpb(COMPARE_WIDE_TAIL);
8033     }//if (VM_Version::supports_avx512vlbw())
8034 #endif // _LP64
8035 
8036 
8037     bind(COMPARE_WIDE_VECTORS_LOOP_AVX2);
8038     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8039       vmovdqu(vec1, Address(str1, result, scale));
8040       vpxor(vec1, Address(str2, result, scale));
8041     } else {
8042       vpmovzxbw(vec1, Address(str1, result, scale1), Assembler::AVX_256bit);
8043       vpxor(vec1, Address(str2, result, scale2));
8044     }
8045     vptest(vec1, vec1);
8046     jcc(Assembler::notZero, VECTOR_NOT_EQUAL);
8047     addptr(result, stride2);
8048     subl(cnt2, stride2);
8049     jccb(Assembler::notZero, COMPARE_WIDE_VECTORS_LOOP);
8050     // clean upper bits of YMM registers
8051     vpxor(vec1, vec1);
8052 
8053     // compare wide vectors tail
8054     bind(COMPARE_WIDE_TAIL);
8055     testptr(result, result);
8056     jccb(Assembler::zero, LENGTH_DIFF_LABEL);
8057 
8058     movl(result, stride2);
8059     movl(cnt2, result);
8060     negptr(result);
8061     jmp(COMPARE_WIDE_VECTORS_LOOP_AVX2);
8062 
8063     // Identifies the mismatching (higher or lower)16-bytes in the 32-byte vectors.
8064     bind(VECTOR_NOT_EQUAL);
8065     // clean upper bits of YMM registers
8066     vpxor(vec1, vec1);
8067     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8068       lea(str1, Address(str1, result, scale));
8069       lea(str2, Address(str2, result, scale));
8070     } else {
8071       lea(str1, Address(str1, result, scale1));
8072       lea(str2, Address(str2, result, scale2));
8073     }
8074     jmp(COMPARE_16_CHARS);
8075 
8076     // Compare tail chars, length between 1 to 15 chars
8077     bind(COMPARE_TAIL_LONG);
8078     movl(cnt2, result);
8079     cmpl(cnt2, stride);
8080     jccb(Assembler::less, COMPARE_SMALL_STR);
8081 
8082     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8083       movdqu(vec1, Address(str1, 0));
8084     } else {
8085       pmovzxbw(vec1, Address(str1, 0));
8086     }
8087     pcmpestri(vec1, Address(str2, 0), pcmpmask);
8088     jcc(Assembler::below, COMPARE_INDEX_CHAR);
8089     subptr(cnt2, stride);
8090     jccb(Assembler::zero, LENGTH_DIFF_LABEL);
8091     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8092       lea(str1, Address(str1, result, scale));
8093       lea(str2, Address(str2, result, scale));
8094     } else {
8095       lea(str1, Address(str1, result, scale1));
8096       lea(str2, Address(str2, result, scale2));
8097     }
8098     negptr(cnt2);
8099     jmpb(WHILE_HEAD_LABEL);
8100 
8101     bind(COMPARE_SMALL_STR);
8102   } else if (UseSSE42Intrinsics) {
8103     assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
8104     Label COMPARE_WIDE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
8105     int pcmpmask = 0x19;
8106     // Setup to compare 8-char (16-byte) vectors,
8107     // start from first character again because it has aligned address.
8108     movl(result, cnt2);
8109     andl(cnt2, ~(stride - 1));   // cnt2 holds the vector count
8110     if (ae == StrIntrinsicNode::LL) {
8111       pcmpmask &= ~0x01;
8112     }
8113     jccb(Assembler::zero, COMPARE_TAIL);
8114     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8115       lea(str1, Address(str1, result, scale));
8116       lea(str2, Address(str2, result, scale));
8117     } else {
8118       lea(str1, Address(str1, result, scale1));
8119       lea(str2, Address(str2, result, scale2));
8120     }
8121     negptr(result);
8122 
8123     // pcmpestri
8124     //   inputs:
8125     //     vec1- substring
8126     //     rax - negative string length (elements count)
8127     //     mem - scanned string
8128     //     rdx - string length (elements count)
8129     //     pcmpmask - cmp mode: 11000 (string compare with negated result)
8130     //               + 00 (unsigned bytes) or  + 01 (unsigned shorts)
8131     //   outputs:
8132     //     rcx - first mismatched element index
8133     assert(result == rax && cnt2 == rdx && cnt1 == rcx, "pcmpestri");
8134 
8135     bind(COMPARE_WIDE_VECTORS);
8136     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8137       movdqu(vec1, Address(str1, result, scale));
8138       pcmpestri(vec1, Address(str2, result, scale), pcmpmask);
8139     } else {
8140       pmovzxbw(vec1, Address(str1, result, scale1));
8141       pcmpestri(vec1, Address(str2, result, scale2), pcmpmask);
8142     }
8143     // After pcmpestri cnt1(rcx) contains mismatched element index
8144 
8145     jccb(Assembler::below, VECTOR_NOT_EQUAL);  // CF==1
8146     addptr(result, stride);
8147     subptr(cnt2, stride);
8148     jccb(Assembler::notZero, COMPARE_WIDE_VECTORS);
8149 
8150     // compare wide vectors tail
8151     testptr(result, result);
8152     jccb(Assembler::zero, LENGTH_DIFF_LABEL);
8153 
8154     movl(cnt2, stride);
8155     movl(result, stride);
8156     negptr(result);
8157     if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8158       movdqu(vec1, Address(str1, result, scale));
8159       pcmpestri(vec1, Address(str2, result, scale), pcmpmask);
8160     } else {
8161       pmovzxbw(vec1, Address(str1, result, scale1));
8162       pcmpestri(vec1, Address(str2, result, scale2), pcmpmask);
8163     }
8164     jccb(Assembler::aboveEqual, LENGTH_DIFF_LABEL);
8165 
8166     // Mismatched characters in the vectors
8167     bind(VECTOR_NOT_EQUAL);
8168     addptr(cnt1, result);
8169     load_next_elements(result, cnt2, str1, str2, scale, scale1, scale2, cnt1, ae);
8170     subl(result, cnt2);
8171     jmpb(POP_LABEL);
8172 
8173     bind(COMPARE_TAIL); // limit is zero
8174     movl(cnt2, result);
8175     // Fallthru to tail compare
8176   }
8177   // Shift str2 and str1 to the end of the arrays, negate min
8178   if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
8179     lea(str1, Address(str1, cnt2, scale));
8180     lea(str2, Address(str2, cnt2, scale));
8181   } else {
8182     lea(str1, Address(str1, cnt2, scale1));
8183     lea(str2, Address(str2, cnt2, scale2));
8184   }
8185   decrementl(cnt2);  // first character was compared already
8186   negptr(cnt2);
8187 
8188   // Compare the rest of the elements
8189   bind(WHILE_HEAD_LABEL);
8190   load_next_elements(result, cnt1, str1, str2, scale, scale1, scale2, cnt2, ae);
8191   subl(result, cnt1);
8192   jccb(Assembler::notZero, POP_LABEL);
8193   increment(cnt2);
8194   jccb(Assembler::notZero, WHILE_HEAD_LABEL);
8195 
8196   // Strings are equal up to min length.  Return the length difference.
8197   bind(LENGTH_DIFF_LABEL);
8198   pop(result);
8199   if (ae == StrIntrinsicNode::UU) {
8200     // Divide diff by 2 to get number of chars
8201     sarl(result, 1);
8202   }
8203   jmpb(DONE_LABEL);
8204 
8205 #ifdef _LP64
8206   if (VM_Version::supports_avx512vlbw()) {
8207 
8208     bind(COMPARE_WIDE_VECTORS_LOOP_FAILED);
8209 
8210     kmovql(cnt1, k7);
8211     notq(cnt1);
8212     bsfq(cnt2, cnt1);
8213     if (ae != StrIntrinsicNode::LL) {
8214       // Divide diff by 2 to get number of chars
8215       sarl(cnt2, 1);
8216     }
8217     addq(result, cnt2);
8218     if (ae == StrIntrinsicNode::LL) {
8219       load_unsigned_byte(cnt1, Address(str2, result));
8220       load_unsigned_byte(result, Address(str1, result));
8221     } else if (ae == StrIntrinsicNode::UU) {
8222       load_unsigned_short(cnt1, Address(str2, result, scale));
8223       load_unsigned_short(result, Address(str1, result, scale));
8224     } else {
8225       load_unsigned_short(cnt1, Address(str2, result, scale2));
8226       load_unsigned_byte(result, Address(str1, result, scale1));
8227     }
8228     subl(result, cnt1);
8229     jmpb(POP_LABEL);
8230   }//if (VM_Version::supports_avx512vlbw())
8231 #endif // _LP64
8232 
8233   // Discard the stored length difference
8234   bind(POP_LABEL);
8235   pop(cnt1);
8236 
8237   // That's it
8238   bind(DONE_LABEL);
8239   if(ae == StrIntrinsicNode::UL) {
8240     negl(result);
8241   }
8242 
8243 }
8244 
8245 // Search for Non-ASCII character (Negative byte value) in a byte array,
8246 // return true if it has any and false otherwise.
8247 void MacroAssembler::has_negatives(Register ary1, Register len,
8248                                    Register result, Register tmp1,
8249                                    XMMRegister vec1, XMMRegister vec2) {
8250 
8251   // rsi: byte array
8252   // rcx: len
8253   // rax: result
8254   ShortBranchVerifier sbv(this);
8255   assert_different_registers(ary1, len, result, tmp1);
8256   assert_different_registers(vec1, vec2);
8257   Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_CHAR, COMPARE_VECTORS, COMPARE_BYTE;
8258 
8259   // len == 0
8260   testl(len, len);
8261   jcc(Assembler::zero, FALSE_LABEL);
8262 
8263   movl(result, len); // copy
8264 
8265   if (UseAVX >= 2 && UseSSE >= 2) {
8266     // With AVX2, use 32-byte vector compare
8267     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
8268 
8269     // Compare 32-byte vectors
8270     andl(result, 0x0000001f);  //   tail count (in bytes)
8271     andl(len, 0xffffffe0);   // vector count (in bytes)
8272     jccb(Assembler::zero, COMPARE_TAIL);
8273 
8274     lea(ary1, Address(ary1, len, Address::times_1));
8275     negptr(len);
8276 
8277     movl(tmp1, 0x80808080);   // create mask to test for Unicode chars in vector
8278     movdl(vec2, tmp1);
8279     vpbroadcastd(vec2, vec2);
8280 
8281     bind(COMPARE_WIDE_VECTORS);
8282     vmovdqu(vec1, Address(ary1, len, Address::times_1));
8283     vptest(vec1, vec2);
8284     jccb(Assembler::notZero, TRUE_LABEL);
8285     addptr(len, 32);
8286     jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
8287 
8288     testl(result, result);
8289     jccb(Assembler::zero, FALSE_LABEL);
8290 
8291     vmovdqu(vec1, Address(ary1, result, Address::times_1, -32));
8292     vptest(vec1, vec2);
8293     jccb(Assembler::notZero, TRUE_LABEL);
8294     jmpb(FALSE_LABEL);
8295 
8296     bind(COMPARE_TAIL); // len is zero
8297     movl(len, result);
8298     // Fallthru to tail compare
8299   } else if (UseSSE42Intrinsics) {
8300     assert(UseSSE >= 4, "SSE4 must be  for SSE4.2 intrinsics to be available");
8301     // With SSE4.2, use double quad vector compare
8302     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
8303 
8304     // Compare 16-byte vectors
8305     andl(result, 0x0000000f);  //   tail count (in bytes)
8306     andl(len, 0xfffffff0);   // vector count (in bytes)
8307     jccb(Assembler::zero, COMPARE_TAIL);
8308 
8309     lea(ary1, Address(ary1, len, Address::times_1));
8310     negptr(len);
8311 
8312     movl(tmp1, 0x80808080);
8313     movdl(vec2, tmp1);
8314     pshufd(vec2, vec2, 0);
8315 
8316     bind(COMPARE_WIDE_VECTORS);
8317     movdqu(vec1, Address(ary1, len, Address::times_1));
8318     ptest(vec1, vec2);
8319     jccb(Assembler::notZero, TRUE_LABEL);
8320     addptr(len, 16);
8321     jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
8322 
8323     testl(result, result);
8324     jccb(Assembler::zero, FALSE_LABEL);
8325 
8326     movdqu(vec1, Address(ary1, result, Address::times_1, -16));
8327     ptest(vec1, vec2);
8328     jccb(Assembler::notZero, TRUE_LABEL);
8329     jmpb(FALSE_LABEL);
8330 
8331     bind(COMPARE_TAIL); // len is zero
8332     movl(len, result);
8333     // Fallthru to tail compare
8334   }
8335 
8336   // Compare 4-byte vectors
8337   andl(len, 0xfffffffc); // vector count (in bytes)
8338   jccb(Assembler::zero, COMPARE_CHAR);
8339 
8340   lea(ary1, Address(ary1, len, Address::times_1));
8341   negptr(len);
8342 
8343   bind(COMPARE_VECTORS);
8344   movl(tmp1, Address(ary1, len, Address::times_1));
8345   andl(tmp1, 0x80808080);
8346   jccb(Assembler::notZero, TRUE_LABEL);
8347   addptr(len, 4);
8348   jcc(Assembler::notZero, COMPARE_VECTORS);
8349 
8350   // Compare trailing char (final 2 bytes), if any
8351   bind(COMPARE_CHAR);
8352   testl(result, 0x2);   // tail  char
8353   jccb(Assembler::zero, COMPARE_BYTE);
8354   load_unsigned_short(tmp1, Address(ary1, 0));
8355   andl(tmp1, 0x00008080);
8356   jccb(Assembler::notZero, TRUE_LABEL);
8357   subptr(result, 2);
8358   lea(ary1, Address(ary1, 2));
8359 
8360   bind(COMPARE_BYTE);
8361   testl(result, 0x1);   // tail  byte
8362   jccb(Assembler::zero, FALSE_LABEL);
8363   load_unsigned_byte(tmp1, Address(ary1, 0));
8364   andl(tmp1, 0x00000080);
8365   jccb(Assembler::notEqual, TRUE_LABEL);
8366   jmpb(FALSE_LABEL);
8367 
8368   bind(TRUE_LABEL);
8369   movl(result, 1);   // return true
8370   jmpb(DONE);
8371 
8372   bind(FALSE_LABEL);
8373   xorl(result, result); // return false
8374 
8375   // That's it
8376   bind(DONE);
8377   if (UseAVX >= 2 && UseSSE >= 2) {
8378     // clean upper bits of YMM registers
8379     vpxor(vec1, vec1);
8380     vpxor(vec2, vec2);
8381   }
8382 }
8383 
8384 // Compare char[] or byte[] arrays aligned to 4 bytes or substrings.
8385 void MacroAssembler::arrays_equals(bool is_array_equ, Register ary1, Register ary2,
8386                                    Register limit, Register result, Register chr,
8387                                    XMMRegister vec1, XMMRegister vec2, bool is_char) {
8388   ShortBranchVerifier sbv(this);
8389   Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR, COMPARE_BYTE;
8390 
8391   int length_offset  = arrayOopDesc::length_offset_in_bytes();
8392   int base_offset    = arrayOopDesc::base_offset_in_bytes(is_char ? T_CHAR : T_BYTE);
8393 
8394   if (is_array_equ) {
8395     // Check the input args
8396     cmpptr(ary1, ary2);
8397     jcc(Assembler::equal, TRUE_LABEL);
8398 
8399     // Need additional checks for arrays_equals.
8400     testptr(ary1, ary1);
8401     jcc(Assembler::zero, FALSE_LABEL);
8402     testptr(ary2, ary2);
8403     jcc(Assembler::zero, FALSE_LABEL);
8404 
8405     // Check the lengths
8406     movl(limit, Address(ary1, length_offset));
8407     cmpl(limit, Address(ary2, length_offset));
8408     jcc(Assembler::notEqual, FALSE_LABEL);
8409   }
8410 
8411   // count == 0
8412   testl(limit, limit);
8413   jcc(Assembler::zero, TRUE_LABEL);
8414 
8415   if (is_array_equ) {
8416     // Load array address
8417     lea(ary1, Address(ary1, base_offset));
8418     lea(ary2, Address(ary2, base_offset));
8419   }
8420 
8421   if (is_array_equ && is_char) {
8422     // arrays_equals when used for char[].
8423     shll(limit, 1);      // byte count != 0
8424   }
8425   movl(result, limit); // copy
8426 
8427   if (UseAVX >= 2) {
8428     // With AVX2, use 32-byte vector compare
8429     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
8430 
8431     // Compare 32-byte vectors
8432     andl(result, 0x0000001f);  //   tail count (in bytes)
8433     andl(limit, 0xffffffe0);   // vector count (in bytes)
8434     jcc(Assembler::zero, COMPARE_TAIL);
8435 
8436     lea(ary1, Address(ary1, limit, Address::times_1));
8437     lea(ary2, Address(ary2, limit, Address::times_1));
8438     negptr(limit);
8439 
8440     bind(COMPARE_WIDE_VECTORS);
8441 
8442 #ifdef _LP64
8443     if (VM_Version::supports_avx512vlbw()) { // trying 64 bytes fast loop
8444       Label COMPARE_WIDE_VECTORS_LOOP_AVX2, COMPARE_WIDE_VECTORS_LOOP_AVX3;
8445 
8446       cmpl(limit, -64);
8447       jccb(Assembler::greater, COMPARE_WIDE_VECTORS_LOOP_AVX2);
8448 
8449       bind(COMPARE_WIDE_VECTORS_LOOP_AVX3); // the hottest loop
8450 
8451       evmovdquq(vec1, Address(ary1, limit, Address::times_1), Assembler::AVX_512bit);
8452       evpcmpeqb(k7, vec1, Address(ary2, limit, Address::times_1), Assembler::AVX_512bit);
8453       kortestql(k7, k7);
8454       jcc(Assembler::aboveEqual, FALSE_LABEL);     // miscompare
8455       addptr(limit, 64);  // update since we already compared at this addr
8456       cmpl(limit, -64);
8457       jccb(Assembler::lessEqual, COMPARE_WIDE_VECTORS_LOOP_AVX3);
8458 
8459       // At this point we may still need to compare -limit+result bytes.
8460       // We could execute the next two instruction and just continue via non-wide path:
8461       //  cmpl(limit, 0);
8462       //  jcc(Assembler::equal, COMPARE_TAIL);  // true
8463       // But since we stopped at the points ary{1,2}+limit which are
8464       // not farther than 64 bytes from the ends of arrays ary{1,2}+result
8465       // (|limit| <= 32 and result < 32),
8466       // we may just compare the last 64 bytes.
8467       //
8468       addptr(result, -64);   // it is safe, bc we just came from this area
8469       evmovdquq(vec1, Address(ary1, result, Address::times_1), Assembler::AVX_512bit);
8470       evpcmpeqb(k7, vec1, Address(ary2, result, Address::times_1), Assembler::AVX_512bit);
8471       kortestql(k7, k7);
8472       jcc(Assembler::aboveEqual, FALSE_LABEL);     // miscompare
8473 
8474       jmp(TRUE_LABEL);
8475 
8476       bind(COMPARE_WIDE_VECTORS_LOOP_AVX2);
8477 
8478     }//if (VM_Version::supports_avx512vlbw())
8479 #endif //_LP64
8480 
8481     vmovdqu(vec1, Address(ary1, limit, Address::times_1));
8482     vmovdqu(vec2, Address(ary2, limit, Address::times_1));
8483     vpxor(vec1, vec2);
8484 
8485     vptest(vec1, vec1);
8486     jccb(Assembler::notZero, FALSE_LABEL);
8487     addptr(limit, 32);
8488     jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
8489 
8490     testl(result, result);
8491     jccb(Assembler::zero, TRUE_LABEL);
8492 
8493     vmovdqu(vec1, Address(ary1, result, Address::times_1, -32));
8494     vmovdqu(vec2, Address(ary2, result, Address::times_1, -32));
8495     vpxor(vec1, vec2);
8496 
8497     vptest(vec1, vec1);
8498     jccb(Assembler::notZero, FALSE_LABEL);
8499     jmpb(TRUE_LABEL);
8500 
8501     bind(COMPARE_TAIL); // limit is zero
8502     movl(limit, result);
8503     // Fallthru to tail compare
8504   } else if (UseSSE42Intrinsics) {
8505     assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
8506     // With SSE4.2, use double quad vector compare
8507     Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
8508 
8509     // Compare 16-byte vectors
8510     andl(result, 0x0000000f);  //   tail count (in bytes)
8511     andl(limit, 0xfffffff0);   // vector count (in bytes)
8512     jccb(Assembler::zero, COMPARE_TAIL);
8513 
8514     lea(ary1, Address(ary1, limit, Address::times_1));
8515     lea(ary2, Address(ary2, limit, Address::times_1));
8516     negptr(limit);
8517 
8518     bind(COMPARE_WIDE_VECTORS);
8519     movdqu(vec1, Address(ary1, limit, Address::times_1));
8520     movdqu(vec2, Address(ary2, limit, Address::times_1));
8521     pxor(vec1, vec2);
8522 
8523     ptest(vec1, vec1);
8524     jccb(Assembler::notZero, FALSE_LABEL);
8525     addptr(limit, 16);
8526     jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
8527 
8528     testl(result, result);
8529     jccb(Assembler::zero, TRUE_LABEL);
8530 
8531     movdqu(vec1, Address(ary1, result, Address::times_1, -16));
8532     movdqu(vec2, Address(ary2, result, Address::times_1, -16));
8533     pxor(vec1, vec2);
8534 
8535     ptest(vec1, vec1);
8536     jccb(Assembler::notZero, FALSE_LABEL);
8537     jmpb(TRUE_LABEL);
8538 
8539     bind(COMPARE_TAIL); // limit is zero
8540     movl(limit, result);
8541     // Fallthru to tail compare
8542   }
8543 
8544   // Compare 4-byte vectors
8545   andl(limit, 0xfffffffc); // vector count (in bytes)
8546   jccb(Assembler::zero, COMPARE_CHAR);
8547 
8548   lea(ary1, Address(ary1, limit, Address::times_1));
8549   lea(ary2, Address(ary2, limit, Address::times_1));
8550   negptr(limit);
8551 
8552   bind(COMPARE_VECTORS);
8553   movl(chr, Address(ary1, limit, Address::times_1));
8554   cmpl(chr, Address(ary2, limit, Address::times_1));
8555   jccb(Assembler::notEqual, FALSE_LABEL);
8556   addptr(limit, 4);
8557   jcc(Assembler::notZero, COMPARE_VECTORS);
8558 
8559   // Compare trailing char (final 2 bytes), if any
8560   bind(COMPARE_CHAR);
8561   testl(result, 0x2);   // tail  char
8562   jccb(Assembler::zero, COMPARE_BYTE);
8563   load_unsigned_short(chr, Address(ary1, 0));
8564   load_unsigned_short(limit, Address(ary2, 0));
8565   cmpl(chr, limit);
8566   jccb(Assembler::notEqual, FALSE_LABEL);
8567 
8568   if (is_array_equ && is_char) {
8569     bind(COMPARE_BYTE);
8570   } else {
8571     lea(ary1, Address(ary1, 2));
8572     lea(ary2, Address(ary2, 2));
8573 
8574     bind(COMPARE_BYTE);
8575     testl(result, 0x1);   // tail  byte
8576     jccb(Assembler::zero, TRUE_LABEL);
8577     load_unsigned_byte(chr, Address(ary1, 0));
8578     load_unsigned_byte(limit, Address(ary2, 0));
8579     cmpl(chr, limit);
8580     jccb(Assembler::notEqual, FALSE_LABEL);
8581   }
8582   bind(TRUE_LABEL);
8583   movl(result, 1);   // return true
8584   jmpb(DONE);
8585 
8586   bind(FALSE_LABEL);
8587   xorl(result, result); // return false
8588 
8589   // That's it
8590   bind(DONE);
8591   if (UseAVX >= 2) {
8592     // clean upper bits of YMM registers
8593     vpxor(vec1, vec1);
8594     vpxor(vec2, vec2);
8595   }
8596 }
8597 
8598 #endif
8599 
8600 void MacroAssembler::generate_fill(BasicType t, bool aligned,
8601                                    Register to, Register value, Register count,
8602                                    Register rtmp, XMMRegister xtmp) {
8603   ShortBranchVerifier sbv(this);
8604   assert_different_registers(to, value, count, rtmp);
8605   Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
8606   Label L_fill_2_bytes, L_fill_4_bytes;
8607 
8608   int shift = -1;
8609   switch (t) {
8610     case T_BYTE:
8611       shift = 2;
8612       break;
8613     case T_SHORT:
8614       shift = 1;
8615       break;
8616     case T_INT:
8617       shift = 0;
8618       break;
8619     default: ShouldNotReachHere();
8620   }
8621 
8622   if (t == T_BYTE) {
8623     andl(value, 0xff);
8624     movl(rtmp, value);
8625     shll(rtmp, 8);
8626     orl(value, rtmp);
8627   }
8628   if (t == T_SHORT) {
8629     andl(value, 0xffff);
8630   }
8631   if (t == T_BYTE || t == T_SHORT) {
8632     movl(rtmp, value);
8633     shll(rtmp, 16);
8634     orl(value, rtmp);
8635   }
8636 
8637   cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
8638   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
8639   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
8640     // align source address at 4 bytes address boundary
8641     if (t == T_BYTE) {
8642       // One byte misalignment happens only for byte arrays
8643       testptr(to, 1);
8644       jccb(Assembler::zero, L_skip_align1);
8645       movb(Address(to, 0), value);
8646       increment(to);
8647       decrement(count);
8648       BIND(L_skip_align1);
8649     }
8650     // Two bytes misalignment happens only for byte and short (char) arrays
8651     testptr(to, 2);
8652     jccb(Assembler::zero, L_skip_align2);
8653     movw(Address(to, 0), value);
8654     addptr(to, 2);
8655     subl(count, 1<<(shift-1));
8656     BIND(L_skip_align2);
8657   }
8658   if (UseSSE < 2) {
8659     Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
8660     // Fill 32-byte chunks
8661     subl(count, 8 << shift);
8662     jcc(Assembler::less, L_check_fill_8_bytes);
8663     align(16);
8664 
8665     BIND(L_fill_32_bytes_loop);
8666 
8667     for (int i = 0; i < 32; i += 4) {
8668       movl(Address(to, i), value);
8669     }
8670 
8671     addptr(to, 32);
8672     subl(count, 8 << shift);
8673     jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
8674     BIND(L_check_fill_8_bytes);
8675     addl(count, 8 << shift);
8676     jccb(Assembler::zero, L_exit);
8677     jmpb(L_fill_8_bytes);
8678 
8679     //
8680     // length is too short, just fill qwords
8681     //
8682     BIND(L_fill_8_bytes_loop);
8683     movl(Address(to, 0), value);
8684     movl(Address(to, 4), value);
8685     addptr(to, 8);
8686     BIND(L_fill_8_bytes);
8687     subl(count, 1 << (shift + 1));
8688     jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
8689     // fall through to fill 4 bytes
8690   } else {
8691     Label L_fill_32_bytes;
8692     if (!UseUnalignedLoadStores) {
8693       // align to 8 bytes, we know we are 4 byte aligned to start
8694       testptr(to, 4);
8695       jccb(Assembler::zero, L_fill_32_bytes);
8696       movl(Address(to, 0), value);
8697       addptr(to, 4);
8698       subl(count, 1<<shift);
8699     }
8700     BIND(L_fill_32_bytes);
8701     {
8702       assert( UseSSE >= 2, "supported cpu only" );
8703       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
8704       if (UseAVX > 2) {
8705         movl(rtmp, 0xffff);
8706         kmovwl(k1, rtmp);
8707       }
8708       movdl(xtmp, value);
8709       if (UseAVX > 2 && UseUnalignedLoadStores) {
8710         // Fill 64-byte chunks
8711         Label L_fill_64_bytes_loop, L_check_fill_32_bytes;
8712         evpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
8713 
8714         subl(count, 16 << shift);
8715         jcc(Assembler::less, L_check_fill_32_bytes);
8716         align(16);
8717 
8718         BIND(L_fill_64_bytes_loop);
8719         evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
8720         addptr(to, 64);
8721         subl(count, 16 << shift);
8722         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
8723 
8724         BIND(L_check_fill_32_bytes);
8725         addl(count, 8 << shift);
8726         jccb(Assembler::less, L_check_fill_8_bytes);
8727         vmovdqu(Address(to, 0), xtmp);
8728         addptr(to, 32);
8729         subl(count, 8 << shift);
8730 
8731         BIND(L_check_fill_8_bytes);
8732       } else if (UseAVX == 2 && UseUnalignedLoadStores) {
8733         // Fill 64-byte chunks
8734         Label L_fill_64_bytes_loop, L_check_fill_32_bytes;
8735         vpbroadcastd(xtmp, xtmp);
8736 
8737         subl(count, 16 << shift);
8738         jcc(Assembler::less, L_check_fill_32_bytes);
8739         align(16);
8740 
8741         BIND(L_fill_64_bytes_loop);
8742         vmovdqu(Address(to, 0), xtmp);
8743         vmovdqu(Address(to, 32), xtmp);
8744         addptr(to, 64);
8745         subl(count, 16 << shift);
8746         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
8747 
8748         BIND(L_check_fill_32_bytes);
8749         addl(count, 8 << shift);
8750         jccb(Assembler::less, L_check_fill_8_bytes);
8751         vmovdqu(Address(to, 0), xtmp);
8752         addptr(to, 32);
8753         subl(count, 8 << shift);
8754 
8755         BIND(L_check_fill_8_bytes);
8756         // clean upper bits of YMM registers
8757         movdl(xtmp, value);
8758         pshufd(xtmp, xtmp, 0);
8759       } else {
8760         // Fill 32-byte chunks
8761         pshufd(xtmp, xtmp, 0);
8762 
8763         subl(count, 8 << shift);
8764         jcc(Assembler::less, L_check_fill_8_bytes);
8765         align(16);
8766 
8767         BIND(L_fill_32_bytes_loop);
8768 
8769         if (UseUnalignedLoadStores) {
8770           movdqu(Address(to, 0), xtmp);
8771           movdqu(Address(to, 16), xtmp);
8772         } else {
8773           movq(Address(to, 0), xtmp);
8774           movq(Address(to, 8), xtmp);
8775           movq(Address(to, 16), xtmp);
8776           movq(Address(to, 24), xtmp);
8777         }
8778 
8779         addptr(to, 32);
8780         subl(count, 8 << shift);
8781         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
8782 
8783         BIND(L_check_fill_8_bytes);
8784       }
8785       addl(count, 8 << shift);
8786       jccb(Assembler::zero, L_exit);
8787       jmpb(L_fill_8_bytes);
8788 
8789       //
8790       // length is too short, just fill qwords
8791       //
8792       BIND(L_fill_8_bytes_loop);
8793       movq(Address(to, 0), xtmp);
8794       addptr(to, 8);
8795       BIND(L_fill_8_bytes);
8796       subl(count, 1 << (shift + 1));
8797       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
8798     }
8799   }
8800   // fill trailing 4 bytes
8801   BIND(L_fill_4_bytes);
8802   testl(count, 1<<shift);
8803   jccb(Assembler::zero, L_fill_2_bytes);
8804   movl(Address(to, 0), value);
8805   if (t == T_BYTE || t == T_SHORT) {
8806     addptr(to, 4);
8807     BIND(L_fill_2_bytes);
8808     // fill trailing 2 bytes
8809     testl(count, 1<<(shift-1));
8810     jccb(Assembler::zero, L_fill_byte);
8811     movw(Address(to, 0), value);
8812     if (t == T_BYTE) {
8813       addptr(to, 2);
8814       BIND(L_fill_byte);
8815       // fill trailing byte
8816       testl(count, 1);
8817       jccb(Assembler::zero, L_exit);
8818       movb(Address(to, 0), value);
8819     } else {
8820       BIND(L_fill_byte);
8821     }
8822   } else {
8823     BIND(L_fill_2_bytes);
8824   }
8825   BIND(L_exit);
8826 }
8827 
8828 // encode char[] to byte[] in ISO_8859_1
8829 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
8830                                       XMMRegister tmp1Reg, XMMRegister tmp2Reg,
8831                                       XMMRegister tmp3Reg, XMMRegister tmp4Reg,
8832                                       Register tmp5, Register result) {
8833   // rsi: src
8834   // rdi: dst
8835   // rdx: len
8836   // rcx: tmp5
8837   // rax: result
8838   ShortBranchVerifier sbv(this);
8839   assert_different_registers(src, dst, len, tmp5, result);
8840   Label L_done, L_copy_1_char, L_copy_1_char_exit;
8841 
8842   // set result
8843   xorl(result, result);
8844   // check for zero length
8845   testl(len, len);
8846   jcc(Assembler::zero, L_done);
8847   movl(result, len);
8848 
8849   // Setup pointers
8850   lea(src, Address(src, len, Address::times_2)); // char[]
8851   lea(dst, Address(dst, len, Address::times_1)); // byte[]
8852   negptr(len);
8853 
8854   if (UseSSE42Intrinsics || UseAVX >= 2) {
8855     assert(UseSSE42Intrinsics ? UseSSE >= 4 : true, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
8856     Label L_chars_8_check, L_copy_8_chars, L_copy_8_chars_exit;
8857     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
8858 
8859     if (UseAVX >= 2) {
8860       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
8861       movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vector
8862       movdl(tmp1Reg, tmp5);
8863       vpbroadcastd(tmp1Reg, tmp1Reg);
8864       jmpb(L_chars_32_check);
8865 
8866       bind(L_copy_32_chars);
8867       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
8868       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
8869       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
8870       vptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in  vector
8871       jccb(Assembler::notZero, L_copy_32_chars_exit);
8872       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
8873       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
8874       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
8875 
8876       bind(L_chars_32_check);
8877       addptr(len, 32);
8878       jccb(Assembler::lessEqual, L_copy_32_chars);
8879 
8880       bind(L_copy_32_chars_exit);
8881       subptr(len, 16);
8882       jccb(Assembler::greater, L_copy_16_chars_exit);
8883 
8884     } else if (UseSSE42Intrinsics) {
8885       movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vector
8886       movdl(tmp1Reg, tmp5);
8887       pshufd(tmp1Reg, tmp1Reg, 0);
8888       jmpb(L_chars_16_check);
8889     }
8890 
8891     bind(L_copy_16_chars);
8892     if (UseAVX >= 2) {
8893       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
8894       vptest(tmp2Reg, tmp1Reg);
8895       jccb(Assembler::notZero, L_copy_16_chars_exit);
8896       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
8897       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
8898     } else {
8899       if (UseAVX > 0) {
8900         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
8901         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
8902         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
8903       } else {
8904         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
8905         por(tmp2Reg, tmp3Reg);
8906         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
8907         por(tmp2Reg, tmp4Reg);
8908       }
8909       ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in  vector
8910       jccb(Assembler::notZero, L_copy_16_chars_exit);
8911       packuswb(tmp3Reg, tmp4Reg);
8912     }
8913     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
8914 
8915     bind(L_chars_16_check);
8916     addptr(len, 16);
8917     jccb(Assembler::lessEqual, L_copy_16_chars);
8918 
8919     bind(L_copy_16_chars_exit);
8920     if (UseAVX >= 2) {
8921       // clean upper bits of YMM registers
8922       vpxor(tmp2Reg, tmp2Reg);
8923       vpxor(tmp3Reg, tmp3Reg);
8924       vpxor(tmp4Reg, tmp4Reg);
8925       movdl(tmp1Reg, tmp5);
8926       pshufd(tmp1Reg, tmp1Reg, 0);
8927     }
8928     subptr(len, 8);
8929     jccb(Assembler::greater, L_copy_8_chars_exit);
8930 
8931     bind(L_copy_8_chars);
8932     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
8933     ptest(tmp3Reg, tmp1Reg);
8934     jccb(Assembler::notZero, L_copy_8_chars_exit);
8935     packuswb(tmp3Reg, tmp1Reg);
8936     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
8937     addptr(len, 8);
8938     jccb(Assembler::lessEqual, L_copy_8_chars);
8939 
8940     bind(L_copy_8_chars_exit);
8941     subptr(len, 8);
8942     jccb(Assembler::zero, L_done);
8943   }
8944 
8945   bind(L_copy_1_char);
8946   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
8947   testl(tmp5, 0xff00);      // check if Unicode char
8948   jccb(Assembler::notZero, L_copy_1_char_exit);
8949   movb(Address(dst, len, Address::times_1, 0), tmp5);
8950   addptr(len, 1);
8951   jccb(Assembler::less, L_copy_1_char);
8952 
8953   bind(L_copy_1_char_exit);
8954   addptr(result, len); // len is negative count of not processed elements
8955   bind(L_done);
8956 }
8957 
8958 #ifdef _LP64
8959 /**
8960  * Helper for multiply_to_len().
8961  */
8962 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
8963   addq(dest_lo, src1);
8964   adcq(dest_hi, 0);
8965   addq(dest_lo, src2);
8966   adcq(dest_hi, 0);
8967 }
8968 
8969 /**
8970  * Multiply 64 bit by 64 bit first loop.
8971  */
8972 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
8973                                            Register y, Register y_idx, Register z,
8974                                            Register carry, Register product,
8975                                            Register idx, Register kdx) {
8976   //
8977   //  jlong carry, x[], y[], z[];
8978   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
8979   //    huge_128 product = y[idx] * x[xstart] + carry;
8980   //    z[kdx] = (jlong)product;
8981   //    carry  = (jlong)(product >>> 64);
8982   //  }
8983   //  z[xstart] = carry;
8984   //
8985 
8986   Label L_first_loop, L_first_loop_exit;
8987   Label L_one_x, L_one_y, L_multiply;
8988 
8989   decrementl(xstart);
8990   jcc(Assembler::negative, L_one_x);
8991 
8992   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
8993   rorq(x_xstart, 32); // convert big-endian to little-endian
8994 
8995   bind(L_first_loop);
8996   decrementl(idx);
8997   jcc(Assembler::negative, L_first_loop_exit);
8998   decrementl(idx);
8999   jcc(Assembler::negative, L_one_y);
9000   movq(y_idx, Address(y, idx, Address::times_4,  0));
9001   rorq(y_idx, 32); // convert big-endian to little-endian
9002   bind(L_multiply);
9003   movq(product, x_xstart);
9004   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
9005   addq(product, carry);
9006   adcq(rdx, 0);
9007   subl(kdx, 2);
9008   movl(Address(z, kdx, Address::times_4,  4), product);
9009   shrq(product, 32);
9010   movl(Address(z, kdx, Address::times_4,  0), product);
9011   movq(carry, rdx);
9012   jmp(L_first_loop);
9013 
9014   bind(L_one_y);
9015   movl(y_idx, Address(y,  0));
9016   jmp(L_multiply);
9017 
9018   bind(L_one_x);
9019   movl(x_xstart, Address(x,  0));
9020   jmp(L_first_loop);
9021 
9022   bind(L_first_loop_exit);
9023 }
9024 
9025 /**
9026  * Multiply 64 bit by 64 bit and add 128 bit.
9027  */
9028 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
9029                                             Register yz_idx, Register idx,
9030                                             Register carry, Register product, int offset) {
9031   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
9032   //     z[kdx] = (jlong)product;
9033 
9034   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
9035   rorq(yz_idx, 32); // convert big-endian to little-endian
9036   movq(product, x_xstart);
9037   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
9038   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
9039   rorq(yz_idx, 32); // convert big-endian to little-endian
9040 
9041   add2_with_carry(rdx, product, carry, yz_idx);
9042 
9043   movl(Address(z, idx, Address::times_4,  offset+4), product);
9044   shrq(product, 32);
9045   movl(Address(z, idx, Address::times_4,  offset), product);
9046 
9047 }
9048 
9049 /**
9050  * Multiply 128 bit by 128 bit. Unrolled inner loop.
9051  */
9052 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
9053                                              Register yz_idx, Register idx, Register jdx,
9054                                              Register carry, Register product,
9055                                              Register carry2) {
9056   //   jlong carry, x[], y[], z[];
9057   //   int kdx = ystart+1;
9058   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
9059   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
9060   //     z[kdx+idx+1] = (jlong)product;
9061   //     jlong carry2  = (jlong)(product >>> 64);
9062   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
9063   //     z[kdx+idx] = (jlong)product;
9064   //     carry  = (jlong)(product >>> 64);
9065   //   }
9066   //   idx += 2;
9067   //   if (idx > 0) {
9068   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
9069   //     z[kdx+idx] = (jlong)product;
9070   //     carry  = (jlong)(product >>> 64);
9071   //   }
9072   //
9073 
9074   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
9075 
9076   movl(jdx, idx);
9077   andl(jdx, 0xFFFFFFFC);
9078   shrl(jdx, 2);
9079 
9080   bind(L_third_loop);
9081   subl(jdx, 1);
9082   jcc(Assembler::negative, L_third_loop_exit);
9083   subl(idx, 4);
9084 
9085   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
9086   movq(carry2, rdx);
9087 
9088   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
9089   movq(carry, rdx);
9090   jmp(L_third_loop);
9091 
9092   bind (L_third_loop_exit);
9093 
9094   andl (idx, 0x3);
9095   jcc(Assembler::zero, L_post_third_loop_done);
9096 
9097   Label L_check_1;
9098   subl(idx, 2);
9099   jcc(Assembler::negative, L_check_1);
9100 
9101   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
9102   movq(carry, rdx);
9103 
9104   bind (L_check_1);
9105   addl (idx, 0x2);
9106   andl (idx, 0x1);
9107   subl(idx, 1);
9108   jcc(Assembler::negative, L_post_third_loop_done);
9109 
9110   movl(yz_idx, Address(y, idx, Address::times_4,  0));
9111   movq(product, x_xstart);
9112   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
9113   movl(yz_idx, Address(z, idx, Address::times_4,  0));
9114 
9115   add2_with_carry(rdx, product, yz_idx, carry);
9116 
9117   movl(Address(z, idx, Address::times_4,  0), product);
9118   shrq(product, 32);
9119 
9120   shlq(rdx, 32);
9121   orq(product, rdx);
9122   movq(carry, product);
9123 
9124   bind(L_post_third_loop_done);
9125 }
9126 
9127 /**
9128  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
9129  *
9130  */
9131 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
9132                                                   Register carry, Register carry2,
9133                                                   Register idx, Register jdx,
9134                                                   Register yz_idx1, Register yz_idx2,
9135                                                   Register tmp, Register tmp3, Register tmp4) {
9136   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
9137 
9138   //   jlong carry, x[], y[], z[];
9139   //   int kdx = ystart+1;
9140   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
9141   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
9142   //     jlong carry2  = (jlong)(tmp3 >>> 64);
9143   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
9144   //     carry  = (jlong)(tmp4 >>> 64);
9145   //     z[kdx+idx+1] = (jlong)tmp3;
9146   //     z[kdx+idx] = (jlong)tmp4;
9147   //   }
9148   //   idx += 2;
9149   //   if (idx > 0) {
9150   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
9151   //     z[kdx+idx] = (jlong)yz_idx1;
9152   //     carry  = (jlong)(yz_idx1 >>> 64);
9153   //   }
9154   //
9155 
9156   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
9157 
9158   movl(jdx, idx);
9159   andl(jdx, 0xFFFFFFFC);
9160   shrl(jdx, 2);
9161 
9162   bind(L_third_loop);
9163   subl(jdx, 1);
9164   jcc(Assembler::negative, L_third_loop_exit);
9165   subl(idx, 4);
9166 
9167   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
9168   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
9169   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
9170   rorxq(yz_idx2, yz_idx2, 32);
9171 
9172   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
9173   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
9174 
9175   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
9176   rorxq(yz_idx1, yz_idx1, 32);
9177   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
9178   rorxq(yz_idx2, yz_idx2, 32);
9179 
9180   if (VM_Version::supports_adx()) {
9181     adcxq(tmp3, carry);
9182     adoxq(tmp3, yz_idx1);
9183 
9184     adcxq(tmp4, tmp);
9185     adoxq(tmp4, yz_idx2);
9186 
9187     movl(carry, 0); // does not affect flags
9188     adcxq(carry2, carry);
9189     adoxq(carry2, carry);
9190   } else {
9191     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
9192     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
9193   }
9194   movq(carry, carry2);
9195 
9196   movl(Address(z, idx, Address::times_4, 12), tmp3);
9197   shrq(tmp3, 32);
9198   movl(Address(z, idx, Address::times_4,  8), tmp3);
9199 
9200   movl(Address(z, idx, Address::times_4,  4), tmp4);
9201   shrq(tmp4, 32);
9202   movl(Address(z, idx, Address::times_4,  0), tmp4);
9203 
9204   jmp(L_third_loop);
9205 
9206   bind (L_third_loop_exit);
9207 
9208   andl (idx, 0x3);
9209   jcc(Assembler::zero, L_post_third_loop_done);
9210 
9211   Label L_check_1;
9212   subl(idx, 2);
9213   jcc(Assembler::negative, L_check_1);
9214 
9215   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
9216   rorxq(yz_idx1, yz_idx1, 32);
9217   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
9218   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
9219   rorxq(yz_idx2, yz_idx2, 32);
9220 
9221   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
9222 
9223   movl(Address(z, idx, Address::times_4,  4), tmp3);
9224   shrq(tmp3, 32);
9225   movl(Address(z, idx, Address::times_4,  0), tmp3);
9226   movq(carry, tmp4);
9227 
9228   bind (L_check_1);
9229   addl (idx, 0x2);
9230   andl (idx, 0x1);
9231   subl(idx, 1);
9232   jcc(Assembler::negative, L_post_third_loop_done);
9233   movl(tmp4, Address(y, idx, Address::times_4,  0));
9234   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
9235   movl(tmp4, Address(z, idx, Address::times_4,  0));
9236 
9237   add2_with_carry(carry2, tmp3, tmp4, carry);
9238 
9239   movl(Address(z, idx, Address::times_4,  0), tmp3);
9240   shrq(tmp3, 32);
9241 
9242   shlq(carry2, 32);
9243   orq(tmp3, carry2);
9244   movq(carry, tmp3);
9245 
9246   bind(L_post_third_loop_done);
9247 }
9248 
9249 /**
9250  * Code for BigInteger::multiplyToLen() instrinsic.
9251  *
9252  * rdi: x
9253  * rax: xlen
9254  * rsi: y
9255  * rcx: ylen
9256  * r8:  z
9257  * r11: zlen
9258  * r12: tmp1
9259  * r13: tmp2
9260  * r14: tmp3
9261  * r15: tmp4
9262  * rbx: tmp5
9263  *
9264  */
9265 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
9266                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
9267   ShortBranchVerifier sbv(this);
9268   assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
9269 
9270   push(tmp1);
9271   push(tmp2);
9272   push(tmp3);
9273   push(tmp4);
9274   push(tmp5);
9275 
9276   push(xlen);
9277   push(zlen);
9278 
9279   const Register idx = tmp1;
9280   const Register kdx = tmp2;
9281   const Register xstart = tmp3;
9282 
9283   const Register y_idx = tmp4;
9284   const Register carry = tmp5;
9285   const Register product  = xlen;
9286   const Register x_xstart = zlen;  // reuse register
9287 
9288   // First Loop.
9289   //
9290   //  final static long LONG_MASK = 0xffffffffL;
9291   //  int xstart = xlen - 1;
9292   //  int ystart = ylen - 1;
9293   //  long carry = 0;
9294   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
9295   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
9296   //    z[kdx] = (int)product;
9297   //    carry = product >>> 32;
9298   //  }
9299   //  z[xstart] = (int)carry;
9300   //
9301 
9302   movl(idx, ylen);      // idx = ylen;
9303   movl(kdx, zlen);      // kdx = xlen+ylen;
9304   xorq(carry, carry);   // carry = 0;
9305 
9306   Label L_done;
9307 
9308   movl(xstart, xlen);
9309   decrementl(xstart);
9310   jcc(Assembler::negative, L_done);
9311 
9312   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
9313 
9314   Label L_second_loop;
9315   testl(kdx, kdx);
9316   jcc(Assembler::zero, L_second_loop);
9317 
9318   Label L_carry;
9319   subl(kdx, 1);
9320   jcc(Assembler::zero, L_carry);
9321 
9322   movl(Address(z, kdx, Address::times_4,  0), carry);
9323   shrq(carry, 32);
9324   subl(kdx, 1);
9325 
9326   bind(L_carry);
9327   movl(Address(z, kdx, Address::times_4,  0), carry);
9328 
9329   // Second and third (nested) loops.
9330   //
9331   // for (int i = xstart-1; i >= 0; i--) { // Second loop
9332   //   carry = 0;
9333   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
9334   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
9335   //                    (z[k] & LONG_MASK) + carry;
9336   //     z[k] = (int)product;
9337   //     carry = product >>> 32;
9338   //   }
9339   //   z[i] = (int)carry;
9340   // }
9341   //
9342   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
9343 
9344   const Register jdx = tmp1;
9345 
9346   bind(L_second_loop);
9347   xorl(carry, carry);    // carry = 0;
9348   movl(jdx, ylen);       // j = ystart+1
9349 
9350   subl(xstart, 1);       // i = xstart-1;
9351   jcc(Assembler::negative, L_done);
9352 
9353   push (z);
9354 
9355   Label L_last_x;
9356   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
9357   subl(xstart, 1);       // i = xstart-1;
9358   jcc(Assembler::negative, L_last_x);
9359 
9360   if (UseBMI2Instructions) {
9361     movq(rdx,  Address(x, xstart, Address::times_4,  0));
9362     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
9363   } else {
9364     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
9365     rorq(x_xstart, 32);  // convert big-endian to little-endian
9366   }
9367 
9368   Label L_third_loop_prologue;
9369   bind(L_third_loop_prologue);
9370 
9371   push (x);
9372   push (xstart);
9373   push (ylen);
9374 
9375 
9376   if (UseBMI2Instructions) {
9377     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
9378   } else { // !UseBMI2Instructions
9379     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
9380   }
9381 
9382   pop(ylen);
9383   pop(xlen);
9384   pop(x);
9385   pop(z);
9386 
9387   movl(tmp3, xlen);
9388   addl(tmp3, 1);
9389   movl(Address(z, tmp3, Address::times_4,  0), carry);
9390   subl(tmp3, 1);
9391   jccb(Assembler::negative, L_done);
9392 
9393   shrq(carry, 32);
9394   movl(Address(z, tmp3, Address::times_4,  0), carry);
9395   jmp(L_second_loop);
9396 
9397   // Next infrequent code is moved outside loops.
9398   bind(L_last_x);
9399   if (UseBMI2Instructions) {
9400     movl(rdx, Address(x,  0));
9401   } else {
9402     movl(x_xstart, Address(x,  0));
9403   }
9404   jmp(L_third_loop_prologue);
9405 
9406   bind(L_done);
9407 
9408   pop(zlen);
9409   pop(xlen);
9410 
9411   pop(tmp5);
9412   pop(tmp4);
9413   pop(tmp3);
9414   pop(tmp2);
9415   pop(tmp1);
9416 }
9417 
9418 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
9419   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
9420   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
9421   Label VECTOR32_LOOP, VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
9422   Label VECTOR16_TAIL, VECTOR8_TAIL, VECTOR4_TAIL;
9423   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
9424   Label SAME_TILL_END, DONE;
9425   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
9426 
9427   //scale is in rcx in both Win64 and Unix
9428   ShortBranchVerifier sbv(this);
9429 
9430   shlq(length);
9431   xorq(result, result);
9432 
9433   cmpq(length, 8);
9434   jcc(Assembler::equal, VECTOR8_LOOP);
9435   jcc(Assembler::less, VECTOR4_TAIL);
9436 
9437   if (UseAVX >= 2){
9438 
9439     cmpq(length, 16);
9440     jcc(Assembler::equal, VECTOR16_LOOP);
9441     jcc(Assembler::less, VECTOR8_LOOP);
9442 
9443     cmpq(length, 32);
9444     jccb(Assembler::less, VECTOR16_TAIL);
9445 
9446     subq(length, 32);
9447     bind(VECTOR32_LOOP);
9448     vmovdqu(rymm0, Address(obja, result));
9449     vmovdqu(rymm1, Address(objb, result));
9450     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
9451     vptest(rymm2, rymm2);
9452     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
9453     addq(result, 32);
9454     subq(length, 32);
9455     jccb(Assembler::greaterEqual, VECTOR32_LOOP);
9456     addq(length, 32);
9457     jcc(Assembler::equal, SAME_TILL_END);
9458     //falling through if less than 32 bytes left //close the branch here.
9459 
9460     bind(VECTOR16_TAIL);
9461     cmpq(length, 16);
9462     jccb(Assembler::less, VECTOR8_TAIL);
9463     bind(VECTOR16_LOOP);
9464     movdqu(rymm0, Address(obja, result));
9465     movdqu(rymm1, Address(objb, result));
9466     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
9467     ptest(rymm2, rymm2);
9468     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
9469     addq(result, 16);
9470     subq(length, 16);
9471     jcc(Assembler::equal, SAME_TILL_END);
9472     //falling through if less than 16 bytes left
9473   } else {//regular intrinsics
9474 
9475     cmpq(length, 16);
9476     jccb(Assembler::less, VECTOR8_TAIL);
9477 
9478     subq(length, 16);
9479     bind(VECTOR16_LOOP);
9480     movdqu(rymm0, Address(obja, result));
9481     movdqu(rymm1, Address(objb, result));
9482     pxor(rymm0, rymm1);
9483     ptest(rymm0, rymm0);
9484     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
9485     addq(result, 16);
9486     subq(length, 16);
9487     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
9488     addq(length, 16);
9489     jcc(Assembler::equal, SAME_TILL_END);
9490     //falling through if less than 16 bytes left
9491   }
9492 
9493   bind(VECTOR8_TAIL);
9494   cmpq(length, 8);
9495   jccb(Assembler::less, VECTOR4_TAIL);
9496   bind(VECTOR8_LOOP);
9497   movq(tmp1, Address(obja, result));
9498   movq(tmp2, Address(objb, result));
9499   xorq(tmp1, tmp2);
9500   testq(tmp1, tmp1);
9501   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
9502   addq(result, 8);
9503   subq(length, 8);
9504   jcc(Assembler::equal, SAME_TILL_END);
9505   //falling through if less than 8 bytes left
9506 
9507   bind(VECTOR4_TAIL);
9508   cmpq(length, 4);
9509   jccb(Assembler::less, BYTES_TAIL);
9510   bind(VECTOR4_LOOP);
9511   movl(tmp1, Address(obja, result));
9512   xorl(tmp1, Address(objb, result));
9513   testl(tmp1, tmp1);
9514   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
9515   addq(result, 4);
9516   subq(length, 4);
9517   jcc(Assembler::equal, SAME_TILL_END);
9518   //falling through if less than 4 bytes left
9519 
9520   bind(BYTES_TAIL);
9521   bind(BYTES_LOOP);
9522   load_unsigned_byte(tmp1, Address(obja, result));
9523   load_unsigned_byte(tmp2, Address(objb, result));
9524   xorl(tmp1, tmp2);
9525   testl(tmp1, tmp1);
9526   jccb(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
9527   decq(length);
9528   jccb(Assembler::zero, SAME_TILL_END);
9529   incq(result);
9530   load_unsigned_byte(tmp1, Address(obja, result));
9531   load_unsigned_byte(tmp2, Address(objb, result));
9532   xorl(tmp1, tmp2);
9533   testl(tmp1, tmp1);
9534   jccb(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
9535   decq(length);
9536   jccb(Assembler::zero, SAME_TILL_END);
9537   incq(result);
9538   load_unsigned_byte(tmp1, Address(obja, result));
9539   load_unsigned_byte(tmp2, Address(objb, result));
9540   xorl(tmp1, tmp2);
9541   testl(tmp1, tmp1);
9542   jccb(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
9543   jmpb(SAME_TILL_END);
9544 
9545   if (UseAVX >= 2){
9546     bind(VECTOR32_NOT_EQUAL);
9547     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
9548     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
9549     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
9550     vpmovmskb(tmp1, rymm0);
9551     bsfq(tmp1, tmp1);
9552     addq(result, tmp1);
9553     shrq(result);
9554     jmpb(DONE);
9555   }
9556 
9557   bind(VECTOR16_NOT_EQUAL);
9558   if (UseAVX >= 2){
9559     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
9560     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
9561     pxor(rymm0, rymm2);
9562   } else {
9563     pcmpeqb(rymm2, rymm2);
9564     pxor(rymm0, rymm1);
9565     pcmpeqb(rymm0, rymm1);
9566     pxor(rymm0, rymm2);
9567   }
9568   pmovmskb(tmp1, rymm0);
9569   bsfq(tmp1, tmp1);
9570   addq(result, tmp1);
9571   shrq(result);
9572   jmpb(DONE);
9573 
9574   bind(VECTOR8_NOT_EQUAL);
9575   bind(VECTOR4_NOT_EQUAL);
9576   bsfq(tmp1, tmp1);
9577   shrq(tmp1, 3);
9578   addq(result, tmp1);
9579   bind(BYTES_NOT_EQUAL);
9580   shrq(result);
9581   jmpb(DONE);
9582 
9583   bind(SAME_TILL_END);
9584   mov64(result, -1);
9585 
9586   bind(DONE);
9587 }
9588 
9589 
9590 //Helper functions for square_to_len()
9591 
9592 /**
9593  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
9594  * Preserves x and z and modifies rest of the registers.
9595  */
9596 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
9597   // Perform square and right shift by 1
9598   // Handle odd xlen case first, then for even xlen do the following
9599   // jlong carry = 0;
9600   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
9601   //     huge_128 product = x[j:j+1] * x[j:j+1];
9602   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
9603   //     z[i+2:i+3] = (jlong)(product >>> 1);
9604   //     carry = (jlong)product;
9605   // }
9606 
9607   xorq(tmp5, tmp5);     // carry
9608   xorq(rdxReg, rdxReg);
9609   xorl(tmp1, tmp1);     // index for x
9610   xorl(tmp4, tmp4);     // index for z
9611 
9612   Label L_first_loop, L_first_loop_exit;
9613 
9614   testl(xlen, 1);
9615   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
9616 
9617   // Square and right shift by 1 the odd element using 32 bit multiply
9618   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
9619   imulq(raxReg, raxReg);
9620   shrq(raxReg, 1);
9621   adcq(tmp5, 0);
9622   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
9623   incrementl(tmp1);
9624   addl(tmp4, 2);
9625 
9626   // Square and  right shift by 1 the rest using 64 bit multiply
9627   bind(L_first_loop);
9628   cmpptr(tmp1, xlen);
9629   jccb(Assembler::equal, L_first_loop_exit);
9630 
9631   // Square
9632   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
9633   rorq(raxReg, 32);    // convert big-endian to little-endian
9634   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
9635 
9636   // Right shift by 1 and save carry
9637   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
9638   rcrq(rdxReg, 1);
9639   rcrq(raxReg, 1);
9640   adcq(tmp5, 0);
9641 
9642   // Store result in z
9643   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
9644   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
9645 
9646   // Update indices for x and z
9647   addl(tmp1, 2);
9648   addl(tmp4, 4);
9649   jmp(L_first_loop);
9650 
9651   bind(L_first_loop_exit);
9652 }
9653 
9654 
9655 /**
9656  * Perform the following multiply add operation using BMI2 instructions
9657  * carry:sum = sum + op1*op2 + carry
9658  * op2 should be in rdx
9659  * op2 is preserved, all other registers are modified
9660  */
9661 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
9662   // assert op2 is rdx
9663   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
9664   addq(sum, carry);
9665   adcq(tmp2, 0);
9666   addq(sum, op1);
9667   adcq(tmp2, 0);
9668   movq(carry, tmp2);
9669 }
9670 
9671 /**
9672  * Perform the following multiply add operation:
9673  * carry:sum = sum + op1*op2 + carry
9674  * Preserves op1, op2 and modifies rest of registers
9675  */
9676 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
9677   // rdx:rax = op1 * op2
9678   movq(raxReg, op2);
9679   mulq(op1);
9680 
9681   //  rdx:rax = sum + carry + rdx:rax
9682   addq(sum, carry);
9683   adcq(rdxReg, 0);
9684   addq(sum, raxReg);
9685   adcq(rdxReg, 0);
9686 
9687   // carry:sum = rdx:sum
9688   movq(carry, rdxReg);
9689 }
9690 
9691 /**
9692  * Add 64 bit long carry into z[] with carry propogation.
9693  * Preserves z and carry register values and modifies rest of registers.
9694  *
9695  */
9696 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
9697   Label L_fourth_loop, L_fourth_loop_exit;
9698 
9699   movl(tmp1, 1);
9700   subl(zlen, 2);
9701   addq(Address(z, zlen, Address::times_4, 0), carry);
9702 
9703   bind(L_fourth_loop);
9704   jccb(Assembler::carryClear, L_fourth_loop_exit);
9705   subl(zlen, 2);
9706   jccb(Assembler::negative, L_fourth_loop_exit);
9707   addq(Address(z, zlen, Address::times_4, 0), tmp1);
9708   jmp(L_fourth_loop);
9709   bind(L_fourth_loop_exit);
9710 }
9711 
9712 /**
9713  * Shift z[] left by 1 bit.
9714  * Preserves x, len, z and zlen registers and modifies rest of the registers.
9715  *
9716  */
9717 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
9718 
9719   Label L_fifth_loop, L_fifth_loop_exit;
9720 
9721   // Fifth loop
9722   // Perform primitiveLeftShift(z, zlen, 1)
9723 
9724   const Register prev_carry = tmp1;
9725   const Register new_carry = tmp4;
9726   const Register value = tmp2;
9727   const Register zidx = tmp3;
9728 
9729   // int zidx, carry;
9730   // long value;
9731   // carry = 0;
9732   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
9733   //    (carry:value)  = (z[i] << 1) | carry ;
9734   //    z[i] = value;
9735   // }
9736 
9737   movl(zidx, zlen);
9738   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
9739 
9740   bind(L_fifth_loop);
9741   decl(zidx);  // Use decl to preserve carry flag
9742   decl(zidx);
9743   jccb(Assembler::negative, L_fifth_loop_exit);
9744 
9745   if (UseBMI2Instructions) {
9746      movq(value, Address(z, zidx, Address::times_4, 0));
9747      rclq(value, 1);
9748      rorxq(value, value, 32);
9749      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
9750   }
9751   else {
9752     // clear new_carry
9753     xorl(new_carry, new_carry);
9754 
9755     // Shift z[i] by 1, or in previous carry and save new carry
9756     movq(value, Address(z, zidx, Address::times_4, 0));
9757     shlq(value, 1);
9758     adcl(new_carry, 0);
9759 
9760     orq(value, prev_carry);
9761     rorq(value, 0x20);
9762     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
9763 
9764     // Set previous carry = new carry
9765     movl(prev_carry, new_carry);
9766   }
9767   jmp(L_fifth_loop);
9768 
9769   bind(L_fifth_loop_exit);
9770 }
9771 
9772 
9773 /**
9774  * Code for BigInteger::squareToLen() intrinsic
9775  *
9776  * rdi: x
9777  * rsi: len
9778  * r8:  z
9779  * rcx: zlen
9780  * r12: tmp1
9781  * r13: tmp2
9782  * r14: tmp3
9783  * r15: tmp4
9784  * rbx: tmp5
9785  *
9786  */
9787 void MacroAssembler::square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
9788 
9789   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, fifth_loop, fifth_loop_exit, L_last_x, L_multiply;
9790   push(tmp1);
9791   push(tmp2);
9792   push(tmp3);
9793   push(tmp4);
9794   push(tmp5);
9795 
9796   // First loop
9797   // Store the squares, right shifted one bit (i.e., divided by 2).
9798   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
9799 
9800   // Add in off-diagonal sums.
9801   //
9802   // Second, third (nested) and fourth loops.
9803   // zlen +=2;
9804   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
9805   //    carry = 0;
9806   //    long op2 = x[xidx:xidx+1];
9807   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
9808   //       k -= 2;
9809   //       long op1 = x[j:j+1];
9810   //       long sum = z[k:k+1];
9811   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
9812   //       z[k:k+1] = sum;
9813   //    }
9814   //    add_one_64(z, k, carry, tmp_regs);
9815   // }
9816 
9817   const Register carry = tmp5;
9818   const Register sum = tmp3;
9819   const Register op1 = tmp4;
9820   Register op2 = tmp2;
9821 
9822   push(zlen);
9823   push(len);
9824   addl(zlen,2);
9825   bind(L_second_loop);
9826   xorq(carry, carry);
9827   subl(zlen, 4);
9828   subl(len, 2);
9829   push(zlen);
9830   push(len);
9831   cmpl(len, 0);
9832   jccb(Assembler::lessEqual, L_second_loop_exit);
9833 
9834   // Multiply an array by one 64 bit long.
9835   if (UseBMI2Instructions) {
9836     op2 = rdxReg;
9837     movq(op2, Address(x, len, Address::times_4,  0));
9838     rorxq(op2, op2, 32);
9839   }
9840   else {
9841     movq(op2, Address(x, len, Address::times_4,  0));
9842     rorq(op2, 32);
9843   }
9844 
9845   bind(L_third_loop);
9846   decrementl(len);
9847   jccb(Assembler::negative, L_third_loop_exit);
9848   decrementl(len);
9849   jccb(Assembler::negative, L_last_x);
9850 
9851   movq(op1, Address(x, len, Address::times_4,  0));
9852   rorq(op1, 32);
9853 
9854   bind(L_multiply);
9855   subl(zlen, 2);
9856   movq(sum, Address(z, zlen, Address::times_4,  0));
9857 
9858   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
9859   if (UseBMI2Instructions) {
9860     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
9861   }
9862   else {
9863     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
9864   }
9865 
9866   movq(Address(z, zlen, Address::times_4, 0), sum);
9867 
9868   jmp(L_third_loop);
9869   bind(L_third_loop_exit);
9870 
9871   // Fourth loop
9872   // Add 64 bit long carry into z with carry propogation.
9873   // Uses offsetted zlen.
9874   add_one_64(z, zlen, carry, tmp1);
9875 
9876   pop(len);
9877   pop(zlen);
9878   jmp(L_second_loop);
9879 
9880   // Next infrequent code is moved outside loops.
9881   bind(L_last_x);
9882   movl(op1, Address(x, 0));
9883   jmp(L_multiply);
9884 
9885   bind(L_second_loop_exit);
9886   pop(len);
9887   pop(zlen);
9888   pop(len);
9889   pop(zlen);
9890 
9891   // Fifth loop
9892   // Shift z left 1 bit.
9893   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
9894 
9895   // z[zlen-1] |= x[len-1] & 1;
9896   movl(tmp3, Address(x, len, Address::times_4, -4));
9897   andl(tmp3, 1);
9898   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
9899 
9900   pop(tmp5);
9901   pop(tmp4);
9902   pop(tmp3);
9903   pop(tmp2);
9904   pop(tmp1);
9905 }
9906 
9907 /**
9908  * Helper function for mul_add()
9909  * Multiply the in[] by int k and add to out[] starting at offset offs using
9910  * 128 bit by 32 bit multiply and return the carry in tmp5.
9911  * Only quad int aligned length of in[] is operated on in this function.
9912  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
9913  * This function preserves out, in and k registers.
9914  * len and offset point to the appropriate index in "in" & "out" correspondingly
9915  * tmp5 has the carry.
9916  * other registers are temporary and are modified.
9917  *
9918  */
9919 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
9920   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
9921   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
9922 
9923   Label L_first_loop, L_first_loop_exit;
9924 
9925   movl(tmp1, len);
9926   shrl(tmp1, 2);
9927 
9928   bind(L_first_loop);
9929   subl(tmp1, 1);
9930   jccb(Assembler::negative, L_first_loop_exit);
9931 
9932   subl(len, 4);
9933   subl(offset, 4);
9934 
9935   Register op2 = tmp2;
9936   const Register sum = tmp3;
9937   const Register op1 = tmp4;
9938   const Register carry = tmp5;
9939 
9940   if (UseBMI2Instructions) {
9941     op2 = rdxReg;
9942   }
9943 
9944   movq(op1, Address(in, len, Address::times_4,  8));
9945   rorq(op1, 32);
9946   movq(sum, Address(out, offset, Address::times_4,  8));
9947   rorq(sum, 32);
9948   if (UseBMI2Instructions) {
9949     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
9950   }
9951   else {
9952     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
9953   }
9954   // Store back in big endian from little endian
9955   rorq(sum, 0x20);
9956   movq(Address(out, offset, Address::times_4,  8), sum);
9957 
9958   movq(op1, Address(in, len, Address::times_4,  0));
9959   rorq(op1, 32);
9960   movq(sum, Address(out, offset, Address::times_4,  0));
9961   rorq(sum, 32);
9962   if (UseBMI2Instructions) {
9963     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
9964   }
9965   else {
9966     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
9967   }
9968   // Store back in big endian from little endian
9969   rorq(sum, 0x20);
9970   movq(Address(out, offset, Address::times_4,  0), sum);
9971 
9972   jmp(L_first_loop);
9973   bind(L_first_loop_exit);
9974 }
9975 
9976 /**
9977  * Code for BigInteger::mulAdd() intrinsic
9978  *
9979  * rdi: out
9980  * rsi: in
9981  * r11: offs (out.length - offset)
9982  * rcx: len
9983  * r8:  k
9984  * r12: tmp1
9985  * r13: tmp2
9986  * r14: tmp3
9987  * r15: tmp4
9988  * rbx: tmp5
9989  * Multiply the in[] by word k and add to out[], return the carry in rax
9990  */
9991 void MacroAssembler::mul_add(Register out, Register in, Register offs,
9992    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
9993    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
9994 
9995   Label L_carry, L_last_in, L_done;
9996 
9997 // carry = 0;
9998 // for (int j=len-1; j >= 0; j--) {
9999 //    long product = (in[j] & LONG_MASK) * kLong +
10000 //                   (out[offs] & LONG_MASK) + carry;
10001 //    out[offs--] = (int)product;
10002 //    carry = product >>> 32;
10003 // }
10004 //
10005   push(tmp1);
10006   push(tmp2);
10007   push(tmp3);
10008   push(tmp4);
10009   push(tmp5);
10010 
10011   Register op2 = tmp2;
10012   const Register sum = tmp3;
10013   const Register op1 = tmp4;
10014   const Register carry =  tmp5;
10015 
10016   if (UseBMI2Instructions) {
10017     op2 = rdxReg;
10018     movl(op2, k);
10019   }
10020   else {
10021     movl(op2, k);
10022   }
10023 
10024   xorq(carry, carry);
10025 
10026   //First loop
10027 
10028   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
10029   //The carry is in tmp5
10030   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
10031 
10032   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
10033   decrementl(len);
10034   jccb(Assembler::negative, L_carry);
10035   decrementl(len);
10036   jccb(Assembler::negative, L_last_in);
10037 
10038   movq(op1, Address(in, len, Address::times_4,  0));
10039   rorq(op1, 32);
10040 
10041   subl(offs, 2);
10042   movq(sum, Address(out, offs, Address::times_4,  0));
10043   rorq(sum, 32);
10044 
10045   if (UseBMI2Instructions) {
10046     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
10047   }
10048   else {
10049     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
10050   }
10051 
10052   // Store back in big endian from little endian
10053   rorq(sum, 0x20);
10054   movq(Address(out, offs, Address::times_4,  0), sum);
10055 
10056   testl(len, len);
10057   jccb(Assembler::zero, L_carry);
10058 
10059   //Multiply the last in[] entry, if any
10060   bind(L_last_in);
10061   movl(op1, Address(in, 0));
10062   movl(sum, Address(out, offs, Address::times_4,  -4));
10063 
10064   movl(raxReg, k);
10065   mull(op1); //tmp4 * eax -> edx:eax
10066   addl(sum, carry);
10067   adcl(rdxReg, 0);
10068   addl(sum, raxReg);
10069   adcl(rdxReg, 0);
10070   movl(carry, rdxReg);
10071 
10072   movl(Address(out, offs, Address::times_4,  -4), sum);
10073 
10074   bind(L_carry);
10075   //return tmp5/carry as carry in rax
10076   movl(rax, carry);
10077 
10078   bind(L_done);
10079   pop(tmp5);
10080   pop(tmp4);
10081   pop(tmp3);
10082   pop(tmp2);
10083   pop(tmp1);
10084 }
10085 #endif
10086 
10087 /**
10088  * Emits code to update CRC-32 with a byte value according to constants in table
10089  *
10090  * @param [in,out]crc   Register containing the crc.
10091  * @param [in]val       Register containing the byte to fold into the CRC.
10092  * @param [in]table     Register containing the table of crc constants.
10093  *
10094  * uint32_t crc;
10095  * val = crc_table[(val ^ crc) & 0xFF];
10096  * crc = val ^ (crc >> 8);
10097  *
10098  */
10099 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
10100   xorl(val, crc);
10101   andl(val, 0xFF);
10102   shrl(crc, 8); // unsigned shift
10103   xorl(crc, Address(table, val, Address::times_4, 0));
10104 }
10105 
10106 /**
10107  * Fold 128-bit data chunk
10108  */
10109 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
10110   if (UseAVX > 0) {
10111     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
10112     vpclmulldq(xcrc, xK, xcrc); // [63:0]
10113     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
10114     pxor(xcrc, xtmp);
10115   } else {
10116     movdqa(xtmp, xcrc);
10117     pclmulhdq(xtmp, xK);   // [123:64]
10118     pclmulldq(xcrc, xK);   // [63:0]
10119     pxor(xcrc, xtmp);
10120     movdqu(xtmp, Address(buf, offset));
10121     pxor(xcrc, xtmp);
10122   }
10123 }
10124 
10125 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
10126   if (UseAVX > 0) {
10127     vpclmulhdq(xtmp, xK, xcrc);
10128     vpclmulldq(xcrc, xK, xcrc);
10129     pxor(xcrc, xbuf);
10130     pxor(xcrc, xtmp);
10131   } else {
10132     movdqa(xtmp, xcrc);
10133     pclmulhdq(xtmp, xK);
10134     pclmulldq(xcrc, xK);
10135     pxor(xcrc, xbuf);
10136     pxor(xcrc, xtmp);
10137   }
10138 }
10139 
10140 /**
10141  * 8-bit folds to compute 32-bit CRC
10142  *
10143  * uint64_t xcrc;
10144  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
10145  */
10146 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
10147   movdl(tmp, xcrc);
10148   andl(tmp, 0xFF);
10149   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
10150   psrldq(xcrc, 1); // unsigned shift one byte
10151   pxor(xcrc, xtmp);
10152 }
10153 
10154 /**
10155  * uint32_t crc;
10156  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
10157  */
10158 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
10159   movl(tmp, crc);
10160   andl(tmp, 0xFF);
10161   shrl(crc, 8);
10162   xorl(crc, Address(table, tmp, Address::times_4, 0));
10163 }
10164 
10165 /**
10166  * @param crc   register containing existing CRC (32-bit)
10167  * @param buf   register pointing to input byte buffer (byte*)
10168  * @param len   register containing number of bytes
10169  * @param table register that will contain address of CRC table
10170  * @param tmp   scratch register
10171  */
10172 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
10173   assert_different_registers(crc, buf, len, table, tmp, rax);
10174 
10175   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
10176   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
10177 
10178   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
10179   // context for the registers used, where all instructions below are using 128-bit mode
10180   // On EVEX without VL and BW, these instructions will all be AVX.
10181   if (VM_Version::supports_avx512vlbw()) {
10182     movl(tmp, 0xffff);
10183     kmovwl(k1, tmp);
10184   }
10185 
10186   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
10187   notl(crc); // ~crc
10188   cmpl(len, 16);
10189   jcc(Assembler::less, L_tail);
10190 
10191   // Align buffer to 16 bytes
10192   movl(tmp, buf);
10193   andl(tmp, 0xF);
10194   jccb(Assembler::zero, L_aligned);
10195   subl(tmp,  16);
10196   addl(len, tmp);
10197 
10198   align(4);
10199   BIND(L_align_loop);
10200   movsbl(rax, Address(buf, 0)); // load byte with sign extension
10201   update_byte_crc32(crc, rax, table);
10202   increment(buf);
10203   incrementl(tmp);
10204   jccb(Assembler::less, L_align_loop);
10205 
10206   BIND(L_aligned);
10207   movl(tmp, len); // save
10208   shrl(len, 4);
10209   jcc(Assembler::zero, L_tail_restore);
10210 
10211   // Fold crc into first bytes of vector
10212   movdqa(xmm1, Address(buf, 0));
10213   movdl(rax, xmm1);
10214   xorl(crc, rax);
10215   pinsrd(xmm1, crc, 0);
10216   addptr(buf, 16);
10217   subl(len, 4); // len > 0
10218   jcc(Assembler::less, L_fold_tail);
10219 
10220   movdqa(xmm2, Address(buf,  0));
10221   movdqa(xmm3, Address(buf, 16));
10222   movdqa(xmm4, Address(buf, 32));
10223   addptr(buf, 48);
10224   subl(len, 3);
10225   jcc(Assembler::lessEqual, L_fold_512b);
10226 
10227   // Fold total 512 bits of polynomial on each iteration,
10228   // 128 bits per each of 4 parallel streams.
10229   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32));
10230 
10231   align(32);
10232   BIND(L_fold_512b_loop);
10233   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
10234   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
10235   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
10236   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
10237   addptr(buf, 64);
10238   subl(len, 4);
10239   jcc(Assembler::greater, L_fold_512b_loop);
10240 
10241   // Fold 512 bits to 128 bits.
10242   BIND(L_fold_512b);
10243   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16));
10244   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
10245   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
10246   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
10247 
10248   // Fold the rest of 128 bits data chunks
10249   BIND(L_fold_tail);
10250   addl(len, 3);
10251   jccb(Assembler::lessEqual, L_fold_128b);
10252   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16));
10253 
10254   BIND(L_fold_tail_loop);
10255   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
10256   addptr(buf, 16);
10257   decrementl(len);
10258   jccb(Assembler::greater, L_fold_tail_loop);
10259 
10260   // Fold 128 bits in xmm1 down into 32 bits in crc register.
10261   BIND(L_fold_128b);
10262   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()));
10263   if (UseAVX > 0) {
10264     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
10265     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
10266     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
10267   } else {
10268     movdqa(xmm2, xmm0);
10269     pclmulqdq(xmm2, xmm1, 0x1);
10270     movdqa(xmm3, xmm0);
10271     pand(xmm3, xmm2);
10272     pclmulqdq(xmm0, xmm3, 0x1);
10273   }
10274   psrldq(xmm1, 8);
10275   psrldq(xmm2, 4);
10276   pxor(xmm0, xmm1);
10277   pxor(xmm0, xmm2);
10278 
10279   // 8 8-bit folds to compute 32-bit CRC.
10280   for (int j = 0; j < 4; j++) {
10281     fold_8bit_crc32(xmm0, table, xmm1, rax);
10282   }
10283   movdl(crc, xmm0); // mov 32 bits to general register
10284   for (int j = 0; j < 4; j++) {
10285     fold_8bit_crc32(crc, table, rax);
10286   }
10287 
10288   BIND(L_tail_restore);
10289   movl(len, tmp); // restore
10290   BIND(L_tail);
10291   andl(len, 0xf);
10292   jccb(Assembler::zero, L_exit);
10293 
10294   // Fold the rest of bytes
10295   align(4);
10296   BIND(L_tail_loop);
10297   movsbl(rax, Address(buf, 0)); // load byte with sign extension
10298   update_byte_crc32(crc, rax, table);
10299   increment(buf);
10300   decrementl(len);
10301   jccb(Assembler::greater, L_tail_loop);
10302 
10303   BIND(L_exit);
10304   notl(crc); // ~c
10305 }
10306 
10307 #ifdef _LP64
10308 // S. Gueron / Information Processing Letters 112 (2012) 184
10309 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
10310 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
10311 // Output: the 64-bit carry-less product of B * CONST
10312 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
10313                                      Register tmp1, Register tmp2, Register tmp3) {
10314   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
10315   if (n > 0) {
10316     addq(tmp3, n * 256 * 8);
10317   }
10318   //    Q1 = TABLEExt[n][B & 0xFF];
10319   movl(tmp1, in);
10320   andl(tmp1, 0x000000FF);
10321   shll(tmp1, 3);
10322   addq(tmp1, tmp3);
10323   movq(tmp1, Address(tmp1, 0));
10324 
10325   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
10326   movl(tmp2, in);
10327   shrl(tmp2, 8);
10328   andl(tmp2, 0x000000FF);
10329   shll(tmp2, 3);
10330   addq(tmp2, tmp3);
10331   movq(tmp2, Address(tmp2, 0));
10332 
10333   shlq(tmp2, 8);
10334   xorq(tmp1, tmp2);
10335 
10336   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
10337   movl(tmp2, in);
10338   shrl(tmp2, 16);
10339   andl(tmp2, 0x000000FF);
10340   shll(tmp2, 3);
10341   addq(tmp2, tmp3);
10342   movq(tmp2, Address(tmp2, 0));
10343 
10344   shlq(tmp2, 16);
10345   xorq(tmp1, tmp2);
10346 
10347   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
10348   shrl(in, 24);
10349   andl(in, 0x000000FF);
10350   shll(in, 3);
10351   addq(in, tmp3);
10352   movq(in, Address(in, 0));
10353 
10354   shlq(in, 24);
10355   xorq(in, tmp1);
10356   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
10357 }
10358 
10359 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
10360                                       Register in_out,
10361                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
10362                                       XMMRegister w_xtmp2,
10363                                       Register tmp1,
10364                                       Register n_tmp2, Register n_tmp3) {
10365   if (is_pclmulqdq_supported) {
10366     movdl(w_xtmp1, in_out); // modified blindly
10367 
10368     movl(tmp1, const_or_pre_comp_const_index);
10369     movdl(w_xtmp2, tmp1);
10370     pclmulqdq(w_xtmp1, w_xtmp2, 0);
10371 
10372     movdq(in_out, w_xtmp1);
10373   } else {
10374     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
10375   }
10376 }
10377 
10378 // Recombination Alternative 2: No bit-reflections
10379 // T1 = (CRC_A * U1) << 1
10380 // T2 = (CRC_B * U2) << 1
10381 // C1 = T1 >> 32
10382 // C2 = T2 >> 32
10383 // T1 = T1 & 0xFFFFFFFF
10384 // T2 = T2 & 0xFFFFFFFF
10385 // T1 = CRC32(0, T1)
10386 // T2 = CRC32(0, T2)
10387 // C1 = C1 ^ T1
10388 // C2 = C2 ^ T2
10389 // CRC = C1 ^ C2 ^ CRC_C
10390 void MacroAssembler::crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2,
10391                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10392                                      Register tmp1, Register tmp2,
10393                                      Register n_tmp3) {
10394   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
10395   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
10396   shlq(in_out, 1);
10397   movl(tmp1, in_out);
10398   shrq(in_out, 32);
10399   xorl(tmp2, tmp2);
10400   crc32(tmp2, tmp1, 4);
10401   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
10402   shlq(in1, 1);
10403   movl(tmp1, in1);
10404   shrq(in1, 32);
10405   xorl(tmp2, tmp2);
10406   crc32(tmp2, tmp1, 4);
10407   xorl(in1, tmp2);
10408   xorl(in_out, in1);
10409   xorl(in_out, in2);
10410 }
10411 
10412 // Set N to predefined value
10413 // Subtract from a lenght of a buffer
10414 // execute in a loop:
10415 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
10416 // for i = 1 to N do
10417 //  CRC_A = CRC32(CRC_A, A[i])
10418 //  CRC_B = CRC32(CRC_B, B[i])
10419 //  CRC_C = CRC32(CRC_C, C[i])
10420 // end for
10421 // Recombine
10422 void MacroAssembler::crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported,
10423                                        Register in_out1, Register in_out2, Register in_out3,
10424                                        Register tmp1, Register tmp2, Register tmp3,
10425                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10426                                        Register tmp4, Register tmp5,
10427                                        Register n_tmp6) {
10428   Label L_processPartitions;
10429   Label L_processPartition;
10430   Label L_exit;
10431 
10432   bind(L_processPartitions);
10433   cmpl(in_out1, 3 * size);
10434   jcc(Assembler::less, L_exit);
10435     xorl(tmp1, tmp1);
10436     xorl(tmp2, tmp2);
10437     movq(tmp3, in_out2);
10438     addq(tmp3, size);
10439 
10440     bind(L_processPartition);
10441       crc32(in_out3, Address(in_out2, 0), 8);
10442       crc32(tmp1, Address(in_out2, size), 8);
10443       crc32(tmp2, Address(in_out2, size * 2), 8);
10444       addq(in_out2, 8);
10445       cmpq(in_out2, tmp3);
10446       jcc(Assembler::less, L_processPartition);
10447     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
10448             w_xtmp1, w_xtmp2, w_xtmp3,
10449             tmp4, tmp5,
10450             n_tmp6);
10451     addq(in_out2, 2 * size);
10452     subl(in_out1, 3 * size);
10453     jmp(L_processPartitions);
10454 
10455   bind(L_exit);
10456 }
10457 #else
10458 void MacroAssembler::crc32c_ipl_alg4(Register in_out, uint32_t n,
10459                                      Register tmp1, Register tmp2, Register tmp3,
10460                                      XMMRegister xtmp1, XMMRegister xtmp2) {
10461   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
10462   if (n > 0) {
10463     addl(tmp3, n * 256 * 8);
10464   }
10465   //    Q1 = TABLEExt[n][B & 0xFF];
10466   movl(tmp1, in_out);
10467   andl(tmp1, 0x000000FF);
10468   shll(tmp1, 3);
10469   addl(tmp1, tmp3);
10470   movq(xtmp1, Address(tmp1, 0));
10471 
10472   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
10473   movl(tmp2, in_out);
10474   shrl(tmp2, 8);
10475   andl(tmp2, 0x000000FF);
10476   shll(tmp2, 3);
10477   addl(tmp2, tmp3);
10478   movq(xtmp2, Address(tmp2, 0));
10479 
10480   psllq(xtmp2, 8);
10481   pxor(xtmp1, xtmp2);
10482 
10483   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
10484   movl(tmp2, in_out);
10485   shrl(tmp2, 16);
10486   andl(tmp2, 0x000000FF);
10487   shll(tmp2, 3);
10488   addl(tmp2, tmp3);
10489   movq(xtmp2, Address(tmp2, 0));
10490 
10491   psllq(xtmp2, 16);
10492   pxor(xtmp1, xtmp2);
10493 
10494   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
10495   shrl(in_out, 24);
10496   andl(in_out, 0x000000FF);
10497   shll(in_out, 3);
10498   addl(in_out, tmp3);
10499   movq(xtmp2, Address(in_out, 0));
10500 
10501   psllq(xtmp2, 24);
10502   pxor(xtmp1, xtmp2); // Result in CXMM
10503   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
10504 }
10505 
10506 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
10507                                       Register in_out,
10508                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
10509                                       XMMRegister w_xtmp2,
10510                                       Register tmp1,
10511                                       Register n_tmp2, Register n_tmp3) {
10512   if (is_pclmulqdq_supported) {
10513     movdl(w_xtmp1, in_out);
10514 
10515     movl(tmp1, const_or_pre_comp_const_index);
10516     movdl(w_xtmp2, tmp1);
10517     pclmulqdq(w_xtmp1, w_xtmp2, 0);
10518     // Keep result in XMM since GPR is 32 bit in length
10519   } else {
10520     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3, w_xtmp1, w_xtmp2);
10521   }
10522 }
10523 
10524 void MacroAssembler::crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2,
10525                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10526                                      Register tmp1, Register tmp2,
10527                                      Register n_tmp3) {
10528   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
10529   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
10530 
10531   psllq(w_xtmp1, 1);
10532   movdl(tmp1, w_xtmp1);
10533   psrlq(w_xtmp1, 32);
10534   movdl(in_out, w_xtmp1);
10535 
10536   xorl(tmp2, tmp2);
10537   crc32(tmp2, tmp1, 4);
10538   xorl(in_out, tmp2);
10539 
10540   psllq(w_xtmp2, 1);
10541   movdl(tmp1, w_xtmp2);
10542   psrlq(w_xtmp2, 32);
10543   movdl(in1, w_xtmp2);
10544 
10545   xorl(tmp2, tmp2);
10546   crc32(tmp2, tmp1, 4);
10547   xorl(in1, tmp2);
10548   xorl(in_out, in1);
10549   xorl(in_out, in2);
10550 }
10551 
10552 void MacroAssembler::crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported,
10553                                        Register in_out1, Register in_out2, Register in_out3,
10554                                        Register tmp1, Register tmp2, Register tmp3,
10555                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10556                                        Register tmp4, Register tmp5,
10557                                        Register n_tmp6) {
10558   Label L_processPartitions;
10559   Label L_processPartition;
10560   Label L_exit;
10561 
10562   bind(L_processPartitions);
10563   cmpl(in_out1, 3 * size);
10564   jcc(Assembler::less, L_exit);
10565     xorl(tmp1, tmp1);
10566     xorl(tmp2, tmp2);
10567     movl(tmp3, in_out2);
10568     addl(tmp3, size);
10569 
10570     bind(L_processPartition);
10571       crc32(in_out3, Address(in_out2, 0), 4);
10572       crc32(tmp1, Address(in_out2, size), 4);
10573       crc32(tmp2, Address(in_out2, size*2), 4);
10574       crc32(in_out3, Address(in_out2, 0+4), 4);
10575       crc32(tmp1, Address(in_out2, size+4), 4);
10576       crc32(tmp2, Address(in_out2, size*2+4), 4);
10577       addl(in_out2, 8);
10578       cmpl(in_out2, tmp3);
10579       jcc(Assembler::less, L_processPartition);
10580 
10581         push(tmp3);
10582         push(in_out1);
10583         push(in_out2);
10584         tmp4 = tmp3;
10585         tmp5 = in_out1;
10586         n_tmp6 = in_out2;
10587 
10588       crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
10589             w_xtmp1, w_xtmp2, w_xtmp3,
10590             tmp4, tmp5,
10591             n_tmp6);
10592 
10593         pop(in_out2);
10594         pop(in_out1);
10595         pop(tmp3);
10596 
10597     addl(in_out2, 2 * size);
10598     subl(in_out1, 3 * size);
10599     jmp(L_processPartitions);
10600 
10601   bind(L_exit);
10602 }
10603 #endif //LP64
10604 
10605 #ifdef _LP64
10606 // Algorithm 2: Pipelined usage of the CRC32 instruction.
10607 // Input: A buffer I of L bytes.
10608 // Output: the CRC32C value of the buffer.
10609 // Notations:
10610 // Write L = 24N + r, with N = floor (L/24).
10611 // r = L mod 24 (0 <= r < 24).
10612 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
10613 // N quadwords, and R consists of r bytes.
10614 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
10615 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
10616 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
10617 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
10618 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
10619                                           Register tmp1, Register tmp2, Register tmp3,
10620                                           Register tmp4, Register tmp5, Register tmp6,
10621                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10622                                           bool is_pclmulqdq_supported) {
10623   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
10624   Label L_wordByWord;
10625   Label L_byteByByteProlog;
10626   Label L_byteByByte;
10627   Label L_exit;
10628 
10629   if (is_pclmulqdq_supported ) {
10630     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
10631     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr+1);
10632 
10633     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
10634     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
10635 
10636     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
10637     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
10638     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
10639   } else {
10640     const_or_pre_comp_const_index[0] = 1;
10641     const_or_pre_comp_const_index[1] = 0;
10642 
10643     const_or_pre_comp_const_index[2] = 3;
10644     const_or_pre_comp_const_index[3] = 2;
10645 
10646     const_or_pre_comp_const_index[4] = 5;
10647     const_or_pre_comp_const_index[5] = 4;
10648    }
10649   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
10650                     in2, in1, in_out,
10651                     tmp1, tmp2, tmp3,
10652                     w_xtmp1, w_xtmp2, w_xtmp3,
10653                     tmp4, tmp5,
10654                     tmp6);
10655   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
10656                     in2, in1, in_out,
10657                     tmp1, tmp2, tmp3,
10658                     w_xtmp1, w_xtmp2, w_xtmp3,
10659                     tmp4, tmp5,
10660                     tmp6);
10661   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
10662                     in2, in1, in_out,
10663                     tmp1, tmp2, tmp3,
10664                     w_xtmp1, w_xtmp2, w_xtmp3,
10665                     tmp4, tmp5,
10666                     tmp6);
10667   movl(tmp1, in2);
10668   andl(tmp1, 0x00000007);
10669   negl(tmp1);
10670   addl(tmp1, in2);
10671   addq(tmp1, in1);
10672 
10673   BIND(L_wordByWord);
10674   cmpq(in1, tmp1);
10675   jcc(Assembler::greaterEqual, L_byteByByteProlog);
10676     crc32(in_out, Address(in1, 0), 4);
10677     addq(in1, 4);
10678     jmp(L_wordByWord);
10679 
10680   BIND(L_byteByByteProlog);
10681   andl(in2, 0x00000007);
10682   movl(tmp2, 1);
10683 
10684   BIND(L_byteByByte);
10685   cmpl(tmp2, in2);
10686   jccb(Assembler::greater, L_exit);
10687     crc32(in_out, Address(in1, 0), 1);
10688     incq(in1);
10689     incl(tmp2);
10690     jmp(L_byteByByte);
10691 
10692   BIND(L_exit);
10693 }
10694 #else
10695 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
10696                                           Register tmp1, Register  tmp2, Register tmp3,
10697                                           Register tmp4, Register  tmp5, Register tmp6,
10698                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
10699                                           bool is_pclmulqdq_supported) {
10700   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
10701   Label L_wordByWord;
10702   Label L_byteByByteProlog;
10703   Label L_byteByByte;
10704   Label L_exit;
10705 
10706   if (is_pclmulqdq_supported) {
10707     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
10708     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 1);
10709 
10710     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
10711     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
10712 
10713     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
10714     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
10715   } else {
10716     const_or_pre_comp_const_index[0] = 1;
10717     const_or_pre_comp_const_index[1] = 0;
10718 
10719     const_or_pre_comp_const_index[2] = 3;
10720     const_or_pre_comp_const_index[3] = 2;
10721 
10722     const_or_pre_comp_const_index[4] = 5;
10723     const_or_pre_comp_const_index[5] = 4;
10724   }
10725   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
10726                     in2, in1, in_out,
10727                     tmp1, tmp2, tmp3,
10728                     w_xtmp1, w_xtmp2, w_xtmp3,
10729                     tmp4, tmp5,
10730                     tmp6);
10731   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
10732                     in2, in1, in_out,
10733                     tmp1, tmp2, tmp3,
10734                     w_xtmp1, w_xtmp2, w_xtmp3,
10735                     tmp4, tmp5,
10736                     tmp6);
10737   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
10738                     in2, in1, in_out,
10739                     tmp1, tmp2, tmp3,
10740                     w_xtmp1, w_xtmp2, w_xtmp3,
10741                     tmp4, tmp5,
10742                     tmp6);
10743   movl(tmp1, in2);
10744   andl(tmp1, 0x00000007);
10745   negl(tmp1);
10746   addl(tmp1, in2);
10747   addl(tmp1, in1);
10748 
10749   BIND(L_wordByWord);
10750   cmpl(in1, tmp1);
10751   jcc(Assembler::greaterEqual, L_byteByByteProlog);
10752     crc32(in_out, Address(in1,0), 4);
10753     addl(in1, 4);
10754     jmp(L_wordByWord);
10755 
10756   BIND(L_byteByByteProlog);
10757   andl(in2, 0x00000007);
10758   movl(tmp2, 1);
10759 
10760   BIND(L_byteByByte);
10761   cmpl(tmp2, in2);
10762   jccb(Assembler::greater, L_exit);
10763     movb(tmp1, Address(in1, 0));
10764     crc32(in_out, tmp1, 1);
10765     incl(in1);
10766     incl(tmp2);
10767     jmp(L_byteByByte);
10768 
10769   BIND(L_exit);
10770 }
10771 #endif // LP64
10772 #undef BIND
10773 #undef BLOCK_COMMENT
10774 
10775 
10776 // Compress char[] array to byte[].
10777 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
10778                                          XMMRegister tmp1Reg, XMMRegister tmp2Reg,
10779                                          XMMRegister tmp3Reg, XMMRegister tmp4Reg,
10780                                          Register tmp5, Register result) {
10781   Label copy_chars_loop, return_length, return_zero, done;
10782 
10783   // rsi: src
10784   // rdi: dst
10785   // rdx: len
10786   // rcx: tmp5
10787   // rax: result
10788 
10789   // rsi holds start addr of source char[] to be compressed
10790   // rdi holds start addr of destination byte[]
10791   // rdx holds length
10792 
10793   assert(len != result, "");
10794 
10795   // save length for return
10796   push(len);
10797 
10798   if (UseSSE42Intrinsics) {
10799     assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
10800     Label copy_32_loop, copy_16, copy_tail;
10801 
10802     movl(result, len);
10803     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
10804 
10805     // vectored compression
10806     andl(len, 0xfffffff0);    // vector count (in chars)
10807     andl(result, 0x0000000f);    // tail count (in chars)
10808     testl(len, len);
10809     jccb(Assembler::zero, copy_16);
10810 
10811     // compress 16 chars per iter
10812     movdl(tmp1Reg, tmp5);
10813     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
10814     pxor(tmp4Reg, tmp4Reg);
10815 
10816     lea(src, Address(src, len, Address::times_2));
10817     lea(dst, Address(dst, len, Address::times_1));
10818     negptr(len);
10819 
10820     bind(copy_32_loop);
10821     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
10822     por(tmp4Reg, tmp2Reg);
10823     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
10824     por(tmp4Reg, tmp3Reg);
10825     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
10826     jcc(Assembler::notZero, return_zero);
10827     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
10828     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
10829     addptr(len, 16);
10830     jcc(Assembler::notZero, copy_32_loop);
10831 
10832     // compress next vector of 8 chars (if any)
10833     bind(copy_16);
10834     movl(len, result);
10835     andl(len, 0xfffffff8);    // vector count (in chars)
10836     andl(result, 0x00000007);    // tail count (in chars)
10837     testl(len, len);
10838     jccb(Assembler::zero, copy_tail);
10839 
10840     movdl(tmp1Reg, tmp5);
10841     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
10842     pxor(tmp3Reg, tmp3Reg);
10843 
10844     movdqu(tmp2Reg, Address(src, 0));
10845     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
10846     jccb(Assembler::notZero, return_zero);
10847     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
10848     movq(Address(dst, 0), tmp2Reg);
10849     addptr(src, 16);
10850     addptr(dst, 8);
10851 
10852     bind(copy_tail);
10853     movl(len, result);
10854   }
10855   // compress 1 char per iter
10856   testl(len, len);
10857   jccb(Assembler::zero, return_length);
10858   lea(src, Address(src, len, Address::times_2));
10859   lea(dst, Address(dst, len, Address::times_1));
10860   negptr(len);
10861 
10862   bind(copy_chars_loop);
10863   load_unsigned_short(result, Address(src, len, Address::times_2));
10864   testl(result, 0xff00);      // check if Unicode char
10865   jccb(Assembler::notZero, return_zero);
10866   movb(Address(dst, len, Address::times_1), result);  // ASCII char; compress to 1 byte
10867   increment(len);
10868   jcc(Assembler::notZero, copy_chars_loop);
10869 
10870   // if compression succeeded, return length
10871   bind(return_length);
10872   pop(result);
10873   jmpb(done);
10874 
10875   // if compression failed, return 0
10876   bind(return_zero);
10877   xorl(result, result);
10878   addptr(rsp, wordSize);
10879 
10880   bind(done);
10881 }
10882 
10883 // Inflate byte[] array to char[].
10884 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
10885                                         XMMRegister tmp1, Register tmp2) {
10886   Label copy_chars_loop, done;
10887 
10888   // rsi: src
10889   // rdi: dst
10890   // rdx: len
10891   // rcx: tmp2
10892 
10893   // rsi holds start addr of source byte[] to be inflated
10894   // rdi holds start addr of destination char[]
10895   // rdx holds length
10896   assert_different_registers(src, dst, len, tmp2);
10897 
10898   if (UseSSE42Intrinsics) {
10899     assert(UseSSE >= 4, "SSE4 must be enabled for SSE4.2 intrinsics to be available");
10900     Label copy_8_loop, copy_bytes, copy_tail;
10901 
10902     movl(tmp2, len);
10903     andl(tmp2, 0x00000007);   // tail count (in chars)
10904     andl(len, 0xfffffff8);    // vector count (in chars)
10905     jccb(Assembler::zero, copy_tail);
10906 
10907     // vectored inflation
10908     lea(src, Address(src, len, Address::times_1));
10909     lea(dst, Address(dst, len, Address::times_2));
10910     negptr(len);
10911 
10912     // inflate 8 chars per iter
10913     bind(copy_8_loop);
10914     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
10915     movdqu(Address(dst, len, Address::times_2), tmp1);
10916     addptr(len, 8);
10917     jcc(Assembler::notZero, copy_8_loop);
10918 
10919     bind(copy_tail);
10920     movl(len, tmp2);
10921 
10922     cmpl(len, 4);
10923     jccb(Assembler::less, copy_bytes);
10924 
10925     movdl(tmp1, Address(src, 0));  // load 4 byte chars
10926     pmovzxbw(tmp1, tmp1);
10927     movq(Address(dst, 0), tmp1);
10928     subptr(len, 4);
10929     addptr(src, 4);
10930     addptr(dst, 8);
10931 
10932     bind(copy_bytes);
10933   }
10934   testl(len, len);
10935   jccb(Assembler::zero, done);
10936   lea(src, Address(src, len, Address::times_1));
10937   lea(dst, Address(dst, len, Address::times_2));
10938   negptr(len);
10939 
10940   // inflate 1 char per iter
10941   bind(copy_chars_loop);
10942   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
10943   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
10944   increment(len);
10945   jcc(Assembler::notZero, copy_chars_loop);
10946 
10947   bind(done);
10948 }
10949 
10950 
10951 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10952   switch (cond) {
10953     // Note some conditions are synonyms for others
10954     case Assembler::zero:         return Assembler::notZero;
10955     case Assembler::notZero:      return Assembler::zero;
10956     case Assembler::less:         return Assembler::greaterEqual;
10957     case Assembler::lessEqual:    return Assembler::greater;
10958     case Assembler::greater:      return Assembler::lessEqual;
10959     case Assembler::greaterEqual: return Assembler::less;
10960     case Assembler::below:        return Assembler::aboveEqual;
10961     case Assembler::belowEqual:   return Assembler::above;
10962     case Assembler::above:        return Assembler::belowEqual;
10963     case Assembler::aboveEqual:   return Assembler::below;
10964     case Assembler::overflow:     return Assembler::noOverflow;
10965     case Assembler::noOverflow:   return Assembler::overflow;
10966     case Assembler::negative:     return Assembler::positive;
10967     case Assembler::positive:     return Assembler::negative;
10968     case Assembler::parity:       return Assembler::noParity;
10969     case Assembler::noParity:     return Assembler::parity;
10970   }
10971   ShouldNotReachHere(); return Assembler::overflow;
10972 }
10973 
10974 SkipIfEqual::SkipIfEqual(
10975     MacroAssembler* masm, const bool* flag_addr, bool value) {
10976   _masm = masm;
10977   _masm->cmp8(ExternalAddress((address)flag_addr), value);
10978   _masm->jcc(Assembler::equal, _label);
10979 }
10980 
10981 SkipIfEqual::~SkipIfEqual() {
10982   _masm->bind(_label);
10983 }
10984 
10985 // 32-bit Windows has its own fast-path implementation
10986 // of get_thread
10987 #if !defined(WIN32) || defined(_LP64)
10988 
10989 // This is simply a call to Thread::current()
10990 void MacroAssembler::get_thread(Register thread) {
10991   if (thread != rax) {
10992     push(rax);
10993   }
10994   LP64_ONLY(push(rdi);)
10995   LP64_ONLY(push(rsi);)
10996   push(rdx);
10997   push(rcx);
10998 #ifdef _LP64
10999   push(r8);
11000   push(r9);
11001   push(r10);
11002   push(r11);
11003 #endif
11004 
11005   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
11006 
11007 #ifdef _LP64
11008   pop(r11);
11009   pop(r10);
11010   pop(r9);
11011   pop(r8);
11012 #endif
11013   pop(rcx);
11014   pop(rdx);
11015   LP64_ONLY(pop(rsi);)
11016   LP64_ONLY(pop(rdi);)
11017   if (thread != rax) {
11018     mov(thread, rax);
11019     pop(rax);
11020   }
11021 }
11022 
11023 #endif