1 /*
    2  * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
    3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
    4  *
    5  * This code is free software; you can redistribute it and/or modify it
    6  * under the terms of the GNU General Public License version 2 only, as
    7  * published by the Free Software Foundation.
    8  *
    9  * This code is distributed in the hope that it will be useful, but WITHOUT
   10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "precompiled.hpp"
   26 #include "asm/assembler.hpp"
   27 #include "asm/assembler.inline.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"
   31 #include "crc32c.h"
   32 #include "gc/shared/barrierSet.hpp"
   33 #include "gc/shared/barrierSetAssembler.hpp"
   34 #include "gc/shared/collectedHeap.inline.hpp"
   35 #include "gc/shared/tlab_globals.hpp"
   36 #include "interpreter/bytecodeHistogram.hpp"
   37 #include "interpreter/interpreter.hpp"
   38 #include "jvm.h"
   39 #include "memory/resourceArea.hpp"
   40 #include "memory/universe.hpp"
   41 #include "oops/accessDecorators.hpp"
   42 #include "oops/compressedKlass.inline.hpp"
   43 #include "oops/compressedOops.inline.hpp"
   44 #include "oops/klass.inline.hpp"
   45 #include "prims/methodHandles.hpp"
   46 #include "runtime/continuation.hpp"
   47 #include "runtime/interfaceSupport.inline.hpp"
   48 #include "runtime/javaThread.hpp"
   49 #include "runtime/jniHandles.hpp"
   50 #include "runtime/objectMonitor.hpp"
   51 #include "runtime/os.hpp"
   52 #include "runtime/safepoint.hpp"
   53 #include "runtime/safepointMechanism.hpp"
   54 #include "runtime/sharedRuntime.hpp"
   55 #include "runtime/stubRoutines.hpp"
   56 #include "utilities/checkedCast.hpp"
   57 #include "utilities/macros.hpp"
   58 
   59 #ifdef PRODUCT
   60 #define BLOCK_COMMENT(str) /* nothing */
   61 #define STOP(error) stop(error)
   62 #else
   63 #define BLOCK_COMMENT(str) block_comment(str)
   64 #define STOP(error) block_comment(error); stop(error)
   65 #endif
   66 
   67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   68 
   69 #ifdef ASSERT
   70 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   71 #endif
   72 
   73 static const Assembler::Condition reverse[] = {
   74     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   75     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   76     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   77     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   78     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   79     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   80     Assembler::above          /* belowEqual    = 0x6 */ ,
   81     Assembler::belowEqual     /* above         = 0x7 */ ,
   82     Assembler::positive       /* negative      = 0x8 */ ,
   83     Assembler::negative       /* positive      = 0x9 */ ,
   84     Assembler::noParity       /* parity        = 0xa */ ,
   85     Assembler::parity         /* noParity      = 0xb */ ,
   86     Assembler::greaterEqual   /* less          = 0xc */ ,
   87     Assembler::less           /* greaterEqual  = 0xd */ ,
   88     Assembler::greater        /* lessEqual     = 0xe */ ,
   89     Assembler::lessEqual      /* greater       = 0xf, */
   90 
   91 };
   92 
   93 
   94 // Implementation of MacroAssembler
   95 
   96 // First all the versions that have distinct versions depending on 32/64 bit
   97 // Unless the difference is trivial (1 line or so).
   98 
   99 #ifndef _LP64
  100 
  101 // 32bit versions
  102 
  103 Address MacroAssembler::as_Address(AddressLiteral adr) {
  104   return Address(adr.target(), adr.rspec());
  105 }
  106 
  107 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  108   assert(rscratch == noreg, "");
  109   return Address::make_array(adr);
  110 }
  111 
  112 void MacroAssembler::call_VM_leaf_base(address entry_point,
  113                                        int number_of_arguments) {
  114   call(RuntimeAddress(entry_point));
  115   increment(rsp, number_of_arguments * wordSize);
  116 }
  117 
  118 void MacroAssembler::cmpklass(Address src1, Metadata* obj) {
  119   cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
  120 }
  121 
  122 
  123 void MacroAssembler::cmpklass(Register src1, Metadata* obj) {
  124   cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
  125 }
  126 
  127 void MacroAssembler::cmpoop(Address src1, jobject obj) {
  128   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
  129 }
  130 
  131 void MacroAssembler::cmpoop(Register src1, jobject obj, Register rscratch) {
  132   assert(rscratch == noreg, "redundant");
  133   cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
  134 }
  135 
  136 void MacroAssembler::extend_sign(Register hi, Register lo) {
  137   // According to Intel Doc. AP-526, "Integer Divide", p.18.
  138   if (VM_Version::is_P6() && hi == rdx && lo == rax) {
  139     cdql();
  140   } else {
  141     movl(hi, lo);
  142     sarl(hi, 31);
  143   }
  144 }
  145 
  146 void MacroAssembler::jC2(Register tmp, Label& L) {
  147   // set parity bit if FPU flag C2 is set (via rax)
  148   save_rax(tmp);
  149   fwait(); fnstsw_ax();
  150   sahf();
  151   restore_rax(tmp);
  152   // branch
  153   jcc(Assembler::parity, L);
  154 }
  155 
  156 void MacroAssembler::jnC2(Register tmp, Label& L) {
  157   // set parity bit if FPU flag C2 is set (via rax)
  158   save_rax(tmp);
  159   fwait(); fnstsw_ax();
  160   sahf();
  161   restore_rax(tmp);
  162   // branch
  163   jcc(Assembler::noParity, L);
  164 }
  165 
  166 // 32bit can do a case table jump in one instruction but we no longer allow the base
  167 // to be installed in the Address class
  168 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  169   assert(rscratch == noreg, "not needed");
  170   jmp(as_Address(entry, noreg));
  171 }
  172 
  173 // Note: y_lo will be destroyed
  174 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  175   // Long compare for Java (semantics as described in JVM spec.)
  176   Label high, low, done;
  177 
  178   cmpl(x_hi, y_hi);
  179   jcc(Assembler::less, low);
  180   jcc(Assembler::greater, high);
  181   // x_hi is the return register
  182   xorl(x_hi, x_hi);
  183   cmpl(x_lo, y_lo);
  184   jcc(Assembler::below, low);
  185   jcc(Assembler::equal, done);
  186 
  187   bind(high);
  188   xorl(x_hi, x_hi);
  189   increment(x_hi);
  190   jmp(done);
  191 
  192   bind(low);
  193   xorl(x_hi, x_hi);
  194   decrementl(x_hi);
  195 
  196   bind(done);
  197 }
  198 
  199 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  200   mov_literal32(dst, (int32_t)src.target(), src.rspec());
  201 }
  202 
  203 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  204   assert(rscratch == noreg, "not needed");
  205 
  206   // leal(dst, as_Address(adr));
  207   // see note in movl as to why we must use a move
  208   mov_literal32(dst, (int32_t)adr.target(), adr.rspec());
  209 }
  210 
  211 void MacroAssembler::leave() {
  212   mov(rsp, rbp);
  213   pop(rbp);
  214 }
  215 
  216 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) {
  217   // Multiplication of two Java long values stored on the stack
  218   // as illustrated below. Result is in rdx:rax.
  219   //
  220   // rsp ---> [  ??  ] \               \
  221   //            ....    | y_rsp_offset  |
  222   //          [ y_lo ] /  (in bytes)    | x_rsp_offset
  223   //          [ y_hi ]                  | (in bytes)
  224   //            ....                    |
  225   //          [ x_lo ]                 /
  226   //          [ x_hi ]
  227   //            ....
  228   //
  229   // Basic idea: lo(result) = lo(x_lo * y_lo)
  230   //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
  231   Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset);
  232   Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset);
  233   Label quick;
  234   // load x_hi, y_hi and check if quick
  235   // multiplication is possible
  236   movl(rbx, x_hi);
  237   movl(rcx, y_hi);
  238   movl(rax, rbx);
  239   orl(rbx, rcx);                                 // rbx, = 0 <=> x_hi = 0 and y_hi = 0
  240   jcc(Assembler::zero, quick);                   // if rbx, = 0 do quick multiply
  241   // do full multiplication
  242   // 1st step
  243   mull(y_lo);                                    // x_hi * y_lo
  244   movl(rbx, rax);                                // save lo(x_hi * y_lo) in rbx,
  245   // 2nd step
  246   movl(rax, x_lo);
  247   mull(rcx);                                     // x_lo * y_hi
  248   addl(rbx, rax);                                // add lo(x_lo * y_hi) to rbx,
  249   // 3rd step
  250   bind(quick);                                   // note: rbx, = 0 if quick multiply!
  251   movl(rax, x_lo);
  252   mull(y_lo);                                    // x_lo * y_lo
  253   addl(rdx, rbx);                                // correct hi(x_lo * y_lo)
  254 }
  255 
  256 void MacroAssembler::lneg(Register hi, Register lo) {
  257   negl(lo);
  258   adcl(hi, 0);
  259   negl(hi);
  260 }
  261 
  262 void MacroAssembler::lshl(Register hi, Register lo) {
  263   // Java shift left long support (semantics as described in JVM spec., p.305)
  264   // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n))
  265   // shift value is in rcx !
  266   assert(hi != rcx, "must not use rcx");
  267   assert(lo != rcx, "must not use rcx");
  268   const Register s = rcx;                        // shift count
  269   const int      n = BitsPerWord;
  270   Label L;
  271   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
  272   cmpl(s, n);                                    // if (s < n)
  273   jcc(Assembler::less, L);                       // else (s >= n)
  274   movl(hi, lo);                                  // x := x << n
  275   xorl(lo, lo);
  276   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
  277   bind(L);                                       // s (mod n) < n
  278   shldl(hi, lo);                                 // x := x << s
  279   shll(lo);
  280 }
  281 
  282 
  283 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) {
  284   // Java shift right long support (semantics as described in JVM spec., p.306 & p.310)
  285   // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n))
  286   assert(hi != rcx, "must not use rcx");
  287   assert(lo != rcx, "must not use rcx");
  288   const Register s = rcx;                        // shift count
  289   const int      n = BitsPerWord;
  290   Label L;
  291   andl(s, 0x3f);                                 // s := s & 0x3f (s < 0x40)
  292   cmpl(s, n);                                    // if (s < n)
  293   jcc(Assembler::less, L);                       // else (s >= n)
  294   movl(lo, hi);                                  // x := x >> n
  295   if (sign_extension) sarl(hi, 31);
  296   else                xorl(hi, hi);
  297   // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
  298   bind(L);                                       // s (mod n) < n
  299   shrdl(lo, hi);                                 // x := x >> s
  300   if (sign_extension) sarl(hi);
  301   else                shrl(hi);
  302 }
  303 
  304 void MacroAssembler::movoop(Register dst, jobject obj) {
  305   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
  306 }
  307 
  308 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  309   assert(rscratch == noreg, "redundant");
  310   mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
  311 }
  312 
  313 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  314   mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
  315 }
  316 
  317 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  318   assert(rscratch == noreg, "redundant");
  319   mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
  320 }
  321 
  322 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  323   if (src.is_lval()) {
  324     mov_literal32(dst, (intptr_t)src.target(), src.rspec());
  325   } else {
  326     movl(dst, as_Address(src));
  327   }
  328 }
  329 
  330 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  331   assert(rscratch == noreg, "redundant");
  332   movl(as_Address(dst, noreg), src);
  333 }
  334 
  335 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  336   movl(dst, as_Address(src, noreg));
  337 }
  338 
  339 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  340   assert(rscratch == noreg, "redundant");
  341   movl(dst, src);
  342 }
  343 
  344 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  345   assert(rscratch == noreg, "redundant");
  346   push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
  347 }
  348 
  349 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  350   assert(rscratch == noreg, "redundant");
  351   push_literal32((int32_t)obj, metadata_Relocation::spec_for_immediate());
  352 }
  353 
  354 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  355   assert(rscratch == noreg, "redundant");
  356   if (src.is_lval()) {
  357     push_literal32((int32_t)src.target(), src.rspec());
  358   } else {
  359     pushl(as_Address(src));
  360   }
  361 }
  362 
  363 static void pass_arg0(MacroAssembler* masm, Register arg) {
  364   masm->push(arg);
  365 }
  366 
  367 static void pass_arg1(MacroAssembler* masm, Register arg) {
  368   masm->push(arg);
  369 }
  370 
  371 static void pass_arg2(MacroAssembler* masm, Register arg) {
  372   masm->push(arg);
  373 }
  374 
  375 static void pass_arg3(MacroAssembler* masm, Register arg) {
  376   masm->push(arg);
  377 }
  378 
  379 #ifndef PRODUCT
  380 extern "C" void findpc(intptr_t x);
  381 #endif
  382 
  383 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) {
  384   // In order to get locks to work, we need to fake a in_VM state
  385   JavaThread* thread = JavaThread::current();
  386   JavaThreadState saved_state = thread->thread_state();
  387   thread->set_thread_state(_thread_in_vm);
  388   if (ShowMessageBoxOnError) {
  389     JavaThread* thread = JavaThread::current();
  390     JavaThreadState saved_state = thread->thread_state();
  391     thread->set_thread_state(_thread_in_vm);
  392     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  393       ttyLocker ttyl;
  394       BytecodeCounter::print();
  395     }
  396     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  397     // This is the value of eip which points to where verify_oop will return.
  398     if (os::message_box(msg, "Execution stopped, print registers?")) {
  399       print_state32(rdi, rsi, rbp, rsp, rbx, rdx, rcx, rax, eip);
  400       BREAKPOINT;
  401     }
  402   }
  403   fatal("DEBUG MESSAGE: %s", msg);
  404 }
  405 
  406 void MacroAssembler::print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip) {
  407   ttyLocker ttyl;
  408   DebuggingContext debugging{};
  409   tty->print_cr("eip = 0x%08x", eip);
  410 #ifndef PRODUCT
  411   if ((WizardMode || Verbose) && PrintMiscellaneous) {
  412     tty->cr();
  413     findpc(eip);
  414     tty->cr();
  415   }
  416 #endif
  417 #define PRINT_REG(rax) \
  418   { tty->print("%s = ", #rax); os::print_location(tty, rax); }
  419   PRINT_REG(rax);
  420   PRINT_REG(rbx);
  421   PRINT_REG(rcx);
  422   PRINT_REG(rdx);
  423   PRINT_REG(rdi);
  424   PRINT_REG(rsi);
  425   PRINT_REG(rbp);
  426   PRINT_REG(rsp);
  427 #undef PRINT_REG
  428   // Print some words near top of staack.
  429   int* dump_sp = (int*) rsp;
  430   for (int col1 = 0; col1 < 8; col1++) {
  431     tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  432     os::print_location(tty, *dump_sp++);
  433   }
  434   for (int row = 0; row < 16; row++) {
  435     tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  436     for (int col = 0; col < 8; col++) {
  437       tty->print(" 0x%08x", *dump_sp++);
  438     }
  439     tty->cr();
  440   }
  441   // Print some instructions around pc:
  442   Disassembler::decode((address)eip-64, (address)eip);
  443   tty->print_cr("--------");
  444   Disassembler::decode((address)eip, (address)eip+32);
  445 }
  446 
  447 void MacroAssembler::stop(const char* msg) {
  448   // push address of message
  449   ExternalAddress message((address)msg);
  450   pushptr(message.addr(), noreg);
  451   { Label L; call(L, relocInfo::none); bind(L); }     // push eip
  452   pusha();                                            // push registers
  453   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
  454   hlt();
  455 }
  456 
  457 void MacroAssembler::warn(const char* msg) {
  458   push_CPU_state();
  459 
  460   // push address of message
  461   ExternalAddress message((address)msg);
  462   pushptr(message.addr(), noreg);
  463 
  464   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  465   addl(rsp, wordSize);       // discard argument
  466   pop_CPU_state();
  467 }
  468 
  469 void MacroAssembler::print_state() {
  470   { Label L; call(L, relocInfo::none); bind(L); }     // push eip
  471   pusha();                                            // push registers
  472 
  473   push_CPU_state();
  474   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::print_state32)));
  475   pop_CPU_state();
  476 
  477   popa();
  478   addl(rsp, wordSize);
  479 }
  480 
  481 #else // _LP64
  482 
  483 // 64 bit versions
  484 
  485 Address MacroAssembler::as_Address(AddressLiteral adr) {
  486   // amd64 always does this as a pc-rel
  487   // we can be absolute or disp based on the instruction type
  488   // jmp/call are displacements others are absolute
  489   assert(!adr.is_lval(), "must be rval");
  490   assert(reachable(adr), "must be");
  491   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  492 
  493 }
  494 
  495 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  496   AddressLiteral base = adr.base();
  497   lea(rscratch, base);
  498   Address index = adr.index();
  499   assert(index._disp == 0, "must not have disp"); // maybe it can?
  500   Address array(rscratch, index._index, index._scale, index._disp);
  501   return array;
  502 }
  503 
  504 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  505   Label L, E;
  506 
  507 #ifdef _WIN64
  508   // Windows always allocates space for it's register args
  509   assert(num_args <= 4, "only register arguments supported");
  510   subq(rsp,  frame::arg_reg_save_area_bytes);
  511 #endif
  512 
  513   // Align stack if necessary
  514   testl(rsp, 15);
  515   jcc(Assembler::zero, L);
  516 
  517   subq(rsp, 8);
  518   call(RuntimeAddress(entry_point));
  519   addq(rsp, 8);
  520   jmp(E);
  521 
  522   bind(L);
  523   call(RuntimeAddress(entry_point));
  524 
  525   bind(E);
  526 
  527 #ifdef _WIN64
  528   // restore stack pointer
  529   addq(rsp, frame::arg_reg_save_area_bytes);
  530 #endif
  531 
  532 }
  533 
  534 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  535   assert(!src2.is_lval(), "should use cmpptr");
  536   assert(rscratch != noreg || always_reachable(src2), "missing");
  537 
  538   if (reachable(src2)) {
  539     cmpq(src1, as_Address(src2));
  540   } else {
  541     lea(rscratch, src2);
  542     Assembler::cmpq(src1, Address(rscratch, 0));
  543   }
  544 }
  545 
  546 int MacroAssembler::corrected_idivq(Register reg) {
  547   // Full implementation of Java ldiv and lrem; checks for special
  548   // case as described in JVM spec., p.243 & p.271.  The function
  549   // returns the (pc) offset of the idivl instruction - may be needed
  550   // for implicit exceptions.
  551   //
  552   //         normal case                           special case
  553   //
  554   // input : rax: dividend                         min_long
  555   //         reg: divisor   (may not be eax/edx)   -1
  556   //
  557   // output: rax: quotient  (= rax idiv reg)       min_long
  558   //         rdx: remainder (= rax irem reg)       0
  559   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  560   static const int64_t min_long = 0x8000000000000000;
  561   Label normal_case, special_case;
  562 
  563   // check for special case
  564   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  565   jcc(Assembler::notEqual, normal_case);
  566   xorl(rdx, rdx); // prepare rdx for possible special case (where
  567                   // remainder = 0)
  568   cmpq(reg, -1);
  569   jcc(Assembler::equal, special_case);
  570 
  571   // handle normal case
  572   bind(normal_case);
  573   cdqq();
  574   int idivq_offset = offset();
  575   idivq(reg);
  576 
  577   // normal and special case exit
  578   bind(special_case);
  579 
  580   return idivq_offset;
  581 }
  582 
  583 void MacroAssembler::decrementq(Register reg, int value) {
  584   if (value == min_jint) { subq(reg, value); return; }
  585   if (value <  0) { incrementq(reg, -value); return; }
  586   if (value == 0) {                        ; return; }
  587   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  588   /* else */      { subq(reg, value)       ; return; }
  589 }
  590 
  591 void MacroAssembler::decrementq(Address dst, int value) {
  592   if (value == min_jint) { subq(dst, value); return; }
  593   if (value <  0) { incrementq(dst, -value); return; }
  594   if (value == 0) {                        ; return; }
  595   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  596   /* else */      { subq(dst, value)       ; return; }
  597 }
  598 
  599 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  600   assert(rscratch != noreg || always_reachable(dst), "missing");
  601 
  602   if (reachable(dst)) {
  603     incrementq(as_Address(dst));
  604   } else {
  605     lea(rscratch, dst);
  606     incrementq(Address(rscratch, 0));
  607   }
  608 }
  609 
  610 void MacroAssembler::incrementq(Register reg, int value) {
  611   if (value == min_jint) { addq(reg, value); return; }
  612   if (value <  0) { decrementq(reg, -value); return; }
  613   if (value == 0) {                        ; return; }
  614   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  615   /* else */      { addq(reg, value)       ; return; }
  616 }
  617 
  618 void MacroAssembler::incrementq(Address dst, int value) {
  619   if (value == min_jint) { addq(dst, value); return; }
  620   if (value <  0) { decrementq(dst, -value); return; }
  621   if (value == 0) {                        ; return; }
  622   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  623   /* else */      { addq(dst, value)       ; return; }
  624 }
  625 
  626 // 32bit can do a case table jump in one instruction but we no longer allow the base
  627 // to be installed in the Address class
  628 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  629   lea(rscratch, entry.base());
  630   Address dispatch = entry.index();
  631   assert(dispatch._base == noreg, "must be");
  632   dispatch._base = rscratch;
  633   jmp(dispatch);
  634 }
  635 
  636 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  637   ShouldNotReachHere(); // 64bit doesn't use two regs
  638   cmpq(x_lo, y_lo);
  639 }
  640 
  641 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  642   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  643 }
  644 
  645 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  646   lea(rscratch, adr);
  647   movptr(dst, rscratch);
  648 }
  649 
  650 void MacroAssembler::leave() {
  651   // %%% is this really better? Why not on 32bit too?
  652   emit_int8((unsigned char)0xC9); // LEAVE
  653 }
  654 
  655 void MacroAssembler::lneg(Register hi, Register lo) {
  656   ShouldNotReachHere(); // 64bit doesn't use two regs
  657   negq(lo);
  658 }
  659 
  660 void MacroAssembler::movoop(Register dst, jobject obj) {
  661   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  662 }
  663 
  664 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  665   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  666   movq(dst, rscratch);
  667 }
  668 
  669 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  670   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  671 }
  672 
  673 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  674   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  675   movq(dst, rscratch);
  676 }
  677 
  678 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  679   if (src.is_lval()) {
  680     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  681   } else {
  682     if (reachable(src)) {
  683       movq(dst, as_Address(src));
  684     } else {
  685       lea(dst, src);
  686       movq(dst, Address(dst, 0));
  687     }
  688   }
  689 }
  690 
  691 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  692   movq(as_Address(dst, rscratch), src);
  693 }
  694 
  695 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  696   movq(dst, as_Address(src, dst /*rscratch*/));
  697 }
  698 
  699 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  700 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  701   if (is_simm32(src)) {
  702     movptr(dst, checked_cast<int32_t>(src));
  703   } else {
  704     mov64(rscratch, src);
  705     movq(dst, rscratch);
  706   }
  707 }
  708 
  709 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  710   movoop(rscratch, obj);
  711   push(rscratch);
  712 }
  713 
  714 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  715   mov_metadata(rscratch, obj);
  716   push(rscratch);
  717 }
  718 
  719 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  720   lea(rscratch, src);
  721   if (src.is_lval()) {
  722     push(rscratch);
  723   } else {
  724     pushq(Address(rscratch, 0));
  725   }
  726 }
  727 
  728 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
  729   reset_last_Java_frame(r15_thread, clear_fp);
  730 }
  731 
  732 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
  733                                          Register last_java_fp,
  734                                          address  last_java_pc,
  735                                          Register rscratch) {
  736   set_last_Java_frame(r15_thread, last_java_sp, last_java_fp, last_java_pc, rscratch);
  737 }
  738 
  739 static void pass_arg0(MacroAssembler* masm, Register arg) {
  740   if (c_rarg0 != arg ) {
  741     masm->mov(c_rarg0, arg);
  742   }
  743 }
  744 
  745 static void pass_arg1(MacroAssembler* masm, Register arg) {
  746   if (c_rarg1 != arg ) {
  747     masm->mov(c_rarg1, arg);
  748   }
  749 }
  750 
  751 static void pass_arg2(MacroAssembler* masm, Register arg) {
  752   if (c_rarg2 != arg ) {
  753     masm->mov(c_rarg2, arg);
  754   }
  755 }
  756 
  757 static void pass_arg3(MacroAssembler* masm, Register arg) {
  758   if (c_rarg3 != arg ) {
  759     masm->mov(c_rarg3, arg);
  760   }
  761 }
  762 
  763 void MacroAssembler::stop(const char* msg) {
  764   if (ShowMessageBoxOnError) {
  765     address rip = pc();
  766     pusha(); // get regs on stack
  767     lea(c_rarg1, InternalAddress(rip));
  768     movq(c_rarg2, rsp); // pass pointer to regs array
  769   }
  770   lea(c_rarg0, ExternalAddress((address) msg));
  771   andq(rsp, -16); // align stack as required by ABI
  772   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  773   hlt();
  774 }
  775 
  776 void MacroAssembler::warn(const char* msg) {
  777   push(rbp);
  778   movq(rbp, rsp);
  779   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  780   push_CPU_state();   // keeps alignment at 16 bytes
  781 
  782   lea(c_rarg0, ExternalAddress((address) msg));
  783   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  784 
  785   pop_CPU_state();
  786   mov(rsp, rbp);
  787   pop(rbp);
  788 }
  789 
  790 void MacroAssembler::print_state() {
  791   address rip = pc();
  792   pusha();            // get regs on stack
  793   push(rbp);
  794   movq(rbp, rsp);
  795   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  796   push_CPU_state();   // keeps alignment at 16 bytes
  797 
  798   lea(c_rarg0, InternalAddress(rip));
  799   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  800   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  801 
  802   pop_CPU_state();
  803   mov(rsp, rbp);
  804   pop(rbp);
  805   popa();
  806 }
  807 
  808 #ifndef PRODUCT
  809 extern "C" void findpc(intptr_t x);
  810 #endif
  811 
  812 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  813   // In order to get locks to work, we need to fake a in_VM state
  814   if (ShowMessageBoxOnError) {
  815     JavaThread* thread = JavaThread::current();
  816     JavaThreadState saved_state = thread->thread_state();
  817     thread->set_thread_state(_thread_in_vm);
  818 #ifndef PRODUCT
  819     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  820       ttyLocker ttyl;
  821       BytecodeCounter::print();
  822     }
  823 #endif
  824     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  825     // XXX correct this offset for amd64
  826     // This is the value of eip which points to where verify_oop will return.
  827     if (os::message_box(msg, "Execution stopped, print registers?")) {
  828       print_state64(pc, regs);
  829       BREAKPOINT;
  830     }
  831   }
  832   fatal("DEBUG MESSAGE: %s", msg);
  833 }
  834 
  835 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  836   ttyLocker ttyl;
  837   DebuggingContext debugging{};
  838   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  839 #ifndef PRODUCT
  840   tty->cr();
  841   findpc(pc);
  842   tty->cr();
  843 #endif
  844 #define PRINT_REG(rax, value) \
  845   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  846   PRINT_REG(rax, regs[15]);
  847   PRINT_REG(rbx, regs[12]);
  848   PRINT_REG(rcx, regs[14]);
  849   PRINT_REG(rdx, regs[13]);
  850   PRINT_REG(rdi, regs[8]);
  851   PRINT_REG(rsi, regs[9]);
  852   PRINT_REG(rbp, regs[10]);
  853   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  854   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  855   PRINT_REG(r8 , regs[7]);
  856   PRINT_REG(r9 , regs[6]);
  857   PRINT_REG(r10, regs[5]);
  858   PRINT_REG(r11, regs[4]);
  859   PRINT_REG(r12, regs[3]);
  860   PRINT_REG(r13, regs[2]);
  861   PRINT_REG(r14, regs[1]);
  862   PRINT_REG(r15, regs[0]);
  863 #undef PRINT_REG
  864   // Print some words near the top of the stack.
  865   int64_t* rsp = &regs[16];
  866   int64_t* dump_sp = rsp;
  867   for (int col1 = 0; col1 < 8; col1++) {
  868     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  869     os::print_location(tty, *dump_sp++);
  870   }
  871   for (int row = 0; row < 25; row++) {
  872     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  873     for (int col = 0; col < 4; col++) {
  874       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  875     }
  876     tty->cr();
  877   }
  878   // Print some instructions around pc:
  879   Disassembler::decode((address)pc-64, (address)pc);
  880   tty->print_cr("--------");
  881   Disassembler::decode((address)pc, (address)pc+32);
  882 }
  883 
  884 // The java_calling_convention describes stack locations as ideal slots on
  885 // a frame with no abi restrictions. Since we must observe abi restrictions
  886 // (like the placement of the register window) the slots must be biased by
  887 // the following value.
  888 static int reg2offset_in(VMReg r) {
  889   // Account for saved rbp and return address
  890   // This should really be in_preserve_stack_slots
  891   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  892 }
  893 
  894 static int reg2offset_out(VMReg r) {
  895   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  896 }
  897 
  898 // A long move
  899 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  900 
  901   // The calling conventions assures us that each VMregpair is either
  902   // all really one physical register or adjacent stack slots.
  903 
  904   if (src.is_single_phys_reg() ) {
  905     if (dst.is_single_phys_reg()) {
  906       if (dst.first() != src.first()) {
  907         mov(dst.first()->as_Register(), src.first()->as_Register());
  908       }
  909     } else {
  910       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  911              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  912       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  913     }
  914   } else if (dst.is_single_phys_reg()) {
  915     assert(src.is_single_reg(),  "not a stack pair");
  916     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  917   } else {
  918     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  919     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  920     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  921   }
  922 }
  923 
  924 // A double move
  925 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  926 
  927   // The calling conventions assures us that each VMregpair is either
  928   // all really one physical register or adjacent stack slots.
  929 
  930   if (src.is_single_phys_reg() ) {
  931     if (dst.is_single_phys_reg()) {
  932       // In theory these overlap but the ordering is such that this is likely a nop
  933       if ( src.first() != dst.first()) {
  934         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  935       }
  936     } else {
  937       assert(dst.is_single_reg(), "not a stack pair");
  938       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  939     }
  940   } else if (dst.is_single_phys_reg()) {
  941     assert(src.is_single_reg(),  "not a stack pair");
  942     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  943   } else {
  944     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  945     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  946     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  947   }
  948 }
  949 
  950 
  951 // A float arg may have to do float reg int reg conversion
  952 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  953   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  954 
  955   // The calling conventions assures us that each VMregpair is either
  956   // all really one physical register or adjacent stack slots.
  957 
  958   if (src.first()->is_stack()) {
  959     if (dst.first()->is_stack()) {
  960       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  961       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  962     } else {
  963       // stack to reg
  964       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  965       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  966     }
  967   } else if (dst.first()->is_stack()) {
  968     // reg to stack
  969     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  970     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  971   } else {
  972     // reg to reg
  973     // In theory these overlap but the ordering is such that this is likely a nop
  974     if ( src.first() != dst.first()) {
  975       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  976     }
  977   }
  978 }
  979 
  980 // On 64 bit we will store integer like items to the stack as
  981 // 64 bits items (x86_32/64 abi) even though java would only store
  982 // 32bits for a parameter. On 32bit it will simply be 32 bits
  983 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  984 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  985   if (src.first()->is_stack()) {
  986     if (dst.first()->is_stack()) {
  987       // stack to stack
  988       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  989       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  990     } else {
  991       // stack to reg
  992       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  993     }
  994   } else if (dst.first()->is_stack()) {
  995     // reg to stack
  996     // Do we really have to sign extend???
  997     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  998     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  999   } else {
 1000     // Do we really have to sign extend???
 1001     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
 1002     if (dst.first() != src.first()) {
 1003       movq(dst.first()->as_Register(), src.first()->as_Register());
 1004     }
 1005   }
 1006 }
 1007 
 1008 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
 1009   if (src.first()->is_stack()) {
 1010     if (dst.first()->is_stack()) {
 1011       // stack to stack
 1012       movq(rax, Address(rbp, reg2offset_in(src.first())));
 1013       movq(Address(rsp, reg2offset_out(dst.first())), rax);
 1014     } else {
 1015       // stack to reg
 1016       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
 1017     }
 1018   } else if (dst.first()->is_stack()) {
 1019     // reg to stack
 1020     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
 1021   } else {
 1022     if (dst.first() != src.first()) {
 1023       movq(dst.first()->as_Register(), src.first()->as_Register());
 1024     }
 1025   }
 1026 }
 1027 
 1028 // An oop arg. Must pass a handle not the oop itself
 1029 void MacroAssembler::object_move(OopMap* map,
 1030                         int oop_handle_offset,
 1031                         int framesize_in_slots,
 1032                         VMRegPair src,
 1033                         VMRegPair dst,
 1034                         bool is_receiver,
 1035                         int* receiver_offset) {
 1036 
 1037   // must pass a handle. First figure out the location we use as a handle
 1038 
 1039   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
 1040 
 1041   // See if oop is null if it is we need no handle
 1042 
 1043   if (src.first()->is_stack()) {
 1044 
 1045     // Oop is already on the stack as an argument
 1046     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
 1047     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
 1048     if (is_receiver) {
 1049       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
 1050     }
 1051 
 1052     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
 1053     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
 1054     // conditionally move a null
 1055     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
 1056   } else {
 1057 
 1058     // Oop is in a register we must store it to the space we reserve
 1059     // on the stack for oop_handles and pass a handle if oop is non-null
 1060 
 1061     const Register rOop = src.first()->as_Register();
 1062     int oop_slot;
 1063     if (rOop == j_rarg0)
 1064       oop_slot = 0;
 1065     else if (rOop == j_rarg1)
 1066       oop_slot = 1;
 1067     else if (rOop == j_rarg2)
 1068       oop_slot = 2;
 1069     else if (rOop == j_rarg3)
 1070       oop_slot = 3;
 1071     else if (rOop == j_rarg4)
 1072       oop_slot = 4;
 1073     else {
 1074       assert(rOop == j_rarg5, "wrong register");
 1075       oop_slot = 5;
 1076     }
 1077 
 1078     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
 1079     int offset = oop_slot*VMRegImpl::stack_slot_size;
 1080 
 1081     map->set_oop(VMRegImpl::stack2reg(oop_slot));
 1082     // Store oop in handle area, may be null
 1083     movptr(Address(rsp, offset), rOop);
 1084     if (is_receiver) {
 1085       *receiver_offset = offset;
 1086     }
 1087 
 1088     cmpptr(rOop, NULL_WORD);
 1089     lea(rHandle, Address(rsp, offset));
 1090     // conditionally move a null from the handle area where it was just stored
 1091     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
 1092   }
 1093 
 1094   // If arg is on the stack then place it otherwise it is already in correct reg.
 1095   if (dst.first()->is_stack()) {
 1096     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
 1097   }
 1098 }
 1099 
 1100 #endif // _LP64
 1101 
 1102 // Now versions that are common to 32/64 bit
 1103 
 1104 void MacroAssembler::addptr(Register dst, int32_t imm32) {
 1105   LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32));
 1106 }
 1107 
 1108 void MacroAssembler::addptr(Register dst, Register src) {
 1109   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
 1110 }
 1111 
 1112 void MacroAssembler::addptr(Address dst, Register src) {
 1113   LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
 1114 }
 1115 
 1116 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1117   assert(rscratch != noreg || always_reachable(src), "missing");
 1118 
 1119   if (reachable(src)) {
 1120     Assembler::addsd(dst, as_Address(src));
 1121   } else {
 1122     lea(rscratch, src);
 1123     Assembler::addsd(dst, Address(rscratch, 0));
 1124   }
 1125 }
 1126 
 1127 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1128   assert(rscratch != noreg || always_reachable(src), "missing");
 1129 
 1130   if (reachable(src)) {
 1131     addss(dst, as_Address(src));
 1132   } else {
 1133     lea(rscratch, src);
 1134     addss(dst, Address(rscratch, 0));
 1135   }
 1136 }
 1137 
 1138 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1139   assert(rscratch != noreg || always_reachable(src), "missing");
 1140 
 1141   if (reachable(src)) {
 1142     Assembler::addpd(dst, as_Address(src));
 1143   } else {
 1144     lea(rscratch, src);
 1145     Assembler::addpd(dst, Address(rscratch, 0));
 1146   }
 1147 }
 1148 
 1149 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
 1150 // Stub code is generated once and never copied.
 1151 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
 1152 void MacroAssembler::align64() {
 1153   align(64, (unsigned long long) pc());
 1154 }
 1155 
 1156 void MacroAssembler::align32() {
 1157   align(32, (unsigned long long) pc());
 1158 }
 1159 
 1160 void MacroAssembler::align(int modulus) {
 1161   // 8273459: Ensure alignment is possible with current segment alignment
 1162   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
 1163   align(modulus, offset());
 1164 }
 1165 
 1166 void MacroAssembler::align(int modulus, int target) {
 1167   if (target % modulus != 0) {
 1168     nop(modulus - (target % modulus));
 1169   }
 1170 }
 1171 
 1172 void MacroAssembler::push_f(XMMRegister r) {
 1173   subptr(rsp, wordSize);
 1174   movflt(Address(rsp, 0), r);
 1175 }
 1176 
 1177 void MacroAssembler::pop_f(XMMRegister r) {
 1178   movflt(r, Address(rsp, 0));
 1179   addptr(rsp, wordSize);
 1180 }
 1181 
 1182 void MacroAssembler::push_d(XMMRegister r) {
 1183   subptr(rsp, 2 * wordSize);
 1184   movdbl(Address(rsp, 0), r);
 1185 }
 1186 
 1187 void MacroAssembler::pop_d(XMMRegister r) {
 1188   movdbl(r, Address(rsp, 0));
 1189   addptr(rsp, 2 * Interpreter::stackElementSize);
 1190 }
 1191 
 1192 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1193   // Used in sign-masking with aligned address.
 1194   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 1195   assert(rscratch != noreg || always_reachable(src), "missing");
 1196 
 1197   if (reachable(src)) {
 1198     Assembler::andpd(dst, as_Address(src));
 1199   } else {
 1200     lea(rscratch, src);
 1201     Assembler::andpd(dst, Address(rscratch, 0));
 1202   }
 1203 }
 1204 
 1205 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1206   // Used in sign-masking with aligned address.
 1207   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 1208   assert(rscratch != noreg || always_reachable(src), "missing");
 1209 
 1210   if (reachable(src)) {
 1211     Assembler::andps(dst, as_Address(src));
 1212   } else {
 1213     lea(rscratch, src);
 1214     Assembler::andps(dst, Address(rscratch, 0));
 1215   }
 1216 }
 1217 
 1218 void MacroAssembler::andptr(Register dst, int32_t imm32) {
 1219   LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32));
 1220 }
 1221 
 1222 #ifdef _LP64
 1223 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
 1224   assert(rscratch != noreg || always_reachable(src), "missing");
 1225 
 1226   if (reachable(src)) {
 1227     andq(dst, as_Address(src));
 1228   } else {
 1229     lea(rscratch, src);
 1230     andq(dst, Address(rscratch, 0));
 1231   }
 1232 }
 1233 #endif
 1234 
 1235 void MacroAssembler::atomic_incl(Address counter_addr) {
 1236   lock();
 1237   incrementl(counter_addr);
 1238 }
 1239 
 1240 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
 1241   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1242 
 1243   if (reachable(counter_addr)) {
 1244     atomic_incl(as_Address(counter_addr));
 1245   } else {
 1246     lea(rscratch, counter_addr);
 1247     atomic_incl(Address(rscratch, 0));
 1248   }
 1249 }
 1250 
 1251 #ifdef _LP64
 1252 void MacroAssembler::atomic_incq(Address counter_addr) {
 1253   lock();
 1254   incrementq(counter_addr);
 1255 }
 1256 
 1257 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
 1258   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1259 
 1260   if (reachable(counter_addr)) {
 1261     atomic_incq(as_Address(counter_addr));
 1262   } else {
 1263     lea(rscratch, counter_addr);
 1264     atomic_incq(Address(rscratch, 0));
 1265   }
 1266 }
 1267 #endif
 1268 
 1269 // Writes to stack successive pages until offset reached to check for
 1270 // stack overflow + shadow pages.  This clobbers tmp.
 1271 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
 1272   movptr(tmp, rsp);
 1273   // Bang stack for total size given plus shadow page size.
 1274   // Bang one page at a time because large size can bang beyond yellow and
 1275   // red zones.
 1276   Label loop;
 1277   bind(loop);
 1278   movl(Address(tmp, (-(int)os::vm_page_size())), size );
 1279   subptr(tmp, (int)os::vm_page_size());
 1280   subl(size, (int)os::vm_page_size());
 1281   jcc(Assembler::greater, loop);
 1282 
 1283   // Bang down shadow pages too.
 1284   // At this point, (tmp-0) is the last address touched, so don't
 1285   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
 1286   // was post-decremented.)  Skip this address by starting at i=1, and
 1287   // touch a few more pages below.  N.B.  It is important to touch all
 1288   // the way down including all pages in the shadow zone.
 1289   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
 1290     // this could be any sized move but this is can be a debugging crumb
 1291     // so the bigger the better.
 1292     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
 1293   }
 1294 }
 1295 
 1296 void MacroAssembler::reserved_stack_check() {
 1297   // testing if reserved zone needs to be enabled
 1298   Label no_reserved_zone_enabling;
 1299   Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 1300   NOT_LP64(get_thread(rsi);)
 1301 
 1302   cmpptr(rsp, Address(thread, JavaThread::reserved_stack_activation_offset()));
 1303   jcc(Assembler::below, no_reserved_zone_enabling);
 1304 
 1305   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), thread);
 1306   jump(RuntimeAddress(StubRoutines::throw_delayed_StackOverflowError_entry()));
 1307   should_not_reach_here();
 1308 
 1309   bind(no_reserved_zone_enabling);
 1310 }
 1311 
 1312 void MacroAssembler::c2bool(Register x) {
 1313   // implements x == 0 ? 0 : 1
 1314   // note: must only look at least-significant byte of x
 1315   //       since C-style booleans are stored in one byte
 1316   //       only! (was bug)
 1317   andl(x, 0xFF);
 1318   setb(Assembler::notZero, x);
 1319 }
 1320 
 1321 // Wouldn't need if AddressLiteral version had new name
 1322 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
 1323   Assembler::call(L, rtype);
 1324 }
 1325 
 1326 void MacroAssembler::call(Register entry) {
 1327   Assembler::call(entry);
 1328 }
 1329 
 1330 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
 1331   assert(rscratch != noreg || always_reachable(entry), "missing");
 1332 
 1333   if (reachable(entry)) {
 1334     Assembler::call_literal(entry.target(), entry.rspec());
 1335   } else {
 1336     lea(rscratch, entry);
 1337     Assembler::call(rscratch);
 1338   }
 1339 }
 1340 
 1341 void MacroAssembler::ic_call(address entry, jint method_index) {
 1342   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 1343 #ifdef _LP64
 1344   // Needs full 64-bit immediate for later patching.
 1345   mov64(rax, (int64_t)Universe::non_oop_word());
 1346 #else
 1347   movptr(rax, (intptr_t)Universe::non_oop_word());
 1348 #endif
 1349   call(AddressLiteral(entry, rh));
 1350 }
 1351 
 1352 int MacroAssembler::ic_check_size() {
 1353   return LP64_ONLY(14) NOT_LP64(12);
 1354 }
 1355 
 1356 int MacroAssembler::ic_check(int end_alignment) {
 1357   Register receiver = LP64_ONLY(j_rarg0) NOT_LP64(rcx);
 1358   Register data = rax;
 1359   Register temp = LP64_ONLY(rscratch1) NOT_LP64(rbx);
 1360 
 1361   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 1362   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 1363   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
 1364   // before the inline cache check here, and not after
 1365   align(end_alignment, offset() + ic_check_size());
 1366 
 1367   int uep_offset = offset();
 1368 
 1369   if (UseCompressedClassPointers) {
 1370     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1371     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1372   } else {
 1373     movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1374     cmpptr(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1375   }
 1376 
 1377   // if inline cache check fails, then jump to runtime routine
 1378   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 1379   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
 1380 
 1381   return uep_offset;
 1382 }
 1383 
 1384 void MacroAssembler::emit_static_call_stub() {
 1385   // Static stub relocation also tags the Method* in the code-stream.
 1386   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1387   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1388   jump(RuntimeAddress(pc()));
 1389 }
 1390 
 1391 // Implementation of call_VM versions
 1392 
 1393 void MacroAssembler::call_VM(Register oop_result,
 1394                              address entry_point,
 1395                              bool check_exceptions) {
 1396   Label C, E;
 1397   call(C, relocInfo::none);
 1398   jmp(E);
 1399 
 1400   bind(C);
 1401   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1402   ret(0);
 1403 
 1404   bind(E);
 1405 }
 1406 
 1407 void MacroAssembler::call_VM(Register oop_result,
 1408                              address entry_point,
 1409                              Register arg_1,
 1410                              bool check_exceptions) {
 1411   Label C, E;
 1412   call(C, relocInfo::none);
 1413   jmp(E);
 1414 
 1415   bind(C);
 1416   pass_arg1(this, arg_1);
 1417   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1418   ret(0);
 1419 
 1420   bind(E);
 1421 }
 1422 
 1423 void MacroAssembler::call_VM(Register oop_result,
 1424                              address entry_point,
 1425                              Register arg_1,
 1426                              Register arg_2,
 1427                              bool check_exceptions) {
 1428   Label C, E;
 1429   call(C, relocInfo::none);
 1430   jmp(E);
 1431 
 1432   bind(C);
 1433 
 1434   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1435 
 1436   pass_arg2(this, arg_2);
 1437   pass_arg1(this, arg_1);
 1438   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1439   ret(0);
 1440 
 1441   bind(E);
 1442 }
 1443 
 1444 void MacroAssembler::call_VM(Register oop_result,
 1445                              address entry_point,
 1446                              Register arg_1,
 1447                              Register arg_2,
 1448                              Register arg_3,
 1449                              bool check_exceptions) {
 1450   Label C, E;
 1451   call(C, relocInfo::none);
 1452   jmp(E);
 1453 
 1454   bind(C);
 1455 
 1456   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1457   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1458   pass_arg3(this, arg_3);
 1459   pass_arg2(this, arg_2);
 1460   pass_arg1(this, arg_1);
 1461   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1462   ret(0);
 1463 
 1464   bind(E);
 1465 }
 1466 
 1467 void MacroAssembler::call_VM(Register oop_result,
 1468                              Register last_java_sp,
 1469                              address entry_point,
 1470                              int number_of_arguments,
 1471                              bool check_exceptions) {
 1472   Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
 1473   call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1474 }
 1475 
 1476 void MacroAssembler::call_VM(Register oop_result,
 1477                              Register last_java_sp,
 1478                              address entry_point,
 1479                              Register arg_1,
 1480                              bool check_exceptions) {
 1481   pass_arg1(this, arg_1);
 1482   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1483 }
 1484 
 1485 void MacroAssembler::call_VM(Register oop_result,
 1486                              Register last_java_sp,
 1487                              address entry_point,
 1488                              Register arg_1,
 1489                              Register arg_2,
 1490                              bool check_exceptions) {
 1491 
 1492   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1493   pass_arg2(this, arg_2);
 1494   pass_arg1(this, arg_1);
 1495   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1496 }
 1497 
 1498 void MacroAssembler::call_VM(Register oop_result,
 1499                              Register last_java_sp,
 1500                              address entry_point,
 1501                              Register arg_1,
 1502                              Register arg_2,
 1503                              Register arg_3,
 1504                              bool check_exceptions) {
 1505   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1506   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1507   pass_arg3(this, arg_3);
 1508   pass_arg2(this, arg_2);
 1509   pass_arg1(this, arg_1);
 1510   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1511 }
 1512 
 1513 void MacroAssembler::super_call_VM(Register oop_result,
 1514                                    Register last_java_sp,
 1515                                    address entry_point,
 1516                                    int number_of_arguments,
 1517                                    bool check_exceptions) {
 1518   Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
 1519   MacroAssembler::call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1520 }
 1521 
 1522 void MacroAssembler::super_call_VM(Register oop_result,
 1523                                    Register last_java_sp,
 1524                                    address entry_point,
 1525                                    Register arg_1,
 1526                                    bool check_exceptions) {
 1527   pass_arg1(this, arg_1);
 1528   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1529 }
 1530 
 1531 void MacroAssembler::super_call_VM(Register oop_result,
 1532                                    Register last_java_sp,
 1533                                    address entry_point,
 1534                                    Register arg_1,
 1535                                    Register arg_2,
 1536                                    bool check_exceptions) {
 1537 
 1538   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1539   pass_arg2(this, arg_2);
 1540   pass_arg1(this, arg_1);
 1541   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1542 }
 1543 
 1544 void MacroAssembler::super_call_VM(Register oop_result,
 1545                                    Register last_java_sp,
 1546                                    address entry_point,
 1547                                    Register arg_1,
 1548                                    Register arg_2,
 1549                                    Register arg_3,
 1550                                    bool check_exceptions) {
 1551   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1552   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1553   pass_arg3(this, arg_3);
 1554   pass_arg2(this, arg_2);
 1555   pass_arg1(this, arg_1);
 1556   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1557 }
 1558 
 1559 void MacroAssembler::call_VM_base(Register oop_result,
 1560                                   Register java_thread,
 1561                                   Register last_java_sp,
 1562                                   address  entry_point,
 1563                                   int      number_of_arguments,
 1564                                   bool     check_exceptions) {
 1565   // determine java_thread register
 1566   if (!java_thread->is_valid()) {
 1567 #ifdef _LP64
 1568     java_thread = r15_thread;
 1569 #else
 1570     java_thread = rdi;
 1571     get_thread(java_thread);
 1572 #endif // LP64
 1573   }
 1574   // determine last_java_sp register
 1575   if (!last_java_sp->is_valid()) {
 1576     last_java_sp = rsp;
 1577   }
 1578   // debugging support
 1579   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1580   LP64_ONLY(assert(java_thread == r15_thread, "unexpected register"));
 1581 #ifdef ASSERT
 1582   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1583   // r12 is the heapbase.
 1584   LP64_ONLY(if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");)
 1585 #endif // ASSERT
 1586 
 1587   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1588   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1589 
 1590   // push java thread (becomes first argument of C function)
 1591 
 1592   NOT_LP64(push(java_thread); number_of_arguments++);
 1593   LP64_ONLY(mov(c_rarg0, r15_thread));
 1594 
 1595   // set last Java frame before call
 1596   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1597 
 1598   // Only interpreter should have to set fp
 1599   set_last_Java_frame(java_thread, last_java_sp, rbp, nullptr, rscratch1);
 1600 
 1601   // do the call, remove parameters
 1602   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1603 
 1604   // restore the thread (cannot use the pushed argument since arguments
 1605   // may be overwritten by C code generated by an optimizing compiler);
 1606   // however can use the register value directly if it is callee saved.
 1607   if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) {
 1608     // rdi & rsi (also r15) are callee saved -> nothing to do
 1609 #ifdef ASSERT
 1610     guarantee(java_thread != rax, "change this code");
 1611     push(rax);
 1612     { Label L;
 1613       get_thread(rax);
 1614       cmpptr(java_thread, rax);
 1615       jcc(Assembler::equal, L);
 1616       STOP("MacroAssembler::call_VM_base: rdi not callee saved?");
 1617       bind(L);
 1618     }
 1619     pop(rax);
 1620 #endif
 1621   } else {
 1622     get_thread(java_thread);
 1623   }
 1624   // reset last Java frame
 1625   // Only interpreter should have to clear fp
 1626   reset_last_Java_frame(java_thread, true);
 1627 
 1628    // C++ interp handles this in the interpreter
 1629   check_and_handle_popframe(java_thread);
 1630   check_and_handle_earlyret(java_thread);
 1631 
 1632   if (check_exceptions) {
 1633     // check for pending exceptions (java_thread is set upon return)
 1634     cmpptr(Address(java_thread, Thread::pending_exception_offset()), NULL_WORD);
 1635 #ifndef _LP64
 1636     jump_cc(Assembler::notEqual,
 1637             RuntimeAddress(StubRoutines::forward_exception_entry()));
 1638 #else
 1639     // This used to conditionally jump to forward_exception however it is
 1640     // possible if we relocate that the branch will not reach. So we must jump
 1641     // around so we can always reach
 1642 
 1643     Label ok;
 1644     jcc(Assembler::equal, ok);
 1645     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1646     bind(ok);
 1647 #endif // LP64
 1648   }
 1649 
 1650   // get oop result if there is one and reset the value in the thread
 1651   if (oop_result->is_valid()) {
 1652     get_vm_result(oop_result, java_thread);
 1653   }
 1654 }
 1655 
 1656 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1657 
 1658   // Calculate the value for last_Java_sp
 1659   // somewhat subtle. call_VM does an intermediate call
 1660   // which places a return address on the stack just under the
 1661   // stack pointer as the user finished with it. This allows
 1662   // use to retrieve last_Java_pc from last_Java_sp[-1].
 1663   // On 32bit we then have to push additional args on the stack to accomplish
 1664   // the actual requested call. On 64bit call_VM only can use register args
 1665   // so the only extra space is the return address that call_VM created.
 1666   // This hopefully explains the calculations here.
 1667 
 1668 #ifdef _LP64
 1669   // We've pushed one address, correct last_Java_sp
 1670   lea(rax, Address(rsp, wordSize));
 1671 #else
 1672   lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize));
 1673 #endif // LP64
 1674 
 1675   call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions);
 1676 
 1677 }
 1678 
 1679 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1680 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1681   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1682 }
 1683 
 1684 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1685   call_VM_leaf_base(entry_point, number_of_arguments);
 1686 }
 1687 
 1688 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1689   pass_arg0(this, arg_0);
 1690   call_VM_leaf(entry_point, 1);
 1691 }
 1692 
 1693 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1694 
 1695   LP64_ONLY(assert_different_registers(arg_0, c_rarg1));
 1696   pass_arg1(this, arg_1);
 1697   pass_arg0(this, arg_0);
 1698   call_VM_leaf(entry_point, 2);
 1699 }
 1700 
 1701 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1702   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1703   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1704   pass_arg2(this, arg_2);
 1705   pass_arg1(this, arg_1);
 1706   pass_arg0(this, arg_0);
 1707   call_VM_leaf(entry_point, 3);
 1708 }
 1709 
 1710 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1711   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3));
 1712   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1713   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1714   pass_arg3(this, arg_3);
 1715   pass_arg2(this, arg_2);
 1716   pass_arg1(this, arg_1);
 1717   pass_arg0(this, arg_0);
 1718   call_VM_leaf(entry_point, 3);
 1719 }
 1720 
 1721 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1722   pass_arg0(this, arg_0);
 1723   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1724 }
 1725 
 1726 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1727   LP64_ONLY(assert_different_registers(arg_0, c_rarg1));
 1728   pass_arg1(this, arg_1);
 1729   pass_arg0(this, arg_0);
 1730   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1731 }
 1732 
 1733 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1734   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2));
 1735   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 1736   pass_arg2(this, arg_2);
 1737   pass_arg1(this, arg_1);
 1738   pass_arg0(this, arg_0);
 1739   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1740 }
 1741 
 1742 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1743   LP64_ONLY(assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3));
 1744   LP64_ONLY(assert_different_registers(arg_1, c_rarg2, c_rarg3));
 1745   LP64_ONLY(assert_different_registers(arg_2, c_rarg3));
 1746   pass_arg3(this, arg_3);
 1747   pass_arg2(this, arg_2);
 1748   pass_arg1(this, arg_1);
 1749   pass_arg0(this, arg_0);
 1750   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1751 }
 1752 
 1753 void MacroAssembler::get_vm_result(Register oop_result, Register java_thread) {
 1754   movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
 1755   movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD);
 1756   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1757 }
 1758 
 1759 void MacroAssembler::get_vm_result_2(Register metadata_result, Register java_thread) {
 1760   movptr(metadata_result, Address(java_thread, JavaThread::vm_result_2_offset()));
 1761   movptr(Address(java_thread, JavaThread::vm_result_2_offset()), NULL_WORD);
 1762 }
 1763 
 1764 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
 1765 }
 1766 
 1767 void MacroAssembler::check_and_handle_popframe(Register java_thread) {
 1768 }
 1769 
 1770 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1771   assert(rscratch != noreg || always_reachable(src1), "missing");
 1772 
 1773   if (reachable(src1)) {
 1774     cmpl(as_Address(src1), imm);
 1775   } else {
 1776     lea(rscratch, src1);
 1777     cmpl(Address(rscratch, 0), imm);
 1778   }
 1779 }
 1780 
 1781 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1782   assert(!src2.is_lval(), "use cmpptr");
 1783   assert(rscratch != noreg || always_reachable(src2), "missing");
 1784 
 1785   if (reachable(src2)) {
 1786     cmpl(src1, as_Address(src2));
 1787   } else {
 1788     lea(rscratch, src2);
 1789     cmpl(src1, Address(rscratch, 0));
 1790   }
 1791 }
 1792 
 1793 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1794   Assembler::cmpl(src1, imm);
 1795 }
 1796 
 1797 void MacroAssembler::cmp32(Register src1, Address src2) {
 1798   Assembler::cmpl(src1, src2);
 1799 }
 1800 
 1801 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1802   ucomisd(opr1, opr2);
 1803 
 1804   Label L;
 1805   if (unordered_is_less) {
 1806     movl(dst, -1);
 1807     jcc(Assembler::parity, L);
 1808     jcc(Assembler::below , L);
 1809     movl(dst, 0);
 1810     jcc(Assembler::equal , L);
 1811     increment(dst);
 1812   } else { // unordered is greater
 1813     movl(dst, 1);
 1814     jcc(Assembler::parity, L);
 1815     jcc(Assembler::above , L);
 1816     movl(dst, 0);
 1817     jcc(Assembler::equal , L);
 1818     decrementl(dst);
 1819   }
 1820   bind(L);
 1821 }
 1822 
 1823 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1824   ucomiss(opr1, opr2);
 1825 
 1826   Label L;
 1827   if (unordered_is_less) {
 1828     movl(dst, -1);
 1829     jcc(Assembler::parity, L);
 1830     jcc(Assembler::below , L);
 1831     movl(dst, 0);
 1832     jcc(Assembler::equal , L);
 1833     increment(dst);
 1834   } else { // unordered is greater
 1835     movl(dst, 1);
 1836     jcc(Assembler::parity, L);
 1837     jcc(Assembler::above , L);
 1838     movl(dst, 0);
 1839     jcc(Assembler::equal , L);
 1840     decrementl(dst);
 1841   }
 1842   bind(L);
 1843 }
 1844 
 1845 
 1846 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1847   assert(rscratch != noreg || always_reachable(src1), "missing");
 1848 
 1849   if (reachable(src1)) {
 1850     cmpb(as_Address(src1), imm);
 1851   } else {
 1852     lea(rscratch, src1);
 1853     cmpb(Address(rscratch, 0), imm);
 1854   }
 1855 }
 1856 
 1857 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1858 #ifdef _LP64
 1859   assert(rscratch != noreg || always_reachable(src2), "missing");
 1860 
 1861   if (src2.is_lval()) {
 1862     movptr(rscratch, src2);
 1863     Assembler::cmpq(src1, rscratch);
 1864   } else if (reachable(src2)) {
 1865     cmpq(src1, as_Address(src2));
 1866   } else {
 1867     lea(rscratch, src2);
 1868     Assembler::cmpq(src1, Address(rscratch, 0));
 1869   }
 1870 #else
 1871   assert(rscratch == noreg, "not needed");
 1872   if (src2.is_lval()) {
 1873     cmp_literal32(src1, (int32_t)src2.target(), src2.rspec());
 1874   } else {
 1875     cmpl(src1, as_Address(src2));
 1876   }
 1877 #endif // _LP64
 1878 }
 1879 
 1880 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1881   assert(src2.is_lval(), "not a mem-mem compare");
 1882 #ifdef _LP64
 1883   // moves src2's literal address
 1884   movptr(rscratch, src2);
 1885   Assembler::cmpq(src1, rscratch);
 1886 #else
 1887   assert(rscratch == noreg, "not needed");
 1888   cmp_literal32(src1, (int32_t)src2.target(), src2.rspec());
 1889 #endif // _LP64
 1890 }
 1891 
 1892 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1893   cmpptr(src1, src2);
 1894 }
 1895 
 1896 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1897   cmpptr(src1, src2);
 1898 }
 1899 
 1900 #ifdef _LP64
 1901 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1902   movoop(rscratch, src2);
 1903   cmpptr(src1, rscratch);
 1904 }
 1905 #endif
 1906 
 1907 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1908   assert(rscratch != noreg || always_reachable(adr), "missing");
 1909 
 1910   if (reachable(adr)) {
 1911     lock();
 1912     cmpxchgptr(reg, as_Address(adr));
 1913   } else {
 1914     lea(rscratch, adr);
 1915     lock();
 1916     cmpxchgptr(reg, Address(rscratch, 0));
 1917   }
 1918 }
 1919 
 1920 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1921   LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr));
 1922 }
 1923 
 1924 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1925   assert(rscratch != noreg || always_reachable(src), "missing");
 1926 
 1927   if (reachable(src)) {
 1928     Assembler::comisd(dst, as_Address(src));
 1929   } else {
 1930     lea(rscratch, src);
 1931     Assembler::comisd(dst, Address(rscratch, 0));
 1932   }
 1933 }
 1934 
 1935 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1936   assert(rscratch != noreg || always_reachable(src), "missing");
 1937 
 1938   if (reachable(src)) {
 1939     Assembler::comiss(dst, as_Address(src));
 1940   } else {
 1941     lea(rscratch, src);
 1942     Assembler::comiss(dst, Address(rscratch, 0));
 1943   }
 1944 }
 1945 
 1946 
 1947 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1948   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1949 
 1950   Condition negated_cond = negate_condition(cond);
 1951   Label L;
 1952   jcc(negated_cond, L);
 1953   pushf(); // Preserve flags
 1954   atomic_incl(counter_addr, rscratch);
 1955   popf();
 1956   bind(L);
 1957 }
 1958 
 1959 int MacroAssembler::corrected_idivl(Register reg) {
 1960   // Full implementation of Java idiv and irem; checks for
 1961   // special case as described in JVM spec., p.243 & p.271.
 1962   // The function returns the (pc) offset of the idivl
 1963   // instruction - may be needed for implicit exceptions.
 1964   //
 1965   //         normal case                           special case
 1966   //
 1967   // input : rax,: dividend                         min_int
 1968   //         reg: divisor   (may not be rax,/rdx)   -1
 1969   //
 1970   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1971   //         rdx: remainder (= rax, irem reg)       0
 1972   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1973   const int min_int = 0x80000000;
 1974   Label normal_case, special_case;
 1975 
 1976   // check for special case
 1977   cmpl(rax, min_int);
 1978   jcc(Assembler::notEqual, normal_case);
 1979   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1980   cmpl(reg, -1);
 1981   jcc(Assembler::equal, special_case);
 1982 
 1983   // handle normal case
 1984   bind(normal_case);
 1985   cdql();
 1986   int idivl_offset = offset();
 1987   idivl(reg);
 1988 
 1989   // normal and special case exit
 1990   bind(special_case);
 1991 
 1992   return idivl_offset;
 1993 }
 1994 
 1995 
 1996 
 1997 void MacroAssembler::decrementl(Register reg, int value) {
 1998   if (value == min_jint) {subl(reg, value) ; return; }
 1999   if (value <  0) { incrementl(reg, -value); return; }
 2000   if (value == 0) {                        ; return; }
 2001   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 2002   /* else */      { subl(reg, value)       ; return; }
 2003 }
 2004 
 2005 void MacroAssembler::decrementl(Address dst, int value) {
 2006   if (value == min_jint) {subl(dst, value) ; return; }
 2007   if (value <  0) { incrementl(dst, -value); return; }
 2008   if (value == 0) {                        ; return; }
 2009   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 2010   /* else */      { subl(dst, value)       ; return; }
 2011 }
 2012 
 2013 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 2014   assert(shift_value > 0, "illegal shift value");
 2015   Label _is_positive;
 2016   testl (reg, reg);
 2017   jcc (Assembler::positive, _is_positive);
 2018   int offset = (1 << shift_value) - 1 ;
 2019 
 2020   if (offset == 1) {
 2021     incrementl(reg);
 2022   } else {
 2023     addl(reg, offset);
 2024   }
 2025 
 2026   bind (_is_positive);
 2027   sarl(reg, shift_value);
 2028 }
 2029 
 2030 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2031   assert(rscratch != noreg || always_reachable(src), "missing");
 2032 
 2033   if (reachable(src)) {
 2034     Assembler::divsd(dst, as_Address(src));
 2035   } else {
 2036     lea(rscratch, src);
 2037     Assembler::divsd(dst, Address(rscratch, 0));
 2038   }
 2039 }
 2040 
 2041 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2042   assert(rscratch != noreg || always_reachable(src), "missing");
 2043 
 2044   if (reachable(src)) {
 2045     Assembler::divss(dst, as_Address(src));
 2046   } else {
 2047     lea(rscratch, src);
 2048     Assembler::divss(dst, Address(rscratch, 0));
 2049   }
 2050 }
 2051 
 2052 void MacroAssembler::enter() {
 2053   push(rbp);
 2054   mov(rbp, rsp);
 2055 }
 2056 
 2057 void MacroAssembler::post_call_nop() {
 2058   if (!Continuations::enabled()) {
 2059     return;
 2060   }
 2061   InstructionMark im(this);
 2062   relocate(post_call_nop_Relocation::spec());
 2063   InlineSkippedInstructionsCounter skipCounter(this);
 2064   emit_int8((uint8_t)0x0f);
 2065   emit_int8((uint8_t)0x1f);
 2066   emit_int8((uint8_t)0x84);
 2067   emit_int8((uint8_t)0x00);
 2068   emit_int32(0x00);
 2069 }
 2070 
 2071 // A 5 byte nop that is safe for patching (see patch_verified_entry)
 2072 void MacroAssembler::fat_nop() {
 2073   if (UseAddressNop) {
 2074     addr_nop_5();
 2075   } else {
 2076     emit_int8((uint8_t)0x26); // es:
 2077     emit_int8((uint8_t)0x2e); // cs:
 2078     emit_int8((uint8_t)0x64); // fs:
 2079     emit_int8((uint8_t)0x65); // gs:
 2080     emit_int8((uint8_t)0x90);
 2081   }
 2082 }
 2083 
 2084 #ifndef _LP64
 2085 void MacroAssembler::fcmp(Register tmp) {
 2086   fcmp(tmp, 1, true, true);
 2087 }
 2088 
 2089 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) {
 2090   assert(!pop_right || pop_left, "usage error");
 2091   if (VM_Version::supports_cmov()) {
 2092     assert(tmp == noreg, "unneeded temp");
 2093     if (pop_left) {
 2094       fucomip(index);
 2095     } else {
 2096       fucomi(index);
 2097     }
 2098     if (pop_right) {
 2099       fpop();
 2100     }
 2101   } else {
 2102     assert(tmp != noreg, "need temp");
 2103     if (pop_left) {
 2104       if (pop_right) {
 2105         fcompp();
 2106       } else {
 2107         fcomp(index);
 2108       }
 2109     } else {
 2110       fcom(index);
 2111     }
 2112     // convert FPU condition into eflags condition via rax,
 2113     save_rax(tmp);
 2114     fwait(); fnstsw_ax();
 2115     sahf();
 2116     restore_rax(tmp);
 2117   }
 2118   // condition codes set as follows:
 2119   //
 2120   // CF (corresponds to C0) if x < y
 2121   // PF (corresponds to C2) if unordered
 2122   // ZF (corresponds to C3) if x = y
 2123 }
 2124 
 2125 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) {
 2126   fcmp2int(dst, unordered_is_less, 1, true, true);
 2127 }
 2128 
 2129 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) {
 2130   fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right);
 2131   Label L;
 2132   if (unordered_is_less) {
 2133     movl(dst, -1);
 2134     jcc(Assembler::parity, L);
 2135     jcc(Assembler::below , L);
 2136     movl(dst, 0);
 2137     jcc(Assembler::equal , L);
 2138     increment(dst);
 2139   } else { // unordered is greater
 2140     movl(dst, 1);
 2141     jcc(Assembler::parity, L);
 2142     jcc(Assembler::above , L);
 2143     movl(dst, 0);
 2144     jcc(Assembler::equal , L);
 2145     decrementl(dst);
 2146   }
 2147   bind(L);
 2148 }
 2149 
 2150 void MacroAssembler::fld_d(AddressLiteral src) {
 2151   fld_d(as_Address(src));
 2152 }
 2153 
 2154 void MacroAssembler::fld_s(AddressLiteral src) {
 2155   fld_s(as_Address(src));
 2156 }
 2157 
 2158 void MacroAssembler::fldcw(AddressLiteral src) {
 2159   fldcw(as_Address(src));
 2160 }
 2161 
 2162 void MacroAssembler::fpop() {
 2163   ffree();
 2164   fincstp();
 2165 }
 2166 
 2167 void MacroAssembler::fremr(Register tmp) {
 2168   save_rax(tmp);
 2169   { Label L;
 2170     bind(L);
 2171     fprem();
 2172     fwait(); fnstsw_ax();
 2173     sahf();
 2174     jcc(Assembler::parity, L);
 2175   }
 2176   restore_rax(tmp);
 2177   // Result is in ST0.
 2178   // Note: fxch & fpop to get rid of ST1
 2179   // (otherwise FPU stack could overflow eventually)
 2180   fxch(1);
 2181   fpop();
 2182 }
 2183 
 2184 void MacroAssembler::empty_FPU_stack() {
 2185   if (VM_Version::supports_mmx()) {
 2186     emms();
 2187   } else {
 2188     for (int i = 8; i-- > 0; ) ffree(i);
 2189   }
 2190 }
 2191 #endif // !LP64
 2192 
 2193 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2194   assert(rscratch != noreg || always_reachable(src), "missing");
 2195   if (reachable(src)) {
 2196     Assembler::mulpd(dst, as_Address(src));
 2197   } else {
 2198     lea(rscratch, src);
 2199     Assembler::mulpd(dst, Address(rscratch, 0));
 2200   }
 2201 }
 2202 
 2203 void MacroAssembler::load_float(Address src) {
 2204 #ifdef _LP64
 2205   movflt(xmm0, src);
 2206 #else
 2207   if (UseSSE >= 1) {
 2208     movflt(xmm0, src);
 2209   } else {
 2210     fld_s(src);
 2211   }
 2212 #endif // LP64
 2213 }
 2214 
 2215 void MacroAssembler::store_float(Address dst) {
 2216 #ifdef _LP64
 2217   movflt(dst, xmm0);
 2218 #else
 2219   if (UseSSE >= 1) {
 2220     movflt(dst, xmm0);
 2221   } else {
 2222     fstp_s(dst);
 2223   }
 2224 #endif // LP64
 2225 }
 2226 
 2227 void MacroAssembler::load_double(Address src) {
 2228 #ifdef _LP64
 2229   movdbl(xmm0, src);
 2230 #else
 2231   if (UseSSE >= 2) {
 2232     movdbl(xmm0, src);
 2233   } else {
 2234     fld_d(src);
 2235   }
 2236 #endif // LP64
 2237 }
 2238 
 2239 void MacroAssembler::store_double(Address dst) {
 2240 #ifdef _LP64
 2241   movdbl(dst, xmm0);
 2242 #else
 2243   if (UseSSE >= 2) {
 2244     movdbl(dst, xmm0);
 2245   } else {
 2246     fstp_d(dst);
 2247   }
 2248 #endif // LP64
 2249 }
 2250 
 2251 // dst = c = a * b + c
 2252 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 2253   Assembler::vfmadd231sd(c, a, b);
 2254   if (dst != c) {
 2255     movdbl(dst, c);
 2256   }
 2257 }
 2258 
 2259 // dst = c = a * b + c
 2260 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 2261   Assembler::vfmadd231ss(c, a, b);
 2262   if (dst != c) {
 2263     movflt(dst, c);
 2264   }
 2265 }
 2266 
 2267 // dst = c = a * b + c
 2268 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 2269   Assembler::vfmadd231pd(c, a, b, vector_len);
 2270   if (dst != c) {
 2271     vmovdqu(dst, c);
 2272   }
 2273 }
 2274 
 2275 // dst = c = a * b + c
 2276 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 2277   Assembler::vfmadd231ps(c, a, b, vector_len);
 2278   if (dst != c) {
 2279     vmovdqu(dst, c);
 2280   }
 2281 }
 2282 
 2283 // dst = c = a * b + c
 2284 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 2285   Assembler::vfmadd231pd(c, a, b, vector_len);
 2286   if (dst != c) {
 2287     vmovdqu(dst, c);
 2288   }
 2289 }
 2290 
 2291 // dst = c = a * b + c
 2292 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 2293   Assembler::vfmadd231ps(c, a, b, vector_len);
 2294   if (dst != c) {
 2295     vmovdqu(dst, c);
 2296   }
 2297 }
 2298 
 2299 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 2300   assert(rscratch != noreg || always_reachable(dst), "missing");
 2301 
 2302   if (reachable(dst)) {
 2303     incrementl(as_Address(dst));
 2304   } else {
 2305     lea(rscratch, dst);
 2306     incrementl(Address(rscratch, 0));
 2307   }
 2308 }
 2309 
 2310 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 2311   incrementl(as_Address(dst, rscratch));
 2312 }
 2313 
 2314 void MacroAssembler::incrementl(Register reg, int value) {
 2315   if (value == min_jint) {addl(reg, value) ; return; }
 2316   if (value <  0) { decrementl(reg, -value); return; }
 2317   if (value == 0) {                        ; return; }
 2318   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 2319   /* else */      { addl(reg, value)       ; return; }
 2320 }
 2321 
 2322 void MacroAssembler::incrementl(Address dst, int value) {
 2323   if (value == min_jint) {addl(dst, value) ; return; }
 2324   if (value <  0) { decrementl(dst, -value); return; }
 2325   if (value == 0) {                        ; return; }
 2326   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 2327   /* else */      { addl(dst, value)       ; return; }
 2328 }
 2329 
 2330 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 2331   assert(rscratch != noreg || always_reachable(dst), "missing");
 2332 
 2333   if (reachable(dst)) {
 2334     jmp_literal(dst.target(), dst.rspec());
 2335   } else {
 2336     lea(rscratch, dst);
 2337     jmp(rscratch);
 2338   }
 2339 }
 2340 
 2341 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 2342   assert(rscratch != noreg || always_reachable(dst), "missing");
 2343 
 2344   if (reachable(dst)) {
 2345     InstructionMark im(this);
 2346     relocate(dst.reloc());
 2347     const int short_size = 2;
 2348     const int long_size = 6;
 2349     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 2350     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 2351       // 0111 tttn #8-bit disp
 2352       emit_int8(0x70 | cc);
 2353       emit_int8((offs - short_size) & 0xFF);
 2354     } else {
 2355       // 0000 1111 1000 tttn #32-bit disp
 2356       emit_int8(0x0F);
 2357       emit_int8((unsigned char)(0x80 | cc));
 2358       emit_int32(offs - long_size);
 2359     }
 2360   } else {
 2361 #ifdef ASSERT
 2362     warning("reversing conditional branch");
 2363 #endif /* ASSERT */
 2364     Label skip;
 2365     jccb(reverse[cc], skip);
 2366     lea(rscratch, dst);
 2367     Assembler::jmp(rscratch);
 2368     bind(skip);
 2369   }
 2370 }
 2371 
 2372 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 2373   assert(rscratch != noreg || always_reachable(src), "missing");
 2374 
 2375   if (reachable(src)) {
 2376     Assembler::ldmxcsr(as_Address(src));
 2377   } else {
 2378     lea(rscratch, src);
 2379     Assembler::ldmxcsr(Address(rscratch, 0));
 2380   }
 2381 }
 2382 
 2383 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 2384   int off;
 2385   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
 2386     off = offset();
 2387     movsbl(dst, src); // movsxb
 2388   } else {
 2389     off = load_unsigned_byte(dst, src);
 2390     shll(dst, 24);
 2391     sarl(dst, 24);
 2392   }
 2393   return off;
 2394 }
 2395 
 2396 // Note: load_signed_short used to be called load_signed_word.
 2397 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 2398 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 2399 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 2400 int MacroAssembler::load_signed_short(Register dst, Address src) {
 2401   int off;
 2402   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
 2403     // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 2404     // version but this is what 64bit has always done. This seems to imply
 2405     // that users are only using 32bits worth.
 2406     off = offset();
 2407     movswl(dst, src); // movsxw
 2408   } else {
 2409     off = load_unsigned_short(dst, src);
 2410     shll(dst, 16);
 2411     sarl(dst, 16);
 2412   }
 2413   return off;
 2414 }
 2415 
 2416 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 2417   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 2418   // and "3.9 Partial Register Penalties", p. 22).
 2419   int off;
 2420   if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) {
 2421     off = offset();
 2422     movzbl(dst, src); // movzxb
 2423   } else {
 2424     xorl(dst, dst);
 2425     off = offset();
 2426     movb(dst, src);
 2427   }
 2428   return off;
 2429 }
 2430 
 2431 // Note: load_unsigned_short used to be called load_unsigned_word.
 2432 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 2433   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 2434   // and "3.9 Partial Register Penalties", p. 22).
 2435   int off;
 2436   if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) {
 2437     off = offset();
 2438     movzwl(dst, src); // movzxw
 2439   } else {
 2440     xorl(dst, dst);
 2441     off = offset();
 2442     movw(dst, src);
 2443   }
 2444   return off;
 2445 }
 2446 
 2447 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 2448   switch (size_in_bytes) {
 2449 #ifndef _LP64
 2450   case  8:
 2451     assert(dst2 != noreg, "second dest register required");
 2452     movl(dst,  src);
 2453     movl(dst2, src.plus_disp(BytesPerInt));
 2454     break;
 2455 #else
 2456   case  8:  movq(dst, src); break;
 2457 #endif
 2458   case  4:  movl(dst, src); break;
 2459   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 2460   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 2461   default:  ShouldNotReachHere();
 2462   }
 2463 }
 2464 
 2465 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 2466   switch (size_in_bytes) {
 2467 #ifndef _LP64
 2468   case  8:
 2469     assert(src2 != noreg, "second source register required");
 2470     movl(dst,                        src);
 2471     movl(dst.plus_disp(BytesPerInt), src2);
 2472     break;
 2473 #else
 2474   case  8:  movq(dst, src); break;
 2475 #endif
 2476   case  4:  movl(dst, src); break;
 2477   case  2:  movw(dst, src); break;
 2478   case  1:  movb(dst, src); break;
 2479   default:  ShouldNotReachHere();
 2480   }
 2481 }
 2482 
 2483 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 2484   assert(rscratch != noreg || always_reachable(dst), "missing");
 2485 
 2486   if (reachable(dst)) {
 2487     movl(as_Address(dst), src);
 2488   } else {
 2489     lea(rscratch, dst);
 2490     movl(Address(rscratch, 0), src);
 2491   }
 2492 }
 2493 
 2494 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 2495   if (reachable(src)) {
 2496     movl(dst, as_Address(src));
 2497   } else {
 2498     lea(dst, src);
 2499     movl(dst, Address(dst, 0));
 2500   }
 2501 }
 2502 
 2503 // C++ bool manipulation
 2504 
 2505 void MacroAssembler::movbool(Register dst, Address src) {
 2506   if(sizeof(bool) == 1)
 2507     movb(dst, src);
 2508   else if(sizeof(bool) == 2)
 2509     movw(dst, src);
 2510   else if(sizeof(bool) == 4)
 2511     movl(dst, src);
 2512   else
 2513     // unsupported
 2514     ShouldNotReachHere();
 2515 }
 2516 
 2517 void MacroAssembler::movbool(Address dst, bool boolconst) {
 2518   if(sizeof(bool) == 1)
 2519     movb(dst, (int) boolconst);
 2520   else if(sizeof(bool) == 2)
 2521     movw(dst, (int) boolconst);
 2522   else if(sizeof(bool) == 4)
 2523     movl(dst, (int) boolconst);
 2524   else
 2525     // unsupported
 2526     ShouldNotReachHere();
 2527 }
 2528 
 2529 void MacroAssembler::movbool(Address dst, Register src) {
 2530   if(sizeof(bool) == 1)
 2531     movb(dst, src);
 2532   else if(sizeof(bool) == 2)
 2533     movw(dst, src);
 2534   else if(sizeof(bool) == 4)
 2535     movl(dst, src);
 2536   else
 2537     // unsupported
 2538     ShouldNotReachHere();
 2539 }
 2540 
 2541 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2542   assert(rscratch != noreg || always_reachable(src), "missing");
 2543 
 2544   if (reachable(src)) {
 2545     movdl(dst, as_Address(src));
 2546   } else {
 2547     lea(rscratch, src);
 2548     movdl(dst, Address(rscratch, 0));
 2549   }
 2550 }
 2551 
 2552 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2553   assert(rscratch != noreg || always_reachable(src), "missing");
 2554 
 2555   if (reachable(src)) {
 2556     movq(dst, as_Address(src));
 2557   } else {
 2558     lea(rscratch, src);
 2559     movq(dst, Address(rscratch, 0));
 2560   }
 2561 }
 2562 
 2563 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2564   assert(rscratch != noreg || always_reachable(src), "missing");
 2565 
 2566   if (reachable(src)) {
 2567     if (UseXmmLoadAndClearUpper) {
 2568       movsd (dst, as_Address(src));
 2569     } else {
 2570       movlpd(dst, as_Address(src));
 2571     }
 2572   } else {
 2573     lea(rscratch, src);
 2574     if (UseXmmLoadAndClearUpper) {
 2575       movsd (dst, Address(rscratch, 0));
 2576     } else {
 2577       movlpd(dst, Address(rscratch, 0));
 2578     }
 2579   }
 2580 }
 2581 
 2582 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2583   assert(rscratch != noreg || always_reachable(src), "missing");
 2584 
 2585   if (reachable(src)) {
 2586     movss(dst, as_Address(src));
 2587   } else {
 2588     lea(rscratch, src);
 2589     movss(dst, Address(rscratch, 0));
 2590   }
 2591 }
 2592 
 2593 void MacroAssembler::movptr(Register dst, Register src) {
 2594   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
 2595 }
 2596 
 2597 void MacroAssembler::movptr(Register dst, Address src) {
 2598   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
 2599 }
 2600 
 2601 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 2602 void MacroAssembler::movptr(Register dst, intptr_t src) {
 2603 #ifdef _LP64
 2604   if (is_uimm32(src)) {
 2605     movl(dst, checked_cast<uint32_t>(src));
 2606   } else if (is_simm32(src)) {
 2607     movq(dst, checked_cast<int32_t>(src));
 2608   } else {
 2609     mov64(dst, src);
 2610   }
 2611 #else
 2612   movl(dst, src);
 2613 #endif
 2614 }
 2615 
 2616 void MacroAssembler::movptr(Address dst, Register src) {
 2617   LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
 2618 }
 2619 
 2620 void MacroAssembler::movptr(Address dst, int32_t src) {
 2621   LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src));
 2622 }
 2623 
 2624 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2625   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2626   Assembler::movdqu(dst, src);
 2627 }
 2628 
 2629 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2630   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2631   Assembler::movdqu(dst, src);
 2632 }
 2633 
 2634 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2635   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2636   Assembler::movdqu(dst, src);
 2637 }
 2638 
 2639 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2640   assert(rscratch != noreg || always_reachable(src), "missing");
 2641 
 2642   if (reachable(src)) {
 2643     movdqu(dst, as_Address(src));
 2644   } else {
 2645     lea(rscratch, src);
 2646     movdqu(dst, Address(rscratch, 0));
 2647   }
 2648 }
 2649 
 2650 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2651   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2652   Assembler::vmovdqu(dst, src);
 2653 }
 2654 
 2655 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2656   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2657   Assembler::vmovdqu(dst, src);
 2658 }
 2659 
 2660 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2661   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2662   Assembler::vmovdqu(dst, src);
 2663 }
 2664 
 2665 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2666   assert(rscratch != noreg || always_reachable(src), "missing");
 2667 
 2668   if (reachable(src)) {
 2669     vmovdqu(dst, as_Address(src));
 2670   }
 2671   else {
 2672     lea(rscratch, src);
 2673     vmovdqu(dst, Address(rscratch, 0));
 2674   }
 2675 }
 2676 
 2677 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2678   assert(rscratch != noreg || always_reachable(src), "missing");
 2679 
 2680   if (vector_len == AVX_512bit) {
 2681     evmovdquq(dst, src, AVX_512bit, rscratch);
 2682   } else if (vector_len == AVX_256bit) {
 2683     vmovdqu(dst, src, rscratch);
 2684   } else {
 2685     movdqu(dst, src, rscratch);
 2686   }
 2687 }
 2688 
 2689 void MacroAssembler::kmov(KRegister dst, Address src) {
 2690   if (VM_Version::supports_avx512bw()) {
 2691     kmovql(dst, src);
 2692   } else {
 2693     assert(VM_Version::supports_evex(), "");
 2694     kmovwl(dst, src);
 2695   }
 2696 }
 2697 
 2698 void MacroAssembler::kmov(Address dst, KRegister src) {
 2699   if (VM_Version::supports_avx512bw()) {
 2700     kmovql(dst, src);
 2701   } else {
 2702     assert(VM_Version::supports_evex(), "");
 2703     kmovwl(dst, src);
 2704   }
 2705 }
 2706 
 2707 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2708   if (VM_Version::supports_avx512bw()) {
 2709     kmovql(dst, src);
 2710   } else {
 2711     assert(VM_Version::supports_evex(), "");
 2712     kmovwl(dst, src);
 2713   }
 2714 }
 2715 
 2716 void MacroAssembler::kmov(Register dst, KRegister src) {
 2717   if (VM_Version::supports_avx512bw()) {
 2718     kmovql(dst, src);
 2719   } else {
 2720     assert(VM_Version::supports_evex(), "");
 2721     kmovwl(dst, src);
 2722   }
 2723 }
 2724 
 2725 void MacroAssembler::kmov(KRegister dst, Register src) {
 2726   if (VM_Version::supports_avx512bw()) {
 2727     kmovql(dst, src);
 2728   } else {
 2729     assert(VM_Version::supports_evex(), "");
 2730     kmovwl(dst, src);
 2731   }
 2732 }
 2733 
 2734 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2735   assert(rscratch != noreg || always_reachable(src), "missing");
 2736 
 2737   if (reachable(src)) {
 2738     kmovql(dst, as_Address(src));
 2739   } else {
 2740     lea(rscratch, src);
 2741     kmovql(dst, Address(rscratch, 0));
 2742   }
 2743 }
 2744 
 2745 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2746   assert(rscratch != noreg || always_reachable(src), "missing");
 2747 
 2748   if (reachable(src)) {
 2749     kmovwl(dst, as_Address(src));
 2750   } else {
 2751     lea(rscratch, src);
 2752     kmovwl(dst, Address(rscratch, 0));
 2753   }
 2754 }
 2755 
 2756 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2757                                int vector_len, Register rscratch) {
 2758   assert(rscratch != noreg || always_reachable(src), "missing");
 2759 
 2760   if (reachable(src)) {
 2761     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2762   } else {
 2763     lea(rscratch, src);
 2764     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2765   }
 2766 }
 2767 
 2768 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2769                                int vector_len, Register rscratch) {
 2770   assert(rscratch != noreg || always_reachable(src), "missing");
 2771 
 2772   if (reachable(src)) {
 2773     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2774   } else {
 2775     lea(rscratch, src);
 2776     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2777   }
 2778 }
 2779 
 2780 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2781   assert(rscratch != noreg || always_reachable(src), "missing");
 2782 
 2783   if (reachable(src)) {
 2784     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2785   } else {
 2786     lea(rscratch, src);
 2787     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2788   }
 2789 }
 2790 
 2791 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2792   assert(rscratch != noreg || always_reachable(src), "missing");
 2793 
 2794   if (reachable(src)) {
 2795     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2796   } else {
 2797     lea(rscratch, src);
 2798     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2799   }
 2800 }
 2801 
 2802 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2803   assert(rscratch != noreg || always_reachable(src), "missing");
 2804 
 2805   if (reachable(src)) {
 2806     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2807   } else {
 2808     lea(rscratch, src);
 2809     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2810   }
 2811 }
 2812 
 2813 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2814   assert(rscratch != noreg || always_reachable(src), "missing");
 2815 
 2816   if (reachable(src)) {
 2817     Assembler::movdqa(dst, as_Address(src));
 2818   } else {
 2819     lea(rscratch, src);
 2820     Assembler::movdqa(dst, Address(rscratch, 0));
 2821   }
 2822 }
 2823 
 2824 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2825   assert(rscratch != noreg || always_reachable(src), "missing");
 2826 
 2827   if (reachable(src)) {
 2828     Assembler::movsd(dst, as_Address(src));
 2829   } else {
 2830     lea(rscratch, src);
 2831     Assembler::movsd(dst, Address(rscratch, 0));
 2832   }
 2833 }
 2834 
 2835 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2836   assert(rscratch != noreg || always_reachable(src), "missing");
 2837 
 2838   if (reachable(src)) {
 2839     Assembler::movss(dst, as_Address(src));
 2840   } else {
 2841     lea(rscratch, src);
 2842     Assembler::movss(dst, Address(rscratch, 0));
 2843   }
 2844 }
 2845 
 2846 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2847   assert(rscratch != noreg || always_reachable(src), "missing");
 2848 
 2849   if (reachable(src)) {
 2850     Assembler::movddup(dst, as_Address(src));
 2851   } else {
 2852     lea(rscratch, src);
 2853     Assembler::movddup(dst, Address(rscratch, 0));
 2854   }
 2855 }
 2856 
 2857 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2858   assert(rscratch != noreg || always_reachable(src), "missing");
 2859 
 2860   if (reachable(src)) {
 2861     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2862   } else {
 2863     lea(rscratch, src);
 2864     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2865   }
 2866 }
 2867 
 2868 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2869   assert(rscratch != noreg || always_reachable(src), "missing");
 2870 
 2871   if (reachable(src)) {
 2872     Assembler::mulsd(dst, as_Address(src));
 2873   } else {
 2874     lea(rscratch, src);
 2875     Assembler::mulsd(dst, Address(rscratch, 0));
 2876   }
 2877 }
 2878 
 2879 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2880   assert(rscratch != noreg || always_reachable(src), "missing");
 2881 
 2882   if (reachable(src)) {
 2883     Assembler::mulss(dst, as_Address(src));
 2884   } else {
 2885     lea(rscratch, src);
 2886     Assembler::mulss(dst, Address(rscratch, 0));
 2887   }
 2888 }
 2889 
 2890 void MacroAssembler::null_check(Register reg, int offset) {
 2891   if (needs_explicit_null_check(offset)) {
 2892     // provoke OS null exception if reg is null by
 2893     // accessing M[reg] w/o changing any (non-CC) registers
 2894     // NOTE: cmpl is plenty here to provoke a segv
 2895     cmpptr(rax, Address(reg, 0));
 2896     // Note: should probably use testl(rax, Address(reg, 0));
 2897     //       may be shorter code (however, this version of
 2898     //       testl needs to be implemented first)
 2899   } else {
 2900     // nothing to do, (later) access of M[reg + offset]
 2901     // will provoke OS null exception if reg is null
 2902   }
 2903 }
 2904 
 2905 void MacroAssembler::os_breakpoint() {
 2906   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2907   // (e.g., MSVC can't call ps() otherwise)
 2908   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2909 }
 2910 
 2911 void MacroAssembler::unimplemented(const char* what) {
 2912   const char* buf = nullptr;
 2913   {
 2914     ResourceMark rm;
 2915     stringStream ss;
 2916     ss.print("unimplemented: %s", what);
 2917     buf = code_string(ss.as_string());
 2918   }
 2919   stop(buf);
 2920 }
 2921 
 2922 #ifdef _LP64
 2923 #define XSTATE_BV 0x200
 2924 #endif
 2925 
 2926 void MacroAssembler::pop_CPU_state() {
 2927   pop_FPU_state();
 2928   pop_IU_state();
 2929 }
 2930 
 2931 void MacroAssembler::pop_FPU_state() {
 2932 #ifndef _LP64
 2933   frstor(Address(rsp, 0));
 2934 #else
 2935   fxrstor(Address(rsp, 0));
 2936 #endif
 2937   addptr(rsp, FPUStateSizeInWords * wordSize);
 2938 }
 2939 
 2940 void MacroAssembler::pop_IU_state() {
 2941   popa();
 2942   LP64_ONLY(addq(rsp, 8));
 2943   popf();
 2944 }
 2945 
 2946 // Save Integer and Float state
 2947 // Warning: Stack must be 16 byte aligned (64bit)
 2948 void MacroAssembler::push_CPU_state() {
 2949   push_IU_state();
 2950   push_FPU_state();
 2951 }
 2952 
 2953 void MacroAssembler::push_FPU_state() {
 2954   subptr(rsp, FPUStateSizeInWords * wordSize);
 2955 #ifndef _LP64
 2956   fnsave(Address(rsp, 0));
 2957   fwait();
 2958 #else
 2959   fxsave(Address(rsp, 0));
 2960 #endif // LP64
 2961 }
 2962 
 2963 void MacroAssembler::push_IU_state() {
 2964   // Push flags first because pusha kills them
 2965   pushf();
 2966   // Make sure rsp stays 16-byte aligned
 2967   LP64_ONLY(subq(rsp, 8));
 2968   pusha();
 2969 }
 2970 
 2971 void MacroAssembler::push_cont_fastpath() {
 2972   if (!Continuations::enabled()) return;
 2973 
 2974 #ifndef _LP64
 2975   Register rthread = rax;
 2976   Register rrealsp = rbx;
 2977   push(rthread);
 2978   push(rrealsp);
 2979 
 2980   get_thread(rthread);
 2981 
 2982   // The code below wants the original RSP.
 2983   // Move it back after the pushes above.
 2984   movptr(rrealsp, rsp);
 2985   addptr(rrealsp, 2*wordSize);
 2986 #else
 2987   Register rthread = r15_thread;
 2988   Register rrealsp = rsp;
 2989 #endif
 2990 
 2991   Label done;
 2992   cmpptr(rrealsp, Address(rthread, JavaThread::cont_fastpath_offset()));
 2993   jccb(Assembler::belowEqual, done);
 2994   movptr(Address(rthread, JavaThread::cont_fastpath_offset()), rrealsp);
 2995   bind(done);
 2996 
 2997 #ifndef _LP64
 2998   pop(rrealsp);
 2999   pop(rthread);
 3000 #endif
 3001 }
 3002 
 3003 void MacroAssembler::pop_cont_fastpath() {
 3004   if (!Continuations::enabled()) return;
 3005 
 3006 #ifndef _LP64
 3007   Register rthread = rax;
 3008   Register rrealsp = rbx;
 3009   push(rthread);
 3010   push(rrealsp);
 3011 
 3012   get_thread(rthread);
 3013 
 3014   // The code below wants the original RSP.
 3015   // Move it back after the pushes above.
 3016   movptr(rrealsp, rsp);
 3017   addptr(rrealsp, 2*wordSize);
 3018 #else
 3019   Register rthread = r15_thread;
 3020   Register rrealsp = rsp;
 3021 #endif
 3022 
 3023   Label done;
 3024   cmpptr(rrealsp, Address(rthread, JavaThread::cont_fastpath_offset()));
 3025   jccb(Assembler::below, done);
 3026   movptr(Address(rthread, JavaThread::cont_fastpath_offset()), 0);
 3027   bind(done);
 3028 
 3029 #ifndef _LP64
 3030   pop(rrealsp);
 3031   pop(rthread);
 3032 #endif
 3033 }
 3034 
 3035 void MacroAssembler::inc_held_monitor_count() {
 3036 #ifndef _LP64
 3037   Register thread = rax;
 3038   push(thread);
 3039   get_thread(thread);
 3040   incrementl(Address(thread, JavaThread::held_monitor_count_offset()));
 3041   pop(thread);
 3042 #else // LP64
 3043   incrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
 3044 #endif
 3045 }
 3046 
 3047 void MacroAssembler::dec_held_monitor_count() {
 3048 #ifndef _LP64
 3049   Register thread = rax;
 3050   push(thread);
 3051   get_thread(thread);
 3052   decrementl(Address(thread, JavaThread::held_monitor_count_offset()));
 3053   pop(thread);
 3054 #else // LP64
 3055   decrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
 3056 #endif
 3057 }
 3058 
 3059 #ifdef ASSERT
 3060 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 3061 #ifdef _LP64
 3062   Label no_cont;
 3063   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 3064   testl(cont, cont);
 3065   jcc(Assembler::zero, no_cont);
 3066   stop(name);
 3067   bind(no_cont);
 3068 #else
 3069   Unimplemented();
 3070 #endif
 3071 }
 3072 #endif
 3073 
 3074 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp) { // determine java_thread register
 3075   if (!java_thread->is_valid()) {
 3076     java_thread = rdi;
 3077     get_thread(java_thread);
 3078   }
 3079   // we must set sp to zero to clear frame
 3080   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 3081   // must clear fp, so that compiled frames are not confused; it is
 3082   // possible that we need it only for debugging
 3083   if (clear_fp) {
 3084     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 3085   }
 3086   // Always clear the pc because it could have been set by make_walkable()
 3087   movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 3088   vzeroupper();
 3089 }
 3090 
 3091 void MacroAssembler::restore_rax(Register tmp) {
 3092   if (tmp == noreg) pop(rax);
 3093   else if (tmp != rax) mov(rax, tmp);
 3094 }
 3095 
 3096 void MacroAssembler::round_to(Register reg, int modulus) {
 3097   addptr(reg, modulus - 1);
 3098   andptr(reg, -modulus);
 3099 }
 3100 
 3101 void MacroAssembler::save_rax(Register tmp) {
 3102   if (tmp == noreg) push(rax);
 3103   else if (tmp != rax) mov(tmp, rax);
 3104 }
 3105 
 3106 void MacroAssembler::safepoint_poll(Label& slow_path, Register thread_reg, bool at_return, bool in_nmethod) {
 3107   if (at_return) {
 3108     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 3109     // we may safely use rsp instead to perform the stack watermark check.
 3110     cmpptr(in_nmethod ? rsp : rbp, Address(thread_reg, JavaThread::polling_word_offset()));
 3111     jcc(Assembler::above, slow_path);
 3112     return;
 3113   }
 3114   testb(Address(thread_reg, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 3115   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 3116 }
 3117 
 3118 // Calls to C land
 3119 //
 3120 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 3121 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 3122 // has to be reset to 0. This is required to allow proper stack traversal.
 3123 void MacroAssembler::set_last_Java_frame(Register java_thread,
 3124                                          Register last_java_sp,
 3125                                          Register last_java_fp,
 3126                                          address  last_java_pc,
 3127                                          Register rscratch) {
 3128   vzeroupper();
 3129   // determine java_thread register
 3130   if (!java_thread->is_valid()) {
 3131     java_thread = rdi;
 3132     get_thread(java_thread);
 3133   }
 3134   // determine last_java_sp register
 3135   if (!last_java_sp->is_valid()) {
 3136     last_java_sp = rsp;
 3137   }
 3138   // last_java_fp is optional
 3139   if (last_java_fp->is_valid()) {
 3140     movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 3141   }
 3142   // last_java_pc is optional
 3143   if (last_java_pc != nullptr) {
 3144     Address java_pc(java_thread,
 3145                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 3146     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 3147   }
 3148   movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 3149 }
 3150 
 3151 void MacroAssembler::shlptr(Register dst, int imm8) {
 3152   LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8));
 3153 }
 3154 
 3155 void MacroAssembler::shrptr(Register dst, int imm8) {
 3156   LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8));
 3157 }
 3158 
 3159 void MacroAssembler::sign_extend_byte(Register reg) {
 3160   if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) {
 3161     movsbl(reg, reg); // movsxb
 3162   } else {
 3163     shll(reg, 24);
 3164     sarl(reg, 24);
 3165   }
 3166 }
 3167 
 3168 void MacroAssembler::sign_extend_short(Register reg) {
 3169   if (LP64_ONLY(true ||) VM_Version::is_P6()) {
 3170     movswl(reg, reg); // movsxw
 3171   } else {
 3172     shll(reg, 16);
 3173     sarl(reg, 16);
 3174   }
 3175 }
 3176 
 3177 void MacroAssembler::testl(Address dst, int32_t imm32) {
 3178   if (imm32 >= 0 && is8bit(imm32)) {
 3179     testb(dst, imm32);
 3180   } else {
 3181     Assembler::testl(dst, imm32);
 3182   }
 3183 }
 3184 
 3185 void MacroAssembler::testl(Register dst, int32_t imm32) {
 3186   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 3187     testb(dst, imm32);
 3188   } else {
 3189     Assembler::testl(dst, imm32);
 3190   }
 3191 }
 3192 
 3193 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 3194   assert(always_reachable(src), "Address should be reachable");
 3195   testl(dst, as_Address(src));
 3196 }
 3197 
 3198 #ifdef _LP64
 3199 
 3200 void MacroAssembler::testq(Address dst, int32_t imm32) {
 3201   if (imm32 >= 0) {
 3202     testl(dst, imm32);
 3203   } else {
 3204     Assembler::testq(dst, imm32);
 3205   }
 3206 }
 3207 
 3208 void MacroAssembler::testq(Register dst, int32_t imm32) {
 3209   if (imm32 >= 0) {
 3210     testl(dst, imm32);
 3211   } else {
 3212     Assembler::testq(dst, imm32);
 3213   }
 3214 }
 3215 
 3216 #endif
 3217 
 3218 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 3219   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3220   Assembler::pcmpeqb(dst, src);
 3221 }
 3222 
 3223 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 3224   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3225   Assembler::pcmpeqw(dst, src);
 3226 }
 3227 
 3228 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 3229   assert((dst->encoding() < 16),"XMM register should be 0-15");
 3230   Assembler::pcmpestri(dst, src, imm8);
 3231 }
 3232 
 3233 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 3234   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3235   Assembler::pcmpestri(dst, src, imm8);
 3236 }
 3237 
 3238 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 3239   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3240   Assembler::pmovzxbw(dst, src);
 3241 }
 3242 
 3243 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 3244   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3245   Assembler::pmovzxbw(dst, src);
 3246 }
 3247 
 3248 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 3249   assert((src->encoding() < 16),"XMM register should be 0-15");
 3250   Assembler::pmovmskb(dst, src);
 3251 }
 3252 
 3253 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 3254   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3255   Assembler::ptest(dst, src);
 3256 }
 3257 
 3258 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3259   assert(rscratch != noreg || always_reachable(src), "missing");
 3260 
 3261   if (reachable(src)) {
 3262     Assembler::sqrtss(dst, as_Address(src));
 3263   } else {
 3264     lea(rscratch, src);
 3265     Assembler::sqrtss(dst, Address(rscratch, 0));
 3266   }
 3267 }
 3268 
 3269 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3270   assert(rscratch != noreg || always_reachable(src), "missing");
 3271 
 3272   if (reachable(src)) {
 3273     Assembler::subsd(dst, as_Address(src));
 3274   } else {
 3275     lea(rscratch, src);
 3276     Assembler::subsd(dst, Address(rscratch, 0));
 3277   }
 3278 }
 3279 
 3280 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 3281   assert(rscratch != noreg || always_reachable(src), "missing");
 3282 
 3283   if (reachable(src)) {
 3284     Assembler::roundsd(dst, as_Address(src), rmode);
 3285   } else {
 3286     lea(rscratch, src);
 3287     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 3288   }
 3289 }
 3290 
 3291 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3292   assert(rscratch != noreg || always_reachable(src), "missing");
 3293 
 3294   if (reachable(src)) {
 3295     Assembler::subss(dst, as_Address(src));
 3296   } else {
 3297     lea(rscratch, src);
 3298     Assembler::subss(dst, Address(rscratch, 0));
 3299   }
 3300 }
 3301 
 3302 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3303   assert(rscratch != noreg || always_reachable(src), "missing");
 3304 
 3305   if (reachable(src)) {
 3306     Assembler::ucomisd(dst, as_Address(src));
 3307   } else {
 3308     lea(rscratch, src);
 3309     Assembler::ucomisd(dst, Address(rscratch, 0));
 3310   }
 3311 }
 3312 
 3313 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3314   assert(rscratch != noreg || always_reachable(src), "missing");
 3315 
 3316   if (reachable(src)) {
 3317     Assembler::ucomiss(dst, as_Address(src));
 3318   } else {
 3319     lea(rscratch, src);
 3320     Assembler::ucomiss(dst, Address(rscratch, 0));
 3321   }
 3322 }
 3323 
 3324 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3325   assert(rscratch != noreg || always_reachable(src), "missing");
 3326 
 3327   // Used in sign-bit flipping with aligned address.
 3328   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 3329   if (reachable(src)) {
 3330     Assembler::xorpd(dst, as_Address(src));
 3331   } else {
 3332     lea(rscratch, src);
 3333     Assembler::xorpd(dst, Address(rscratch, 0));
 3334   }
 3335 }
 3336 
 3337 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 3338   if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
 3339     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 3340   }
 3341   else {
 3342     Assembler::xorpd(dst, src);
 3343   }
 3344 }
 3345 
 3346 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 3347   if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
 3348     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 3349   } else {
 3350     Assembler::xorps(dst, src);
 3351   }
 3352 }
 3353 
 3354 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3355   assert(rscratch != noreg || always_reachable(src), "missing");
 3356 
 3357   // Used in sign-bit flipping with aligned address.
 3358   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 3359   if (reachable(src)) {
 3360     Assembler::xorps(dst, as_Address(src));
 3361   } else {
 3362     lea(rscratch, src);
 3363     Assembler::xorps(dst, Address(rscratch, 0));
 3364   }
 3365 }
 3366 
 3367 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 3368   assert(rscratch != noreg || always_reachable(src), "missing");
 3369 
 3370   // Used in sign-bit flipping with aligned address.
 3371   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 3372   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 3373   if (reachable(src)) {
 3374     Assembler::pshufb(dst, as_Address(src));
 3375   } else {
 3376     lea(rscratch, src);
 3377     Assembler::pshufb(dst, Address(rscratch, 0));
 3378   }
 3379 }
 3380 
 3381 // AVX 3-operands instructions
 3382 
 3383 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3384   assert(rscratch != noreg || always_reachable(src), "missing");
 3385 
 3386   if (reachable(src)) {
 3387     vaddsd(dst, nds, as_Address(src));
 3388   } else {
 3389     lea(rscratch, src);
 3390     vaddsd(dst, nds, Address(rscratch, 0));
 3391   }
 3392 }
 3393 
 3394 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3395   assert(rscratch != noreg || always_reachable(src), "missing");
 3396 
 3397   if (reachable(src)) {
 3398     vaddss(dst, nds, as_Address(src));
 3399   } else {
 3400     lea(rscratch, src);
 3401     vaddss(dst, nds, Address(rscratch, 0));
 3402   }
 3403 }
 3404 
 3405 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3406   assert(UseAVX > 0, "requires some form of AVX");
 3407   assert(rscratch != noreg || always_reachable(src), "missing");
 3408 
 3409   if (reachable(src)) {
 3410     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 3411   } else {
 3412     lea(rscratch, src);
 3413     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 3414   }
 3415 }
 3416 
 3417 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3418   assert(UseAVX > 0, "requires some form of AVX");
 3419   assert(rscratch != noreg || always_reachable(src), "missing");
 3420 
 3421   if (reachable(src)) {
 3422     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 3423   } else {
 3424     lea(rscratch, src);
 3425     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 3426   }
 3427 }
 3428 
 3429 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 3430   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3431   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 3432 
 3433   vandps(dst, nds, negate_field, vector_len, rscratch);
 3434 }
 3435 
 3436 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 3437   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3438   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 3439 
 3440   vandpd(dst, nds, negate_field, vector_len, rscratch);
 3441 }
 3442 
 3443 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3444   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3445   Assembler::vpaddb(dst, nds, src, vector_len);
 3446 }
 3447 
 3448 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3449   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3450   Assembler::vpaddb(dst, nds, src, vector_len);
 3451 }
 3452 
 3453 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3454   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3455   Assembler::vpaddw(dst, nds, src, vector_len);
 3456 }
 3457 
 3458 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3459   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3460   Assembler::vpaddw(dst, nds, src, vector_len);
 3461 }
 3462 
 3463 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3464   assert(rscratch != noreg || always_reachable(src), "missing");
 3465 
 3466   if (reachable(src)) {
 3467     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 3468   } else {
 3469     lea(rscratch, src);
 3470     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 3471   }
 3472 }
 3473 
 3474 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3475   assert(rscratch != noreg || always_reachable(src), "missing");
 3476 
 3477   if (reachable(src)) {
 3478     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 3479   } else {
 3480     lea(rscratch, src);
 3481     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 3482   }
 3483 }
 3484 
 3485 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3486   assert(rscratch != noreg || always_reachable(src), "missing");
 3487 
 3488   if (reachable(src)) {
 3489     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3490   } else {
 3491     lea(rscratch, src);
 3492     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3493   }
 3494 }
 3495 
 3496 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3497   assert(rscratch != noreg || always_reachable(src), "missing");
 3498 
 3499   if (reachable(src)) {
 3500     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3501   } else {
 3502     lea(rscratch, src);
 3503     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3504   }
 3505 }
 3506 
 3507 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3508   assert(rscratch != noreg || always_reachable(src), "missing");
 3509 
 3510   if (reachable(src)) {
 3511     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3512   } else {
 3513     lea(rscratch, src);
 3514     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3515   }
 3516 }
 3517 
 3518 // Vector float blend
 3519 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3520 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3521   // WARN: Allow dst == (src1|src2), mask == scratch
 3522   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
 3523   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3524   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3525   if (blend_emulation && scratch_available && dst_available) {
 3526     if (compute_mask) {
 3527       vpsrad(scratch, mask, 32, vector_len);
 3528       mask = scratch;
 3529     }
 3530     if (dst == src1) {
 3531       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3532       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3533     } else {
 3534       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3535       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3536     }
 3537     vpor(dst, dst, scratch, vector_len);
 3538   } else {
 3539     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3540   }
 3541 }
 3542 
 3543 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3544 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3545   // WARN: Allow dst == (src1|src2), mask == scratch
 3546   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
 3547   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3548   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3549   if (blend_emulation && scratch_available && dst_available) {
 3550     if (compute_mask) {
 3551       vpxor(scratch, scratch, scratch, vector_len);
 3552       vpcmpgtq(scratch, scratch, mask, vector_len);
 3553       mask = scratch;
 3554     }
 3555     if (dst == src1) {
 3556       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3557       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3558     } else {
 3559       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3560       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3561     }
 3562     vpor(dst, dst, scratch, vector_len);
 3563   } else {
 3564     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3565   }
 3566 }
 3567 
 3568 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3569   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3570   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3571 }
 3572 
 3573 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3574   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3575   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3576 }
 3577 
 3578 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3579   assert(rscratch != noreg || always_reachable(src), "missing");
 3580 
 3581   if (reachable(src)) {
 3582     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3583   } else {
 3584     lea(rscratch, src);
 3585     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3586   }
 3587 }
 3588 
 3589 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3590                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3591   assert(rscratch != noreg || always_reachable(src), "missing");
 3592 
 3593   if (reachable(src)) {
 3594     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3595   } else {
 3596     lea(rscratch, src);
 3597     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3598   }
 3599 }
 3600 
 3601 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3602                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3603   assert(rscratch != noreg || always_reachable(src), "missing");
 3604 
 3605   if (reachable(src)) {
 3606     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3607   } else {
 3608     lea(rscratch, src);
 3609     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3610   }
 3611 }
 3612 
 3613 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3614                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3615   assert(rscratch != noreg || always_reachable(src), "missing");
 3616 
 3617   if (reachable(src)) {
 3618     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3619   } else {
 3620     lea(rscratch, src);
 3621     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3622   }
 3623 }
 3624 
 3625 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3626                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3627   assert(rscratch != noreg || always_reachable(src), "missing");
 3628 
 3629   if (reachable(src)) {
 3630     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3631   } else {
 3632     lea(rscratch, src);
 3633     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3634   }
 3635 }
 3636 
 3637 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3638   if (width == Assembler::Q) {
 3639     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3640   } else {
 3641     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3642   }
 3643 }
 3644 
 3645 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3646   int eq_cond_enc = 0x29;
 3647   int gt_cond_enc = 0x37;
 3648   if (width != Assembler::Q) {
 3649     eq_cond_enc = 0x74 + width;
 3650     gt_cond_enc = 0x64 + width;
 3651   }
 3652   switch (cond) {
 3653   case eq:
 3654     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3655     break;
 3656   case neq:
 3657     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3658     vallones(xtmp, vector_len);
 3659     vpxor(dst, xtmp, dst, vector_len);
 3660     break;
 3661   case le:
 3662     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3663     vallones(xtmp, vector_len);
 3664     vpxor(dst, xtmp, dst, vector_len);
 3665     break;
 3666   case nlt:
 3667     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3668     vallones(xtmp, vector_len);
 3669     vpxor(dst, xtmp, dst, vector_len);
 3670     break;
 3671   case lt:
 3672     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3673     break;
 3674   case nle:
 3675     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3676     break;
 3677   default:
 3678     assert(false, "Should not reach here");
 3679   }
 3680 }
 3681 
 3682 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3683   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3684   Assembler::vpmovzxbw(dst, src, vector_len);
 3685 }
 3686 
 3687 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3688   assert((src->encoding() < 16),"XMM register should be 0-15");
 3689   Assembler::vpmovmskb(dst, src, vector_len);
 3690 }
 3691 
 3692 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3693   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3694   Assembler::vpmullw(dst, nds, src, vector_len);
 3695 }
 3696 
 3697 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3698   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3699   Assembler::vpmullw(dst, nds, src, vector_len);
 3700 }
 3701 
 3702 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3703   assert((UseAVX > 0), "AVX support is needed");
 3704   assert(rscratch != noreg || always_reachable(src), "missing");
 3705 
 3706   if (reachable(src)) {
 3707     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3708   } else {
 3709     lea(rscratch, src);
 3710     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3711   }
 3712 }
 3713 
 3714 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3715   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3716   Assembler::vpsubb(dst, nds, src, vector_len);
 3717 }
 3718 
 3719 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3720   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3721   Assembler::vpsubb(dst, nds, src, vector_len);
 3722 }
 3723 
 3724 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3725   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3726   Assembler::vpsubw(dst, nds, src, vector_len);
 3727 }
 3728 
 3729 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3730   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3731   Assembler::vpsubw(dst, nds, src, vector_len);
 3732 }
 3733 
 3734 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3735   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3736   Assembler::vpsraw(dst, nds, shift, vector_len);
 3737 }
 3738 
 3739 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3740   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3741   Assembler::vpsraw(dst, nds, shift, vector_len);
 3742 }
 3743 
 3744 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3745   assert(UseAVX > 2,"");
 3746   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3747      vector_len = 2;
 3748   }
 3749   Assembler::evpsraq(dst, nds, shift, vector_len);
 3750 }
 3751 
 3752 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3753   assert(UseAVX > 2,"");
 3754   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3755      vector_len = 2;
 3756   }
 3757   Assembler::evpsraq(dst, nds, shift, vector_len);
 3758 }
 3759 
 3760 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3761   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3762   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3763 }
 3764 
 3765 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3766   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3767   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3768 }
 3769 
 3770 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3771   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3772   Assembler::vpsllw(dst, nds, shift, vector_len);
 3773 }
 3774 
 3775 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3776   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3777   Assembler::vpsllw(dst, nds, shift, vector_len);
 3778 }
 3779 
 3780 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3781   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3782   Assembler::vptest(dst, src);
 3783 }
 3784 
 3785 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3786   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3787   Assembler::punpcklbw(dst, src);
 3788 }
 3789 
 3790 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3791   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3792   Assembler::pshufd(dst, src, mode);
 3793 }
 3794 
 3795 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3796   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3797   Assembler::pshuflw(dst, src, mode);
 3798 }
 3799 
 3800 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3801   assert(rscratch != noreg || always_reachable(src), "missing");
 3802 
 3803   if (reachable(src)) {
 3804     vandpd(dst, nds, as_Address(src), vector_len);
 3805   } else {
 3806     lea(rscratch, src);
 3807     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3808   }
 3809 }
 3810 
 3811 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3812   assert(rscratch != noreg || always_reachable(src), "missing");
 3813 
 3814   if (reachable(src)) {
 3815     vandps(dst, nds, as_Address(src), vector_len);
 3816   } else {
 3817     lea(rscratch, src);
 3818     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3819   }
 3820 }
 3821 
 3822 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3823                             bool merge, int vector_len, Register rscratch) {
 3824   assert(rscratch != noreg || always_reachable(src), "missing");
 3825 
 3826   if (reachable(src)) {
 3827     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3828   } else {
 3829     lea(rscratch, src);
 3830     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3831   }
 3832 }
 3833 
 3834 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3835   assert(rscratch != noreg || always_reachable(src), "missing");
 3836 
 3837   if (reachable(src)) {
 3838     vdivsd(dst, nds, as_Address(src));
 3839   } else {
 3840     lea(rscratch, src);
 3841     vdivsd(dst, nds, Address(rscratch, 0));
 3842   }
 3843 }
 3844 
 3845 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3846   assert(rscratch != noreg || always_reachable(src), "missing");
 3847 
 3848   if (reachable(src)) {
 3849     vdivss(dst, nds, as_Address(src));
 3850   } else {
 3851     lea(rscratch, src);
 3852     vdivss(dst, nds, Address(rscratch, 0));
 3853   }
 3854 }
 3855 
 3856 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3857   assert(rscratch != noreg || always_reachable(src), "missing");
 3858 
 3859   if (reachable(src)) {
 3860     vmulsd(dst, nds, as_Address(src));
 3861   } else {
 3862     lea(rscratch, src);
 3863     vmulsd(dst, nds, Address(rscratch, 0));
 3864   }
 3865 }
 3866 
 3867 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3868   assert(rscratch != noreg || always_reachable(src), "missing");
 3869 
 3870   if (reachable(src)) {
 3871     vmulss(dst, nds, as_Address(src));
 3872   } else {
 3873     lea(rscratch, src);
 3874     vmulss(dst, nds, Address(rscratch, 0));
 3875   }
 3876 }
 3877 
 3878 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3879   assert(rscratch != noreg || always_reachable(src), "missing");
 3880 
 3881   if (reachable(src)) {
 3882     vsubsd(dst, nds, as_Address(src));
 3883   } else {
 3884     lea(rscratch, src);
 3885     vsubsd(dst, nds, Address(rscratch, 0));
 3886   }
 3887 }
 3888 
 3889 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3890   assert(rscratch != noreg || always_reachable(src), "missing");
 3891 
 3892   if (reachable(src)) {
 3893     vsubss(dst, nds, as_Address(src));
 3894   } else {
 3895     lea(rscratch, src);
 3896     vsubss(dst, nds, Address(rscratch, 0));
 3897   }
 3898 }
 3899 
 3900 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3901   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3902   assert(rscratch != noreg || always_reachable(src), "missing");
 3903 
 3904   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3905 }
 3906 
 3907 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3908   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3909   assert(rscratch != noreg || always_reachable(src), "missing");
 3910 
 3911   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3912 }
 3913 
 3914 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3915   assert(rscratch != noreg || always_reachable(src), "missing");
 3916 
 3917   if (reachable(src)) {
 3918     vxorpd(dst, nds, as_Address(src), vector_len);
 3919   } else {
 3920     lea(rscratch, src);
 3921     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3922   }
 3923 }
 3924 
 3925 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3926   assert(rscratch != noreg || always_reachable(src), "missing");
 3927 
 3928   if (reachable(src)) {
 3929     vxorps(dst, nds, as_Address(src), vector_len);
 3930   } else {
 3931     lea(rscratch, src);
 3932     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3933   }
 3934 }
 3935 
 3936 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3937   assert(rscratch != noreg || always_reachable(src), "missing");
 3938 
 3939   if (UseAVX > 1 || (vector_len < 1)) {
 3940     if (reachable(src)) {
 3941       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3942     } else {
 3943       lea(rscratch, src);
 3944       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3945     }
 3946   } else {
 3947     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3948   }
 3949 }
 3950 
 3951 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3952   assert(rscratch != noreg || always_reachable(src), "missing");
 3953 
 3954   if (reachable(src)) {
 3955     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3956   } else {
 3957     lea(rscratch, src);
 3958     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3959   }
 3960 }
 3961 
 3962 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3963   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3964   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3965   // The inverted mask is sign-extended
 3966   andptr(possibly_non_local, inverted_mask);
 3967 }
 3968 
 3969 void MacroAssembler::resolve_jobject(Register value,
 3970                                      Register thread,
 3971                                      Register tmp) {
 3972   assert_different_registers(value, thread, tmp);
 3973   Label done, tagged, weak_tagged;
 3974   testptr(value, value);
 3975   jcc(Assembler::zero, done);           // Use null as-is.
 3976   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3977   jcc(Assembler::notZero, tagged);
 3978 
 3979   // Resolve local handle
 3980   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp, thread);
 3981   verify_oop(value);
 3982   jmp(done);
 3983 
 3984   bind(tagged);
 3985   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3986   jcc(Assembler::notZero, weak_tagged);
 3987 
 3988   // Resolve global handle
 3989   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp, thread);
 3990   verify_oop(value);
 3991   jmp(done);
 3992 
 3993   bind(weak_tagged);
 3994   // Resolve jweak.
 3995   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3996                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp, thread);
 3997   verify_oop(value);
 3998 
 3999   bind(done);
 4000 }
 4001 
 4002 void MacroAssembler::resolve_global_jobject(Register value,
 4003                                             Register thread,
 4004                                             Register tmp) {
 4005   assert_different_registers(value, thread, tmp);
 4006   Label done;
 4007 
 4008   testptr(value, value);
 4009   jcc(Assembler::zero, done);           // Use null as-is.
 4010 
 4011 #ifdef ASSERT
 4012   {
 4013     Label valid_global_tag;
 4014     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 4015     jcc(Assembler::notZero, valid_global_tag);
 4016     stop("non global jobject using resolve_global_jobject");
 4017     bind(valid_global_tag);
 4018   }
 4019 #endif
 4020 
 4021   // Resolve global handle
 4022   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp, thread);
 4023   verify_oop(value);
 4024 
 4025   bind(done);
 4026 }
 4027 
 4028 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 4029   LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32));
 4030 }
 4031 
 4032 // Force generation of a 4 byte immediate value even if it fits into 8bit
 4033 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 4034   LP64_ONLY(subq_imm32(dst, imm32)) NOT_LP64(subl_imm32(dst, imm32));
 4035 }
 4036 
 4037 void MacroAssembler::subptr(Register dst, Register src) {
 4038   LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src));
 4039 }
 4040 
 4041 // C++ bool manipulation
 4042 void MacroAssembler::testbool(Register dst) {
 4043   if(sizeof(bool) == 1)
 4044     testb(dst, 0xff);
 4045   else if(sizeof(bool) == 2) {
 4046     // testw implementation needed for two byte bools
 4047     ShouldNotReachHere();
 4048   } else if(sizeof(bool) == 4)
 4049     testl(dst, dst);
 4050   else
 4051     // unsupported
 4052     ShouldNotReachHere();
 4053 }
 4054 
 4055 void MacroAssembler::testptr(Register dst, Register src) {
 4056   LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
 4057 }
 4058 
 4059 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 4060 void MacroAssembler::tlab_allocate(Register thread, Register obj,
 4061                                    Register var_size_in_bytes,
 4062                                    int con_size_in_bytes,
 4063                                    Register t1,
 4064                                    Register t2,
 4065                                    Label& slow_case) {
 4066   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 4067   bs->tlab_allocate(this, thread, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 4068 }
 4069 
 4070 RegSet MacroAssembler::call_clobbered_gp_registers() {
 4071   RegSet regs;
 4072 #ifdef _LP64
 4073   regs += RegSet::of(rax, rcx, rdx);
 4074 #ifndef WINDOWS
 4075   regs += RegSet::of(rsi, rdi);
 4076 #endif
 4077   regs += RegSet::range(r8, r11);
 4078 #else
 4079   regs += RegSet::of(rax, rcx, rdx);
 4080 #endif
 4081   return regs;
 4082 }
 4083 
 4084 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 4085   int num_xmm_registers = XMMRegister::available_xmm_registers();
 4086 #if defined(WINDOWS) && defined(_LP64)
 4087   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 4088   if (num_xmm_registers > 16) {
 4089      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 4090   }
 4091   return result;
 4092 #else
 4093   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 4094 #endif
 4095 }
 4096 
 4097 static int FPUSaveAreaSize = align_up(108, StackAlignmentInBytes); // 108 bytes needed for FPU state by fsave/frstor
 4098 
 4099 #ifndef _LP64
 4100 static bool use_x87_registers() { return UseSSE < 2; }
 4101 #endif
 4102 static bool use_xmm_registers() { return UseSSE >= 1; }
 4103 
 4104 // C1 only ever uses the first double/float of the XMM register.
 4105 static int xmm_save_size() { return UseSSE >= 2 ? sizeof(double) : sizeof(float); }
 4106 
 4107 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 4108   if (UseSSE == 1) {
 4109     masm->movflt(Address(rsp, offset), reg);
 4110   } else {
 4111     masm->movdbl(Address(rsp, offset), reg);
 4112   }
 4113 }
 4114 
 4115 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 4116   if (UseSSE == 1) {
 4117     masm->movflt(reg, Address(rsp, offset));
 4118   } else {
 4119     masm->movdbl(reg, Address(rsp, offset));
 4120   }
 4121 }
 4122 
 4123 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 4124                                   bool save_fpu, int& gp_area_size,
 4125                                   int& fp_area_size, int& xmm_area_size) {
 4126 
 4127   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 4128                          StackAlignmentInBytes);
 4129 #ifdef _LP64
 4130   fp_area_size = 0;
 4131 #else
 4132   fp_area_size = (save_fpu && use_x87_registers()) ? FPUSaveAreaSize : 0;
 4133 #endif
 4134   xmm_area_size = (save_fpu && use_xmm_registers()) ? xmm_registers.size() * xmm_save_size() : 0;
 4135 
 4136   return gp_area_size + fp_area_size + xmm_area_size;
 4137 }
 4138 
 4139 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 4140   block_comment("push_call_clobbered_registers start");
 4141   // Regular registers
 4142   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 4143 
 4144   int gp_area_size;
 4145   int fp_area_size;
 4146   int xmm_area_size;
 4147   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 4148                                                gp_area_size, fp_area_size, xmm_area_size);
 4149   subptr(rsp, total_save_size);
 4150 
 4151   push_set(gp_registers_to_push, 0);
 4152 
 4153 #ifndef _LP64
 4154   if (save_fpu && use_x87_registers()) {
 4155     fnsave(Address(rsp, gp_area_size));
 4156     fwait();
 4157   }
 4158 #endif
 4159   if (save_fpu && use_xmm_registers()) {
 4160     push_set(call_clobbered_xmm_registers(), gp_area_size + fp_area_size);
 4161   }
 4162 
 4163   block_comment("push_call_clobbered_registers end");
 4164 }
 4165 
 4166 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 4167   block_comment("pop_call_clobbered_registers start");
 4168 
 4169   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 4170 
 4171   int gp_area_size;
 4172   int fp_area_size;
 4173   int xmm_area_size;
 4174   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 4175                                                gp_area_size, fp_area_size, xmm_area_size);
 4176 
 4177   if (restore_fpu && use_xmm_registers()) {
 4178     pop_set(call_clobbered_xmm_registers(), gp_area_size + fp_area_size);
 4179   }
 4180 #ifndef _LP64
 4181   if (restore_fpu && use_x87_registers()) {
 4182     frstor(Address(rsp, gp_area_size));
 4183   }
 4184 #endif
 4185 
 4186   pop_set(gp_registers_to_pop, 0);
 4187 
 4188   addptr(rsp, total_save_size);
 4189 
 4190   vzeroupper();
 4191 
 4192   block_comment("pop_call_clobbered_registers end");
 4193 }
 4194 
 4195 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 4196   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 4197   int spill_offset = offset;
 4198 
 4199   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 4200     save_xmm_register(this, spill_offset, *it);
 4201     spill_offset += xmm_save_size();
 4202   }
 4203 }
 4204 
 4205 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 4206   int restore_size = set.size() * xmm_save_size();
 4207   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 4208 
 4209   int restore_offset = offset + restore_size - xmm_save_size();
 4210 
 4211   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 4212     restore_xmm_register(this, restore_offset, *it);
 4213     restore_offset -= xmm_save_size();
 4214   }
 4215 }
 4216 
 4217 void MacroAssembler::push_set(RegSet set, int offset) {
 4218   int spill_offset;
 4219   if (offset == -1) {
 4220     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4221     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4222     subptr(rsp, aligned_size);
 4223     spill_offset = 0;
 4224   } else {
 4225     spill_offset = offset;
 4226   }
 4227 
 4228   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 4229     movptr(Address(rsp, spill_offset), *it);
 4230     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4231   }
 4232 }
 4233 
 4234 void MacroAssembler::pop_set(RegSet set, int offset) {
 4235 
 4236   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4237   int restore_size = set.size() * gp_reg_size;
 4238   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 4239 
 4240   int restore_offset;
 4241   if (offset == -1) {
 4242     restore_offset = restore_size - gp_reg_size;
 4243   } else {
 4244     restore_offset = offset + restore_size - gp_reg_size;
 4245   }
 4246   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 4247     movptr(*it, Address(rsp, restore_offset));
 4248     restore_offset -= gp_reg_size;
 4249   }
 4250 
 4251   if (offset == -1) {
 4252     addptr(rsp, aligned_size);
 4253   }
 4254 }
 4255 
 4256 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 4257 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 4258   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 4259   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 4260   Label done;
 4261 
 4262   testptr(length_in_bytes, length_in_bytes);
 4263   jcc(Assembler::zero, done);
 4264 
 4265   // initialize topmost word, divide index by 2, check if odd and test if zero
 4266   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 4267 #ifdef ASSERT
 4268   {
 4269     Label L;
 4270     testptr(length_in_bytes, BytesPerWord - 1);
 4271     jcc(Assembler::zero, L);
 4272     stop("length must be a multiple of BytesPerWord");
 4273     bind(L);
 4274   }
 4275 #endif
 4276   Register index = length_in_bytes;
 4277   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 4278   if (UseIncDec) {
 4279     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 4280   } else {
 4281     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 4282     shrptr(index, 1);
 4283   }
 4284 #ifndef _LP64
 4285   // index could have not been a multiple of 8 (i.e., bit 2 was set)
 4286   {
 4287     Label even;
 4288     // note: if index was a multiple of 8, then it cannot
 4289     //       be 0 now otherwise it must have been 0 before
 4290     //       => if it is even, we don't need to check for 0 again
 4291     jcc(Assembler::carryClear, even);
 4292     // clear topmost word (no jump would be needed if conditional assignment worked here)
 4293     movptr(Address(address, index, Address::times_8, offset_in_bytes - 0*BytesPerWord), temp);
 4294     // index could be 0 now, must check again
 4295     jcc(Assembler::zero, done);
 4296     bind(even);
 4297   }
 4298 #endif // !_LP64
 4299   // initialize remaining object fields: index is a multiple of 2 now
 4300   {
 4301     Label loop;
 4302     bind(loop);
 4303     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 4304     NOT_LP64(movptr(Address(address, index, Address::times_8, offset_in_bytes - 2*BytesPerWord), temp);)
 4305     decrement(index);
 4306     jcc(Assembler::notZero, loop);
 4307   }
 4308 
 4309   bind(done);
 4310 }
 4311 
 4312 // Look up the method for a megamorphic invokeinterface call.
 4313 // The target method is determined by <intf_klass, itable_index>.
 4314 // The receiver klass is in recv_klass.
 4315 // On success, the result will be in method_result, and execution falls through.
 4316 // On failure, execution transfers to the given label.
 4317 void MacroAssembler::lookup_interface_method(Register recv_klass,
 4318                                              Register intf_klass,
 4319                                              RegisterOrConstant itable_index,
 4320                                              Register method_result,
 4321                                              Register scan_temp,
 4322                                              Label& L_no_such_interface,
 4323                                              bool return_method) {
 4324   assert_different_registers(recv_klass, intf_klass, scan_temp);
 4325   assert_different_registers(method_result, intf_klass, scan_temp);
 4326   assert(recv_klass != method_result || !return_method,
 4327          "recv_klass can be destroyed when method isn't needed");
 4328 
 4329   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 4330          "caller must use same register for non-constant itable index as for method");
 4331 
 4332   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 4333   int vtable_base = in_bytes(Klass::vtable_start_offset());
 4334   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 4335   int scan_step   = itableOffsetEntry::size() * wordSize;
 4336   int vte_size    = vtableEntry::size_in_bytes();
 4337   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 4338   assert(vte_size == wordSize, "else adjust times_vte_scale");
 4339 
 4340   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 4341 
 4342   // %%% Could store the aligned, prescaled offset in the klassoop.
 4343   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 4344 
 4345   if (return_method) {
 4346     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 4347     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4348     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 4349   }
 4350 
 4351   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 4352   //   if (scan->interface() == intf) {
 4353   //     result = (klass + scan->offset() + itable_index);
 4354   //   }
 4355   // }
 4356   Label search, found_method;
 4357 
 4358   for (int peel = 1; peel >= 0; peel--) {
 4359     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 4360     cmpptr(intf_klass, method_result);
 4361 
 4362     if (peel) {
 4363       jccb(Assembler::equal, found_method);
 4364     } else {
 4365       jccb(Assembler::notEqual, search);
 4366       // (invert the test to fall through to found_method...)
 4367     }
 4368 
 4369     if (!peel)  break;
 4370 
 4371     bind(search);
 4372 
 4373     // Check that the previous entry is non-null.  A null entry means that
 4374     // the receiver class doesn't implement the interface, and wasn't the
 4375     // same as when the caller was compiled.
 4376     testptr(method_result, method_result);
 4377     jcc(Assembler::zero, L_no_such_interface);
 4378     addptr(scan_temp, scan_step);
 4379   }
 4380 
 4381   bind(found_method);
 4382 
 4383   if (return_method) {
 4384     // Got a hit.
 4385     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 4386     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 4387   }
 4388 }
 4389 
 4390 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 4391 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 4392 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 4393 // The target method is determined by <holder_klass, itable_index>.
 4394 // The receiver klass is in recv_klass.
 4395 // On success, the result will be in method_result, and execution falls through.
 4396 // On failure, execution transfers to the given label.
 4397 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 4398                                                   Register holder_klass,
 4399                                                   Register resolved_klass,
 4400                                                   Register method_result,
 4401                                                   Register scan_temp,
 4402                                                   Register temp_reg2,
 4403                                                   Register receiver,
 4404                                                   int itable_index,
 4405                                                   Label& L_no_such_interface) {
 4406   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 4407   Register temp_itbl_klass = method_result;
 4408   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 4409 
 4410   int vtable_base = in_bytes(Klass::vtable_start_offset());
 4411   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 4412   int scan_step = itableOffsetEntry::size() * wordSize;
 4413   int vte_size = vtableEntry::size_in_bytes();
 4414   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 4415   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 4416   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 4417   assert(vte_size == wordSize, "adjust times_vte_scale");
 4418 
 4419   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 4420 
 4421   // temp_itbl_klass = recv_klass.itable[0]
 4422   // scan_temp = &recv_klass.itable[0] + step
 4423   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 4424   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 4425   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 4426   xorptr(temp_reg, temp_reg);
 4427 
 4428   // Initial checks:
 4429   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 4430   //   - if (itable[0] == 0), no such interface
 4431   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 4432   cmpptr(holder_klass, resolved_klass);
 4433   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 4434   testptr(temp_itbl_klass, temp_itbl_klass);
 4435   jccb(Assembler::zero, L_no_such_interface);
 4436   cmpptr(holder_klass, temp_itbl_klass);
 4437   jccb(Assembler::equal, L_holder_found);
 4438 
 4439   // Loop: Look for holder_klass record in itable
 4440   //   do {
 4441   //     tmp = itable[index];
 4442   //     index += step;
 4443   //     if (tmp == holder_klass) {
 4444   //       goto L_holder_found; // Found!
 4445   //     }
 4446   //   } while (tmp != 0);
 4447   //   goto L_no_such_interface // Not found.
 4448   Label L_scan_holder;
 4449   bind(L_scan_holder);
 4450     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4451     addptr(scan_temp, scan_step);
 4452     cmpptr(holder_klass, temp_itbl_klass);
 4453     jccb(Assembler::equal, L_holder_found);
 4454     testptr(temp_itbl_klass, temp_itbl_klass);
 4455     jccb(Assembler::notZero, L_scan_holder);
 4456 
 4457   jmpb(L_no_such_interface);
 4458 
 4459   // Loop: Look for resolved_class record in itable
 4460   //   do {
 4461   //     tmp = itable[index];
 4462   //     index += step;
 4463   //     if (tmp == holder_klass) {
 4464   //        // Also check if we have met a holder klass
 4465   //        holder_tmp = itable[index-step-ioffset];
 4466   //     }
 4467   //     if (tmp == resolved_klass) {
 4468   //        goto L_resolved_found;  // Found!
 4469   //     }
 4470   //   } while (tmp != 0);
 4471   //   goto L_no_such_interface // Not found.
 4472   //
 4473   Label L_loop_scan_resolved;
 4474   bind(L_loop_scan_resolved);
 4475     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4476     addptr(scan_temp, scan_step);
 4477     bind(L_loop_scan_resolved_entry);
 4478     cmpptr(holder_klass, temp_itbl_klass);
 4479     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4480     cmpptr(resolved_klass, temp_itbl_klass);
 4481     jccb(Assembler::equal, L_resolved_found);
 4482     testptr(temp_itbl_klass, temp_itbl_klass);
 4483     jccb(Assembler::notZero, L_loop_scan_resolved);
 4484 
 4485   jmpb(L_no_such_interface);
 4486 
 4487   Label L_ready;
 4488 
 4489   // See if we already have a holder klass. If not, go and scan for it.
 4490   bind(L_resolved_found);
 4491   testptr(temp_reg, temp_reg);
 4492   jccb(Assembler::zero, L_scan_holder);
 4493   jmpb(L_ready);
 4494 
 4495   bind(L_holder_found);
 4496   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4497 
 4498   // Finally, temp_reg contains holder_klass vtable offset
 4499   bind(L_ready);
 4500   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4501   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4502     load_klass(scan_temp, receiver, noreg);
 4503     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4504   } else {
 4505     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4506   }
 4507 }
 4508 
 4509 
 4510 // virtual method calling
 4511 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4512                                            RegisterOrConstant vtable_index,
 4513                                            Register method_result) {
 4514   const ByteSize base = Klass::vtable_start_offset();
 4515   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4516   Address vtable_entry_addr(recv_klass,
 4517                             vtable_index, Address::times_ptr,
 4518                             base + vtableEntry::method_offset());
 4519   movptr(method_result, vtable_entry_addr);
 4520 }
 4521 
 4522 
 4523 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4524                            Register super_klass,
 4525                            Register temp_reg,
 4526                            Label& L_success) {
 4527   Label L_failure;
 4528   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4529   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4530   bind(L_failure);
 4531 }
 4532 
 4533 
 4534 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4535                                                    Register super_klass,
 4536                                                    Register temp_reg,
 4537                                                    Label* L_success,
 4538                                                    Label* L_failure,
 4539                                                    Label* L_slow_path,
 4540                                         RegisterOrConstant super_check_offset) {
 4541   assert_different_registers(sub_klass, super_klass, temp_reg);
 4542   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4543   if (super_check_offset.is_register()) {
 4544     assert_different_registers(sub_klass, super_klass,
 4545                                super_check_offset.as_register());
 4546   } else if (must_load_sco) {
 4547     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4548   }
 4549 
 4550   Label L_fallthrough;
 4551   int label_nulls = 0;
 4552   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4553   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4554   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4555   assert(label_nulls <= 1, "at most one null in the batch");
 4556 
 4557   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4558   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4559   Address super_check_offset_addr(super_klass, sco_offset);
 4560 
 4561   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4562   // range of a jccb.  If this routine grows larger, reconsider at
 4563   // least some of these.
 4564 #define local_jcc(assembler_cond, label)                                \
 4565   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4566   else                             jcc( assembler_cond, label) /*omit semi*/
 4567 
 4568   // Hacked jmp, which may only be used just before L_fallthrough.
 4569 #define final_jmp(label)                                                \
 4570   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4571   else                            jmp(label)                /*omit semi*/
 4572 
 4573   // If the pointers are equal, we are done (e.g., String[] elements).
 4574   // This self-check enables sharing of secondary supertype arrays among
 4575   // non-primary types such as array-of-interface.  Otherwise, each such
 4576   // type would need its own customized SSA.
 4577   // We move this check to the front of the fast path because many
 4578   // type checks are in fact trivially successful in this manner,
 4579   // so we get a nicely predicted branch right at the start of the check.
 4580   cmpptr(sub_klass, super_klass);
 4581   local_jcc(Assembler::equal, *L_success);
 4582 
 4583   // Check the supertype display:
 4584   if (must_load_sco) {
 4585     // Positive movl does right thing on LP64.
 4586     movl(temp_reg, super_check_offset_addr);
 4587     super_check_offset = RegisterOrConstant(temp_reg);
 4588   }
 4589   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4590   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4591 
 4592   // This check has worked decisively for primary supers.
 4593   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4594   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4595   // This works in the same check above because of a tricky aliasing
 4596   // between the super_cache and the primary super display elements.
 4597   // (The 'super_check_addr' can address either, as the case requires.)
 4598   // Note that the cache is updated below if it does not help us find
 4599   // what we need immediately.
 4600   // So if it was a primary super, we can just fail immediately.
 4601   // Otherwise, it's the slow path for us (no success at this point).
 4602 
 4603   if (super_check_offset.is_register()) {
 4604     local_jcc(Assembler::equal, *L_success);
 4605     cmpl(super_check_offset.as_register(), sc_offset);
 4606     if (L_failure == &L_fallthrough) {
 4607       local_jcc(Assembler::equal, *L_slow_path);
 4608     } else {
 4609       local_jcc(Assembler::notEqual, *L_failure);
 4610       final_jmp(*L_slow_path);
 4611     }
 4612   } else if (super_check_offset.as_constant() == sc_offset) {
 4613     // Need a slow path; fast failure is impossible.
 4614     if (L_slow_path == &L_fallthrough) {
 4615       local_jcc(Assembler::equal, *L_success);
 4616     } else {
 4617       local_jcc(Assembler::notEqual, *L_slow_path);
 4618       final_jmp(*L_success);
 4619     }
 4620   } else {
 4621     // No slow path; it's a fast decision.
 4622     if (L_failure == &L_fallthrough) {
 4623       local_jcc(Assembler::equal, *L_success);
 4624     } else {
 4625       local_jcc(Assembler::notEqual, *L_failure);
 4626       final_jmp(*L_success);
 4627     }
 4628   }
 4629 
 4630   bind(L_fallthrough);
 4631 
 4632 #undef local_jcc
 4633 #undef final_jmp
 4634 }
 4635 
 4636 
 4637 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4638                                                    Register super_klass,
 4639                                                    Register temp_reg,
 4640                                                    Register temp2_reg,
 4641                                                    Label* L_success,
 4642                                                    Label* L_failure,
 4643                                                    bool set_cond_codes) {
 4644   assert_different_registers(sub_klass, super_klass, temp_reg);
 4645   if (temp2_reg != noreg)
 4646     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4647 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4648 
 4649   Label L_fallthrough;
 4650   int label_nulls = 0;
 4651   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4652   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4653   assert(label_nulls <= 1, "at most one null in the batch");
 4654 
 4655   // a couple of useful fields in sub_klass:
 4656   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4657   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4658   Address secondary_supers_addr(sub_klass, ss_offset);
 4659   Address super_cache_addr(     sub_klass, sc_offset);
 4660 
 4661   // Do a linear scan of the secondary super-klass chain.
 4662   // This code is rarely used, so simplicity is a virtue here.
 4663   // The repne_scan instruction uses fixed registers, which we must spill.
 4664   // Don't worry too much about pre-existing connections with the input regs.
 4665 
 4666   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4667   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4668 
 4669   // Get super_klass value into rax (even if it was in rdi or rcx).
 4670   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4671   if (super_klass != rax) {
 4672     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4673     mov(rax, super_klass);
 4674   }
 4675   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4676   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4677 
 4678 #ifndef PRODUCT
 4679   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4680   ExternalAddress pst_counter_addr((address) pst_counter);
 4681   NOT_LP64(  incrementl(pst_counter_addr) );
 4682   LP64_ONLY( lea(rcx, pst_counter_addr) );
 4683   LP64_ONLY( incrementl(Address(rcx, 0)) );
 4684 #endif //PRODUCT
 4685 
 4686   // We will consult the secondary-super array.
 4687   movptr(rdi, secondary_supers_addr);
 4688   // Load the array length.  (Positive movl does right thing on LP64.)
 4689   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4690   // Skip to start of data.
 4691   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4692 
 4693   // Scan RCX words at [RDI] for an occurrence of RAX.
 4694   // Set NZ/Z based on last compare.
 4695   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4696   // not change flags (only scas instruction which is repeated sets flags).
 4697   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4698 
 4699     testptr(rax,rax); // Set Z = 0
 4700     repne_scan();
 4701 
 4702   // Unspill the temp. registers:
 4703   if (pushed_rdi)  pop(rdi);
 4704   if (pushed_rcx)  pop(rcx);
 4705   if (pushed_rax)  pop(rax);
 4706 
 4707   if (set_cond_codes) {
 4708     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4709     assert(!pushed_rdi, "rdi must be left non-null");
 4710     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4711   }
 4712 
 4713   if (L_failure == &L_fallthrough)
 4714         jccb(Assembler::notEqual, *L_failure);
 4715   else  jcc(Assembler::notEqual, *L_failure);
 4716 
 4717   // Success.  Cache the super we found and proceed in triumph.
 4718   movptr(super_cache_addr, super_klass);
 4719 
 4720   if (L_success != &L_fallthrough) {
 4721     jmp(*L_success);
 4722   }
 4723 
 4724 #undef IS_A_TEMP
 4725 
 4726   bind(L_fallthrough);
 4727 }
 4728 
 4729 void MacroAssembler::clinit_barrier(Register klass, Register thread, Label* L_fast_path, Label* L_slow_path) {
 4730   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4731 
 4732   Label L_fallthrough;
 4733   if (L_fast_path == nullptr) {
 4734     L_fast_path = &L_fallthrough;
 4735   } else if (L_slow_path == nullptr) {
 4736     L_slow_path = &L_fallthrough;
 4737   }
 4738 
 4739   // Fast path check: class is fully initialized
 4740   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4741   jcc(Assembler::equal, *L_fast_path);
 4742 
 4743   // Fast path check: current thread is initializer thread
 4744   cmpptr(thread, Address(klass, InstanceKlass::init_thread_offset()));
 4745   if (L_slow_path == &L_fallthrough) {
 4746     jcc(Assembler::equal, *L_fast_path);
 4747     bind(*L_slow_path);
 4748   } else if (L_fast_path == &L_fallthrough) {
 4749     jcc(Assembler::notEqual, *L_slow_path);
 4750     bind(*L_fast_path);
 4751   } else {
 4752     Unimplemented();
 4753   }
 4754 }
 4755 
 4756 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4757   if (VM_Version::supports_cmov()) {
 4758     cmovl(cc, dst, src);
 4759   } else {
 4760     Label L;
 4761     jccb(negate_condition(cc), L);
 4762     movl(dst, src);
 4763     bind(L);
 4764   }
 4765 }
 4766 
 4767 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4768   if (VM_Version::supports_cmov()) {
 4769     cmovl(cc, dst, src);
 4770   } else {
 4771     Label L;
 4772     jccb(negate_condition(cc), L);
 4773     movl(dst, src);
 4774     bind(L);
 4775   }
 4776 }
 4777 
 4778 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4779   if (!VerifyOops) return;
 4780 
 4781   BLOCK_COMMENT("verify_oop {");
 4782 #ifdef _LP64
 4783   push(rscratch1);
 4784 #endif
 4785   push(rax);                          // save rax
 4786   push(reg);                          // pass register argument
 4787 
 4788   // Pass register number to verify_oop_subroutine
 4789   const char* b = nullptr;
 4790   {
 4791     ResourceMark rm;
 4792     stringStream ss;
 4793     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4794     b = code_string(ss.as_string());
 4795   }
 4796   ExternalAddress buffer((address) b);
 4797   pushptr(buffer.addr(), rscratch1);
 4798 
 4799   // call indirectly to solve generation ordering problem
 4800   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4801   call(rax);
 4802   // Caller pops the arguments (oop, message) and restores rax, r10
 4803   BLOCK_COMMENT("} verify_oop");
 4804 }
 4805 
 4806 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4807   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4808     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4809     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4810     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4811   } else if (VM_Version::supports_avx()) {
 4812     vpcmpeqd(dst, dst, dst, vector_len);
 4813   } else {
 4814     assert(VM_Version::supports_sse2(), "");
 4815     pcmpeqd(dst, dst);
 4816   }
 4817 }
 4818 
 4819 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4820                                          int extra_slot_offset) {
 4821   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4822   int stackElementSize = Interpreter::stackElementSize;
 4823   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4824 #ifdef ASSERT
 4825   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4826   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4827 #endif
 4828   Register             scale_reg    = noreg;
 4829   Address::ScaleFactor scale_factor = Address::no_scale;
 4830   if (arg_slot.is_constant()) {
 4831     offset += arg_slot.as_constant() * stackElementSize;
 4832   } else {
 4833     scale_reg    = arg_slot.as_register();
 4834     scale_factor = Address::times(stackElementSize);
 4835   }
 4836   offset += wordSize;           // return PC is on stack
 4837   return Address(rsp, scale_reg, scale_factor, offset);
 4838 }
 4839 
 4840 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 4841   if (!VerifyOops) return;
 4842 
 4843 #ifdef _LP64
 4844   push(rscratch1);
 4845 #endif
 4846   push(rax); // save rax,
 4847   // addr may contain rsp so we will have to adjust it based on the push
 4848   // we just did (and on 64 bit we do two pushes)
 4849   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 4850   // stores rax into addr which is backwards of what was intended.
 4851   if (addr.uses(rsp)) {
 4852     lea(rax, addr);
 4853     pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
 4854   } else {
 4855     pushptr(addr);
 4856   }
 4857 
 4858   // Pass register number to verify_oop_subroutine
 4859   const char* b = nullptr;
 4860   {
 4861     ResourceMark rm;
 4862     stringStream ss;
 4863     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 4864     b = code_string(ss.as_string());
 4865   }
 4866   ExternalAddress buffer((address) b);
 4867   pushptr(buffer.addr(), rscratch1);
 4868 
 4869   // call indirectly to solve generation ordering problem
 4870   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4871   call(rax);
 4872   // Caller pops the arguments (addr, message) and restores rax, r10.
 4873 }
 4874 
 4875 void MacroAssembler::verify_tlab() {
 4876 #ifdef ASSERT
 4877   if (UseTLAB && VerifyOops) {
 4878     Label next, ok;
 4879     Register t1 = rsi;
 4880     Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread);
 4881 
 4882     push(t1);
 4883     NOT_LP64(push(thread_reg));
 4884     NOT_LP64(get_thread(thread_reg));
 4885 
 4886     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
 4887     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
 4888     jcc(Assembler::aboveEqual, next);
 4889     STOP("assert(top >= start)");
 4890     should_not_reach_here();
 4891 
 4892     bind(next);
 4893     movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
 4894     cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
 4895     jcc(Assembler::aboveEqual, ok);
 4896     STOP("assert(top <= end)");
 4897     should_not_reach_here();
 4898 
 4899     bind(ok);
 4900     NOT_LP64(pop(thread_reg));
 4901     pop(t1);
 4902   }
 4903 #endif
 4904 }
 4905 
 4906 class ControlWord {
 4907  public:
 4908   int32_t _value;
 4909 
 4910   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 4911   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 4912   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 4913   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 4914   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 4915   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 4916   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 4917   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 4918 
 4919   void print() const {
 4920     // rounding control
 4921     const char* rc;
 4922     switch (rounding_control()) {
 4923       case 0: rc = "round near"; break;
 4924       case 1: rc = "round down"; break;
 4925       case 2: rc = "round up  "; break;
 4926       case 3: rc = "chop      "; break;
 4927       default:
 4928         rc = nullptr; // silence compiler warnings
 4929         fatal("Unknown rounding control: %d", rounding_control());
 4930     };
 4931     // precision control
 4932     const char* pc;
 4933     switch (precision_control()) {
 4934       case 0: pc = "24 bits "; break;
 4935       case 1: pc = "reserved"; break;
 4936       case 2: pc = "53 bits "; break;
 4937       case 3: pc = "64 bits "; break;
 4938       default:
 4939         pc = nullptr; // silence compiler warnings
 4940         fatal("Unknown precision control: %d", precision_control());
 4941     };
 4942     // flags
 4943     char f[9];
 4944     f[0] = ' ';
 4945     f[1] = ' ';
 4946     f[2] = (precision   ()) ? 'P' : 'p';
 4947     f[3] = (underflow   ()) ? 'U' : 'u';
 4948     f[4] = (overflow    ()) ? 'O' : 'o';
 4949     f[5] = (zero_divide ()) ? 'Z' : 'z';
 4950     f[6] = (denormalized()) ? 'D' : 'd';
 4951     f[7] = (invalid     ()) ? 'I' : 'i';
 4952     f[8] = '\x0';
 4953     // output
 4954     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 4955   }
 4956 
 4957 };
 4958 
 4959 class StatusWord {
 4960  public:
 4961   int32_t _value;
 4962 
 4963   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 4964   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 4965   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 4966   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 4967   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 4968   int  top() const                     { return  (_value >> 11) & 7      ; }
 4969   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 4970   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 4971   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 4972   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 4973   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 4974   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 4975   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 4976   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 4977 
 4978   void print() const {
 4979     // condition codes
 4980     char c[5];
 4981     c[0] = (C3()) ? '3' : '-';
 4982     c[1] = (C2()) ? '2' : '-';
 4983     c[2] = (C1()) ? '1' : '-';
 4984     c[3] = (C0()) ? '0' : '-';
 4985     c[4] = '\x0';
 4986     // flags
 4987     char f[9];
 4988     f[0] = (error_status()) ? 'E' : '-';
 4989     f[1] = (stack_fault ()) ? 'S' : '-';
 4990     f[2] = (precision   ()) ? 'P' : '-';
 4991     f[3] = (underflow   ()) ? 'U' : '-';
 4992     f[4] = (overflow    ()) ? 'O' : '-';
 4993     f[5] = (zero_divide ()) ? 'Z' : '-';
 4994     f[6] = (denormalized()) ? 'D' : '-';
 4995     f[7] = (invalid     ()) ? 'I' : '-';
 4996     f[8] = '\x0';
 4997     // output
 4998     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 4999   }
 5000 
 5001 };
 5002 
 5003 class TagWord {
 5004  public:
 5005   int32_t _value;
 5006 
 5007   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5008 
 5009   void print() const {
 5010     printf("%04x", _value & 0xFFFF);
 5011   }
 5012 
 5013 };
 5014 
 5015 class FPU_Register {
 5016  public:
 5017   int32_t _m0;
 5018   int32_t _m1;
 5019   int16_t _ex;
 5020 
 5021   bool is_indefinite() const           {
 5022     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5023   }
 5024 
 5025   void print() const {
 5026     char  sign = (_ex < 0) ? '-' : '+';
 5027     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5028     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5029   };
 5030 
 5031 };
 5032 
 5033 class FPU_State {
 5034  public:
 5035   enum {
 5036     register_size       = 10,
 5037     number_of_registers =  8,
 5038     register_mask       =  7
 5039   };
 5040 
 5041   ControlWord  _control_word;
 5042   StatusWord   _status_word;
 5043   TagWord      _tag_word;
 5044   int32_t      _error_offset;
 5045   int32_t      _error_selector;
 5046   int32_t      _data_offset;
 5047   int32_t      _data_selector;
 5048   int8_t       _register[register_size * number_of_registers];
 5049 
 5050   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5051   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5052 
 5053   const char* tag_as_string(int tag) const {
 5054     switch (tag) {
 5055       case 0: return "valid";
 5056       case 1: return "zero";
 5057       case 2: return "special";
 5058       case 3: return "empty";
 5059     }
 5060     ShouldNotReachHere();
 5061     return nullptr;
 5062   }
 5063 
 5064   void print() const {
 5065     // print computation registers
 5066     { int t = _status_word.top();
 5067       for (int i = 0; i < number_of_registers; i++) {
 5068         int j = (i - t) & register_mask;
 5069         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5070         st(j)->print();
 5071         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5072       }
 5073     }
 5074     printf("\n");
 5075     // print control registers
 5076     printf("ctrl = "); _control_word.print(); printf("\n");
 5077     printf("stat = "); _status_word .print(); printf("\n");
 5078     printf("tags = "); _tag_word    .print(); printf("\n");
 5079   }
 5080 
 5081 };
 5082 
 5083 class Flag_Register {
 5084  public:
 5085   int32_t _value;
 5086 
 5087   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5088   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5089   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5090   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5091   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5092   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5093   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5094 
 5095   void print() const {
 5096     // flags
 5097     char f[8];
 5098     f[0] = (overflow       ()) ? 'O' : '-';
 5099     f[1] = (direction      ()) ? 'D' : '-';
 5100     f[2] = (sign           ()) ? 'S' : '-';
 5101     f[3] = (zero           ()) ? 'Z' : '-';
 5102     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5103     f[5] = (parity         ()) ? 'P' : '-';
 5104     f[6] = (carry          ()) ? 'C' : '-';
 5105     f[7] = '\x0';
 5106     // output
 5107     printf("%08x  flags = %s", _value, f);
 5108   }
 5109 
 5110 };
 5111 
 5112 class IU_Register {
 5113  public:
 5114   int32_t _value;
 5115 
 5116   void print() const {
 5117     printf("%08x  %11d", _value, _value);
 5118   }
 5119 
 5120 };
 5121 
 5122 class IU_State {
 5123  public:
 5124   Flag_Register _eflags;
 5125   IU_Register   _rdi;
 5126   IU_Register   _rsi;
 5127   IU_Register   _rbp;
 5128   IU_Register   _rsp;
 5129   IU_Register   _rbx;
 5130   IU_Register   _rdx;
 5131   IU_Register   _rcx;
 5132   IU_Register   _rax;
 5133 
 5134   void print() const {
 5135     // computation registers
 5136     printf("rax,  = "); _rax.print(); printf("\n");
 5137     printf("rbx,  = "); _rbx.print(); printf("\n");
 5138     printf("rcx  = "); _rcx.print(); printf("\n");
 5139     printf("rdx  = "); _rdx.print(); printf("\n");
 5140     printf("rdi  = "); _rdi.print(); printf("\n");
 5141     printf("rsi  = "); _rsi.print(); printf("\n");
 5142     printf("rbp,  = "); _rbp.print(); printf("\n");
 5143     printf("rsp  = "); _rsp.print(); printf("\n");
 5144     printf("\n");
 5145     // control registers
 5146     printf("flgs = "); _eflags.print(); printf("\n");
 5147   }
 5148 };
 5149 
 5150 
 5151 class CPU_State {
 5152  public:
 5153   FPU_State _fpu_state;
 5154   IU_State  _iu_state;
 5155 
 5156   void print() const {
 5157     printf("--------------------------------------------------\n");
 5158     _iu_state .print();
 5159     printf("\n");
 5160     _fpu_state.print();
 5161     printf("--------------------------------------------------\n");
 5162   }
 5163 
 5164 };
 5165 
 5166 
 5167 static void _print_CPU_state(CPU_State* state) {
 5168   state->print();
 5169 };
 5170 
 5171 
 5172 void MacroAssembler::print_CPU_state() {
 5173   push_CPU_state();
 5174   push(rsp);                // pass CPU state
 5175   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5176   addptr(rsp, wordSize);       // discard argument
 5177   pop_CPU_state();
 5178 }
 5179 
 5180 
 5181 #ifndef _LP64
 5182 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) {
 5183   static int counter = 0;
 5184   FPU_State* fs = &state->_fpu_state;
 5185   counter++;
 5186   // For leaf calls, only verify that the top few elements remain empty.
 5187   // We only need 1 empty at the top for C2 code.
 5188   if( stack_depth < 0 ) {
 5189     if( fs->tag_for_st(7) != 3 ) {
 5190       printf("FPR7 not empty\n");
 5191       state->print();
 5192       assert(false, "error");
 5193       return false;
 5194     }
 5195     return true;                // All other stack states do not matter
 5196   }
 5197 
 5198   assert((fs->_control_word._value & 0xffff) == StubRoutines::x86::fpu_cntrl_wrd_std(),
 5199          "bad FPU control word");
 5200 
 5201   // compute stack depth
 5202   int i = 0;
 5203   while (i < FPU_State::number_of_registers && fs->tag_for_st(i)  < 3) i++;
 5204   int d = i;
 5205   while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++;
 5206   // verify findings
 5207   if (i != FPU_State::number_of_registers) {
 5208     // stack not contiguous
 5209     printf("%s: stack not contiguous at ST%d\n", s, i);
 5210     state->print();
 5211     assert(false, "error");
 5212     return false;
 5213   }
 5214   // check if computed stack depth corresponds to expected stack depth
 5215   if (stack_depth < 0) {
 5216     // expected stack depth is -stack_depth or less
 5217     if (d > -stack_depth) {
 5218       // too many elements on the stack
 5219       printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d);
 5220       state->print();
 5221       assert(false, "error");
 5222       return false;
 5223     }
 5224   } else {
 5225     // expected stack depth is stack_depth
 5226     if (d != stack_depth) {
 5227       // wrong stack depth
 5228       printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d);
 5229       state->print();
 5230       assert(false, "error");
 5231       return false;
 5232     }
 5233   }
 5234   // everything is cool
 5235   return true;
 5236 }
 5237 
 5238 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
 5239   if (!VerifyFPU) return;
 5240   push_CPU_state();
 5241   push(rsp);                // pass CPU state
 5242   ExternalAddress msg((address) s);
 5243   // pass message string s
 5244   pushptr(msg.addr(), noreg);
 5245   push(stack_depth);        // pass stack depth
 5246   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU)));
 5247   addptr(rsp, 3 * wordSize);   // discard arguments
 5248   // check for error
 5249   { Label L;
 5250     testl(rax, rax);
 5251     jcc(Assembler::notZero, L);
 5252     int3();                  // break if error condition
 5253     bind(L);
 5254   }
 5255   pop_CPU_state();
 5256 }
 5257 #endif // _LP64
 5258 
 5259 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5260   // Either restore the MXCSR register after returning from the JNI Call
 5261   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5262   if (VM_Version::supports_sse()) {
 5263     if (RestoreMXCSROnJNICalls) {
 5264       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5265     } else if (CheckJNICalls) {
 5266       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5267     }
 5268   }
 5269   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5270   vzeroupper();
 5271 
 5272 #ifndef _LP64
 5273   // Either restore the x87 floating pointer control word after returning
 5274   // from the JNI call or verify that it wasn't changed.
 5275   if (CheckJNICalls) {
 5276     call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
 5277   }
 5278 #endif // _LP64
 5279 }
 5280 
 5281 // ((OopHandle)result).resolve();
 5282 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5283   assert_different_registers(result, tmp);
 5284 
 5285   // Only 64 bit platforms support GCs that require a tmp register
 5286   // Only IN_HEAP loads require a thread_tmp register
 5287   // OopHandle::resolve is an indirection like jobject.
 5288   access_load_at(T_OBJECT, IN_NATIVE,
 5289                  result, Address(result, 0), tmp, /*tmp_thread*/noreg);
 5290 }
 5291 
 5292 // ((WeakHandle)result).resolve();
 5293 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5294   assert_different_registers(rresult, rtmp);
 5295   Label resolved;
 5296 
 5297   // A null weak handle resolves to null.
 5298   cmpptr(rresult, 0);
 5299   jcc(Assembler::equal, resolved);
 5300 
 5301   // Only 64 bit platforms support GCs that require a tmp register
 5302   // Only IN_HEAP loads require a thread_tmp register
 5303   // WeakHandle::resolve is an indirection like jweak.
 5304   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5305                  rresult, Address(rresult, 0), rtmp, /*tmp_thread*/noreg);
 5306   bind(resolved);
 5307 }
 5308 
 5309 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5310   // get mirror
 5311   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5312   load_method_holder(mirror, method);
 5313   movptr(mirror, Address(mirror, mirror_offset));
 5314   resolve_oop_handle(mirror, tmp);
 5315 }
 5316 
 5317 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5318   load_method_holder(rresult, rmethod);
 5319   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5320 }
 5321 
 5322 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5323   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5324   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5325   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5326 }
 5327 
 5328 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5329   assert_different_registers(src, tmp);
 5330   assert_different_registers(dst, tmp);
 5331 #ifdef _LP64
 5332   if (UseCompressedClassPointers) {
 5333     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5334     decode_klass_not_null(dst, tmp);
 5335   } else
 5336 #endif
 5337     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5338 }
 5339 
 5340 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5341   assert_different_registers(src, tmp);
 5342   assert_different_registers(dst, tmp);
 5343 #ifdef _LP64
 5344   if (UseCompressedClassPointers) {
 5345     encode_klass_not_null(src, tmp);
 5346     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5347   } else
 5348 #endif
 5349     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5350 }
 5351 
 5352 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5353                                     Register tmp1, Register thread_tmp) {
 5354   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5355   decorators = AccessInternal::decorator_fixup(decorators, type);
 5356   bool as_raw = (decorators & AS_RAW) != 0;
 5357   if (as_raw) {
 5358     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 5359   } else {
 5360     bs->load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
 5361   }
 5362 }
 5363 
 5364 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5365                                      Register tmp1, Register tmp2, Register tmp3) {
 5366   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5367   decorators = AccessInternal::decorator_fixup(decorators, type);
 5368   bool as_raw = (decorators & AS_RAW) != 0;
 5369   if (as_raw) {
 5370     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5371   } else {
 5372     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5373   }
 5374 }
 5375 
 5376 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
 5377                                    Register thread_tmp, DecoratorSet decorators) {
 5378   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
 5379 }
 5380 
 5381 // Doesn't do verification, generates fixed size code
 5382 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 5383                                             Register thread_tmp, DecoratorSet decorators) {
 5384   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, thread_tmp);
 5385 }
 5386 
 5387 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5388                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5389   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5390 }
 5391 
 5392 // Used for storing nulls.
 5393 void MacroAssembler::store_heap_oop_null(Address dst) {
 5394   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5395 }
 5396 
 5397 #ifdef _LP64
 5398 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5399   if (UseCompressedClassPointers) {
 5400     // Store to klass gap in destination
 5401     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5402   }
 5403 }
 5404 
 5405 #ifdef ASSERT
 5406 void MacroAssembler::verify_heapbase(const char* msg) {
 5407   assert (UseCompressedOops, "should be compressed");
 5408   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5409   if (CheckCompressedOops) {
 5410     Label ok;
 5411     ExternalAddress src2(CompressedOops::ptrs_base_addr());
 5412     const bool is_src2_reachable = reachable(src2);
 5413     if (!is_src2_reachable) {
 5414       push(rscratch1);  // cmpptr trashes rscratch1
 5415     }
 5416     cmpptr(r12_heapbase, src2, rscratch1);
 5417     jcc(Assembler::equal, ok);
 5418     STOP(msg);
 5419     bind(ok);
 5420     if (!is_src2_reachable) {
 5421       pop(rscratch1);
 5422     }
 5423   }
 5424 }
 5425 #endif
 5426 
 5427 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5428 void MacroAssembler::encode_heap_oop(Register r) {
 5429 #ifdef ASSERT
 5430   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5431 #endif
 5432   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5433   if (CompressedOops::base() == nullptr) {
 5434     if (CompressedOops::shift() != 0) {
 5435       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5436       shrq(r, LogMinObjAlignmentInBytes);
 5437     }
 5438     return;
 5439   }
 5440   testq(r, r);
 5441   cmovq(Assembler::equal, r, r12_heapbase);
 5442   subq(r, r12_heapbase);
 5443   shrq(r, LogMinObjAlignmentInBytes);
 5444 }
 5445 
 5446 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5447 #ifdef ASSERT
 5448   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5449   if (CheckCompressedOops) {
 5450     Label ok;
 5451     testq(r, r);
 5452     jcc(Assembler::notEqual, ok);
 5453     STOP("null oop passed to encode_heap_oop_not_null");
 5454     bind(ok);
 5455   }
 5456 #endif
 5457   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5458   if (CompressedOops::base() != nullptr) {
 5459     subq(r, r12_heapbase);
 5460   }
 5461   if (CompressedOops::shift() != 0) {
 5462     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5463     shrq(r, LogMinObjAlignmentInBytes);
 5464   }
 5465 }
 5466 
 5467 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5468 #ifdef ASSERT
 5469   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5470   if (CheckCompressedOops) {
 5471     Label ok;
 5472     testq(src, src);
 5473     jcc(Assembler::notEqual, ok);
 5474     STOP("null oop passed to encode_heap_oop_not_null2");
 5475     bind(ok);
 5476   }
 5477 #endif
 5478   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5479   if (dst != src) {
 5480     movq(dst, src);
 5481   }
 5482   if (CompressedOops::base() != nullptr) {
 5483     subq(dst, r12_heapbase);
 5484   }
 5485   if (CompressedOops::shift() != 0) {
 5486     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5487     shrq(dst, LogMinObjAlignmentInBytes);
 5488   }
 5489 }
 5490 
 5491 void  MacroAssembler::decode_heap_oop(Register r) {
 5492 #ifdef ASSERT
 5493   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5494 #endif
 5495   if (CompressedOops::base() == nullptr) {
 5496     if (CompressedOops::shift() != 0) {
 5497       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5498       shlq(r, LogMinObjAlignmentInBytes);
 5499     }
 5500   } else {
 5501     Label done;
 5502     shlq(r, LogMinObjAlignmentInBytes);
 5503     jccb(Assembler::equal, done);
 5504     addq(r, r12_heapbase);
 5505     bind(done);
 5506   }
 5507   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5508 }
 5509 
 5510 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5511   // Note: it will change flags
 5512   assert (UseCompressedOops, "should only be used for compressed headers");
 5513   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5514   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5515   // vtableStubs also counts instructions in pd_code_size_limit.
 5516   // Also do not verify_oop as this is called by verify_oop.
 5517   if (CompressedOops::shift() != 0) {
 5518     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5519     shlq(r, LogMinObjAlignmentInBytes);
 5520     if (CompressedOops::base() != nullptr) {
 5521       addq(r, r12_heapbase);
 5522     }
 5523   } else {
 5524     assert (CompressedOops::base() == nullptr, "sanity");
 5525   }
 5526 }
 5527 
 5528 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5529   // Note: it will change flags
 5530   assert (UseCompressedOops, "should only be used for compressed headers");
 5531   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5532   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5533   // vtableStubs also counts instructions in pd_code_size_limit.
 5534   // Also do not verify_oop as this is called by verify_oop.
 5535   if (CompressedOops::shift() != 0) {
 5536     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5537     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5538       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5539     } else {
 5540       if (dst != src) {
 5541         movq(dst, src);
 5542       }
 5543       shlq(dst, LogMinObjAlignmentInBytes);
 5544       if (CompressedOops::base() != nullptr) {
 5545         addq(dst, r12_heapbase);
 5546       }
 5547     }
 5548   } else {
 5549     assert (CompressedOops::base() == nullptr, "sanity");
 5550     if (dst != src) {
 5551       movq(dst, src);
 5552     }
 5553   }
 5554 }
 5555 
 5556 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5557   assert_different_registers(r, tmp);
 5558   if (CompressedKlassPointers::base() != nullptr) {
 5559     mov64(tmp, (int64_t)CompressedKlassPointers::base());
 5560     subq(r, tmp);
 5561   }
 5562   if (CompressedKlassPointers::shift() != 0) {
 5563     assert (LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
 5564     shrq(r, LogKlassAlignmentInBytes);
 5565   }
 5566 }
 5567 
 5568 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5569   assert_different_registers(src, dst);
 5570   if (CompressedKlassPointers::base() != nullptr) {
 5571     mov64(dst, -(int64_t)CompressedKlassPointers::base());
 5572     addq(dst, src);
 5573   } else {
 5574     movptr(dst, src);
 5575   }
 5576   if (CompressedKlassPointers::shift() != 0) {
 5577     assert (LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
 5578     shrq(dst, LogKlassAlignmentInBytes);
 5579   }
 5580 }
 5581 
 5582 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5583   assert_different_registers(r, tmp);
 5584   // Note: it will change flags
 5585   assert(UseCompressedClassPointers, "should only be used for compressed headers");
 5586   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5587   // vtableStubs also counts instructions in pd_code_size_limit.
 5588   // Also do not verify_oop as this is called by verify_oop.
 5589   if (CompressedKlassPointers::shift() != 0) {
 5590     assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
 5591     shlq(r, LogKlassAlignmentInBytes);
 5592   }
 5593   if (CompressedKlassPointers::base() != nullptr) {
 5594     mov64(tmp, (int64_t)CompressedKlassPointers::base());
 5595     addq(r, tmp);
 5596   }
 5597 }
 5598 
 5599 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5600   assert_different_registers(src, dst);
 5601   // Note: it will change flags
 5602   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5603   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5604   // vtableStubs also counts instructions in pd_code_size_limit.
 5605   // Also do not verify_oop as this is called by verify_oop.
 5606 
 5607   if (CompressedKlassPointers::base() == nullptr &&
 5608       CompressedKlassPointers::shift() == 0) {
 5609     // The best case scenario is that there is no base or shift. Then it is already
 5610     // a pointer that needs nothing but a register rename.
 5611     movl(dst, src);
 5612   } else {
 5613     if (CompressedKlassPointers::base() != nullptr) {
 5614       mov64(dst, (int64_t)CompressedKlassPointers::base());
 5615     } else {
 5616       xorq(dst, dst);
 5617     }
 5618     if (CompressedKlassPointers::shift() != 0) {
 5619       assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
 5620       assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
 5621       leaq(dst, Address(dst, src, Address::times_8, 0));
 5622     } else {
 5623       addq(dst, src);
 5624     }
 5625   }
 5626 }
 5627 
 5628 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5629   assert (UseCompressedOops, "should only be used for compressed headers");
 5630   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5631   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5632   int oop_index = oop_recorder()->find_index(obj);
 5633   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5634   mov_narrow_oop(dst, oop_index, rspec);
 5635 }
 5636 
 5637 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5638   assert (UseCompressedOops, "should only be used for compressed headers");
 5639   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5640   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5641   int oop_index = oop_recorder()->find_index(obj);
 5642   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5643   mov_narrow_oop(dst, oop_index, rspec);
 5644 }
 5645 
 5646 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5647   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5648   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5649   int klass_index = oop_recorder()->find_index(k);
 5650   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5651   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5652 }
 5653 
 5654 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5655   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5656   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5657   int klass_index = oop_recorder()->find_index(k);
 5658   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5659   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5660 }
 5661 
 5662 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5663   assert (UseCompressedOops, "should only be used for compressed headers");
 5664   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5665   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5666   int oop_index = oop_recorder()->find_index(obj);
 5667   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5668   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5669 }
 5670 
 5671 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5672   assert (UseCompressedOops, "should only be used for compressed headers");
 5673   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5674   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5675   int oop_index = oop_recorder()->find_index(obj);
 5676   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5677   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5678 }
 5679 
 5680 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5681   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5682   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5683   int klass_index = oop_recorder()->find_index(k);
 5684   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5685   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5686 }
 5687 
 5688 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5689   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5690   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5691   int klass_index = oop_recorder()->find_index(k);
 5692   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5693   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5694 }
 5695 
 5696 void MacroAssembler::reinit_heapbase() {
 5697   if (UseCompressedOops) {
 5698     if (Universe::heap() != nullptr) {
 5699       if (CompressedOops::base() == nullptr) {
 5700         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5701       } else {
 5702         mov64(r12_heapbase, (int64_t)CompressedOops::ptrs_base());
 5703       }
 5704     } else {
 5705       movptr(r12_heapbase, ExternalAddress(CompressedOops::ptrs_base_addr()));
 5706     }
 5707   }
 5708 }
 5709 
 5710 #endif // _LP64
 5711 
 5712 #if COMPILER2_OR_JVMCI
 5713 
 5714 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5715 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5716   // cnt - number of qwords (8-byte words).
 5717   // base - start address, qword aligned.
 5718   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5719   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 5720   if (use64byteVector) {
 5721     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5722   } else if (MaxVectorSize >= 32) {
 5723     vpxor(xtmp, xtmp, xtmp, AVX_256bit);
 5724   } else {
 5725     pxor(xtmp, xtmp);
 5726   }
 5727   jmp(L_zero_64_bytes);
 5728 
 5729   BIND(L_loop);
 5730   if (MaxVectorSize >= 32) {
 5731     fill64(base, 0, xtmp, use64byteVector);
 5732   } else {
 5733     movdqu(Address(base,  0), xtmp);
 5734     movdqu(Address(base, 16), xtmp);
 5735     movdqu(Address(base, 32), xtmp);
 5736     movdqu(Address(base, 48), xtmp);
 5737   }
 5738   addptr(base, 64);
 5739 
 5740   BIND(L_zero_64_bytes);
 5741   subptr(cnt, 8);
 5742   jccb(Assembler::greaterEqual, L_loop);
 5743 
 5744   // Copy trailing 64 bytes
 5745   if (use64byteVector) {
 5746     addptr(cnt, 8);
 5747     jccb(Assembler::equal, L_end);
 5748     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5749     jmp(L_end);
 5750   } else {
 5751     addptr(cnt, 4);
 5752     jccb(Assembler::less, L_tail);
 5753     if (MaxVectorSize >= 32) {
 5754       vmovdqu(Address(base, 0), xtmp);
 5755     } else {
 5756       movdqu(Address(base,  0), xtmp);
 5757       movdqu(Address(base, 16), xtmp);
 5758     }
 5759   }
 5760   addptr(base, 32);
 5761   subptr(cnt, 4);
 5762 
 5763   BIND(L_tail);
 5764   addptr(cnt, 4);
 5765   jccb(Assembler::lessEqual, L_end);
 5766   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5767     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5768   } else {
 5769     decrement(cnt);
 5770 
 5771     BIND(L_sloop);
 5772     movq(Address(base, 0), xtmp);
 5773     addptr(base, 8);
 5774     decrement(cnt);
 5775     jccb(Assembler::greaterEqual, L_sloop);
 5776   }
 5777   BIND(L_end);
 5778 }
 5779 
 5780 // Clearing constant sized memory using YMM/ZMM registers.
 5781 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5782   assert(UseAVX > 2 && VM_Version::supports_avx512vlbw(), "");
 5783   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 5784 
 5785   int vector64_count = (cnt & (~0x7)) >> 3;
 5786   cnt = cnt & 0x7;
 5787   const int fill64_per_loop = 4;
 5788   const int max_unrolled_fill64 = 8;
 5789 
 5790   // 64 byte initialization loop.
 5791   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5792   int start64 = 0;
 5793   if (vector64_count > max_unrolled_fill64) {
 5794     Label LOOP;
 5795     Register index = rtmp;
 5796 
 5797     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5798 
 5799     movl(index, 0);
 5800     BIND(LOOP);
 5801     for (int i = 0; i < fill64_per_loop; i++) {
 5802       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5803     }
 5804     addl(index, fill64_per_loop * 64);
 5805     cmpl(index, start64 * 64);
 5806     jccb(Assembler::less, LOOP);
 5807   }
 5808   for (int i = start64; i < vector64_count; i++) {
 5809     fill64(base, i * 64, xtmp, use64byteVector);
 5810   }
 5811 
 5812   // Clear remaining 64 byte tail.
 5813   int disp = vector64_count * 64;
 5814   if (cnt) {
 5815     switch (cnt) {
 5816       case 1:
 5817         movq(Address(base, disp), xtmp);
 5818         break;
 5819       case 2:
 5820         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5821         break;
 5822       case 3:
 5823         movl(rtmp, 0x7);
 5824         kmovwl(mask, rtmp);
 5825         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5826         break;
 5827       case 4:
 5828         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5829         break;
 5830       case 5:
 5831         if (use64byteVector) {
 5832           movl(rtmp, 0x1F);
 5833           kmovwl(mask, rtmp);
 5834           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5835         } else {
 5836           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5837           movq(Address(base, disp + 32), xtmp);
 5838         }
 5839         break;
 5840       case 6:
 5841         if (use64byteVector) {
 5842           movl(rtmp, 0x3F);
 5843           kmovwl(mask, rtmp);
 5844           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5845         } else {
 5846           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5847           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5848         }
 5849         break;
 5850       case 7:
 5851         if (use64byteVector) {
 5852           movl(rtmp, 0x7F);
 5853           kmovwl(mask, rtmp);
 5854           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5855         } else {
 5856           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5857           movl(rtmp, 0x7);
 5858           kmovwl(mask, rtmp);
 5859           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5860         }
 5861         break;
 5862       default:
 5863         fatal("Unexpected length : %d\n",cnt);
 5864         break;
 5865     }
 5866   }
 5867 }
 5868 
 5869 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5870                                bool is_large, KRegister mask) {
 5871   // cnt      - number of qwords (8-byte words).
 5872   // base     - start address, qword aligned.
 5873   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 5874   assert(base==rdi, "base register must be edi for rep stos");
 5875   assert(tmp==rax,   "tmp register must be eax for rep stos");
 5876   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 5877   assert(InitArrayShortSize % BytesPerLong == 0,
 5878     "InitArrayShortSize should be the multiple of BytesPerLong");
 5879 
 5880   Label DONE;
 5881   if (!is_large || !UseXMMForObjInit) {
 5882     xorptr(tmp, tmp);
 5883   }
 5884 
 5885   if (!is_large) {
 5886     Label LOOP, LONG;
 5887     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 5888     jccb(Assembler::greater, LONG);
 5889 
 5890     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 5891 
 5892     decrement(cnt);
 5893     jccb(Assembler::negative, DONE); // Zero length
 5894 
 5895     // Use individual pointer-sized stores for small counts:
 5896     BIND(LOOP);
 5897     movptr(Address(base, cnt, Address::times_ptr), tmp);
 5898     decrement(cnt);
 5899     jccb(Assembler::greaterEqual, LOOP);
 5900     jmpb(DONE);
 5901 
 5902     BIND(LONG);
 5903   }
 5904 
 5905   // Use longer rep-prefixed ops for non-small counts:
 5906   if (UseFastStosb) {
 5907     shlptr(cnt, 3); // convert to number of bytes
 5908     rep_stosb();
 5909   } else if (UseXMMForObjInit) {
 5910     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 5911   } else {
 5912     NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
 5913     rep_stos();
 5914   }
 5915 
 5916   BIND(DONE);
 5917 }
 5918 
 5919 #endif //COMPILER2_OR_JVMCI
 5920 
 5921 
 5922 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 5923                                    Register to, Register value, Register count,
 5924                                    Register rtmp, XMMRegister xtmp) {
 5925   ShortBranchVerifier sbv(this);
 5926   assert_different_registers(to, value, count, rtmp);
 5927   Label L_exit;
 5928   Label L_fill_2_bytes, L_fill_4_bytes;
 5929 
 5930 #if defined(COMPILER2) && defined(_LP64)
 5931   if(MaxVectorSize >=32 &&
 5932      VM_Version::supports_avx512vlbw() &&
 5933      VM_Version::supports_bmi2()) {
 5934     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 5935     return;
 5936   }
 5937 #endif
 5938 
 5939   int shift = -1;
 5940   switch (t) {
 5941     case T_BYTE:
 5942       shift = 2;
 5943       break;
 5944     case T_SHORT:
 5945       shift = 1;
 5946       break;
 5947     case T_INT:
 5948       shift = 0;
 5949       break;
 5950     default: ShouldNotReachHere();
 5951   }
 5952 
 5953   if (t == T_BYTE) {
 5954     andl(value, 0xff);
 5955     movl(rtmp, value);
 5956     shll(rtmp, 8);
 5957     orl(value, rtmp);
 5958   }
 5959   if (t == T_SHORT) {
 5960     andl(value, 0xffff);
 5961   }
 5962   if (t == T_BYTE || t == T_SHORT) {
 5963     movl(rtmp, value);
 5964     shll(rtmp, 16);
 5965     orl(value, rtmp);
 5966   }
 5967 
 5968   cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
 5969   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 5970   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 5971     Label L_skip_align2;
 5972     // align source address at 4 bytes address boundary
 5973     if (t == T_BYTE) {
 5974       Label L_skip_align1;
 5975       // One byte misalignment happens only for byte arrays
 5976       testptr(to, 1);
 5977       jccb(Assembler::zero, L_skip_align1);
 5978       movb(Address(to, 0), value);
 5979       increment(to);
 5980       decrement(count);
 5981       BIND(L_skip_align1);
 5982     }
 5983     // Two bytes misalignment happens only for byte and short (char) arrays
 5984     testptr(to, 2);
 5985     jccb(Assembler::zero, L_skip_align2);
 5986     movw(Address(to, 0), value);
 5987     addptr(to, 2);
 5988     subl(count, 1<<(shift-1));
 5989     BIND(L_skip_align2);
 5990   }
 5991   if (UseSSE < 2) {
 5992     Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 5993     // Fill 32-byte chunks
 5994     subl(count, 8 << shift);
 5995     jcc(Assembler::less, L_check_fill_8_bytes);
 5996     align(16);
 5997 
 5998     BIND(L_fill_32_bytes_loop);
 5999 
 6000     for (int i = 0; i < 32; i += 4) {
 6001       movl(Address(to, i), value);
 6002     }
 6003 
 6004     addptr(to, 32);
 6005     subl(count, 8 << shift);
 6006     jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6007     BIND(L_check_fill_8_bytes);
 6008     addl(count, 8 << shift);
 6009     jccb(Assembler::zero, L_exit);
 6010     jmpb(L_fill_8_bytes);
 6011 
 6012     //
 6013     // length is too short, just fill qwords
 6014     //
 6015     BIND(L_fill_8_bytes_loop);
 6016     movl(Address(to, 0), value);
 6017     movl(Address(to, 4), value);
 6018     addptr(to, 8);
 6019     BIND(L_fill_8_bytes);
 6020     subl(count, 1 << (shift + 1));
 6021     jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6022     // fall through to fill 4 bytes
 6023   } else {
 6024     Label L_fill_32_bytes;
 6025     if (!UseUnalignedLoadStores) {
 6026       // align to 8 bytes, we know we are 4 byte aligned to start
 6027       testptr(to, 4);
 6028       jccb(Assembler::zero, L_fill_32_bytes);
 6029       movl(Address(to, 0), value);
 6030       addptr(to, 4);
 6031       subl(count, 1<<shift);
 6032     }
 6033     BIND(L_fill_32_bytes);
 6034     {
 6035       assert( UseSSE >= 2, "supported cpu only" );
 6036       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6037       movdl(xtmp, value);
 6038       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6039         Label L_check_fill_32_bytes;
 6040         if (UseAVX > 2) {
 6041           // Fill 64-byte chunks
 6042           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6043 
 6044           // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
 6045           cmpl(count, VM_Version::avx3_threshold());
 6046           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6047 
 6048           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6049 
 6050           subl(count, 16 << shift);
 6051           jccb(Assembler::less, L_check_fill_32_bytes);
 6052           align(16);
 6053 
 6054           BIND(L_fill_64_bytes_loop_avx3);
 6055           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6056           addptr(to, 64);
 6057           subl(count, 16 << shift);
 6058           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6059           jmpb(L_check_fill_32_bytes);
 6060 
 6061           BIND(L_check_fill_64_bytes_avx2);
 6062         }
 6063         // Fill 64-byte chunks
 6064         Label L_fill_64_bytes_loop;
 6065         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6066 
 6067         subl(count, 16 << shift);
 6068         jcc(Assembler::less, L_check_fill_32_bytes);
 6069         align(16);
 6070 
 6071         BIND(L_fill_64_bytes_loop);
 6072         vmovdqu(Address(to, 0), xtmp);
 6073         vmovdqu(Address(to, 32), xtmp);
 6074         addptr(to, 64);
 6075         subl(count, 16 << shift);
 6076         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6077 
 6078         BIND(L_check_fill_32_bytes);
 6079         addl(count, 8 << shift);
 6080         jccb(Assembler::less, L_check_fill_8_bytes);
 6081         vmovdqu(Address(to, 0), xtmp);
 6082         addptr(to, 32);
 6083         subl(count, 8 << shift);
 6084 
 6085         BIND(L_check_fill_8_bytes);
 6086         // clean upper bits of YMM registers
 6087         movdl(xtmp, value);
 6088         pshufd(xtmp, xtmp, 0);
 6089       } else {
 6090         // Fill 32-byte chunks
 6091         pshufd(xtmp, xtmp, 0);
 6092 
 6093         subl(count, 8 << shift);
 6094         jcc(Assembler::less, L_check_fill_8_bytes);
 6095         align(16);
 6096 
 6097         BIND(L_fill_32_bytes_loop);
 6098 
 6099         if (UseUnalignedLoadStores) {
 6100           movdqu(Address(to, 0), xtmp);
 6101           movdqu(Address(to, 16), xtmp);
 6102         } else {
 6103           movq(Address(to, 0), xtmp);
 6104           movq(Address(to, 8), xtmp);
 6105           movq(Address(to, 16), xtmp);
 6106           movq(Address(to, 24), xtmp);
 6107         }
 6108 
 6109         addptr(to, 32);
 6110         subl(count, 8 << shift);
 6111         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6112 
 6113         BIND(L_check_fill_8_bytes);
 6114       }
 6115       addl(count, 8 << shift);
 6116       jccb(Assembler::zero, L_exit);
 6117       jmpb(L_fill_8_bytes);
 6118 
 6119       //
 6120       // length is too short, just fill qwords
 6121       //
 6122       BIND(L_fill_8_bytes_loop);
 6123       movq(Address(to, 0), xtmp);
 6124       addptr(to, 8);
 6125       BIND(L_fill_8_bytes);
 6126       subl(count, 1 << (shift + 1));
 6127       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6128     }
 6129   }
 6130   // fill trailing 4 bytes
 6131   BIND(L_fill_4_bytes);
 6132   testl(count, 1<<shift);
 6133   jccb(Assembler::zero, L_fill_2_bytes);
 6134   movl(Address(to, 0), value);
 6135   if (t == T_BYTE || t == T_SHORT) {
 6136     Label L_fill_byte;
 6137     addptr(to, 4);
 6138     BIND(L_fill_2_bytes);
 6139     // fill trailing 2 bytes
 6140     testl(count, 1<<(shift-1));
 6141     jccb(Assembler::zero, L_fill_byte);
 6142     movw(Address(to, 0), value);
 6143     if (t == T_BYTE) {
 6144       addptr(to, 2);
 6145       BIND(L_fill_byte);
 6146       // fill trailing byte
 6147       testl(count, 1);
 6148       jccb(Assembler::zero, L_exit);
 6149       movb(Address(to, 0), value);
 6150     } else {
 6151       BIND(L_fill_byte);
 6152     }
 6153   } else {
 6154     BIND(L_fill_2_bytes);
 6155   }
 6156   BIND(L_exit);
 6157 }
 6158 
 6159 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6160   switch(type) {
 6161     case T_BYTE:
 6162     case T_BOOLEAN:
 6163       evpbroadcastb(dst, src, vector_len);
 6164       break;
 6165     case T_SHORT:
 6166     case T_CHAR:
 6167       evpbroadcastw(dst, src, vector_len);
 6168       break;
 6169     case T_INT:
 6170     case T_FLOAT:
 6171       evpbroadcastd(dst, src, vector_len);
 6172       break;
 6173     case T_LONG:
 6174     case T_DOUBLE:
 6175       evpbroadcastq(dst, src, vector_len);
 6176       break;
 6177     default:
 6178       fatal("Unhandled type : %s", type2name(type));
 6179       break;
 6180   }
 6181 }
 6182 
 6183 // encode char[] to byte[] in ISO_8859_1 or ASCII
 6184    //@IntrinsicCandidate
 6185    //private static int implEncodeISOArray(byte[] sa, int sp,
 6186    //byte[] da, int dp, int len) {
 6187    //  int i = 0;
 6188    //  for (; i < len; i++) {
 6189    //    char c = StringUTF16.getChar(sa, sp++);
 6190    //    if (c > '\u00FF')
 6191    //      break;
 6192    //    da[dp++] = (byte)c;
 6193    //  }
 6194    //  return i;
 6195    //}
 6196    //
 6197    //@IntrinsicCandidate
 6198    //private static int implEncodeAsciiArray(char[] sa, int sp,
 6199    //    byte[] da, int dp, int len) {
 6200    //  int i = 0;
 6201    //  for (; i < len; i++) {
 6202    //    char c = sa[sp++];
 6203    //    if (c >= '\u0080')
 6204    //      break;
 6205    //    da[dp++] = (byte)c;
 6206    //  }
 6207    //  return i;
 6208    //}
 6209 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6210   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6211   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6212   Register tmp5, Register result, bool ascii) {
 6213 
 6214   // rsi: src
 6215   // rdi: dst
 6216   // rdx: len
 6217   // rcx: tmp5
 6218   // rax: result
 6219   ShortBranchVerifier sbv(this);
 6220   assert_different_registers(src, dst, len, tmp5, result);
 6221   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6222 
 6223   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6224   int short_mask = ascii ? 0xff80 : 0xff00;
 6225 
 6226   // set result
 6227   xorl(result, result);
 6228   // check for zero length
 6229   testl(len, len);
 6230   jcc(Assembler::zero, L_done);
 6231 
 6232   movl(result, len);
 6233 
 6234   // Setup pointers
 6235   lea(src, Address(src, len, Address::times_2)); // char[]
 6236   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6237   negptr(len);
 6238 
 6239   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6240     Label L_copy_8_chars, L_copy_8_chars_exit;
 6241     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6242 
 6243     if (UseAVX >= 2) {
 6244       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6245       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6246       movdl(tmp1Reg, tmp5);
 6247       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6248       jmp(L_chars_32_check);
 6249 
 6250       bind(L_copy_32_chars);
 6251       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6252       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6253       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6254       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6255       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6256       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6257       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6258       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6259 
 6260       bind(L_chars_32_check);
 6261       addptr(len, 32);
 6262       jcc(Assembler::lessEqual, L_copy_32_chars);
 6263 
 6264       bind(L_copy_32_chars_exit);
 6265       subptr(len, 16);
 6266       jccb(Assembler::greater, L_copy_16_chars_exit);
 6267 
 6268     } else if (UseSSE42Intrinsics) {
 6269       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6270       movdl(tmp1Reg, tmp5);
 6271       pshufd(tmp1Reg, tmp1Reg, 0);
 6272       jmpb(L_chars_16_check);
 6273     }
 6274 
 6275     bind(L_copy_16_chars);
 6276     if (UseAVX >= 2) {
 6277       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6278       vptest(tmp2Reg, tmp1Reg);
 6279       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6280       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6281       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6282     } else {
 6283       if (UseAVX > 0) {
 6284         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6285         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6286         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6287       } else {
 6288         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6289         por(tmp2Reg, tmp3Reg);
 6290         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6291         por(tmp2Reg, tmp4Reg);
 6292       }
 6293       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6294       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6295       packuswb(tmp3Reg, tmp4Reg);
 6296     }
 6297     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6298 
 6299     bind(L_chars_16_check);
 6300     addptr(len, 16);
 6301     jcc(Assembler::lessEqual, L_copy_16_chars);
 6302 
 6303     bind(L_copy_16_chars_exit);
 6304     if (UseAVX >= 2) {
 6305       // clean upper bits of YMM registers
 6306       vpxor(tmp2Reg, tmp2Reg);
 6307       vpxor(tmp3Reg, tmp3Reg);
 6308       vpxor(tmp4Reg, tmp4Reg);
 6309       movdl(tmp1Reg, tmp5);
 6310       pshufd(tmp1Reg, tmp1Reg, 0);
 6311     }
 6312     subptr(len, 8);
 6313     jccb(Assembler::greater, L_copy_8_chars_exit);
 6314 
 6315     bind(L_copy_8_chars);
 6316     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6317     ptest(tmp3Reg, tmp1Reg);
 6318     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6319     packuswb(tmp3Reg, tmp1Reg);
 6320     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6321     addptr(len, 8);
 6322     jccb(Assembler::lessEqual, L_copy_8_chars);
 6323 
 6324     bind(L_copy_8_chars_exit);
 6325     subptr(len, 8);
 6326     jccb(Assembler::zero, L_done);
 6327   }
 6328 
 6329   bind(L_copy_1_char);
 6330   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6331   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6332   jccb(Assembler::notZero, L_copy_1_char_exit);
 6333   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6334   addptr(len, 1);
 6335   jccb(Assembler::less, L_copy_1_char);
 6336 
 6337   bind(L_copy_1_char_exit);
 6338   addptr(result, len); // len is negative count of not processed elements
 6339 
 6340   bind(L_done);
 6341 }
 6342 
 6343 #ifdef _LP64
 6344 /**
 6345  * Helper for multiply_to_len().
 6346  */
 6347 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6348   addq(dest_lo, src1);
 6349   adcq(dest_hi, 0);
 6350   addq(dest_lo, src2);
 6351   adcq(dest_hi, 0);
 6352 }
 6353 
 6354 /**
 6355  * Multiply 64 bit by 64 bit first loop.
 6356  */
 6357 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6358                                            Register y, Register y_idx, Register z,
 6359                                            Register carry, Register product,
 6360                                            Register idx, Register kdx) {
 6361   //
 6362   //  jlong carry, x[], y[], z[];
 6363   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6364   //    huge_128 product = y[idx] * x[xstart] + carry;
 6365   //    z[kdx] = (jlong)product;
 6366   //    carry  = (jlong)(product >>> 64);
 6367   //  }
 6368   //  z[xstart] = carry;
 6369   //
 6370 
 6371   Label L_first_loop, L_first_loop_exit;
 6372   Label L_one_x, L_one_y, L_multiply;
 6373 
 6374   decrementl(xstart);
 6375   jcc(Assembler::negative, L_one_x);
 6376 
 6377   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6378   rorq(x_xstart, 32); // convert big-endian to little-endian
 6379 
 6380   bind(L_first_loop);
 6381   decrementl(idx);
 6382   jcc(Assembler::negative, L_first_loop_exit);
 6383   decrementl(idx);
 6384   jcc(Assembler::negative, L_one_y);
 6385   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6386   rorq(y_idx, 32); // convert big-endian to little-endian
 6387   bind(L_multiply);
 6388   movq(product, x_xstart);
 6389   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6390   addq(product, carry);
 6391   adcq(rdx, 0);
 6392   subl(kdx, 2);
 6393   movl(Address(z, kdx, Address::times_4,  4), product);
 6394   shrq(product, 32);
 6395   movl(Address(z, kdx, Address::times_4,  0), product);
 6396   movq(carry, rdx);
 6397   jmp(L_first_loop);
 6398 
 6399   bind(L_one_y);
 6400   movl(y_idx, Address(y,  0));
 6401   jmp(L_multiply);
 6402 
 6403   bind(L_one_x);
 6404   movl(x_xstart, Address(x,  0));
 6405   jmp(L_first_loop);
 6406 
 6407   bind(L_first_loop_exit);
 6408 }
 6409 
 6410 /**
 6411  * Multiply 64 bit by 64 bit and add 128 bit.
 6412  */
 6413 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6414                                             Register yz_idx, Register idx,
 6415                                             Register carry, Register product, int offset) {
 6416   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6417   //     z[kdx] = (jlong)product;
 6418 
 6419   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6420   rorq(yz_idx, 32); // convert big-endian to little-endian
 6421   movq(product, x_xstart);
 6422   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6423   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6424   rorq(yz_idx, 32); // convert big-endian to little-endian
 6425 
 6426   add2_with_carry(rdx, product, carry, yz_idx);
 6427 
 6428   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6429   shrq(product, 32);
 6430   movl(Address(z, idx, Address::times_4,  offset), product);
 6431 
 6432 }
 6433 
 6434 /**
 6435  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6436  */
 6437 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6438                                              Register yz_idx, Register idx, Register jdx,
 6439                                              Register carry, Register product,
 6440                                              Register carry2) {
 6441   //   jlong carry, x[], y[], z[];
 6442   //   int kdx = ystart+1;
 6443   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6444   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6445   //     z[kdx+idx+1] = (jlong)product;
 6446   //     jlong carry2  = (jlong)(product >>> 64);
 6447   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6448   //     z[kdx+idx] = (jlong)product;
 6449   //     carry  = (jlong)(product >>> 64);
 6450   //   }
 6451   //   idx += 2;
 6452   //   if (idx > 0) {
 6453   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6454   //     z[kdx+idx] = (jlong)product;
 6455   //     carry  = (jlong)(product >>> 64);
 6456   //   }
 6457   //
 6458 
 6459   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6460 
 6461   movl(jdx, idx);
 6462   andl(jdx, 0xFFFFFFFC);
 6463   shrl(jdx, 2);
 6464 
 6465   bind(L_third_loop);
 6466   subl(jdx, 1);
 6467   jcc(Assembler::negative, L_third_loop_exit);
 6468   subl(idx, 4);
 6469 
 6470   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6471   movq(carry2, rdx);
 6472 
 6473   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6474   movq(carry, rdx);
 6475   jmp(L_third_loop);
 6476 
 6477   bind (L_third_loop_exit);
 6478 
 6479   andl (idx, 0x3);
 6480   jcc(Assembler::zero, L_post_third_loop_done);
 6481 
 6482   Label L_check_1;
 6483   subl(idx, 2);
 6484   jcc(Assembler::negative, L_check_1);
 6485 
 6486   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6487   movq(carry, rdx);
 6488 
 6489   bind (L_check_1);
 6490   addl (idx, 0x2);
 6491   andl (idx, 0x1);
 6492   subl(idx, 1);
 6493   jcc(Assembler::negative, L_post_third_loop_done);
 6494 
 6495   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6496   movq(product, x_xstart);
 6497   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6498   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6499 
 6500   add2_with_carry(rdx, product, yz_idx, carry);
 6501 
 6502   movl(Address(z, idx, Address::times_4,  0), product);
 6503   shrq(product, 32);
 6504 
 6505   shlq(rdx, 32);
 6506   orq(product, rdx);
 6507   movq(carry, product);
 6508 
 6509   bind(L_post_third_loop_done);
 6510 }
 6511 
 6512 /**
 6513  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6514  *
 6515  */
 6516 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6517                                                   Register carry, Register carry2,
 6518                                                   Register idx, Register jdx,
 6519                                                   Register yz_idx1, Register yz_idx2,
 6520                                                   Register tmp, Register tmp3, Register tmp4) {
 6521   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6522 
 6523   //   jlong carry, x[], y[], z[];
 6524   //   int kdx = ystart+1;
 6525   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6526   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6527   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6528   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6529   //     carry  = (jlong)(tmp4 >>> 64);
 6530   //     z[kdx+idx+1] = (jlong)tmp3;
 6531   //     z[kdx+idx] = (jlong)tmp4;
 6532   //   }
 6533   //   idx += 2;
 6534   //   if (idx > 0) {
 6535   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6536   //     z[kdx+idx] = (jlong)yz_idx1;
 6537   //     carry  = (jlong)(yz_idx1 >>> 64);
 6538   //   }
 6539   //
 6540 
 6541   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6542 
 6543   movl(jdx, idx);
 6544   andl(jdx, 0xFFFFFFFC);
 6545   shrl(jdx, 2);
 6546 
 6547   bind(L_third_loop);
 6548   subl(jdx, 1);
 6549   jcc(Assembler::negative, L_third_loop_exit);
 6550   subl(idx, 4);
 6551 
 6552   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6553   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6554   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6555   rorxq(yz_idx2, yz_idx2, 32);
 6556 
 6557   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6558   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6559 
 6560   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6561   rorxq(yz_idx1, yz_idx1, 32);
 6562   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6563   rorxq(yz_idx2, yz_idx2, 32);
 6564 
 6565   if (VM_Version::supports_adx()) {
 6566     adcxq(tmp3, carry);
 6567     adoxq(tmp3, yz_idx1);
 6568 
 6569     adcxq(tmp4, tmp);
 6570     adoxq(tmp4, yz_idx2);
 6571 
 6572     movl(carry, 0); // does not affect flags
 6573     adcxq(carry2, carry);
 6574     adoxq(carry2, carry);
 6575   } else {
 6576     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6577     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6578   }
 6579   movq(carry, carry2);
 6580 
 6581   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6582   shrq(tmp3, 32);
 6583   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6584 
 6585   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6586   shrq(tmp4, 32);
 6587   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6588 
 6589   jmp(L_third_loop);
 6590 
 6591   bind (L_third_loop_exit);
 6592 
 6593   andl (idx, 0x3);
 6594   jcc(Assembler::zero, L_post_third_loop_done);
 6595 
 6596   Label L_check_1;
 6597   subl(idx, 2);
 6598   jcc(Assembler::negative, L_check_1);
 6599 
 6600   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6601   rorxq(yz_idx1, yz_idx1, 32);
 6602   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6603   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6604   rorxq(yz_idx2, yz_idx2, 32);
 6605 
 6606   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6607 
 6608   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6609   shrq(tmp3, 32);
 6610   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6611   movq(carry, tmp4);
 6612 
 6613   bind (L_check_1);
 6614   addl (idx, 0x2);
 6615   andl (idx, 0x1);
 6616   subl(idx, 1);
 6617   jcc(Assembler::negative, L_post_third_loop_done);
 6618   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6619   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6620   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6621 
 6622   add2_with_carry(carry2, tmp3, tmp4, carry);
 6623 
 6624   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6625   shrq(tmp3, 32);
 6626 
 6627   shlq(carry2, 32);
 6628   orq(tmp3, carry2);
 6629   movq(carry, tmp3);
 6630 
 6631   bind(L_post_third_loop_done);
 6632 }
 6633 
 6634 /**
 6635  * Code for BigInteger::multiplyToLen() intrinsic.
 6636  *
 6637  * rdi: x
 6638  * rax: xlen
 6639  * rsi: y
 6640  * rcx: ylen
 6641  * r8:  z
 6642  * r11: zlen
 6643  * r12: tmp1
 6644  * r13: tmp2
 6645  * r14: tmp3
 6646  * r15: tmp4
 6647  * rbx: tmp5
 6648  *
 6649  */
 6650 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
 6651                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6652   ShortBranchVerifier sbv(this);
 6653   assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6654 
 6655   push(tmp1);
 6656   push(tmp2);
 6657   push(tmp3);
 6658   push(tmp4);
 6659   push(tmp5);
 6660 
 6661   push(xlen);
 6662   push(zlen);
 6663 
 6664   const Register idx = tmp1;
 6665   const Register kdx = tmp2;
 6666   const Register xstart = tmp3;
 6667 
 6668   const Register y_idx = tmp4;
 6669   const Register carry = tmp5;
 6670   const Register product  = xlen;
 6671   const Register x_xstart = zlen;  // reuse register
 6672 
 6673   // First Loop.
 6674   //
 6675   //  final static long LONG_MASK = 0xffffffffL;
 6676   //  int xstart = xlen - 1;
 6677   //  int ystart = ylen - 1;
 6678   //  long carry = 0;
 6679   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6680   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6681   //    z[kdx] = (int)product;
 6682   //    carry = product >>> 32;
 6683   //  }
 6684   //  z[xstart] = (int)carry;
 6685   //
 6686 
 6687   movl(idx, ylen);      // idx = ylen;
 6688   movl(kdx, zlen);      // kdx = xlen+ylen;
 6689   xorq(carry, carry);   // carry = 0;
 6690 
 6691   Label L_done;
 6692 
 6693   movl(xstart, xlen);
 6694   decrementl(xstart);
 6695   jcc(Assembler::negative, L_done);
 6696 
 6697   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6698 
 6699   Label L_second_loop;
 6700   testl(kdx, kdx);
 6701   jcc(Assembler::zero, L_second_loop);
 6702 
 6703   Label L_carry;
 6704   subl(kdx, 1);
 6705   jcc(Assembler::zero, L_carry);
 6706 
 6707   movl(Address(z, kdx, Address::times_4,  0), carry);
 6708   shrq(carry, 32);
 6709   subl(kdx, 1);
 6710 
 6711   bind(L_carry);
 6712   movl(Address(z, kdx, Address::times_4,  0), carry);
 6713 
 6714   // Second and third (nested) loops.
 6715   //
 6716   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6717   //   carry = 0;
 6718   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6719   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6720   //                    (z[k] & LONG_MASK) + carry;
 6721   //     z[k] = (int)product;
 6722   //     carry = product >>> 32;
 6723   //   }
 6724   //   z[i] = (int)carry;
 6725   // }
 6726   //
 6727   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6728 
 6729   const Register jdx = tmp1;
 6730 
 6731   bind(L_second_loop);
 6732   xorl(carry, carry);    // carry = 0;
 6733   movl(jdx, ylen);       // j = ystart+1
 6734 
 6735   subl(xstart, 1);       // i = xstart-1;
 6736   jcc(Assembler::negative, L_done);
 6737 
 6738   push (z);
 6739 
 6740   Label L_last_x;
 6741   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6742   subl(xstart, 1);       // i = xstart-1;
 6743   jcc(Assembler::negative, L_last_x);
 6744 
 6745   if (UseBMI2Instructions) {
 6746     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6747     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6748   } else {
 6749     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6750     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6751   }
 6752 
 6753   Label L_third_loop_prologue;
 6754   bind(L_third_loop_prologue);
 6755 
 6756   push (x);
 6757   push (xstart);
 6758   push (ylen);
 6759 
 6760 
 6761   if (UseBMI2Instructions) {
 6762     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6763   } else { // !UseBMI2Instructions
 6764     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6765   }
 6766 
 6767   pop(ylen);
 6768   pop(xlen);
 6769   pop(x);
 6770   pop(z);
 6771 
 6772   movl(tmp3, xlen);
 6773   addl(tmp3, 1);
 6774   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6775   subl(tmp3, 1);
 6776   jccb(Assembler::negative, L_done);
 6777 
 6778   shrq(carry, 32);
 6779   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6780   jmp(L_second_loop);
 6781 
 6782   // Next infrequent code is moved outside loops.
 6783   bind(L_last_x);
 6784   if (UseBMI2Instructions) {
 6785     movl(rdx, Address(x,  0));
 6786   } else {
 6787     movl(x_xstart, Address(x,  0));
 6788   }
 6789   jmp(L_third_loop_prologue);
 6790 
 6791   bind(L_done);
 6792 
 6793   pop(zlen);
 6794   pop(xlen);
 6795 
 6796   pop(tmp5);
 6797   pop(tmp4);
 6798   pop(tmp3);
 6799   pop(tmp2);
 6800   pop(tmp1);
 6801 }
 6802 
 6803 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6804   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6805   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6806   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6807   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6808   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6809   Label SAME_TILL_END, DONE;
 6810   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6811 
 6812   //scale is in rcx in both Win64 and Unix
 6813   ShortBranchVerifier sbv(this);
 6814 
 6815   shlq(length);
 6816   xorq(result, result);
 6817 
 6818   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6819       VM_Version::supports_avx512vlbw()) {
 6820     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6821 
 6822     cmpq(length, 64);
 6823     jcc(Assembler::less, VECTOR32_TAIL);
 6824 
 6825     movq(tmp1, length);
 6826     andq(tmp1, 0x3F);      // tail count
 6827     andq(length, ~(0x3F)); //vector count
 6828 
 6829     bind(VECTOR64_LOOP);
 6830     // AVX512 code to compare 64 byte vectors.
 6831     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6832     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6833     kortestql(k7, k7);
 6834     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6835     addq(result, 64);
 6836     subq(length, 64);
 6837     jccb(Assembler::notZero, VECTOR64_LOOP);
 6838 
 6839     //bind(VECTOR64_TAIL);
 6840     testq(tmp1, tmp1);
 6841     jcc(Assembler::zero, SAME_TILL_END);
 6842 
 6843     //bind(VECTOR64_TAIL);
 6844     // AVX512 code to compare up to 63 byte vectors.
 6845     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6846     shlxq(tmp2, tmp2, tmp1);
 6847     notq(tmp2);
 6848     kmovql(k3, tmp2);
 6849 
 6850     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 6851     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6852 
 6853     ktestql(k7, k3);
 6854     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 6855 
 6856     bind(VECTOR64_NOT_EQUAL);
 6857     kmovql(tmp1, k7);
 6858     notq(tmp1);
 6859     tzcntq(tmp1, tmp1);
 6860     addq(result, tmp1);
 6861     shrq(result);
 6862     jmp(DONE);
 6863     bind(VECTOR32_TAIL);
 6864   }
 6865 
 6866   cmpq(length, 8);
 6867   jcc(Assembler::equal, VECTOR8_LOOP);
 6868   jcc(Assembler::less, VECTOR4_TAIL);
 6869 
 6870   if (UseAVX >= 2) {
 6871     Label VECTOR16_TAIL, VECTOR32_LOOP;
 6872 
 6873     cmpq(length, 16);
 6874     jcc(Assembler::equal, VECTOR16_LOOP);
 6875     jcc(Assembler::less, VECTOR8_LOOP);
 6876 
 6877     cmpq(length, 32);
 6878     jccb(Assembler::less, VECTOR16_TAIL);
 6879 
 6880     subq(length, 32);
 6881     bind(VECTOR32_LOOP);
 6882     vmovdqu(rymm0, Address(obja, result));
 6883     vmovdqu(rymm1, Address(objb, result));
 6884     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 6885     vptest(rymm2, rymm2);
 6886     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 6887     addq(result, 32);
 6888     subq(length, 32);
 6889     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 6890     addq(length, 32);
 6891     jcc(Assembler::equal, SAME_TILL_END);
 6892     //falling through if less than 32 bytes left //close the branch here.
 6893 
 6894     bind(VECTOR16_TAIL);
 6895     cmpq(length, 16);
 6896     jccb(Assembler::less, VECTOR8_TAIL);
 6897     bind(VECTOR16_LOOP);
 6898     movdqu(rymm0, Address(obja, result));
 6899     movdqu(rymm1, Address(objb, result));
 6900     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 6901     ptest(rymm2, rymm2);
 6902     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 6903     addq(result, 16);
 6904     subq(length, 16);
 6905     jcc(Assembler::equal, SAME_TILL_END);
 6906     //falling through if less than 16 bytes left
 6907   } else {//regular intrinsics
 6908 
 6909     cmpq(length, 16);
 6910     jccb(Assembler::less, VECTOR8_TAIL);
 6911 
 6912     subq(length, 16);
 6913     bind(VECTOR16_LOOP);
 6914     movdqu(rymm0, Address(obja, result));
 6915     movdqu(rymm1, Address(objb, result));
 6916     pxor(rymm0, rymm1);
 6917     ptest(rymm0, rymm0);
 6918     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 6919     addq(result, 16);
 6920     subq(length, 16);
 6921     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 6922     addq(length, 16);
 6923     jcc(Assembler::equal, SAME_TILL_END);
 6924     //falling through if less than 16 bytes left
 6925   }
 6926 
 6927   bind(VECTOR8_TAIL);
 6928   cmpq(length, 8);
 6929   jccb(Assembler::less, VECTOR4_TAIL);
 6930   bind(VECTOR8_LOOP);
 6931   movq(tmp1, Address(obja, result));
 6932   movq(tmp2, Address(objb, result));
 6933   xorq(tmp1, tmp2);
 6934   testq(tmp1, tmp1);
 6935   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 6936   addq(result, 8);
 6937   subq(length, 8);
 6938   jcc(Assembler::equal, SAME_TILL_END);
 6939   //falling through if less than 8 bytes left
 6940 
 6941   bind(VECTOR4_TAIL);
 6942   cmpq(length, 4);
 6943   jccb(Assembler::less, BYTES_TAIL);
 6944   bind(VECTOR4_LOOP);
 6945   movl(tmp1, Address(obja, result));
 6946   xorl(tmp1, Address(objb, result));
 6947   testl(tmp1, tmp1);
 6948   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 6949   addq(result, 4);
 6950   subq(length, 4);
 6951   jcc(Assembler::equal, SAME_TILL_END);
 6952   //falling through if less than 4 bytes left
 6953 
 6954   bind(BYTES_TAIL);
 6955   bind(BYTES_LOOP);
 6956   load_unsigned_byte(tmp1, Address(obja, result));
 6957   load_unsigned_byte(tmp2, Address(objb, result));
 6958   xorl(tmp1, tmp2);
 6959   testl(tmp1, tmp1);
 6960   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 6961   decq(length);
 6962   jcc(Assembler::zero, SAME_TILL_END);
 6963   incq(result);
 6964   load_unsigned_byte(tmp1, Address(obja, result));
 6965   load_unsigned_byte(tmp2, Address(objb, result));
 6966   xorl(tmp1, tmp2);
 6967   testl(tmp1, tmp1);
 6968   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 6969   decq(length);
 6970   jcc(Assembler::zero, SAME_TILL_END);
 6971   incq(result);
 6972   load_unsigned_byte(tmp1, Address(obja, result));
 6973   load_unsigned_byte(tmp2, Address(objb, result));
 6974   xorl(tmp1, tmp2);
 6975   testl(tmp1, tmp1);
 6976   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 6977   jmp(SAME_TILL_END);
 6978 
 6979   if (UseAVX >= 2) {
 6980     bind(VECTOR32_NOT_EQUAL);
 6981     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 6982     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 6983     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 6984     vpmovmskb(tmp1, rymm0);
 6985     bsfq(tmp1, tmp1);
 6986     addq(result, tmp1);
 6987     shrq(result);
 6988     jmp(DONE);
 6989   }
 6990 
 6991   bind(VECTOR16_NOT_EQUAL);
 6992   if (UseAVX >= 2) {
 6993     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 6994     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 6995     pxor(rymm0, rymm2);
 6996   } else {
 6997     pcmpeqb(rymm2, rymm2);
 6998     pxor(rymm0, rymm1);
 6999     pcmpeqb(rymm0, rymm1);
 7000     pxor(rymm0, rymm2);
 7001   }
 7002   pmovmskb(tmp1, rymm0);
 7003   bsfq(tmp1, tmp1);
 7004   addq(result, tmp1);
 7005   shrq(result);
 7006   jmpb(DONE);
 7007 
 7008   bind(VECTOR8_NOT_EQUAL);
 7009   bind(VECTOR4_NOT_EQUAL);
 7010   bsfq(tmp1, tmp1);
 7011   shrq(tmp1, 3);
 7012   addq(result, tmp1);
 7013   bind(BYTES_NOT_EQUAL);
 7014   shrq(result);
 7015   jmpb(DONE);
 7016 
 7017   bind(SAME_TILL_END);
 7018   mov64(result, -1);
 7019 
 7020   bind(DONE);
 7021 }
 7022 
 7023 //Helper functions for square_to_len()
 7024 
 7025 /**
 7026  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7027  * Preserves x and z and modifies rest of the registers.
 7028  */
 7029 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7030   // Perform square and right shift by 1
 7031   // Handle odd xlen case first, then for even xlen do the following
 7032   // jlong carry = 0;
 7033   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7034   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7035   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7036   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7037   //     carry = (jlong)product;
 7038   // }
 7039 
 7040   xorq(tmp5, tmp5);     // carry
 7041   xorq(rdxReg, rdxReg);
 7042   xorl(tmp1, tmp1);     // index for x
 7043   xorl(tmp4, tmp4);     // index for z
 7044 
 7045   Label L_first_loop, L_first_loop_exit;
 7046 
 7047   testl(xlen, 1);
 7048   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7049 
 7050   // Square and right shift by 1 the odd element using 32 bit multiply
 7051   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7052   imulq(raxReg, raxReg);
 7053   shrq(raxReg, 1);
 7054   adcq(tmp5, 0);
 7055   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7056   incrementl(tmp1);
 7057   addl(tmp4, 2);
 7058 
 7059   // Square and  right shift by 1 the rest using 64 bit multiply
 7060   bind(L_first_loop);
 7061   cmpptr(tmp1, xlen);
 7062   jccb(Assembler::equal, L_first_loop_exit);
 7063 
 7064   // Square
 7065   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7066   rorq(raxReg, 32);    // convert big-endian to little-endian
 7067   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7068 
 7069   // Right shift by 1 and save carry
 7070   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7071   rcrq(rdxReg, 1);
 7072   rcrq(raxReg, 1);
 7073   adcq(tmp5, 0);
 7074 
 7075   // Store result in z
 7076   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7077   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7078 
 7079   // Update indices for x and z
 7080   addl(tmp1, 2);
 7081   addl(tmp4, 4);
 7082   jmp(L_first_loop);
 7083 
 7084   bind(L_first_loop_exit);
 7085 }
 7086 
 7087 
 7088 /**
 7089  * Perform the following multiply add operation using BMI2 instructions
 7090  * carry:sum = sum + op1*op2 + carry
 7091  * op2 should be in rdx
 7092  * op2 is preserved, all other registers are modified
 7093  */
 7094 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7095   // assert op2 is rdx
 7096   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7097   addq(sum, carry);
 7098   adcq(tmp2, 0);
 7099   addq(sum, op1);
 7100   adcq(tmp2, 0);
 7101   movq(carry, tmp2);
 7102 }
 7103 
 7104 /**
 7105  * Perform the following multiply add operation:
 7106  * carry:sum = sum + op1*op2 + carry
 7107  * Preserves op1, op2 and modifies rest of registers
 7108  */
 7109 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7110   // rdx:rax = op1 * op2
 7111   movq(raxReg, op2);
 7112   mulq(op1);
 7113 
 7114   //  rdx:rax = sum + carry + rdx:rax
 7115   addq(sum, carry);
 7116   adcq(rdxReg, 0);
 7117   addq(sum, raxReg);
 7118   adcq(rdxReg, 0);
 7119 
 7120   // carry:sum = rdx:sum
 7121   movq(carry, rdxReg);
 7122 }
 7123 
 7124 /**
 7125  * Add 64 bit long carry into z[] with carry propagation.
 7126  * Preserves z and carry register values and modifies rest of registers.
 7127  *
 7128  */
 7129 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7130   Label L_fourth_loop, L_fourth_loop_exit;
 7131 
 7132   movl(tmp1, 1);
 7133   subl(zlen, 2);
 7134   addq(Address(z, zlen, Address::times_4, 0), carry);
 7135 
 7136   bind(L_fourth_loop);
 7137   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7138   subl(zlen, 2);
 7139   jccb(Assembler::negative, L_fourth_loop_exit);
 7140   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7141   jmp(L_fourth_loop);
 7142   bind(L_fourth_loop_exit);
 7143 }
 7144 
 7145 /**
 7146  * Shift z[] left by 1 bit.
 7147  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7148  *
 7149  */
 7150 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7151 
 7152   Label L_fifth_loop, L_fifth_loop_exit;
 7153 
 7154   // Fifth loop
 7155   // Perform primitiveLeftShift(z, zlen, 1)
 7156 
 7157   const Register prev_carry = tmp1;
 7158   const Register new_carry = tmp4;
 7159   const Register value = tmp2;
 7160   const Register zidx = tmp3;
 7161 
 7162   // int zidx, carry;
 7163   // long value;
 7164   // carry = 0;
 7165   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7166   //    (carry:value)  = (z[i] << 1) | carry ;
 7167   //    z[i] = value;
 7168   // }
 7169 
 7170   movl(zidx, zlen);
 7171   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7172 
 7173   bind(L_fifth_loop);
 7174   decl(zidx);  // Use decl to preserve carry flag
 7175   decl(zidx);
 7176   jccb(Assembler::negative, L_fifth_loop_exit);
 7177 
 7178   if (UseBMI2Instructions) {
 7179      movq(value, Address(z, zidx, Address::times_4, 0));
 7180      rclq(value, 1);
 7181      rorxq(value, value, 32);
 7182      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7183   }
 7184   else {
 7185     // clear new_carry
 7186     xorl(new_carry, new_carry);
 7187 
 7188     // Shift z[i] by 1, or in previous carry and save new carry
 7189     movq(value, Address(z, zidx, Address::times_4, 0));
 7190     shlq(value, 1);
 7191     adcl(new_carry, 0);
 7192 
 7193     orq(value, prev_carry);
 7194     rorq(value, 0x20);
 7195     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7196 
 7197     // Set previous carry = new carry
 7198     movl(prev_carry, new_carry);
 7199   }
 7200   jmp(L_fifth_loop);
 7201 
 7202   bind(L_fifth_loop_exit);
 7203 }
 7204 
 7205 
 7206 /**
 7207  * Code for BigInteger::squareToLen() intrinsic
 7208  *
 7209  * rdi: x
 7210  * rsi: len
 7211  * r8:  z
 7212  * rcx: zlen
 7213  * r12: tmp1
 7214  * r13: tmp2
 7215  * r14: tmp3
 7216  * r15: tmp4
 7217  * rbx: tmp5
 7218  *
 7219  */
 7220 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) {
 7221 
 7222   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7223   push(tmp1);
 7224   push(tmp2);
 7225   push(tmp3);
 7226   push(tmp4);
 7227   push(tmp5);
 7228 
 7229   // First loop
 7230   // Store the squares, right shifted one bit (i.e., divided by 2).
 7231   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7232 
 7233   // Add in off-diagonal sums.
 7234   //
 7235   // Second, third (nested) and fourth loops.
 7236   // zlen +=2;
 7237   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7238   //    carry = 0;
 7239   //    long op2 = x[xidx:xidx+1];
 7240   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7241   //       k -= 2;
 7242   //       long op1 = x[j:j+1];
 7243   //       long sum = z[k:k+1];
 7244   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7245   //       z[k:k+1] = sum;
 7246   //    }
 7247   //    add_one_64(z, k, carry, tmp_regs);
 7248   // }
 7249 
 7250   const Register carry = tmp5;
 7251   const Register sum = tmp3;
 7252   const Register op1 = tmp4;
 7253   Register op2 = tmp2;
 7254 
 7255   push(zlen);
 7256   push(len);
 7257   addl(zlen,2);
 7258   bind(L_second_loop);
 7259   xorq(carry, carry);
 7260   subl(zlen, 4);
 7261   subl(len, 2);
 7262   push(zlen);
 7263   push(len);
 7264   cmpl(len, 0);
 7265   jccb(Assembler::lessEqual, L_second_loop_exit);
 7266 
 7267   // Multiply an array by one 64 bit long.
 7268   if (UseBMI2Instructions) {
 7269     op2 = rdxReg;
 7270     movq(op2, Address(x, len, Address::times_4,  0));
 7271     rorxq(op2, op2, 32);
 7272   }
 7273   else {
 7274     movq(op2, Address(x, len, Address::times_4,  0));
 7275     rorq(op2, 32);
 7276   }
 7277 
 7278   bind(L_third_loop);
 7279   decrementl(len);
 7280   jccb(Assembler::negative, L_third_loop_exit);
 7281   decrementl(len);
 7282   jccb(Assembler::negative, L_last_x);
 7283 
 7284   movq(op1, Address(x, len, Address::times_4,  0));
 7285   rorq(op1, 32);
 7286 
 7287   bind(L_multiply);
 7288   subl(zlen, 2);
 7289   movq(sum, Address(z, zlen, Address::times_4,  0));
 7290 
 7291   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7292   if (UseBMI2Instructions) {
 7293     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7294   }
 7295   else {
 7296     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7297   }
 7298 
 7299   movq(Address(z, zlen, Address::times_4, 0), sum);
 7300 
 7301   jmp(L_third_loop);
 7302   bind(L_third_loop_exit);
 7303 
 7304   // Fourth loop
 7305   // Add 64 bit long carry into z with carry propagation.
 7306   // Uses offsetted zlen.
 7307   add_one_64(z, zlen, carry, tmp1);
 7308 
 7309   pop(len);
 7310   pop(zlen);
 7311   jmp(L_second_loop);
 7312 
 7313   // Next infrequent code is moved outside loops.
 7314   bind(L_last_x);
 7315   movl(op1, Address(x, 0));
 7316   jmp(L_multiply);
 7317 
 7318   bind(L_second_loop_exit);
 7319   pop(len);
 7320   pop(zlen);
 7321   pop(len);
 7322   pop(zlen);
 7323 
 7324   // Fifth loop
 7325   // Shift z left 1 bit.
 7326   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7327 
 7328   // z[zlen-1] |= x[len-1] & 1;
 7329   movl(tmp3, Address(x, len, Address::times_4, -4));
 7330   andl(tmp3, 1);
 7331   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7332 
 7333   pop(tmp5);
 7334   pop(tmp4);
 7335   pop(tmp3);
 7336   pop(tmp2);
 7337   pop(tmp1);
 7338 }
 7339 
 7340 /**
 7341  * Helper function for mul_add()
 7342  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7343  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7344  * Only quad int aligned length of in[] is operated on in this function.
 7345  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7346  * This function preserves out, in and k registers.
 7347  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7348  * tmp5 has the carry.
 7349  * other registers are temporary and are modified.
 7350  *
 7351  */
 7352 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7353   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7354   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7355 
 7356   Label L_first_loop, L_first_loop_exit;
 7357 
 7358   movl(tmp1, len);
 7359   shrl(tmp1, 2);
 7360 
 7361   bind(L_first_loop);
 7362   subl(tmp1, 1);
 7363   jccb(Assembler::negative, L_first_loop_exit);
 7364 
 7365   subl(len, 4);
 7366   subl(offset, 4);
 7367 
 7368   Register op2 = tmp2;
 7369   const Register sum = tmp3;
 7370   const Register op1 = tmp4;
 7371   const Register carry = tmp5;
 7372 
 7373   if (UseBMI2Instructions) {
 7374     op2 = rdxReg;
 7375   }
 7376 
 7377   movq(op1, Address(in, len, Address::times_4,  8));
 7378   rorq(op1, 32);
 7379   movq(sum, Address(out, offset, Address::times_4,  8));
 7380   rorq(sum, 32);
 7381   if (UseBMI2Instructions) {
 7382     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7383   }
 7384   else {
 7385     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7386   }
 7387   // Store back in big endian from little endian
 7388   rorq(sum, 0x20);
 7389   movq(Address(out, offset, Address::times_4,  8), sum);
 7390 
 7391   movq(op1, Address(in, len, Address::times_4,  0));
 7392   rorq(op1, 32);
 7393   movq(sum, Address(out, offset, Address::times_4,  0));
 7394   rorq(sum, 32);
 7395   if (UseBMI2Instructions) {
 7396     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7397   }
 7398   else {
 7399     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7400   }
 7401   // Store back in big endian from little endian
 7402   rorq(sum, 0x20);
 7403   movq(Address(out, offset, Address::times_4,  0), sum);
 7404 
 7405   jmp(L_first_loop);
 7406   bind(L_first_loop_exit);
 7407 }
 7408 
 7409 /**
 7410  * Code for BigInteger::mulAdd() intrinsic
 7411  *
 7412  * rdi: out
 7413  * rsi: in
 7414  * r11: offs (out.length - offset)
 7415  * rcx: len
 7416  * r8:  k
 7417  * r12: tmp1
 7418  * r13: tmp2
 7419  * r14: tmp3
 7420  * r15: tmp4
 7421  * rbx: tmp5
 7422  * Multiply the in[] by word k and add to out[], return the carry in rax
 7423  */
 7424 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7425    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7426    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7427 
 7428   Label L_carry, L_last_in, L_done;
 7429 
 7430 // carry = 0;
 7431 // for (int j=len-1; j >= 0; j--) {
 7432 //    long product = (in[j] & LONG_MASK) * kLong +
 7433 //                   (out[offs] & LONG_MASK) + carry;
 7434 //    out[offs--] = (int)product;
 7435 //    carry = product >>> 32;
 7436 // }
 7437 //
 7438   push(tmp1);
 7439   push(tmp2);
 7440   push(tmp3);
 7441   push(tmp4);
 7442   push(tmp5);
 7443 
 7444   Register op2 = tmp2;
 7445   const Register sum = tmp3;
 7446   const Register op1 = tmp4;
 7447   const Register carry =  tmp5;
 7448 
 7449   if (UseBMI2Instructions) {
 7450     op2 = rdxReg;
 7451     movl(op2, k);
 7452   }
 7453   else {
 7454     movl(op2, k);
 7455   }
 7456 
 7457   xorq(carry, carry);
 7458 
 7459   //First loop
 7460 
 7461   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7462   //The carry is in tmp5
 7463   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7464 
 7465   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7466   decrementl(len);
 7467   jccb(Assembler::negative, L_carry);
 7468   decrementl(len);
 7469   jccb(Assembler::negative, L_last_in);
 7470 
 7471   movq(op1, Address(in, len, Address::times_4,  0));
 7472   rorq(op1, 32);
 7473 
 7474   subl(offs, 2);
 7475   movq(sum, Address(out, offs, Address::times_4,  0));
 7476   rorq(sum, 32);
 7477 
 7478   if (UseBMI2Instructions) {
 7479     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7480   }
 7481   else {
 7482     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7483   }
 7484 
 7485   // Store back in big endian from little endian
 7486   rorq(sum, 0x20);
 7487   movq(Address(out, offs, Address::times_4,  0), sum);
 7488 
 7489   testl(len, len);
 7490   jccb(Assembler::zero, L_carry);
 7491 
 7492   //Multiply the last in[] entry, if any
 7493   bind(L_last_in);
 7494   movl(op1, Address(in, 0));
 7495   movl(sum, Address(out, offs, Address::times_4,  -4));
 7496 
 7497   movl(raxReg, k);
 7498   mull(op1); //tmp4 * eax -> edx:eax
 7499   addl(sum, carry);
 7500   adcl(rdxReg, 0);
 7501   addl(sum, raxReg);
 7502   adcl(rdxReg, 0);
 7503   movl(carry, rdxReg);
 7504 
 7505   movl(Address(out, offs, Address::times_4,  -4), sum);
 7506 
 7507   bind(L_carry);
 7508   //return tmp5/carry as carry in rax
 7509   movl(rax, carry);
 7510 
 7511   bind(L_done);
 7512   pop(tmp5);
 7513   pop(tmp4);
 7514   pop(tmp3);
 7515   pop(tmp2);
 7516   pop(tmp1);
 7517 }
 7518 #endif
 7519 
 7520 /**
 7521  * Emits code to update CRC-32 with a byte value according to constants in table
 7522  *
 7523  * @param [in,out]crc   Register containing the crc.
 7524  * @param [in]val       Register containing the byte to fold into the CRC.
 7525  * @param [in]table     Register containing the table of crc constants.
 7526  *
 7527  * uint32_t crc;
 7528  * val = crc_table[(val ^ crc) & 0xFF];
 7529  * crc = val ^ (crc >> 8);
 7530  *
 7531  */
 7532 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7533   xorl(val, crc);
 7534   andl(val, 0xFF);
 7535   shrl(crc, 8); // unsigned shift
 7536   xorl(crc, Address(table, val, Address::times_4, 0));
 7537 }
 7538 
 7539 /**
 7540  * Fold 128-bit data chunk
 7541  */
 7542 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7543   if (UseAVX > 0) {
 7544     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7545     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7546     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7547     pxor(xcrc, xtmp);
 7548   } else {
 7549     movdqa(xtmp, xcrc);
 7550     pclmulhdq(xtmp, xK);   // [123:64]
 7551     pclmulldq(xcrc, xK);   // [63:0]
 7552     pxor(xcrc, xtmp);
 7553     movdqu(xtmp, Address(buf, offset));
 7554     pxor(xcrc, xtmp);
 7555   }
 7556 }
 7557 
 7558 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7559   if (UseAVX > 0) {
 7560     vpclmulhdq(xtmp, xK, xcrc);
 7561     vpclmulldq(xcrc, xK, xcrc);
 7562     pxor(xcrc, xbuf);
 7563     pxor(xcrc, xtmp);
 7564   } else {
 7565     movdqa(xtmp, xcrc);
 7566     pclmulhdq(xtmp, xK);
 7567     pclmulldq(xcrc, xK);
 7568     pxor(xcrc, xbuf);
 7569     pxor(xcrc, xtmp);
 7570   }
 7571 }
 7572 
 7573 /**
 7574  * 8-bit folds to compute 32-bit CRC
 7575  *
 7576  * uint64_t xcrc;
 7577  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7578  */
 7579 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7580   movdl(tmp, xcrc);
 7581   andl(tmp, 0xFF);
 7582   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7583   psrldq(xcrc, 1); // unsigned shift one byte
 7584   pxor(xcrc, xtmp);
 7585 }
 7586 
 7587 /**
 7588  * uint32_t crc;
 7589  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7590  */
 7591 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7592   movl(tmp, crc);
 7593   andl(tmp, 0xFF);
 7594   shrl(crc, 8);
 7595   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7596 }
 7597 
 7598 /**
 7599  * @param crc   register containing existing CRC (32-bit)
 7600  * @param buf   register pointing to input byte buffer (byte*)
 7601  * @param len   register containing number of bytes
 7602  * @param table register that will contain address of CRC table
 7603  * @param tmp   scratch register
 7604  */
 7605 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7606   assert_different_registers(crc, buf, len, table, tmp, rax);
 7607 
 7608   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7609   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7610 
 7611   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7612   // context for the registers used, where all instructions below are using 128-bit mode
 7613   // On EVEX without VL and BW, these instructions will all be AVX.
 7614   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7615   notl(crc); // ~crc
 7616   cmpl(len, 16);
 7617   jcc(Assembler::less, L_tail);
 7618 
 7619   // Align buffer to 16 bytes
 7620   movl(tmp, buf);
 7621   andl(tmp, 0xF);
 7622   jccb(Assembler::zero, L_aligned);
 7623   subl(tmp,  16);
 7624   addl(len, tmp);
 7625 
 7626   align(4);
 7627   BIND(L_align_loop);
 7628   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7629   update_byte_crc32(crc, rax, table);
 7630   increment(buf);
 7631   incrementl(tmp);
 7632   jccb(Assembler::less, L_align_loop);
 7633 
 7634   BIND(L_aligned);
 7635   movl(tmp, len); // save
 7636   shrl(len, 4);
 7637   jcc(Assembler::zero, L_tail_restore);
 7638 
 7639   // Fold crc into first bytes of vector
 7640   movdqa(xmm1, Address(buf, 0));
 7641   movdl(rax, xmm1);
 7642   xorl(crc, rax);
 7643   if (VM_Version::supports_sse4_1()) {
 7644     pinsrd(xmm1, crc, 0);
 7645   } else {
 7646     pinsrw(xmm1, crc, 0);
 7647     shrl(crc, 16);
 7648     pinsrw(xmm1, crc, 1);
 7649   }
 7650   addptr(buf, 16);
 7651   subl(len, 4); // len > 0
 7652   jcc(Assembler::less, L_fold_tail);
 7653 
 7654   movdqa(xmm2, Address(buf,  0));
 7655   movdqa(xmm3, Address(buf, 16));
 7656   movdqa(xmm4, Address(buf, 32));
 7657   addptr(buf, 48);
 7658   subl(len, 3);
 7659   jcc(Assembler::lessEqual, L_fold_512b);
 7660 
 7661   // Fold total 512 bits of polynomial on each iteration,
 7662   // 128 bits per each of 4 parallel streams.
 7663   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7664 
 7665   align32();
 7666   BIND(L_fold_512b_loop);
 7667   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7668   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7669   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7670   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7671   addptr(buf, 64);
 7672   subl(len, 4);
 7673   jcc(Assembler::greater, L_fold_512b_loop);
 7674 
 7675   // Fold 512 bits to 128 bits.
 7676   BIND(L_fold_512b);
 7677   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7678   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7679   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7680   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7681 
 7682   // Fold the rest of 128 bits data chunks
 7683   BIND(L_fold_tail);
 7684   addl(len, 3);
 7685   jccb(Assembler::lessEqual, L_fold_128b);
 7686   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7687 
 7688   BIND(L_fold_tail_loop);
 7689   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7690   addptr(buf, 16);
 7691   decrementl(len);
 7692   jccb(Assembler::greater, L_fold_tail_loop);
 7693 
 7694   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7695   BIND(L_fold_128b);
 7696   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7697   if (UseAVX > 0) {
 7698     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7699     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7700     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7701   } else {
 7702     movdqa(xmm2, xmm0);
 7703     pclmulqdq(xmm2, xmm1, 0x1);
 7704     movdqa(xmm3, xmm0);
 7705     pand(xmm3, xmm2);
 7706     pclmulqdq(xmm0, xmm3, 0x1);
 7707   }
 7708   psrldq(xmm1, 8);
 7709   psrldq(xmm2, 4);
 7710   pxor(xmm0, xmm1);
 7711   pxor(xmm0, xmm2);
 7712 
 7713   // 8 8-bit folds to compute 32-bit CRC.
 7714   for (int j = 0; j < 4; j++) {
 7715     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7716   }
 7717   movdl(crc, xmm0); // mov 32 bits to general register
 7718   for (int j = 0; j < 4; j++) {
 7719     fold_8bit_crc32(crc, table, rax);
 7720   }
 7721 
 7722   BIND(L_tail_restore);
 7723   movl(len, tmp); // restore
 7724   BIND(L_tail);
 7725   andl(len, 0xf);
 7726   jccb(Assembler::zero, L_exit);
 7727 
 7728   // Fold the rest of bytes
 7729   align(4);
 7730   BIND(L_tail_loop);
 7731   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7732   update_byte_crc32(crc, rax, table);
 7733   increment(buf);
 7734   decrementl(len);
 7735   jccb(Assembler::greater, L_tail_loop);
 7736 
 7737   BIND(L_exit);
 7738   notl(crc); // ~c
 7739 }
 7740 
 7741 #ifdef _LP64
 7742 // Helper function for AVX 512 CRC32
 7743 // Fold 512-bit data chunks
 7744 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7745                                              Register pos, int offset) {
 7746   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7747   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7748   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7749   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7750   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7751 }
 7752 
 7753 // Helper function for AVX 512 CRC32
 7754 // Compute CRC32 for < 256B buffers
 7755 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7756                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7757                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7758 
 7759   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7760   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7761   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7762 
 7763   // check if there is enough buffer to be able to fold 16B at a time
 7764   cmpl(len, 32);
 7765   jcc(Assembler::less, L_less_than_32);
 7766 
 7767   // if there is, load the constants
 7768   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7769   movdl(xmm0, crc);                        // get the initial crc value
 7770   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7771   pxor(xmm7, xmm0);
 7772 
 7773   // update the buffer pointer
 7774   addl(pos, 16);
 7775   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7776   subl(len, 32);
 7777   jmp(L_16B_reduction_loop);
 7778 
 7779   bind(L_less_than_32);
 7780   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7781   movl(rax, crc);
 7782   testl(len, len);
 7783   jcc(Assembler::equal, L_cleanup);
 7784 
 7785   movdl(xmm0, crc);                        //get the initial crc value
 7786 
 7787   cmpl(len, 16);
 7788   jcc(Assembler::equal, L_exact_16_left);
 7789   jcc(Assembler::less, L_less_than_16_left);
 7790 
 7791   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7792   pxor(xmm7, xmm0);                       //xor the initial crc value
 7793   addl(pos, 16);
 7794   subl(len, 16);
 7795   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7796   jmp(L_get_last_two_xmms);
 7797 
 7798   bind(L_less_than_16_left);
 7799   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7800   pxor(xmm1, xmm1);
 7801   movptr(tmp1, rsp);
 7802   movdqu(Address(tmp1, 0 * 16), xmm1);
 7803 
 7804   cmpl(len, 4);
 7805   jcc(Assembler::less, L_only_less_than_4);
 7806 
 7807   //backup the counter value
 7808   movl(tmp2, len);
 7809   cmpl(len, 8);
 7810   jcc(Assembler::less, L_less_than_8_left);
 7811 
 7812   //load 8 Bytes
 7813   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7814   movq(Address(tmp1, 0 * 16), rax);
 7815   addptr(tmp1, 8);
 7816   subl(len, 8);
 7817   addl(pos, 8);
 7818 
 7819   bind(L_less_than_8_left);
 7820   cmpl(len, 4);
 7821   jcc(Assembler::less, L_less_than_4_left);
 7822 
 7823   //load 4 Bytes
 7824   movl(rax, Address(buf, pos, Address::times_1, 0));
 7825   movl(Address(tmp1, 0 * 16), rax);
 7826   addptr(tmp1, 4);
 7827   subl(len, 4);
 7828   addl(pos, 4);
 7829 
 7830   bind(L_less_than_4_left);
 7831   cmpl(len, 2);
 7832   jcc(Assembler::less, L_less_than_2_left);
 7833 
 7834   // load 2 Bytes
 7835   movw(rax, Address(buf, pos, Address::times_1, 0));
 7836   movl(Address(tmp1, 0 * 16), rax);
 7837   addptr(tmp1, 2);
 7838   subl(len, 2);
 7839   addl(pos, 2);
 7840 
 7841   bind(L_less_than_2_left);
 7842   cmpl(len, 1);
 7843   jcc(Assembler::less, L_zero_left);
 7844 
 7845   // load 1 Byte
 7846   movb(rax, Address(buf, pos, Address::times_1, 0));
 7847   movb(Address(tmp1, 0 * 16), rax);
 7848 
 7849   bind(L_zero_left);
 7850   movdqu(xmm7, Address(rsp, 0));
 7851   pxor(xmm7, xmm0);                       //xor the initial crc value
 7852 
 7853   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 7854   movdqu(xmm0, Address(rax, tmp2));
 7855   pshufb(xmm7, xmm0);
 7856   jmp(L_128_done);
 7857 
 7858   bind(L_exact_16_left);
 7859   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 7860   pxor(xmm7, xmm0);                       //xor the initial crc value
 7861   jmp(L_128_done);
 7862 
 7863   bind(L_only_less_than_4);
 7864   cmpl(len, 3);
 7865   jcc(Assembler::less, L_only_less_than_3);
 7866 
 7867   // load 3 Bytes
 7868   movb(rax, Address(buf, pos, Address::times_1, 0));
 7869   movb(Address(tmp1, 0), rax);
 7870 
 7871   movb(rax, Address(buf, pos, Address::times_1, 1));
 7872   movb(Address(tmp1, 1), rax);
 7873 
 7874   movb(rax, Address(buf, pos, Address::times_1, 2));
 7875   movb(Address(tmp1, 2), rax);
 7876 
 7877   movdqu(xmm7, Address(rsp, 0));
 7878   pxor(xmm7, xmm0);                     //xor the initial crc value
 7879 
 7880   pslldq(xmm7, 0x5);
 7881   jmp(L_barrett);
 7882   bind(L_only_less_than_3);
 7883   cmpl(len, 2);
 7884   jcc(Assembler::less, L_only_less_than_2);
 7885 
 7886   // load 2 Bytes
 7887   movb(rax, Address(buf, pos, Address::times_1, 0));
 7888   movb(Address(tmp1, 0), rax);
 7889 
 7890   movb(rax, Address(buf, pos, Address::times_1, 1));
 7891   movb(Address(tmp1, 1), rax);
 7892 
 7893   movdqu(xmm7, Address(rsp, 0));
 7894   pxor(xmm7, xmm0);                     //xor the initial crc value
 7895 
 7896   pslldq(xmm7, 0x6);
 7897   jmp(L_barrett);
 7898 
 7899   bind(L_only_less_than_2);
 7900   //load 1 Byte
 7901   movb(rax, Address(buf, pos, Address::times_1, 0));
 7902   movb(Address(tmp1, 0), rax);
 7903 
 7904   movdqu(xmm7, Address(rsp, 0));
 7905   pxor(xmm7, xmm0);                     //xor the initial crc value
 7906 
 7907   pslldq(xmm7, 0x7);
 7908 }
 7909 
 7910 /**
 7911 * Compute CRC32 using AVX512 instructions
 7912 * param crc   register containing existing CRC (32-bit)
 7913 * param buf   register pointing to input byte buffer (byte*)
 7914 * param len   register containing number of bytes
 7915 * param table address of crc or crc32c table
 7916 * param tmp1  scratch register
 7917 * param tmp2  scratch register
 7918 * return rax  result register
 7919 *
 7920 * This routine is identical for crc32c with the exception of the precomputed constant
 7921 * table which will be passed as the table argument.  The calculation steps are
 7922 * the same for both variants.
 7923 */
 7924 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 7925   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 7926 
 7927   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7928   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7929   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 7930   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 7931   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 7932 
 7933   const Register pos = r12;
 7934   push(r12);
 7935   subptr(rsp, 16 * 2 + 8);
 7936 
 7937   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7938   // context for the registers used, where all instructions below are using 128-bit mode
 7939   // On EVEX without VL and BW, these instructions will all be AVX.
 7940   movl(pos, 0);
 7941 
 7942   // check if smaller than 256B
 7943   cmpl(len, 256);
 7944   jcc(Assembler::less, L_less_than_256);
 7945 
 7946   // load the initial crc value
 7947   movdl(xmm10, crc);
 7948 
 7949   // receive the initial 64B data, xor the initial crc value
 7950   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 7951   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 7952   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 7953   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 7954 
 7955   subl(len, 256);
 7956   cmpl(len, 256);
 7957   jcc(Assembler::less, L_fold_128_B_loop);
 7958 
 7959   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 7960   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 7961   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 7962   subl(len, 256);
 7963 
 7964   bind(L_fold_256_B_loop);
 7965   addl(pos, 256);
 7966   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 7967   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 7968   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 7969   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 7970 
 7971   subl(len, 256);
 7972   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 7973 
 7974   // Fold 256 into 128
 7975   addl(pos, 256);
 7976   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 7977   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 7978   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 7979 
 7980   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 7981   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 7982   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 7983 
 7984   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 7985   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 7986 
 7987   addl(len, 128);
 7988   jmp(L_fold_128_B_register);
 7989 
 7990   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 7991   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 7992 
 7993   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 7994   bind(L_fold_128_B_loop);
 7995   addl(pos, 128);
 7996   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 7997   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 7998 
 7999   subl(len, 128);
 8000   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8001 
 8002   addl(pos, 128);
 8003 
 8004   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8005   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8006   bind(L_fold_128_B_register);
 8007   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8008   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8009   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8010   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8011   // save last that has no multiplicand
 8012   vextracti64x2(xmm7, xmm4, 3);
 8013 
 8014   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8015   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8016   // Needed later in reduction loop
 8017   movdqu(xmm10, Address(table, 1 * 16));
 8018   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8019   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8020 
 8021   // Swap 1,0,3,2 - 01 00 11 10
 8022   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8023   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8024   vextracti128(xmm5, xmm8, 1);
 8025   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8026 
 8027   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8028   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8029   addl(len, 128 - 16);
 8030   jcc(Assembler::less, L_final_reduction_for_128);
 8031 
 8032   bind(L_16B_reduction_loop);
 8033   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8034   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8035   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8036   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8037   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8038   addl(pos, 16);
 8039   subl(len, 16);
 8040   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8041 
 8042   bind(L_final_reduction_for_128);
 8043   addl(len, 16);
 8044   jcc(Assembler::equal, L_128_done);
 8045 
 8046   bind(L_get_last_two_xmms);
 8047   movdqu(xmm2, xmm7);
 8048   addl(pos, len);
 8049   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8050   subl(pos, len);
 8051 
 8052   // get rid of the extra data that was loaded before
 8053   // load the shift constant
 8054   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8055   movdqu(xmm0, Address(rax, len));
 8056   addl(rax, len);
 8057 
 8058   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8059   //Change mask to 512
 8060   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8061   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8062 
 8063   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8064   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8065   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8066   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8067   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8068 
 8069   bind(L_128_done);
 8070   // compute crc of a 128-bit value
 8071   movdqu(xmm10, Address(table, 3 * 16));
 8072   movdqu(xmm0, xmm7);
 8073 
 8074   // 64b fold
 8075   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8076   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8077   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8078 
 8079   // 32b fold
 8080   movdqu(xmm0, xmm7);
 8081   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8082   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8083   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8084   jmp(L_barrett);
 8085 
 8086   bind(L_less_than_256);
 8087   kernel_crc32_avx512_256B(crc, buf, len, table, pos, tmp1, tmp2, L_barrett, L_16B_reduction_loop, L_get_last_two_xmms, L_128_done, L_cleanup);
 8088 
 8089   //barrett reduction
 8090   bind(L_barrett);
 8091   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8092   movdqu(xmm1, xmm7);
 8093   movdqu(xmm2, xmm7);
 8094   movdqu(xmm10, Address(table, 4 * 16));
 8095 
 8096   pclmulqdq(xmm7, xmm10, 0x0);
 8097   pxor(xmm7, xmm2);
 8098   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8099   movdqu(xmm2, xmm7);
 8100   pclmulqdq(xmm7, xmm10, 0x10);
 8101   pxor(xmm7, xmm2);
 8102   pxor(xmm7, xmm1);
 8103   pextrd(crc, xmm7, 2);
 8104 
 8105   bind(L_cleanup);
 8106   addptr(rsp, 16 * 2 + 8);
 8107   pop(r12);
 8108 }
 8109 
 8110 // S. Gueron / Information Processing Letters 112 (2012) 184
 8111 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8112 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8113 // Output: the 64-bit carry-less product of B * CONST
 8114 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8115                                      Register tmp1, Register tmp2, Register tmp3) {
 8116   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8117   if (n > 0) {
 8118     addq(tmp3, n * 256 * 8);
 8119   }
 8120   //    Q1 = TABLEExt[n][B & 0xFF];
 8121   movl(tmp1, in);
 8122   andl(tmp1, 0x000000FF);
 8123   shll(tmp1, 3);
 8124   addq(tmp1, tmp3);
 8125   movq(tmp1, Address(tmp1, 0));
 8126 
 8127   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8128   movl(tmp2, in);
 8129   shrl(tmp2, 8);
 8130   andl(tmp2, 0x000000FF);
 8131   shll(tmp2, 3);
 8132   addq(tmp2, tmp3);
 8133   movq(tmp2, Address(tmp2, 0));
 8134 
 8135   shlq(tmp2, 8);
 8136   xorq(tmp1, tmp2);
 8137 
 8138   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8139   movl(tmp2, in);
 8140   shrl(tmp2, 16);
 8141   andl(tmp2, 0x000000FF);
 8142   shll(tmp2, 3);
 8143   addq(tmp2, tmp3);
 8144   movq(tmp2, Address(tmp2, 0));
 8145 
 8146   shlq(tmp2, 16);
 8147   xorq(tmp1, tmp2);
 8148 
 8149   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8150   shrl(in, 24);
 8151   andl(in, 0x000000FF);
 8152   shll(in, 3);
 8153   addq(in, tmp3);
 8154   movq(in, Address(in, 0));
 8155 
 8156   shlq(in, 24);
 8157   xorq(in, tmp1);
 8158   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8159 }
 8160 
 8161 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8162                                       Register in_out,
 8163                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8164                                       XMMRegister w_xtmp2,
 8165                                       Register tmp1,
 8166                                       Register n_tmp2, Register n_tmp3) {
 8167   if (is_pclmulqdq_supported) {
 8168     movdl(w_xtmp1, in_out); // modified blindly
 8169 
 8170     movl(tmp1, const_or_pre_comp_const_index);
 8171     movdl(w_xtmp2, tmp1);
 8172     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8173 
 8174     movdq(in_out, w_xtmp1);
 8175   } else {
 8176     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8177   }
 8178 }
 8179 
 8180 // Recombination Alternative 2: No bit-reflections
 8181 // T1 = (CRC_A * U1) << 1
 8182 // T2 = (CRC_B * U2) << 1
 8183 // C1 = T1 >> 32
 8184 // C2 = T2 >> 32
 8185 // T1 = T1 & 0xFFFFFFFF
 8186 // T2 = T2 & 0xFFFFFFFF
 8187 // T1 = CRC32(0, T1)
 8188 // T2 = CRC32(0, T2)
 8189 // C1 = C1 ^ T1
 8190 // C2 = C2 ^ T2
 8191 // CRC = C1 ^ C2 ^ CRC_C
 8192 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,
 8193                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8194                                      Register tmp1, Register tmp2,
 8195                                      Register n_tmp3) {
 8196   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8197   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8198   shlq(in_out, 1);
 8199   movl(tmp1, in_out);
 8200   shrq(in_out, 32);
 8201   xorl(tmp2, tmp2);
 8202   crc32(tmp2, tmp1, 4);
 8203   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8204   shlq(in1, 1);
 8205   movl(tmp1, in1);
 8206   shrq(in1, 32);
 8207   xorl(tmp2, tmp2);
 8208   crc32(tmp2, tmp1, 4);
 8209   xorl(in1, tmp2);
 8210   xorl(in_out, in1);
 8211   xorl(in_out, in2);
 8212 }
 8213 
 8214 // Set N to predefined value
 8215 // Subtract from a length of a buffer
 8216 // execute in a loop:
 8217 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8218 // for i = 1 to N do
 8219 //  CRC_A = CRC32(CRC_A, A[i])
 8220 //  CRC_B = CRC32(CRC_B, B[i])
 8221 //  CRC_C = CRC32(CRC_C, C[i])
 8222 // end for
 8223 // Recombine
 8224 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,
 8225                                        Register in_out1, Register in_out2, Register in_out3,
 8226                                        Register tmp1, Register tmp2, Register tmp3,
 8227                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8228                                        Register tmp4, Register tmp5,
 8229                                        Register n_tmp6) {
 8230   Label L_processPartitions;
 8231   Label L_processPartition;
 8232   Label L_exit;
 8233 
 8234   bind(L_processPartitions);
 8235   cmpl(in_out1, 3 * size);
 8236   jcc(Assembler::less, L_exit);
 8237     xorl(tmp1, tmp1);
 8238     xorl(tmp2, tmp2);
 8239     movq(tmp3, in_out2);
 8240     addq(tmp3, size);
 8241 
 8242     bind(L_processPartition);
 8243       crc32(in_out3, Address(in_out2, 0), 8);
 8244       crc32(tmp1, Address(in_out2, size), 8);
 8245       crc32(tmp2, Address(in_out2, size * 2), 8);
 8246       addq(in_out2, 8);
 8247       cmpq(in_out2, tmp3);
 8248       jcc(Assembler::less, L_processPartition);
 8249     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8250             w_xtmp1, w_xtmp2, w_xtmp3,
 8251             tmp4, tmp5,
 8252             n_tmp6);
 8253     addq(in_out2, 2 * size);
 8254     subl(in_out1, 3 * size);
 8255     jmp(L_processPartitions);
 8256 
 8257   bind(L_exit);
 8258 }
 8259 #else
 8260 void MacroAssembler::crc32c_ipl_alg4(Register in_out, uint32_t n,
 8261                                      Register tmp1, Register tmp2, Register tmp3,
 8262                                      XMMRegister xtmp1, XMMRegister xtmp2) {
 8263   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8264   if (n > 0) {
 8265     addl(tmp3, n * 256 * 8);
 8266   }
 8267   //    Q1 = TABLEExt[n][B & 0xFF];
 8268   movl(tmp1, in_out);
 8269   andl(tmp1, 0x000000FF);
 8270   shll(tmp1, 3);
 8271   addl(tmp1, tmp3);
 8272   movq(xtmp1, Address(tmp1, 0));
 8273 
 8274   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8275   movl(tmp2, in_out);
 8276   shrl(tmp2, 8);
 8277   andl(tmp2, 0x000000FF);
 8278   shll(tmp2, 3);
 8279   addl(tmp2, tmp3);
 8280   movq(xtmp2, Address(tmp2, 0));
 8281 
 8282   psllq(xtmp2, 8);
 8283   pxor(xtmp1, xtmp2);
 8284 
 8285   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8286   movl(tmp2, in_out);
 8287   shrl(tmp2, 16);
 8288   andl(tmp2, 0x000000FF);
 8289   shll(tmp2, 3);
 8290   addl(tmp2, tmp3);
 8291   movq(xtmp2, Address(tmp2, 0));
 8292 
 8293   psllq(xtmp2, 16);
 8294   pxor(xtmp1, xtmp2);
 8295 
 8296   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8297   shrl(in_out, 24);
 8298   andl(in_out, 0x000000FF);
 8299   shll(in_out, 3);
 8300   addl(in_out, tmp3);
 8301   movq(xtmp2, Address(in_out, 0));
 8302 
 8303   psllq(xtmp2, 24);
 8304   pxor(xtmp1, xtmp2); // Result in CXMM
 8305   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8306 }
 8307 
 8308 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8309                                       Register in_out,
 8310                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8311                                       XMMRegister w_xtmp2,
 8312                                       Register tmp1,
 8313                                       Register n_tmp2, Register n_tmp3) {
 8314   if (is_pclmulqdq_supported) {
 8315     movdl(w_xtmp1, in_out);
 8316 
 8317     movl(tmp1, const_or_pre_comp_const_index);
 8318     movdl(w_xtmp2, tmp1);
 8319     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8320     // Keep result in XMM since GPR is 32 bit in length
 8321   } else {
 8322     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3, w_xtmp1, w_xtmp2);
 8323   }
 8324 }
 8325 
 8326 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,
 8327                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8328                                      Register tmp1, Register tmp2,
 8329                                      Register n_tmp3) {
 8330   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8331   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8332 
 8333   psllq(w_xtmp1, 1);
 8334   movdl(tmp1, w_xtmp1);
 8335   psrlq(w_xtmp1, 32);
 8336   movdl(in_out, w_xtmp1);
 8337 
 8338   xorl(tmp2, tmp2);
 8339   crc32(tmp2, tmp1, 4);
 8340   xorl(in_out, tmp2);
 8341 
 8342   psllq(w_xtmp2, 1);
 8343   movdl(tmp1, w_xtmp2);
 8344   psrlq(w_xtmp2, 32);
 8345   movdl(in1, w_xtmp2);
 8346 
 8347   xorl(tmp2, tmp2);
 8348   crc32(tmp2, tmp1, 4);
 8349   xorl(in1, tmp2);
 8350   xorl(in_out, in1);
 8351   xorl(in_out, in2);
 8352 }
 8353 
 8354 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,
 8355                                        Register in_out1, Register in_out2, Register in_out3,
 8356                                        Register tmp1, Register tmp2, Register tmp3,
 8357                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8358                                        Register tmp4, Register tmp5,
 8359                                        Register n_tmp6) {
 8360   Label L_processPartitions;
 8361   Label L_processPartition;
 8362   Label L_exit;
 8363 
 8364   bind(L_processPartitions);
 8365   cmpl(in_out1, 3 * size);
 8366   jcc(Assembler::less, L_exit);
 8367     xorl(tmp1, tmp1);
 8368     xorl(tmp2, tmp2);
 8369     movl(tmp3, in_out2);
 8370     addl(tmp3, size);
 8371 
 8372     bind(L_processPartition);
 8373       crc32(in_out3, Address(in_out2, 0), 4);
 8374       crc32(tmp1, Address(in_out2, size), 4);
 8375       crc32(tmp2, Address(in_out2, size*2), 4);
 8376       crc32(in_out3, Address(in_out2, 0+4), 4);
 8377       crc32(tmp1, Address(in_out2, size+4), 4);
 8378       crc32(tmp2, Address(in_out2, size*2+4), 4);
 8379       addl(in_out2, 8);
 8380       cmpl(in_out2, tmp3);
 8381       jcc(Assembler::less, L_processPartition);
 8382 
 8383         push(tmp3);
 8384         push(in_out1);
 8385         push(in_out2);
 8386         tmp4 = tmp3;
 8387         tmp5 = in_out1;
 8388         n_tmp6 = in_out2;
 8389 
 8390       crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8391             w_xtmp1, w_xtmp2, w_xtmp3,
 8392             tmp4, tmp5,
 8393             n_tmp6);
 8394 
 8395         pop(in_out2);
 8396         pop(in_out1);
 8397         pop(tmp3);
 8398 
 8399     addl(in_out2, 2 * size);
 8400     subl(in_out1, 3 * size);
 8401     jmp(L_processPartitions);
 8402 
 8403   bind(L_exit);
 8404 }
 8405 #endif //LP64
 8406 
 8407 #ifdef _LP64
 8408 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8409 // Input: A buffer I of L bytes.
 8410 // Output: the CRC32C value of the buffer.
 8411 // Notations:
 8412 // Write L = 24N + r, with N = floor (L/24).
 8413 // r = L mod 24 (0 <= r < 24).
 8414 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8415 // N quadwords, and R consists of r bytes.
 8416 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8417 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8418 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8419 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8420 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8421                                           Register tmp1, Register tmp2, Register tmp3,
 8422                                           Register tmp4, Register tmp5, Register tmp6,
 8423                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8424                                           bool is_pclmulqdq_supported) {
 8425   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8426   Label L_wordByWord;
 8427   Label L_byteByByteProlog;
 8428   Label L_byteByByte;
 8429   Label L_exit;
 8430 
 8431   if (is_pclmulqdq_supported ) {
 8432     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
 8433     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr+1);
 8434 
 8435     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
 8436     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
 8437 
 8438     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
 8439     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
 8440     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8441   } else {
 8442     const_or_pre_comp_const_index[0] = 1;
 8443     const_or_pre_comp_const_index[1] = 0;
 8444 
 8445     const_or_pre_comp_const_index[2] = 3;
 8446     const_or_pre_comp_const_index[3] = 2;
 8447 
 8448     const_or_pre_comp_const_index[4] = 5;
 8449     const_or_pre_comp_const_index[5] = 4;
 8450    }
 8451   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8452                     in2, in1, in_out,
 8453                     tmp1, tmp2, tmp3,
 8454                     w_xtmp1, w_xtmp2, w_xtmp3,
 8455                     tmp4, tmp5,
 8456                     tmp6);
 8457   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8458                     in2, in1, in_out,
 8459                     tmp1, tmp2, tmp3,
 8460                     w_xtmp1, w_xtmp2, w_xtmp3,
 8461                     tmp4, tmp5,
 8462                     tmp6);
 8463   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8464                     in2, in1, in_out,
 8465                     tmp1, tmp2, tmp3,
 8466                     w_xtmp1, w_xtmp2, w_xtmp3,
 8467                     tmp4, tmp5,
 8468                     tmp6);
 8469   movl(tmp1, in2);
 8470   andl(tmp1, 0x00000007);
 8471   negl(tmp1);
 8472   addl(tmp1, in2);
 8473   addq(tmp1, in1);
 8474 
 8475   cmpq(in1, tmp1);
 8476   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8477   align(16);
 8478   BIND(L_wordByWord);
 8479     crc32(in_out, Address(in1, 0), 8);
 8480     addq(in1, 8);
 8481     cmpq(in1, tmp1);
 8482     jcc(Assembler::less, L_wordByWord);
 8483 
 8484   BIND(L_byteByByteProlog);
 8485   andl(in2, 0x00000007);
 8486   movl(tmp2, 1);
 8487 
 8488   cmpl(tmp2, in2);
 8489   jccb(Assembler::greater, L_exit);
 8490   BIND(L_byteByByte);
 8491     crc32(in_out, Address(in1, 0), 1);
 8492     incq(in1);
 8493     incl(tmp2);
 8494     cmpl(tmp2, in2);
 8495     jcc(Assembler::lessEqual, L_byteByByte);
 8496 
 8497   BIND(L_exit);
 8498 }
 8499 #else
 8500 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8501                                           Register tmp1, Register  tmp2, Register tmp3,
 8502                                           Register tmp4, Register  tmp5, Register tmp6,
 8503                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8504                                           bool is_pclmulqdq_supported) {
 8505   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8506   Label L_wordByWord;
 8507   Label L_byteByByteProlog;
 8508   Label L_byteByByte;
 8509   Label L_exit;
 8510 
 8511   if (is_pclmulqdq_supported) {
 8512     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
 8513     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 1);
 8514 
 8515     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
 8516     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
 8517 
 8518     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
 8519     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
 8520   } else {
 8521     const_or_pre_comp_const_index[0] = 1;
 8522     const_or_pre_comp_const_index[1] = 0;
 8523 
 8524     const_or_pre_comp_const_index[2] = 3;
 8525     const_or_pre_comp_const_index[3] = 2;
 8526 
 8527     const_or_pre_comp_const_index[4] = 5;
 8528     const_or_pre_comp_const_index[5] = 4;
 8529   }
 8530   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8531                     in2, in1, in_out,
 8532                     tmp1, tmp2, tmp3,
 8533                     w_xtmp1, w_xtmp2, w_xtmp3,
 8534                     tmp4, tmp5,
 8535                     tmp6);
 8536   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8537                     in2, in1, in_out,
 8538                     tmp1, tmp2, tmp3,
 8539                     w_xtmp1, w_xtmp2, w_xtmp3,
 8540                     tmp4, tmp5,
 8541                     tmp6);
 8542   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8543                     in2, in1, in_out,
 8544                     tmp1, tmp2, tmp3,
 8545                     w_xtmp1, w_xtmp2, w_xtmp3,
 8546                     tmp4, tmp5,
 8547                     tmp6);
 8548   movl(tmp1, in2);
 8549   andl(tmp1, 0x00000007);
 8550   negl(tmp1);
 8551   addl(tmp1, in2);
 8552   addl(tmp1, in1);
 8553 
 8554   BIND(L_wordByWord);
 8555   cmpl(in1, tmp1);
 8556   jcc(Assembler::greaterEqual, L_byteByByteProlog);
 8557     crc32(in_out, Address(in1,0), 4);
 8558     addl(in1, 4);
 8559     jmp(L_wordByWord);
 8560 
 8561   BIND(L_byteByByteProlog);
 8562   andl(in2, 0x00000007);
 8563   movl(tmp2, 1);
 8564 
 8565   BIND(L_byteByByte);
 8566   cmpl(tmp2, in2);
 8567   jccb(Assembler::greater, L_exit);
 8568     movb(tmp1, Address(in1, 0));
 8569     crc32(in_out, tmp1, 1);
 8570     incl(in1);
 8571     incl(tmp2);
 8572     jmp(L_byteByByte);
 8573 
 8574   BIND(L_exit);
 8575 }
 8576 #endif // LP64
 8577 #undef BIND
 8578 #undef BLOCK_COMMENT
 8579 
 8580 // Compress char[] array to byte[].
 8581 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8582 // Return the array length if every element in array can be encoded,
 8583 // otherwise, the index of first non-latin1 (> 0xff) character.
 8584 //   @IntrinsicCandidate
 8585 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8586 //     for (int i = 0; i < len; i++) {
 8587 //       char c = src[srcOff];
 8588 //       if (c > 0xff) {
 8589 //           return i;  // return index of non-latin1 char
 8590 //       }
 8591 //       dst[dstOff] = (byte)c;
 8592 //       srcOff++;
 8593 //       dstOff++;
 8594 //     }
 8595 //     return len;
 8596 //   }
 8597 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8598   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8599   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8600   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8601   Label copy_chars_loop, done, reset_sp, copy_tail;
 8602 
 8603   // rsi: src
 8604   // rdi: dst
 8605   // rdx: len
 8606   // rcx: tmp5
 8607   // rax: result
 8608 
 8609   // rsi holds start addr of source char[] to be compressed
 8610   // rdi holds start addr of destination byte[]
 8611   // rdx holds length
 8612 
 8613   assert(len != result, "");
 8614 
 8615   // save length for return
 8616   movl(result, len);
 8617 
 8618   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8619     VM_Version::supports_avx512vlbw() &&
 8620     VM_Version::supports_bmi2()) {
 8621 
 8622     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8623 
 8624     // alignment
 8625     Label post_alignment;
 8626 
 8627     // if length of the string is less than 32, handle it the old fashioned way
 8628     testl(len, -32);
 8629     jcc(Assembler::zero, below_threshold);
 8630 
 8631     // First check whether a character is compressible ( <= 0xFF).
 8632     // Create mask to test for Unicode chars inside zmm vector
 8633     movl(tmp5, 0x00FF);
 8634     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8635 
 8636     testl(len, -64);
 8637     jccb(Assembler::zero, post_alignment);
 8638 
 8639     movl(tmp5, dst);
 8640     andl(tmp5, (32 - 1));
 8641     negl(tmp5);
 8642     andl(tmp5, (32 - 1));
 8643 
 8644     // bail out when there is nothing to be done
 8645     testl(tmp5, 0xFFFFFFFF);
 8646     jccb(Assembler::zero, post_alignment);
 8647 
 8648     // ~(~0 << len), where len is the # of remaining elements to process
 8649     movl(len, 0xFFFFFFFF);
 8650     shlxl(len, len, tmp5);
 8651     notl(len);
 8652     kmovdl(mask2, len);
 8653     movl(len, result);
 8654 
 8655     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8656     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8657     ktestd(mask1, mask2);
 8658     jcc(Assembler::carryClear, copy_tail);
 8659 
 8660     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8661 
 8662     addptr(src, tmp5);
 8663     addptr(src, tmp5);
 8664     addptr(dst, tmp5);
 8665     subl(len, tmp5);
 8666 
 8667     bind(post_alignment);
 8668     // end of alignment
 8669 
 8670     movl(tmp5, len);
 8671     andl(tmp5, (32 - 1));    // tail count (in chars)
 8672     andl(len, ~(32 - 1));    // vector count (in chars)
 8673     jccb(Assembler::zero, copy_loop_tail);
 8674 
 8675     lea(src, Address(src, len, Address::times_2));
 8676     lea(dst, Address(dst, len, Address::times_1));
 8677     negptr(len);
 8678 
 8679     bind(copy_32_loop);
 8680     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8681     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8682     kortestdl(mask1, mask1);
 8683     jccb(Assembler::carryClear, reset_for_copy_tail);
 8684 
 8685     // All elements in current processed chunk are valid candidates for
 8686     // compression. Write a truncated byte elements to the memory.
 8687     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8688     addptr(len, 32);
 8689     jccb(Assembler::notZero, copy_32_loop);
 8690 
 8691     bind(copy_loop_tail);
 8692     // bail out when there is nothing to be done
 8693     testl(tmp5, 0xFFFFFFFF);
 8694     jcc(Assembler::zero, done);
 8695 
 8696     movl(len, tmp5);
 8697 
 8698     // ~(~0 << len), where len is the # of remaining elements to process
 8699     movl(tmp5, 0xFFFFFFFF);
 8700     shlxl(tmp5, tmp5, len);
 8701     notl(tmp5);
 8702 
 8703     kmovdl(mask2, tmp5);
 8704 
 8705     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8706     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8707     ktestd(mask1, mask2);
 8708     jcc(Assembler::carryClear, copy_tail);
 8709 
 8710     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8711     jmp(done);
 8712 
 8713     bind(reset_for_copy_tail);
 8714     lea(src, Address(src, tmp5, Address::times_2));
 8715     lea(dst, Address(dst, tmp5, Address::times_1));
 8716     subptr(len, tmp5);
 8717     jmp(copy_chars_loop);
 8718 
 8719     bind(below_threshold);
 8720   }
 8721 
 8722   if (UseSSE42Intrinsics) {
 8723     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8724 
 8725     // vectored compression
 8726     testl(len, 0xfffffff8);
 8727     jcc(Assembler::zero, copy_tail);
 8728 
 8729     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8730     movdl(tmp1Reg, tmp5);
 8731     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8732 
 8733     andl(len, 0xfffffff0);
 8734     jccb(Assembler::zero, copy_16);
 8735 
 8736     // compress 16 chars per iter
 8737     pxor(tmp4Reg, tmp4Reg);
 8738 
 8739     lea(src, Address(src, len, Address::times_2));
 8740     lea(dst, Address(dst, len, Address::times_1));
 8741     negptr(len);
 8742 
 8743     bind(copy_32_loop);
 8744     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8745     por(tmp4Reg, tmp2Reg);
 8746     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8747     por(tmp4Reg, tmp3Reg);
 8748     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8749     jccb(Assembler::notZero, reset_for_copy_tail);
 8750     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8751     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8752     addptr(len, 16);
 8753     jccb(Assembler::notZero, copy_32_loop);
 8754 
 8755     // compress next vector of 8 chars (if any)
 8756     bind(copy_16);
 8757     // len = 0
 8758     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8759     jccb(Assembler::zero, copy_tail_sse);
 8760 
 8761     pxor(tmp3Reg, tmp3Reg);
 8762 
 8763     movdqu(tmp2Reg, Address(src, 0));
 8764     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8765     jccb(Assembler::notZero, reset_for_copy_tail);
 8766     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8767     movq(Address(dst, 0), tmp2Reg);
 8768     addptr(src, 16);
 8769     addptr(dst, 8);
 8770     jmpb(copy_tail_sse);
 8771 
 8772     bind(reset_for_copy_tail);
 8773     movl(tmp5, result);
 8774     andl(tmp5, 0x0000000f);
 8775     lea(src, Address(src, tmp5, Address::times_2));
 8776     lea(dst, Address(dst, tmp5, Address::times_1));
 8777     subptr(len, tmp5);
 8778     jmpb(copy_chars_loop);
 8779 
 8780     bind(copy_tail_sse);
 8781     movl(len, result);
 8782     andl(len, 0x00000007);    // tail count (in chars)
 8783   }
 8784   // compress 1 char per iter
 8785   bind(copy_tail);
 8786   testl(len, len);
 8787   jccb(Assembler::zero, done);
 8788   lea(src, Address(src, len, Address::times_2));
 8789   lea(dst, Address(dst, len, Address::times_1));
 8790   negptr(len);
 8791 
 8792   bind(copy_chars_loop);
 8793   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8794   testl(tmp5, 0xff00);      // check if Unicode char
 8795   jccb(Assembler::notZero, reset_sp);
 8796   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8797   increment(len);
 8798   jccb(Assembler::notZero, copy_chars_loop);
 8799 
 8800   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8801   bind(reset_sp);
 8802   addl(result, len);
 8803 
 8804   bind(done);
 8805 }
 8806 
 8807 // Inflate byte[] array to char[].
 8808 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8809 //   @IntrinsicCandidate
 8810 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8811 //     for (int i = 0; i < len; i++) {
 8812 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8813 //     }
 8814 //   }
 8815 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8816   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8817   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8818   // rsi: src
 8819   // rdi: dst
 8820   // rdx: len
 8821   // rcx: tmp2
 8822 
 8823   // rsi holds start addr of source byte[] to be inflated
 8824   // rdi holds start addr of destination char[]
 8825   // rdx holds length
 8826   assert_different_registers(src, dst, len, tmp2);
 8827   movl(tmp2, len);
 8828   if ((UseAVX > 2) && // AVX512
 8829     VM_Version::supports_avx512vlbw() &&
 8830     VM_Version::supports_bmi2()) {
 8831 
 8832     Label copy_32_loop, copy_tail;
 8833     Register tmp3_aliased = len;
 8834 
 8835     // if length of the string is less than 16, handle it in an old fashioned way
 8836     testl(len, -16);
 8837     jcc(Assembler::zero, below_threshold);
 8838 
 8839     testl(len, -1 * AVX3Threshold);
 8840     jcc(Assembler::zero, avx3_threshold);
 8841 
 8842     // In order to use only one arithmetic operation for the main loop we use
 8843     // this pre-calculation
 8844     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8845     andl(len, -32);     // vector count
 8846     jccb(Assembler::zero, copy_tail);
 8847 
 8848     lea(src, Address(src, len, Address::times_1));
 8849     lea(dst, Address(dst, len, Address::times_2));
 8850     negptr(len);
 8851 
 8852 
 8853     // inflate 32 chars per iter
 8854     bind(copy_32_loop);
 8855     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8856     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8857     addptr(len, 32);
 8858     jcc(Assembler::notZero, copy_32_loop);
 8859 
 8860     bind(copy_tail);
 8861     // bail out when there is nothing to be done
 8862     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8863     jcc(Assembler::zero, done);
 8864 
 8865     // ~(~0 << length), where length is the # of remaining elements to process
 8866     movl(tmp3_aliased, -1);
 8867     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8868     notl(tmp3_aliased);
 8869     kmovdl(mask, tmp3_aliased);
 8870     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8871     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8872 
 8873     jmp(done);
 8874     bind(avx3_threshold);
 8875   }
 8876   if (UseSSE42Intrinsics) {
 8877     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8878 
 8879     if (UseAVX > 1) {
 8880       andl(tmp2, (16 - 1));
 8881       andl(len, -16);
 8882       jccb(Assembler::zero, copy_new_tail);
 8883     } else {
 8884       andl(tmp2, 0x00000007);   // tail count (in chars)
 8885       andl(len, 0xfffffff8);    // vector count (in chars)
 8886       jccb(Assembler::zero, copy_tail);
 8887     }
 8888 
 8889     // vectored inflation
 8890     lea(src, Address(src, len, Address::times_1));
 8891     lea(dst, Address(dst, len, Address::times_2));
 8892     negptr(len);
 8893 
 8894     if (UseAVX > 1) {
 8895       bind(copy_16_loop);
 8896       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8897       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8898       addptr(len, 16);
 8899       jcc(Assembler::notZero, copy_16_loop);
 8900 
 8901       bind(below_threshold);
 8902       bind(copy_new_tail);
 8903       movl(len, tmp2);
 8904       andl(tmp2, 0x00000007);
 8905       andl(len, 0xFFFFFFF8);
 8906       jccb(Assembler::zero, copy_tail);
 8907 
 8908       pmovzxbw(tmp1, Address(src, 0));
 8909       movdqu(Address(dst, 0), tmp1);
 8910       addptr(src, 8);
 8911       addptr(dst, 2 * 8);
 8912 
 8913       jmp(copy_tail, true);
 8914     }
 8915 
 8916     // inflate 8 chars per iter
 8917     bind(copy_8_loop);
 8918     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8919     movdqu(Address(dst, len, Address::times_2), tmp1);
 8920     addptr(len, 8);
 8921     jcc(Assembler::notZero, copy_8_loop);
 8922 
 8923     bind(copy_tail);
 8924     movl(len, tmp2);
 8925 
 8926     cmpl(len, 4);
 8927     jccb(Assembler::less, copy_bytes);
 8928 
 8929     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8930     pmovzxbw(tmp1, tmp1);
 8931     movq(Address(dst, 0), tmp1);
 8932     subptr(len, 4);
 8933     addptr(src, 4);
 8934     addptr(dst, 8);
 8935 
 8936     bind(copy_bytes);
 8937   } else {
 8938     bind(below_threshold);
 8939   }
 8940 
 8941   testl(len, len);
 8942   jccb(Assembler::zero, done);
 8943   lea(src, Address(src, len, Address::times_1));
 8944   lea(dst, Address(dst, len, Address::times_2));
 8945   negptr(len);
 8946 
 8947   // inflate 1 char per iter
 8948   bind(copy_chars_loop);
 8949   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8950   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8951   increment(len);
 8952   jcc(Assembler::notZero, copy_chars_loop);
 8953 
 8954   bind(done);
 8955 }
 8956 
 8957 
 8958 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8959   switch(type) {
 8960     case T_BYTE:
 8961     case T_BOOLEAN:
 8962       evmovdqub(dst, kmask, src, merge, vector_len);
 8963       break;
 8964     case T_CHAR:
 8965     case T_SHORT:
 8966       evmovdquw(dst, kmask, src, merge, vector_len);
 8967       break;
 8968     case T_INT:
 8969     case T_FLOAT:
 8970       evmovdqul(dst, kmask, src, merge, vector_len);
 8971       break;
 8972     case T_LONG:
 8973     case T_DOUBLE:
 8974       evmovdquq(dst, kmask, src, merge, vector_len);
 8975       break;
 8976     default:
 8977       fatal("Unexpected type argument %s", type2name(type));
 8978       break;
 8979   }
 8980 }
 8981 
 8982 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8983   switch(type) {
 8984     case T_BYTE:
 8985     case T_BOOLEAN:
 8986       evmovdqub(dst, kmask, src, merge, vector_len);
 8987       break;
 8988     case T_CHAR:
 8989     case T_SHORT:
 8990       evmovdquw(dst, kmask, src, merge, vector_len);
 8991       break;
 8992     case T_INT:
 8993     case T_FLOAT:
 8994       evmovdqul(dst, kmask, src, merge, vector_len);
 8995       break;
 8996     case T_LONG:
 8997     case T_DOUBLE:
 8998       evmovdquq(dst, kmask, src, merge, vector_len);
 8999       break;
 9000     default:
 9001       fatal("Unexpected type argument %s", type2name(type));
 9002       break;
 9003   }
 9004 }
 9005 
 9006 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9007   switch(masklen) {
 9008     case 2:
 9009        knotbl(dst, src);
 9010        movl(rtmp, 3);
 9011        kmovbl(ktmp, rtmp);
 9012        kandbl(dst, ktmp, dst);
 9013        break;
 9014     case 4:
 9015        knotbl(dst, src);
 9016        movl(rtmp, 15);
 9017        kmovbl(ktmp, rtmp);
 9018        kandbl(dst, ktmp, dst);
 9019        break;
 9020     case 8:
 9021        knotbl(dst, src);
 9022        break;
 9023     case 16:
 9024        knotwl(dst, src);
 9025        break;
 9026     case 32:
 9027        knotdl(dst, src);
 9028        break;
 9029     case 64:
 9030        knotql(dst, src);
 9031        break;
 9032     default:
 9033       fatal("Unexpected vector length %d", masklen);
 9034       break;
 9035   }
 9036 }
 9037 
 9038 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9039   switch(type) {
 9040     case T_BOOLEAN:
 9041     case T_BYTE:
 9042        kandbl(dst, src1, src2);
 9043        break;
 9044     case T_CHAR:
 9045     case T_SHORT:
 9046        kandwl(dst, src1, src2);
 9047        break;
 9048     case T_INT:
 9049     case T_FLOAT:
 9050        kanddl(dst, src1, src2);
 9051        break;
 9052     case T_LONG:
 9053     case T_DOUBLE:
 9054        kandql(dst, src1, src2);
 9055        break;
 9056     default:
 9057       fatal("Unexpected type argument %s", type2name(type));
 9058       break;
 9059   }
 9060 }
 9061 
 9062 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9063   switch(type) {
 9064     case T_BOOLEAN:
 9065     case T_BYTE:
 9066        korbl(dst, src1, src2);
 9067        break;
 9068     case T_CHAR:
 9069     case T_SHORT:
 9070        korwl(dst, src1, src2);
 9071        break;
 9072     case T_INT:
 9073     case T_FLOAT:
 9074        kordl(dst, src1, src2);
 9075        break;
 9076     case T_LONG:
 9077     case T_DOUBLE:
 9078        korql(dst, src1, src2);
 9079        break;
 9080     default:
 9081       fatal("Unexpected type argument %s", type2name(type));
 9082       break;
 9083   }
 9084 }
 9085 
 9086 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9087   switch(type) {
 9088     case T_BOOLEAN:
 9089     case T_BYTE:
 9090        kxorbl(dst, src1, src2);
 9091        break;
 9092     case T_CHAR:
 9093     case T_SHORT:
 9094        kxorwl(dst, src1, src2);
 9095        break;
 9096     case T_INT:
 9097     case T_FLOAT:
 9098        kxordl(dst, src1, src2);
 9099        break;
 9100     case T_LONG:
 9101     case T_DOUBLE:
 9102        kxorql(dst, src1, src2);
 9103        break;
 9104     default:
 9105       fatal("Unexpected type argument %s", type2name(type));
 9106       break;
 9107   }
 9108 }
 9109 
 9110 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9111   switch(type) {
 9112     case T_BOOLEAN:
 9113     case T_BYTE:
 9114       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9115     case T_CHAR:
 9116     case T_SHORT:
 9117       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9118     case T_INT:
 9119     case T_FLOAT:
 9120       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9121     case T_LONG:
 9122     case T_DOUBLE:
 9123       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9124     default:
 9125       fatal("Unexpected type argument %s", type2name(type)); break;
 9126   }
 9127 }
 9128 
 9129 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9130   switch(type) {
 9131     case T_BOOLEAN:
 9132     case T_BYTE:
 9133       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9134     case T_CHAR:
 9135     case T_SHORT:
 9136       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9137     case T_INT:
 9138     case T_FLOAT:
 9139       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9140     case T_LONG:
 9141     case T_DOUBLE:
 9142       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9143     default:
 9144       fatal("Unexpected type argument %s", type2name(type)); break;
 9145   }
 9146 }
 9147 
 9148 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9149   switch(type) {
 9150     case T_BYTE:
 9151       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9152     case T_SHORT:
 9153       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9154     case T_INT:
 9155       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9156     case T_LONG:
 9157       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9158     default:
 9159       fatal("Unexpected type argument %s", type2name(type)); break;
 9160   }
 9161 }
 9162 
 9163 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9164   switch(type) {
 9165     case T_BYTE:
 9166       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9167     case T_SHORT:
 9168       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9169     case T_INT:
 9170       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9171     case T_LONG:
 9172       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9173     default:
 9174       fatal("Unexpected type argument %s", type2name(type)); break;
 9175   }
 9176 }
 9177 
 9178 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9179   switch(type) {
 9180     case T_BYTE:
 9181       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9182     case T_SHORT:
 9183       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9184     case T_INT:
 9185       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9186     case T_LONG:
 9187       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9188     default:
 9189       fatal("Unexpected type argument %s", type2name(type)); break;
 9190   }
 9191 }
 9192 
 9193 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9194   switch(type) {
 9195     case T_BYTE:
 9196       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9197     case T_SHORT:
 9198       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9199     case T_INT:
 9200       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9201     case T_LONG:
 9202       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9203     default:
 9204       fatal("Unexpected type argument %s", type2name(type)); break;
 9205   }
 9206 }
 9207 
 9208 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9209   switch(type) {
 9210     case T_INT:
 9211       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9212     case T_LONG:
 9213       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9214     default:
 9215       fatal("Unexpected type argument %s", type2name(type)); break;
 9216   }
 9217 }
 9218 
 9219 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9220   switch(type) {
 9221     case T_INT:
 9222       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9223     case T_LONG:
 9224       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9225     default:
 9226       fatal("Unexpected type argument %s", type2name(type)); break;
 9227   }
 9228 }
 9229 
 9230 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9231   switch(type) {
 9232     case T_INT:
 9233       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9234     case T_LONG:
 9235       evporq(dst, mask, nds, src, merge, vector_len); break;
 9236     default:
 9237       fatal("Unexpected type argument %s", type2name(type)); break;
 9238   }
 9239 }
 9240 
 9241 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9242   switch(type) {
 9243     case T_INT:
 9244       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9245     case T_LONG:
 9246       evporq(dst, mask, nds, src, merge, vector_len); break;
 9247     default:
 9248       fatal("Unexpected type argument %s", type2name(type)); break;
 9249   }
 9250 }
 9251 
 9252 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9253   switch(type) {
 9254     case T_INT:
 9255       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9256     case T_LONG:
 9257       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9258     default:
 9259       fatal("Unexpected type argument %s", type2name(type)); break;
 9260   }
 9261 }
 9262 
 9263 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9264   switch(type) {
 9265     case T_INT:
 9266       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9267     case T_LONG:
 9268       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9269     default:
 9270       fatal("Unexpected type argument %s", type2name(type)); break;
 9271   }
 9272 }
 9273 
 9274 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9275   switch(masklen) {
 9276     case 8:
 9277        kortestbl(src1, src2);
 9278        break;
 9279     case 16:
 9280        kortestwl(src1, src2);
 9281        break;
 9282     case 32:
 9283        kortestdl(src1, src2);
 9284        break;
 9285     case 64:
 9286        kortestql(src1, src2);
 9287        break;
 9288     default:
 9289       fatal("Unexpected mask length %d", masklen);
 9290       break;
 9291   }
 9292 }
 9293 
 9294 
 9295 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9296   switch(masklen)  {
 9297     case 8:
 9298        ktestbl(src1, src2);
 9299        break;
 9300     case 16:
 9301        ktestwl(src1, src2);
 9302        break;
 9303     case 32:
 9304        ktestdl(src1, src2);
 9305        break;
 9306     case 64:
 9307        ktestql(src1, src2);
 9308        break;
 9309     default:
 9310       fatal("Unexpected mask length %d", masklen);
 9311       break;
 9312   }
 9313 }
 9314 
 9315 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9316   switch(type) {
 9317     case T_INT:
 9318       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9319     case T_LONG:
 9320       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9321     default:
 9322       fatal("Unexpected type argument %s", type2name(type)); break;
 9323       break;
 9324   }
 9325 }
 9326 
 9327 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9328   switch(type) {
 9329     case T_INT:
 9330       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9331     case T_LONG:
 9332       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9333     default:
 9334       fatal("Unexpected type argument %s", type2name(type)); break;
 9335   }
 9336 }
 9337 
 9338 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9339   switch(type) {
 9340     case T_INT:
 9341       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9342     case T_LONG:
 9343       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9344     default:
 9345       fatal("Unexpected type argument %s", type2name(type)); break;
 9346   }
 9347 }
 9348 
 9349 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9350   switch(type) {
 9351     case T_INT:
 9352       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9353     case T_LONG:
 9354       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9355     default:
 9356       fatal("Unexpected type argument %s", type2name(type)); break;
 9357   }
 9358 }
 9359 
 9360 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9361   assert(rscratch != noreg || always_reachable(src), "missing");
 9362 
 9363   if (reachable(src)) {
 9364     evpandq(dst, nds, as_Address(src), vector_len);
 9365   } else {
 9366     lea(rscratch, src);
 9367     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9368   }
 9369 }
 9370 
 9371 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9372   assert(rscratch != noreg || always_reachable(src), "missing");
 9373 
 9374   if (reachable(src)) {
 9375     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9376   } else {
 9377     lea(rscratch, src);
 9378     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9379   }
 9380 }
 9381 
 9382 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9383   assert(rscratch != noreg || always_reachable(src), "missing");
 9384 
 9385   if (reachable(src)) {
 9386     evporq(dst, nds, as_Address(src), vector_len);
 9387   } else {
 9388     lea(rscratch, src);
 9389     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9390   }
 9391 }
 9392 
 9393 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9394   assert(rscratch != noreg || always_reachable(src), "missing");
 9395 
 9396   if (reachable(src)) {
 9397     vpshufb(dst, nds, as_Address(src), vector_len);
 9398   } else {
 9399     lea(rscratch, src);
 9400     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9401   }
 9402 }
 9403 
 9404 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9405   assert(rscratch != noreg || always_reachable(src3), "missing");
 9406 
 9407   if (reachable(src3)) {
 9408     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9409   } else {
 9410     lea(rscratch, src3);
 9411     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9412   }
 9413 }
 9414 
 9415 #if COMPILER2_OR_JVMCI
 9416 
 9417 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9418                                  Register length, Register temp, int vec_enc) {
 9419   // Computing mask for predicated vector store.
 9420   movptr(temp, -1);
 9421   bzhiq(temp, temp, length);
 9422   kmov(mask, temp);
 9423   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9424 }
 9425 
 9426 // Set memory operation for length "less than" 64 bytes.
 9427 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9428                                        XMMRegister xmm, KRegister mask, Register length,
 9429                                        Register temp, bool use64byteVector) {
 9430   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9431   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9432   if (!use64byteVector) {
 9433     fill32(dst, disp, xmm);
 9434     subptr(length, 32 >> shift);
 9435     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9436   } else {
 9437     assert(MaxVectorSize == 64, "vector length != 64");
 9438     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9439   }
 9440 }
 9441 
 9442 
 9443 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9444                                        XMMRegister xmm, KRegister mask, Register length,
 9445                                        Register temp) {
 9446   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9447   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9448   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9449 }
 9450 
 9451 
 9452 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9453   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9454   vmovdqu(dst, xmm);
 9455 }
 9456 
 9457 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9458   fill32(Address(dst, disp), xmm);
 9459 }
 9460 
 9461 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9462   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9463   if (!use64byteVector) {
 9464     fill32(dst, xmm);
 9465     fill32(dst.plus_disp(32), xmm);
 9466   } else {
 9467     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9468   }
 9469 }
 9470 
 9471 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9472   fill64(Address(dst, disp), xmm, use64byteVector);
 9473 }
 9474 
 9475 #ifdef _LP64
 9476 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9477                                         Register count, Register rtmp, XMMRegister xtmp) {
 9478   Label L_exit;
 9479   Label L_fill_start;
 9480   Label L_fill_64_bytes;
 9481   Label L_fill_96_bytes;
 9482   Label L_fill_128_bytes;
 9483   Label L_fill_128_bytes_loop;
 9484   Label L_fill_128_loop_header;
 9485   Label L_fill_128_bytes_loop_header;
 9486   Label L_fill_128_bytes_loop_pre_header;
 9487   Label L_fill_zmm_sequence;
 9488 
 9489   int shift = -1;
 9490   int avx3threshold = VM_Version::avx3_threshold();
 9491   switch(type) {
 9492     case T_BYTE:  shift = 0;
 9493       break;
 9494     case T_SHORT: shift = 1;
 9495       break;
 9496     case T_INT:   shift = 2;
 9497       break;
 9498     /* Uncomment when LONG fill stubs are supported.
 9499     case T_LONG:  shift = 3;
 9500       break;
 9501     */
 9502     default:
 9503       fatal("Unhandled type: %s\n", type2name(type));
 9504   }
 9505 
 9506   if ((avx3threshold != 0)  || (MaxVectorSize == 32)) {
 9507 
 9508     if (MaxVectorSize == 64) {
 9509       cmpq(count, avx3threshold >> shift);
 9510       jcc(Assembler::greater, L_fill_zmm_sequence);
 9511     }
 9512 
 9513     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9514 
 9515     bind(L_fill_start);
 9516 
 9517     cmpq(count, 32 >> shift);
 9518     jccb(Assembler::greater, L_fill_64_bytes);
 9519     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9520     jmp(L_exit);
 9521 
 9522     bind(L_fill_64_bytes);
 9523     cmpq(count, 64 >> shift);
 9524     jccb(Assembler::greater, L_fill_96_bytes);
 9525     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9526     jmp(L_exit);
 9527 
 9528     bind(L_fill_96_bytes);
 9529     cmpq(count, 96 >> shift);
 9530     jccb(Assembler::greater, L_fill_128_bytes);
 9531     fill64(to, 0, xtmp);
 9532     subq(count, 64 >> shift);
 9533     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9534     jmp(L_exit);
 9535 
 9536     bind(L_fill_128_bytes);
 9537     cmpq(count, 128 >> shift);
 9538     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9539     fill64(to, 0, xtmp);
 9540     fill32(to, 64, xtmp);
 9541     subq(count, 96 >> shift);
 9542     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9543     jmp(L_exit);
 9544 
 9545     bind(L_fill_128_bytes_loop_pre_header);
 9546     {
 9547       mov(rtmp, to);
 9548       andq(rtmp, 31);
 9549       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9550       negq(rtmp);
 9551       addq(rtmp, 32);
 9552       mov64(r8, -1L);
 9553       bzhiq(r8, r8, rtmp);
 9554       kmovql(k2, r8);
 9555       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9556       addq(to, rtmp);
 9557       shrq(rtmp, shift);
 9558       subq(count, rtmp);
 9559     }
 9560 
 9561     cmpq(count, 128 >> shift);
 9562     jcc(Assembler::less, L_fill_start);
 9563 
 9564     bind(L_fill_128_bytes_loop_header);
 9565     subq(count, 128 >> shift);
 9566 
 9567     align32();
 9568     bind(L_fill_128_bytes_loop);
 9569       fill64(to, 0, xtmp);
 9570       fill64(to, 64, xtmp);
 9571       addq(to, 128);
 9572       subq(count, 128 >> shift);
 9573       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9574 
 9575     addq(count, 128 >> shift);
 9576     jcc(Assembler::zero, L_exit);
 9577     jmp(L_fill_start);
 9578   }
 9579 
 9580   if (MaxVectorSize == 64) {
 9581     // Sequence using 64 byte ZMM register.
 9582     Label L_fill_128_bytes_zmm;
 9583     Label L_fill_192_bytes_zmm;
 9584     Label L_fill_192_bytes_loop_zmm;
 9585     Label L_fill_192_bytes_loop_header_zmm;
 9586     Label L_fill_192_bytes_loop_pre_header_zmm;
 9587     Label L_fill_start_zmm_sequence;
 9588 
 9589     bind(L_fill_zmm_sequence);
 9590     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9591 
 9592     bind(L_fill_start_zmm_sequence);
 9593     cmpq(count, 64 >> shift);
 9594     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9595     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9596     jmp(L_exit);
 9597 
 9598     bind(L_fill_128_bytes_zmm);
 9599     cmpq(count, 128 >> shift);
 9600     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9601     fill64(to, 0, xtmp, true);
 9602     subq(count, 64 >> shift);
 9603     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9604     jmp(L_exit);
 9605 
 9606     bind(L_fill_192_bytes_zmm);
 9607     cmpq(count, 192 >> shift);
 9608     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9609     fill64(to, 0, xtmp, true);
 9610     fill64(to, 64, xtmp, true);
 9611     subq(count, 128 >> shift);
 9612     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9613     jmp(L_exit);
 9614 
 9615     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9616     {
 9617       movq(rtmp, to);
 9618       andq(rtmp, 63);
 9619       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9620       negq(rtmp);
 9621       addq(rtmp, 64);
 9622       mov64(r8, -1L);
 9623       bzhiq(r8, r8, rtmp);
 9624       kmovql(k2, r8);
 9625       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9626       addq(to, rtmp);
 9627       shrq(rtmp, shift);
 9628       subq(count, rtmp);
 9629     }
 9630 
 9631     cmpq(count, 192 >> shift);
 9632     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9633 
 9634     bind(L_fill_192_bytes_loop_header_zmm);
 9635     subq(count, 192 >> shift);
 9636 
 9637     align32();
 9638     bind(L_fill_192_bytes_loop_zmm);
 9639       fill64(to, 0, xtmp, true);
 9640       fill64(to, 64, xtmp, true);
 9641       fill64(to, 128, xtmp, true);
 9642       addq(to, 192);
 9643       subq(count, 192 >> shift);
 9644       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9645 
 9646     addq(count, 192 >> shift);
 9647     jcc(Assembler::zero, L_exit);
 9648     jmp(L_fill_start_zmm_sequence);
 9649   }
 9650   bind(L_exit);
 9651 }
 9652 #endif
 9653 #endif //COMPILER2_OR_JVMCI
 9654 
 9655 
 9656 #ifdef _LP64
 9657 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9658   Label done;
 9659   cvttss2sil(dst, src);
 9660   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9661   cmpl(dst, 0x80000000); // float_sign_flip
 9662   jccb(Assembler::notEqual, done);
 9663   subptr(rsp, 8);
 9664   movflt(Address(rsp, 0), src);
 9665   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9666   pop(dst);
 9667   bind(done);
 9668 }
 9669 
 9670 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9671   Label done;
 9672   cvttsd2sil(dst, src);
 9673   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9674   cmpl(dst, 0x80000000); // float_sign_flip
 9675   jccb(Assembler::notEqual, done);
 9676   subptr(rsp, 8);
 9677   movdbl(Address(rsp, 0), src);
 9678   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9679   pop(dst);
 9680   bind(done);
 9681 }
 9682 
 9683 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9684   Label done;
 9685   cvttss2siq(dst, src);
 9686   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9687   jccb(Assembler::notEqual, done);
 9688   subptr(rsp, 8);
 9689   movflt(Address(rsp, 0), src);
 9690   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9691   pop(dst);
 9692   bind(done);
 9693 }
 9694 
 9695 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9696   // Following code is line by line assembly translation rounding algorithm.
 9697   // Please refer to java.lang.Math.round(float) algorithm for details.
 9698   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9699   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9700   const int32_t FloatConsts_EXP_BIAS = 127;
 9701   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9702   const int32_t MINUS_32 = 0xFFFFFFE0;
 9703   Label L_special_case, L_block1, L_exit;
 9704   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9705   movdl(dst, src);
 9706   andl(dst, rtmp);
 9707   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9708   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9709   subl(rtmp, dst);
 9710   movl(rcx, rtmp);
 9711   movl(dst, MINUS_32);
 9712   testl(rtmp, dst);
 9713   jccb(Assembler::notEqual, L_special_case);
 9714   movdl(dst, src);
 9715   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9716   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9717   movdl(rtmp, src);
 9718   testl(rtmp, rtmp);
 9719   jccb(Assembler::greaterEqual, L_block1);
 9720   negl(dst);
 9721   bind(L_block1);
 9722   sarl(dst);
 9723   addl(dst, 0x1);
 9724   sarl(dst, 0x1);
 9725   jmp(L_exit);
 9726   bind(L_special_case);
 9727   convert_f2i(dst, src);
 9728   bind(L_exit);
 9729 }
 9730 
 9731 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9732   // Following code is line by line assembly translation rounding algorithm.
 9733   // Please refer to java.lang.Math.round(double) algorithm for details.
 9734   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9735   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9736   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9737   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9738   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9739   Label L_special_case, L_block1, L_exit;
 9740   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9741   movq(dst, src);
 9742   andq(dst, rtmp);
 9743   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9744   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9745   subq(rtmp, dst);
 9746   movq(rcx, rtmp);
 9747   mov64(dst, MINUS_64);
 9748   testq(rtmp, dst);
 9749   jccb(Assembler::notEqual, L_special_case);
 9750   movq(dst, src);
 9751   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9752   andq(dst, rtmp);
 9753   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9754   orq(dst, rtmp);
 9755   movq(rtmp, src);
 9756   testq(rtmp, rtmp);
 9757   jccb(Assembler::greaterEqual, L_block1);
 9758   negq(dst);
 9759   bind(L_block1);
 9760   sarq(dst);
 9761   addq(dst, 0x1);
 9762   sarq(dst, 0x1);
 9763   jmp(L_exit);
 9764   bind(L_special_case);
 9765   convert_d2l(dst, src);
 9766   bind(L_exit);
 9767 }
 9768 
 9769 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9770   Label done;
 9771   cvttsd2siq(dst, src);
 9772   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9773   jccb(Assembler::notEqual, done);
 9774   subptr(rsp, 8);
 9775   movdbl(Address(rsp, 0), src);
 9776   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9777   pop(dst);
 9778   bind(done);
 9779 }
 9780 
 9781 void MacroAssembler::cache_wb(Address line)
 9782 {
 9783   // 64 bit cpus always support clflush
 9784   assert(VM_Version::supports_clflush(), "clflush should be available");
 9785   bool optimized = VM_Version::supports_clflushopt();
 9786   bool no_evict = VM_Version::supports_clwb();
 9787 
 9788   // prefer clwb (writeback without evict) otherwise
 9789   // prefer clflushopt (potentially parallel writeback with evict)
 9790   // otherwise fallback on clflush (serial writeback with evict)
 9791 
 9792   if (optimized) {
 9793     if (no_evict) {
 9794       clwb(line);
 9795     } else {
 9796       clflushopt(line);
 9797     }
 9798   } else {
 9799     // no need for fence when using CLFLUSH
 9800     clflush(line);
 9801   }
 9802 }
 9803 
 9804 void MacroAssembler::cache_wbsync(bool is_pre)
 9805 {
 9806   assert(VM_Version::supports_clflush(), "clflush should be available");
 9807   bool optimized = VM_Version::supports_clflushopt();
 9808   bool no_evict = VM_Version::supports_clwb();
 9809 
 9810   // pick the correct implementation
 9811 
 9812   if (!is_pre && (optimized || no_evict)) {
 9813     // need an sfence for post flush when using clflushopt or clwb
 9814     // otherwise no no need for any synchroniaztion
 9815 
 9816     sfence();
 9817   }
 9818 }
 9819 
 9820 #endif // _LP64
 9821 
 9822 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9823   switch (cond) {
 9824     // Note some conditions are synonyms for others
 9825     case Assembler::zero:         return Assembler::notZero;
 9826     case Assembler::notZero:      return Assembler::zero;
 9827     case Assembler::less:         return Assembler::greaterEqual;
 9828     case Assembler::lessEqual:    return Assembler::greater;
 9829     case Assembler::greater:      return Assembler::lessEqual;
 9830     case Assembler::greaterEqual: return Assembler::less;
 9831     case Assembler::below:        return Assembler::aboveEqual;
 9832     case Assembler::belowEqual:   return Assembler::above;
 9833     case Assembler::above:        return Assembler::belowEqual;
 9834     case Assembler::aboveEqual:   return Assembler::below;
 9835     case Assembler::overflow:     return Assembler::noOverflow;
 9836     case Assembler::noOverflow:   return Assembler::overflow;
 9837     case Assembler::negative:     return Assembler::positive;
 9838     case Assembler::positive:     return Assembler::negative;
 9839     case Assembler::parity:       return Assembler::noParity;
 9840     case Assembler::noParity:     return Assembler::parity;
 9841   }
 9842   ShouldNotReachHere(); return Assembler::overflow;
 9843 }
 9844 
 9845 SkipIfEqual::SkipIfEqual(
 9846     MacroAssembler* masm, const bool* flag_addr, bool value, Register rscratch) {
 9847   _masm = masm;
 9848   _masm->cmp8(ExternalAddress((address)flag_addr), value, rscratch);
 9849   _masm->jcc(Assembler::equal, _label);
 9850 }
 9851 
 9852 SkipIfEqual::~SkipIfEqual() {
 9853   _masm->bind(_label);
 9854 }
 9855 
 9856 // 32-bit Windows has its own fast-path implementation
 9857 // of get_thread
 9858 #if !defined(WIN32) || defined(_LP64)
 9859 
 9860 // This is simply a call to Thread::current()
 9861 void MacroAssembler::get_thread(Register thread) {
 9862   if (thread != rax) {
 9863     push(rax);
 9864   }
 9865   LP64_ONLY(push(rdi);)
 9866   LP64_ONLY(push(rsi);)
 9867   push(rdx);
 9868   push(rcx);
 9869 #ifdef _LP64
 9870   push(r8);
 9871   push(r9);
 9872   push(r10);
 9873   push(r11);
 9874 #endif
 9875 
 9876   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9877 
 9878 #ifdef _LP64
 9879   pop(r11);
 9880   pop(r10);
 9881   pop(r9);
 9882   pop(r8);
 9883 #endif
 9884   pop(rcx);
 9885   pop(rdx);
 9886   LP64_ONLY(pop(rsi);)
 9887   LP64_ONLY(pop(rdi);)
 9888   if (thread != rax) {
 9889     mov(thread, rax);
 9890     pop(rax);
 9891   }
 9892 }
 9893 
 9894 
 9895 #endif // !WIN32 || _LP64
 9896 
 9897 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9898   Label L_stack_ok;
 9899   if (bias == 0) {
 9900     testptr(sp, 2 * wordSize - 1);
 9901   } else {
 9902     // lea(tmp, Address(rsp, bias);
 9903     mov(tmp, sp);
 9904     addptr(tmp, bias);
 9905     testptr(tmp, 2 * wordSize - 1);
 9906   }
 9907   jcc(Assembler::equal, L_stack_ok);
 9908   block_comment(msg);
 9909   stop(msg);
 9910   bind(L_stack_ok);
 9911 }
 9912 
 9913 // Implements lightweight-locking.
 9914 //
 9915 // obj: the object to be locked
 9916 // reg_rax: rax
 9917 // thread: the thread which attempts to lock obj
 9918 // tmp: a temporary register
 9919 void MacroAssembler::lightweight_lock(Register obj, Register reg_rax, Register thread, Register tmp, Label& slow) {
 9920   assert(reg_rax == rax, "");
 9921   assert_different_registers(obj, reg_rax, thread, tmp);
 9922 
 9923   Label push;
 9924   const Register top = tmp;
 9925 
 9926   // Preload the markWord. It is important that this is the first
 9927   // instruction emitted as it is part of C1's null check semantics.
 9928   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9929 
 9930   // Load top.
 9931   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9932 
 9933   // Check if the lock-stack is full.
 9934   cmpl(top, LockStack::end_offset());
 9935   jcc(Assembler::greaterEqual, slow);
 9936 
 9937   // Check for recursion.
 9938   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9939   jcc(Assembler::equal, push);
 9940 
 9941   // Check header for monitor (0b10).
 9942   testptr(reg_rax, markWord::monitor_value);
 9943   jcc(Assembler::notZero, slow);
 9944 
 9945   // Try to lock. Transition lock bits 0b01 => 0b00
 9946   movptr(tmp, reg_rax);
 9947   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9948   orptr(reg_rax, markWord::unlocked_value);
 9949   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9950   jcc(Assembler::notEqual, slow);
 9951 
 9952   // Restore top, CAS clobbers register.
 9953   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9954 
 9955   bind(push);
 9956   // After successful lock, push object on lock-stack.
 9957   movptr(Address(thread, top), obj);
 9958   incrementl(top, oopSize);
 9959   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9960 }
 9961 
 9962 // Implements lightweight-unlocking.
 9963 //
 9964 // obj: the object to be unlocked
 9965 // reg_rax: rax
 9966 // thread: the thread
 9967 // tmp: a temporary register
 9968 //
 9969 // x86_32 Note: reg_rax and thread may alias each other due to limited register
 9970 //              availiability.
 9971 void MacroAssembler::lightweight_unlock(Register obj, Register reg_rax, Register thread, Register tmp, Label& slow) {
 9972   assert(reg_rax == rax, "");
 9973   assert_different_registers(obj, reg_rax, tmp);
 9974   LP64_ONLY(assert_different_registers(obj, reg_rax, thread, tmp);)
 9975 
 9976   Label unlocked, push_and_slow;
 9977   const Register top = tmp;
 9978 
 9979   // Check if obj is top of lock-stack.
 9980   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9981   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9982   jcc(Assembler::notEqual, slow);
 9983 
 9984   // Pop lock-stack.
 9985   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9986   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9987 
 9988   // Check if recursive.
 9989   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
 9990   jcc(Assembler::equal, unlocked);
 9991 
 9992   // Not recursive. Check header for monitor (0b10).
 9993   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9994   testptr(reg_rax, markWord::monitor_value);
 9995   jcc(Assembler::notZero, push_and_slow);
 9996 
 9997 #ifdef ASSERT
 9998   // Check header not unlocked (0b01).
 9999   Label not_unlocked;
10000   testptr(reg_rax, markWord::unlocked_value);
10001   jcc(Assembler::zero, not_unlocked);
10002   stop("lightweight_unlock already unlocked");
10003   bind(not_unlocked);
10004 #endif
10005 
10006   // Try to unlock. Transition lock bits 0b00 => 0b01
10007   movptr(tmp, reg_rax);
10008   orptr(tmp, markWord::unlocked_value);
10009   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10010   jcc(Assembler::equal, unlocked);
10011 
10012   bind(push_and_slow);
10013   // Restore lock-stack and handle the unlock in runtime.
10014   if (thread == reg_rax) {
10015     // On x86_32 we may lose the thread.
10016     get_thread(thread);
10017   }
10018 #ifdef ASSERT
10019   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10020   movptr(Address(thread, top), obj);
10021 #endif
10022   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10023   jmp(slow);
10024 
10025   bind(unlocked);
10026 }